Journal of Algebra 253 (2002) 50–99

www.academicpress.com

Affine Kac–Moody algebras

graded by affine root systems
Josiane Nervi
Institut de Recherche Mathématique Avancée, Université Louis Pasteur, 7, rue René Descartes,
F-67084 Strasbourg cedex, France
Received 18 December 2000
Communicated by Georgia Benkart

0. Introduction

The aim of this paper is to give the complete classification of all affine Kac–
Moody algebras graded by affine root systems.
An affine Lie algebra ĝ is said to be graded by the affine root system Σ if ĝ
contains an affine subalgebra a whose
root system relative to a Cartan subalgebra
ha is equal to Σ and such that ĝ = λ∈Σ∪{0} Vλ where Vλ = {x ∈ ĝ: [h, x] =
λ(h)x for all h ∈ ha }.
The Lie algebra a is then called the grading subalgebra. This is slight
modification of the notion of grading by a finite root system which was introduced
by Berman and Moody [1] and developed by Benkart and Zelmanov [2],
Neher [3], Allison et al. [4]. In the preceding paper [5] we gave the complete
classification of all simple finite-dimensional Lie algebras graded by a finite root
system.
Let us give now a description of the contents of this paper.
In Section 1 we construct a large class of affine subalgebras of a given affine
algebra ĝ, the so-called admissible subalgebras. In the finite-dimensional case, the
admissible subalgebras were constructed by H. Rubenthaler [6] and they played
a crucial role in the classification of dual pairs in reductive Lie algebras [7]. These
algebras are also important for the classification of the simple finite-dimensional
Lie algebras graded by finite root systems [5].

E-mail address: [email protected].

0021-8693/02/$ – see front matter  2002 Elsevier Science (USA). All rights reserved.
PII: S 0 0 2 1 - 8 6 9 3 ( 0 2 ) 0 0 0 5 3 - 4
 J. Nervi / Journal of Algebra 253 (2002) 50–99 51

The admissible subalgebras are parameterized by some “nice” subsets Γ of

a set Π of simple roots of the root system ∆ = (ĝ, h) where h is some Cartan
subalgebra of ĝ (Definition 1.4.1.1). Such “nice” pairs (Π, Γ ) are also called
admissible.
Roughly speaking if (Π, Γ ) is admissible then it is possible to associate to Γ
a set of sl2 -triples (Xi , Hi , Yi ), i = 0, 1, . . . , k (k = |Π\Γ |). These sl2 -triples
arise classically from the regularity of some prehomogeneous vector spaces (see
Section 1.1) related to Z-gradings of certain finite-dimensional subalgebras of ĝ
(see Section 1.3). If (Π, Γ ) is admissible then the related sl2 -triples (Xi , Hi , Yi )
(i = 0, 1, . . . , k) and the space hΓ = {h ∈ h: α(h) = 0 for all α ∈ Γ } generate the
admissible subalgebra ĝΓ (Proposition 1.4.2.1), which is affine. It is worthwhile
to note that the pair (Π, ∅) is admissible, in that case the corresponding sl2 -triples
are the usual generators of ĝ, and therefore ĝ∅ = ĝ.
From the definition, the classification of the admissible subalgebras relies on
a Dynkin diagram combinatorics. This classification is given in Appendix A.
We also describe an important subfamily of the set of admissible subalgebras,
the so-called C-admissible subalgebras. The root system of a C-admissible
subalgebra ĝΓ is studied in Section 1.5. In this section we prove also that ĝ is
an integrable ĝΓ -module.
In Section 2.2 we prove that any C-admissible subalgebra ĝΓ defines a grading
of ĝ relative to its root system ∆Γ (Proposition 2.2.1). This gives a large family
of gradings of ĝ.
In Section 2.4 we prove that the grading defined by C-admissible subalgebras
can be characterized by the irreducibility of the representations (V0 , Vλ ) for every
simple root λ ∈ ∆Γ (Theorem 2.4.3).
After this, we study the converse problem and start with an arbitrary grading
of ĝ by an affine root system Σ.
In Sections 2.3 and 2.5 we associate to such a data a C-admissible pair (Π, Γ )
such that the corresponding C-admissible subalgebra ĝΓ is again graded by Σ.
(Propositions 2.3.4, 2.3.7, and Theorem 2.5.5). Moreover, the grading of ĝΓ by Σ
is “maximal” in the sense of Definition 2.5.7 and we show that the classification
of all gradings of ĝ reduces to the classification of all maximal Σ-gradings of
C-admissible subalgebras (Theorem 2.5.10).
In Section 2.6 we first prove a technical lemma (Lemma 2.6.1) which allows
us, after some computations, to achieve in Section 2.7 the classification of the
maximal gradations. The list of these maximal gradations is given in Appendix B.
We summarize the classification of gradings in Theorem 2.7.9.
If one takes the underlying finite-dimensional subalgebra å of the grading
subalgebra a then å gives rise to a finite grading (in the original sense of [1]) of
the algebra ĝ. We will study more precisely this aspect in a forthcoming paper. In
another paper we will investigate the relationship between the grading subalgebras
of ĝ (especially the C-admissible ones) and the dual pairs in ĝ (in the spirit of [7]).
52 J. Nervi / Journal of Algebra 253 (2002) 50–99

1. C-admissible subalgebras of affine Kac–Moody algebras

1.1. Prehomogeneous vector spaces of parabolic type

Let us recall here some basic facts from the theory of parabolic prehomoge-
neous vector spaces (details can be found in [8,9]). If G is an algebraic group
over C and (ρ, V ) a rational representation of G in a finite-dimensional vector
space, the triple (G, ρ, V ), or briefly (G, V ), is called a prehomogeneous vec-
tor space if G has a Zariski open orbit in V . This is, of course, an infinitesimal
condition; more precisely if g is the Lie algebra of G and if (g, dρ, V ) is the
corresponding derived representation of g, the condition of prehomogeneity is
equivalent to the existence of an element v0 ∈ V such that the map
g −→ V
X −→ dρ(X) v0
is surjective.
For that reason, a finite-dimensional representation (g, V ) verifying the above
mentioned surjectivity condition will also be called a prehomogeneous vector
space. A prehomogeneous vector space (G, V ), where the group G is reductive
is called regular if the isotropy subgroup of the open orbit is reductive. This is
known to be equivalent to the fact that the complementary set of the open orbit is
an hypersurface.

Let now g = gk be a Z-gradation of a finite-dimensional reductive Lie
algebra. By a result of Vinberg [10], the representation (g0 , gi ) (i = 0), is always
a prehomogeneous vector space.
Let now g be a simple Lie algebra. Let h be a Cartan subalgebra of g and let R
be the root system of the pair (g, h). We choose a basis Ψ of R and denote by θ
a subset of Ψ . Let Hθ the unique element defined by the equations:
α(Hθ ) = 0 ∀α ∈ θ, α(Hθ ) = 2 ∀α ∈ Ψ \θ.


Let dp (θ ) = {X ∈ g: [Hθ , X] = 2pX}. Then g = p∈Z dp (θ ) is a Z-gradation
of g and the prehomogeneous vector spaces (d0 (θ ), dp (θ )) are said to be of
parabolic type. Moreover, the representation (d0 (θ ), d1 (θ )) is irreducible if and
only if |Ψ \θ | = 1 (see, for example, [9]). The following theorem will be used
later.

Theorem 1.1.1 [9]. Suppose that the prehomogeneous vector space of parabolic
type (d0 (θ ), d1 (θ )) is irreducible. Then it is regular if and only if there exist
Y ∈ d−1 (θ ) and X ∈ d1 (θ ) such that (X, Hθ , Y ) is a sl2 -triple.

If [d1 (θ ), d1 (θ )] = {0} then we will say that the prehomogeneous vector space
of parabolic type is commutative (and this is equivalent to say that dp (θ ) = {0}
for any p
2).
 J. Nervi / Journal of Algebra 253 (2002) 50–99 53

The regular prehomogeneous vector spaces of parabolic type have been

classified by H. Rubenthaler [9, Table 2, p. 42].

1.2. Affine algebras

Let us first recall some facts about affine algebras and fix some notations
(see [11] for details).
An affine generalized Cartan matrix A = (aij )i,j =0,...,k indecomposable matrix
of integers satisfying: aii = 2, aij  0 if i = j ; aij = 0 implies aij = 0.
Furthermore all the proper principal minors of A are positive and det A = 0.
A realization of the matrix A is a triple (h, Π, Π ∨ ), unique up to isomorphism,
where h is a complex vector space, Π = {α0 , . . . , αn } ⊂ h , and Π ∨ =
{α0∨ , . . . , αn∨ } ⊂ h satisfying the conditions:

• both sets Π and Π ∨ are linearly independent;

• αj , αi∨  = aij (i, j = 0, . . . , n);
• dim h = n + 2.

The affine Kac–Moody algebra ĝ is the Lie algebra over C generated by h

and the symbols ei , fi with defining relations: [h, h] = 0, [ei , fj ] = δij αi∨ ,
[h, ei ] = αi (h)ei , [h, fi ] = −αi (h)fi (for all i, j, h), (adei )1−aij (ej ) = 0 =
(adfi )1−aij (fj ) (i = j ). The last two relations are called Serre’s relations and
the quadruple (ĝ, h, Π, Π ∨ ) is called the quadruple associated to the matrix A.
We have the canonical embedding h ⊂ ĝ and linear independent Chevalley
generators for the derived algebra g of ĝ. 
The one-dimensional center Z(ĝ) of ĝ lies in h = h ∩ g = ni=0 Cα
∨
i and
n
consists of the scalar multiples of the (canonical) central element K = i=0 ai∨ αi∨
such that tAt (a0∨ , . . . , an∨ ) = 0 and the ai∨ are positive relatively prime integers.
Every non-zero ideal which is different of CK contains g [11, Proposition 1.7.b].
Fix an element d ∈ h defined modulo K by the following conditions:
αi , d = 0 for i = 1, . . . , n, α0 , d = 1. The element d is called the scaling
element and we have h = h ⊕ Cd.

The root space decomposition is given by ĝ = α∈h ĝα where
 
ĝα = x ∈ ĝ: [h, x] = α(h)x for all h ∈ h .
 
Put Q = ni=0 Zαi (respectively Q+ = ni=0 Z+ αi ). The set ∆ ⊂ Q of roots
(respectively the set ∆+ of positive roots) is defined by ∆ = {α ∈ h : α = 0,
ĝα = {0}} (respectively ∆+ = ∆ ∩ Q+ ).
The Weyl group W is the subgroup of Aut(h ) generated by the fundamental
reflections sαj (αj ∈ Π ) defined by sαj (λ) = λ − λ, αj∨ αj , for all λ ∈ h .
The Weyl group W preserves the root system ∆ and a real root (respectively
an imaginary root) is an element of ∆Re = W (Π) (respectively ∆Im = ∆\∆Re ).
Let us consider α ∈ ∆Re such that α = w(αi ) for some αi ∈ Π and w ∈ W . Let
54 J. Nervi / Journal of Algebra 253 (2002) 50–99

us put α ∨ = w(αi∨ ). One has α, α ∨  = 2 and there exist Xα ∈ ĝα and Yα ∈ ĝ−α
such that (Xα , α ∨ , Yα ) is a sl2 -triple. For a real root α, we also have dim ĝα = 1
and ∆ ∩ Zα = {α, −α}. 
Let us denote by δ = ni=0 ai αi the unique element of h such that
At (a0 , . . . , an ) = 0 and ai are positive relatively prime integers; then ∆Im = {j δ:
j ∈ Z }. The imaginary root spaces are finite-dimensional too [11, Corollary 8.3].
Denote by g̊ the underlying simple  subalgebra of ĝ generated by the triples
(ei , αi∨ , fi ), i = 1, . . . , n and by h̊ = ni=1 Cαi∨ the Cartan subalgebra of g̊. One
has h = h̊⊕CK ⊕Cd and the root system ∆ = (ĝ, h) can be completely described
in terms of δ and the root system ∆˚ of the pair (g̊, h̊) [11, Proposition 6.3]. The
set Π̊ = {α1 , . . . , αn } is a basis of ∆. ˚ The Cartan matrix of finite type Å of Π̊
is obtained from the matrix A by deleting the 0th row and column. The Dynkin
diagram S(Å) is obtained by deleting the 0th vertex on the Dynkin diagram, S(A),
of the matrix A.

1.3. Construction of a Z-gradation of ĝ

Let Γ be a proper subset (possibly empty) of the root basis Π . We set

hΓ = {h ∈ h: α(h) = 0 ∀α ∈ Γ }.

Remark 1.3.1. The one-dimensional center of the Lie algebra ĝ defined by

 
Z(ĝ) = CK = h ∈ h: αi (h) = 0 ∀i = 0, . . . , n
lies in hΓ .

Remark 1.3.2. Since the roots in Γ are linearly independent, we have

dim hΓ = dim h − card Γ.

Proposition 1.3.3.

• If α0 ∈/ Γ (i.e. Γ ⊂ Π̊ ) then hΓ = h̊Γ ⊕ CK ⊕ Cd where h̊Γ = {h ∈ h̊:

α(h) = 0 ∀α ∈ Γ }.
• If α0 ∈ Γ then hΓ = {h − α0 (h)d + λK: λ ∈ C, h ∈ h̃Γ } where h̃Γ = {h ∈ h̊:
α(h) = 0 ∀α ∈ Γ \{α0 }}.

Proof. An element H ∈ h can be written H = h + λK + µd (h ∈ h̊, λ, µ ∈ C).

• If α0 ∈
/ Γ , one has
h ∈ hΓ ⇔ α(h) = α(H ) = 0 ∀α ∈ Γ.
• If α0 ∈ Γ , one has

α(h) = α(H ) = 0 ∀α ∈ Γ \{α0 },
h ∈ hΓ ⇔ ✷
µ = −α0 (H ).
 J. Nervi / Journal of Algebra 253 (2002) 50–99 55


 For an element α = αi ∈Π nαi αi in the root lattice Q, the integer htΓ (α) =
αi ∈Π\Γ nαi is called the Γ -height of α.
For any integer p we set


  α∈Ωp ĝ
α if Ω = ∅,
p
Ωp = α ∈ ∆ ∪ {0}: htΓ (α) = p , dp =
{0} if Ωp = ∅.

Remark 1.3.4. (1) Since htΓ (δ)
1, we have Ω0 ∩ ∆Im = ∅.

(2) From the fact that for any integer j one has htΓ (j δ) = j htΓ (δ), we deduce
that |Ωp ∩ ∆Im |  1 for any integer p.
(3) The subalgebra d0 is equal to h ⊕ ĝΓ  where Γ  is the set of roots which
are a linear combination of elements of Γ . Since Γ = Π , we have Γ  ⊂ ∆Re .

Proposition 1.3.5. For any integer p, dim dp < ∞.

Proof. Using Remark 1.2.4(2) and since dim ĝα < ∞ for any α ∈ ∆ ∪ {0}, it is
sufficient to prove that |Ωp ∩ ∆Re | < ∞. Let α be a real root in Ωp . Suppose
(2)
that ĝ is not of type A2l , then α = j δ + γ , j ∈ Z, γ ∈ ∆˚ [11, Proposition 6.3].
Hence we have htΓ (α) = j htΓ (δ) + htΓ (γ ) = p. Since the set {htΓ (γ ), γ ∈ ∆} ˚ is
(2)
finite, the result follows. If ĝ is of type A2l one can also have α = 2 (2j − 1)δ + γ ,
1

j ∈ Z, γ ∈ ∆˚ and the proof is similar. ✷

The proof of the following two propositions is straightforward.

Proposition 1.3.6. The decomposition ĝ = p∈Z dp is a Z-gradation of ĝ.

Proposition 1.3.7. The finite-dimensional subalgebra d0 is reductive and its

center is equal to hΓ . The Dynkin diagram of the semisimple algebra d0 = [d0, d0 ]
is obtained from the Dynkin diagram of ĝ by removing the vertices of Π\Γ .

Let us assume from now on that |Π\Γ |
2.

In this section, we will decompose the d0 -module d1 (under the adjoint action)
into a direct sum of irreducible d0 -modules.
Let us make the convention which consists of circling the vertices ω0 , . . . , ωk
of Π\Γ in the Dynkin diagram S(A). We denote by Πj the connected component
of Γ ∪ {ωj } which contains ωj . As |Π\Γ |
2, the set Πj , can be identified with
a proper subdiagram of S(A); hence it is of finite type. Denote by ∆j = Πj  the
set of roots in ∆ which are linear combinations of elements in Πj . Let gj be the
associated
 simple algebra which can be viewed as a subalgebra of ĝ. The algebra
hj = α∈Πj Cα ∨ is a Cartan subalgebra of gj and ∆j is the set of roots of the
pair (gj , hj ).
56 J. Nervi / Journal of Algebra 253 (2002) 50–99

The subset Γj = Πj \{ωj } of Πj defines again a Z-gradation of the finite-

dimensional algebra gj as seen in Section 1.1. To be more precise, let Hj be the
unique element in hj satisfying the equations
α(Hj ) = 0 ∀α ∈ Γj , ωj (Hj ) = 2.
j
The spaces of this gradation are then defined by dp = {x ∈ gj : [Hj , x] = 2px}
j j
and the representations (d0 , dp ) are known to be irreducible (see Section 1.1).
Let us illustrate the construction of the Πj ’s by the following example:

(1)
The algebra ĝ is of type B14 . The connected components Πj are of type D5 ,
A8 , and B9 respectively:

Proposition 1.3.8.

k j
• We have d1 = j =0 d1 .
j
• For any j , the space d1 is an irreducible d0 -module for the adjoint action.


Proof. Recall that d1 = α∈Ω1 ĝα where Ω1 = {α ∈ ∆: htΓ (α) = 1}. We also
j
j
have d1 = α∈Ω j ĝα where Ω1 = {α ∈ ∆j : α = ωj (mod Γj )}. Since the
1
j
support of a root is a connected subgraph of S(A), one has Ω1 = Ω1 ∩ ∆j and Ω1
j
j j
is the union of the disjoint sets Ω1 . Therefore d1 = kj =0 d1 . The space d1 is an
j j j
irreducible d0 -module which is d0 -stable. As d0 ⊂ d0 , the space d1 is also an
irreducible d0 -module. ✷

1.4. The admissible subalgebras

1.4.1. A construction of generalized Cartan matrices

The key point of our construction is the following choice of the subset Γ (or
Π\Γ ). Using the same notations as in the preceding sections, we will choose
j j
the simple roots ω0 , . . . , ωk in Π\Γ such that (d0 , d1 ) is regular (respectively
regular and commutative) for any j = 0, . . . , k. This leads to the following
definition.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 57

Definition 1.4.1.1. The pair (Π, Γ ) is called admissible (respectively C-admis-

j j
sible) if (d0 , d1 ) is regular (respectively regular and commutative) for any j =
0, . . . , k.

The classification of admissible (respectively C-admissible) pairs (Π, Γ ) is

now easy. From above, such pairs are in one to one correspondence with weighted
Dynkin diagrams such that each connected subdiagram with only one circled root
(by the procedure explained in Section 1.1) lies in the table of regular (respectively
regular commutative) irreducible parabolic prehomogeneous spaces (see the left
column of Table A).

We assume now that (Π, Γ ) is admissible.

Proposition 1.4.1.2.
 j j
(i) For any j = 0, . . . , k, we have Hj = α∈Πj xα α ∨ where each xα is a positive
integer.
(ii) The elements H0 , . . . , Hk are linearly independent in hΓ .

Proof. Let us fix j in {0, . . . , k}; let Mj denote the Cartan matrix of finite type
 j
associated to the quadruple (gj , hj , Πj , Πj∨ ). Let us set Hj = α∈Πj xα α ∨ . Since
Hj verifies

α(Hj ) = 0 ∀α ∈ Γj , ωj (Hj ) = 2,
j
the column X = t((xα )α∈Πj ) is the solution of the linear system tMj X = B
where B is a column whose coefficients are all zero except one (say the qth
one) which is equal to 2. Using [11, Theorem 4.3] we deduce that X > 0 (i.e.
xα > 0 ∀α ∈ Πj ) and we have X = (tMj )−1 B; more precisely X = 2Λq where
j

Λq is the qth column of the matrix (tMj )−1 . These matrices (tMj )−1 are well
known (they give the fundamental weights in terms of the simple roots and we
find these relations in tables in [12]). It is easy to check that for every regular
j j j
space (d0 , d1 ) we have the expected result, namely that each xα is a positive
integer. More precisely:

• For the types B, C, E7 , E8 , F4 , G2 , the coefficients of the matrix 2(tMj )−1

are always integers.
j j
• For the type A, the space (d0 , d1 ) is regular if gj is of type A2p+1 and ωj
is the central simple root in the Dynkin diagram (i.e. q = p + 1). Hence we
have X = 2Λp+1 = t(1, 2, . . . , p, p + 1, p, . . . , 2, 1).
• For the type Dn :
 if ωj is different from both αn and αn−1 then 2Λq is integer;
58 J. Nervi / Journal of Algebra 253 (2002) 50–99

 if ωj = αn (respectively ωj = αn−1 ) then n is even and 2Λn (respectively

2Λn−1 ) is an integer.
• For the type E6 it is sufficient to check that 2Λ2 and 2Λ4 are integers.

It is convenient to resume the cases A, D, E6 by the following diagrams:

 j
Let us prove now that Hj = xα α ∨ lies in hΓ . Since for any γ ∈ Γ \Γj
α∈Πj
 j
and any α ∈ Πj one has (γ , α) = 0, then γ (Hj ) = α∈Πj xα γ (α ∨ ) = 0. For
any γ ∈ Γj , one also has α(Hj ) = 0 by definition of the element Hj . Therefore
Hj ∈ hΓ .

Let us set now H = kj =0 aj Hj = 0 for some (a0 , . . . , ak ). For any j , the
j j
coefficient of ωj∨ in H is aj xωj (where xωj is the coefficient of ωj∨ in Hj ). Since
j j
the elements of Π ∨ are linearly independent, we have aj xωj = 0. Since xωj > 0,
we have aj = 0 for any j , and therefore the elements H0 , . . . , Hk are linearly
independent. ✷

Remark 1.4.1.3. As dim hΓ = dim h − |Γ | = k + 2, the set {H0 , . . . , Hk } is not

a basis of hΓ .

Proposition 1.4.1.4. The matrix M = (ωj (Hi ))ij is a generalized Cartan matrix.

Proof. For any (i, j ), put mij = ωj (Hi ). Then for all j we have mjj = ωj (Hj ) =
2. If i = j , we have
  
i ∨
mij = ωj xγ γ = xγi ωj (γ ∨ ). ()
γ ∈Πi γ ∈Πi

On the other hand, the constants in the preceding sum are positive integers
(Proposition 1.4.1.2) and the constants ωj , γ ∨  are non-positive integers (since
ωj ∈
/ Πi ). Therefore the mij are non-positive integers.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 59

Suppose now that mij = 0 for some i = j . For all γ ∈ Πi , one has
ωj , γ ∨   0. As xγi > 0 for all γ ∈ Πi , the equality () above implies that
ωj , γ ∨  = 0 for all γ ∈ Πi . Then Πj ∩ Πi = ∅ and furthermore (ωi , ωj ) = 0.
Hence ωi , γ ∨  = 0 for all γ ∈ Πj , therefore mj i = 0. ✷

Theorem 1.4.1.5. The generalized Cartan matrix M is of affine type.

We will need the following lemma.

Lemma 1.4.1.6. The set {H0 , . . . , Hk } is a basis of the subspace h ∩ hΓ .

Proof. We remark first that Hj = [Xj , Yj ] is an element of h for any j . Using

Remark 1.4.1.3, it is sufficient to prove that the space h ∩ hΓ is a proper subspace
of hΓ .
If α0 ∈
/ Γ thend ∈ hΓ (Proposition 1.3.3) and (d, H0 , . . . , Hk ) is a basis of hΓ ,
hence h ∩ hΓ = kj =0 CHj .
Observe that if S(A) is of (circular) type A(1) n , we can always reorder the
indices of the simple roots such that α0 ∈ /Γ.
If α0 ∈ Γ then we can assume that S(A) is not of type A(1) n and we know
that hΓ is the set of the elements H in form of H = h − α0 (h)d + λK where
λ ∈ C and h varies in the orthogonal of Γ \{α0 } in h̊ (Proposition 1.3.3). Let us
show now that there exists an element in hΓ which is not in h .
Among the connected components defined in S(A) as described in Section 1.1,
choose one that contains α0 and denote it by Π0 . Since α0 ∈ Γ , Π0 contains more
than one simple root and we will call ω0 the unique vertex of Π\Γ which is
in Π0 . Put now Π0 = {α0 , . . . , αl = ω0 , . . . , αp }. The simple algebra g0 associated
to Π0 (and to the corresponding Cartan matrix denoted by M0 ) contains the simple
subalgebra g̃0 corresponding to the connected set Π0 \{α0 }. The invertible Cartan
matrix M 0 associated to g̃0 is obtained from M0 by deleting the row and column
corresponding to α0 .
Let h be an element of the Cartan subalgebra of g̃0 such that h = x1 α1∨ + · · · +
xp αp∨ . Set X = t(x1 , . . . , xp ). The system

αj (h) = 0 for all j = l, ω0 (h) = 1

is equivalent to the equation M 0 is of finite

0 X = t(0, . . . , 0, 1, 0, . . ., 0). Since M
type, X is unique and X > 0 [11, Theorem 4.3].
This element h is obviously in the orthogonal of Γ \{α0 } in h̊. Since the
p α0 is connected
vertex to only one vertex in Π0 \{α0 }, one deduces that α0 (h) =
j =1 x j α0 (αj
∨
) < 0. Hence h − α0 (h)d is an element of hΓ that does not lie in h

and hΓ ∩ h is a proper subspace of hΓ . Then hΓ ∩ h = kj =0 CHj . ✷
60 J. Nervi / Journal of Algebra 253 (2002) 50–99

We can illustrate this by the following example:

Proof of Theorem 1.4.1.5. The matrix M is not invertible. In fact, the non-
zero canonical central element K which is obviously in h ∩ hΓ , can be written
K = x0 H0 + · · · + xn Hn (Lemma 1.4.1.6) and one has M t(x0 , . . . , xn ) = 0.
The matrix M is indecomposable. In fact, recall that for any j = 0, . . . , k the
coefficients of the element Hj in the basis Πj∨ are positive (Proposition 1.4.1.2).
As S(A) is a connected graph, one deduces that it is impossible to decompose the
set {0, . . . , k} into I1 ∪ I2 with ωj (Hi ) = 0 for all j ∈ I1 and i ∈ I2 .
Let M  be a proper principal submatrix of M associated to a proper subset I  of
{0, . . . , k}. Then M  corresponds to the C-admissible diagram of semisimple type
constituted by the union of Πj for j ∈ I  . Using [6, Theorem 3.1], we deduce that
det M  > 0. We conclude by applying [11, Proposition 4.7.b]. ✷

We give in terms of Dynkin diagrams the generalized Cartan matrices

associated to admissible (respectively C-admissible) pairs (Π, Γ ) (see the right
column of Table A). Note that this classification allows to give the type of the
admissible (respectively C-admissible) algebras that we define in the following
Section 1.4.2.
These Cartan matrices are obtained in the following way. We first calculate
the coefficients of the elements Hj in the basis Πj∨ by the procedure described
in the proof of Proposition 1.4.1.2. Then it is easy to calculate the coefficients
mij = ωj (Hi ) of the corresponding affine matrix M.
Let us illustrate by the following example. In an affine algebra ĝ of type Bn(1) ,
we consider the set Γ defined by the following diagram and such the pair (Π, Γ )
is admissible.

For any j , the numerical labels in the following diagrams are coefficients of Hj
in the coroot basis Πj∨ .
The connected component Π0 is of type Dl :
 J. Nervi / Journal of Algebra 253 (2002) 50–99 61

The connected components Π1 , . . . Πk−1 are of type A2p+1 :

The connected component Πk is of type Bl :

One remarks that only the circled coefficients are useful to calculate the
ωj (Hi )’s. The Cartan matrix associated to the pair (Π, Γ ) is then
 
2 −2 0 . . . 0
 −1 2 −1 . . . 0 
 . .. 
M =  .
. ..
.
..
.
..
. 
. 
 0 . . . −1 2 −1 
0 . . . 0 −2 2
and the Dynkin diagram is

1.4.2. Construction of the admissible subalgebras

Let us consider the subalgebra ĝΓ of ĝ, generated by hΓ and the elements
Xj , Yj for j = 0, . . . , k. Put Ψ = {ω0 , . . . , ωk } where ωj is the restriction to hΓ
of the simple root ωj and put Ψ ∨ = {H0 , . . . , Hk }.

Proposition 1.4.2.1. The quadruple (ĝΓ , hΓ , Ψ, Ψ ∨ ) is the quadruple associated

to the affine matrix M.

Proof. The fact that the triple (hΓ , Ψ, Ψ ∨ ) is a realization of M is straightfor-

ward. Furthermore one has clearly the relations:

• [hΓ , hΓ ] = {0};
j
• as [d1i , d−1 ] = {0} for i = j , one has [Xi , Yj ] = δij Hj for all i, j ;



• as d1 = α=ωj (mod Γj ) ĝα (respectively d−1 = α=ωj (mod Γj ) ĝ−α ) one has
j j

[h, Xj ] = ωj (h)Xj (respectively [h, Yj ] = −ωj (h)Yj for all h ∈ hΓ .

Let us prove now that Serre’s relations are verified.

Fix i ∈ {0, . . . , k} and consider the sl2 -triple (Xi , Hi , Yi ).
If k
1 then for any j = i the proper subset Πi ∪ Πj of S(A) corresponds
to a semisimple subalgebra g̃ of ĝ. This finite-dimensional subalgebra is a sl2 -
module under the action of (Xi , Hi , Yi ). The element Yj of g̃ satisfies the
62 J. Nervi / Journal of Algebra 253 (2002) 50–99

relations [Xi , Yj ] = 0, [Hi , Yj ] = −ωj (Hi )Yj = −mij Yj . Hence Yj generates an

irreducible (1 −mij )-dimensional submodule with basis {vm , vm = (ad(Yi ))m (Yj )
for 0  m  1 − mij } and one has (ad(Yi ))(1−mij ) (Yj ) = 0 [13]. The second
relation, (ad(Xi ))(1−mij ) (Xj ) = 0 (i = j ), is proved the same way.
If k = 1 then the argument does not hold because one has Π0 ∪ Π1 = S(A).
Consider ĝ as a sl2 -module under the action of (X0 , H0 , Y0 ). Then Y1 is a
primitive element of weight −m01 . The vectors vm = m! 1
(ad(Y0 ))m (Y1 ), m
0,
are eigen vectors whose weights are −m01 − 2m and one has the well-known
relations:
ad(Y0 )(vm ) = (m + 1)vm+1 , (1)
ad(X0 )(vm ) = (−m01 − m + 1)vm−1 . (2)

−ω −β
−ω −β
Since Y1 ∈ β ĝ 1 and Y0 ∈ β ĝ 0 where β varies in the set of positive


linear combination of elements of Γ , we have (ad(Y0 ))m (Y1 ) ∈ β ĝ−mω0 −ω1 −β .
For m great enough, −mω0 − ω1 − β is not a root, therefore vm = 0. Put
N = min{m
0: vm = 0}. One has N
1 and (1) implies that vm = 0 for
all m
N . Since vN−1 = 0, (2) implies that −m01 − N + 1 = 0. Hence
(ad(Y0 ))1−m01 (Y1 ) = 0. By similar arguments we prove the other Serre’s relations
and the theorem follows. ✷

Definition 1.4.2.2. The subalgebra ĝΓ is called the admissible (respectively C-

admissible) subalgebra associated to the admissible (respectively C-admissible)
pair (Π, Γ ).

Denote by ∆Γ the root system of the pair (ĝΓ , hΓ ).

Remark 1.4.2.3. The set Ψ = {ω0 , . . . , ωk } is a root basis and Ψ ∨ = {H0 , . . . , Hk }

is the coroot basis.

Denote by δΓ the unique root of ∆Γ such that δΓ = kj =0 bj ωj where bj
are positive relatively prime integers satisfying M t(b0 , . . . , bk ) = 0. One has
(2)
Γ = {j δ: j ∈ Z }. We recall that b0 = 2 if ĝΓ is of type A2l and b0 = 1 if
∆Im 

not.

Remark 1.4.2.4. Since K is an element of hΓ , the one-dimensional center of ĝΓ

is Z(ĝΓ ) = CK.

Remark 1.4.2.5. The subalgebra ĝΓ depends on the choice of the elements
Xj , Yj . Obviously all admissible subalgebras corresponding to the same subset Γ
are isomorphic.

Remark 1.4.2.6. If the subset Γ = ∅ then one has hΓ = h and ĝΓ = ĝ.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 63

1.5. The root system of the C-admissible subalgebras

In this section we will assume that the pair (Π, Γ ) is C-admissible.

Remark 1.5.1. The algebras of type A(2)

2l do not contain any C-admissible
subalgebra (see Appendix A).

Except in the two following cases:

there exists in Π\Γ a unique root such the associated connected component of
S(A) (as defined in Section 1.2) contains α0 . Such a root will be denoted by ω0
(possibly ω0 = α0 ). The root ω0 of Ψ corresponds to the so-called 0th vertex on
(1)
the Dynkin diagram of ĝΓ . In the following example ĝ is of type E7 , ĝΓ is of
(2)
type A5 and one has ω0 = α0 :

E7(1) A(2)
5

In the two exceptional cases before, one can consider that ω0 = α0 by

reordering the indices.
Choose now in hΓ an element denoted by dΓ , unique up to a summand
proportional to K and satisfying
ω0 (dΓ ) = 1, ωj (dΓ ) = 0 for all j = 1, . . . , k.
The set {dΓ , H0 , . . . , Hk } is a basis of hΓ [11, 6.2, p. 81].

Remark 1.5.2. If ω0 = α0 then dΓ = d.

Proposition 1.5.3. The restriction to hΓ of the root δ of ∆, denoted by δ, is equal

to kδΓ where k is a positive integer. Furthermore if ω0 = α0 then δ = δΓ .

Proof. Recall that δΓ is also defined by the relations

 k

δΓ  CHj = 0, δΓ , dΓ  = b0 ,
j =0
(2)
where b0 = 1 if ĝΓ is of type A2l and b0 = 2 if not.
64 J. Nervi / Journal of Algebra 253 (2002) 50–99


One has δ = ni=0 aαi αi where aα0 = 1 (see Remark 1.5.1). Then δ =
k 
j =0 aωj ωj . As Hj ∈ h for any j , one has δ(Hj ) = 0 and


k
δ(dΓ ) = aωj ωj (dΓ ) = aω0 .
j =0

Hence
aω0
δ= δΓ .
b0
(2)
If ĝΓ is not of type A2l then b0 = 1 and the result follows.
If ĝΓ is of type A(2)
2l then b0 = 2. One checks that for each case where the
(2)
C-admissible algebra ĝΓ is of type A2l the number aω0 is an even integer. From
Table A the list of such (Π, Γ ) is given by the following diagrams (where the
numerical labels are the coefficients of the root δ).

(1) (2)
E7 A2

(1) (2)
E8 A2

E8(1) A(2)
4

(1) (2)
E8 A2

E6(2) A(2)
4

In any case where ω0 = α0 we have aω0 = aα0 = 1 and δ = δΓ . ✷

Corollary 1.5.4. ∆Im ⊂ ∆Im

Γ .

The following proposition is standard.

 J. Nervi / Journal of Algebra 253 (2002) 50–99 65

Proposition 1.5.5. ∆Γ ⊂ ∆.

Proposition 1.5.6. The algebra ĝ is an integrable ĝΓ -module.

Proof.

One recall that a ĝΓ -module V is called hΓ -diagonalizable if V =
λ∈hΓ Vλ where Vλ = {v ∈ V : h(v) = λ(h)v for h ∈ hΓ } and it is called
integrable if all Xj , Yj for j = 0, . . . , k are locally nilpotent on V .

The ĝΓ -module ĝ is clearly hΓ -diagonalizable since one has ĝ = α∈∆∪{0} Vα
where
  
Vα = ĝβ = x ∈ ĝ: [h, x] = α(h)x, h ∈ hΓ .
β=α

Since ĝα ⊂ Vα , then α is a weight of the ĝΓ -module ĝ.

Let us prove now that for any j = 0, . . . , k, ad Xj and ad Yj are locally
nilpotent on ĝ. It is sufficient to prove they are locally nilpotent on h and on
every ei and fi (i = 0, . . . , n) [11, Lemme 3.4]. 
Fix j ∈ {0, . . . , k}. Let h be an element of h. Since h = hΓ ⊕ α∈Γ Cα ∨ ,
we can write h = H + hj + hk where H ∈ hΓ , hj ∈ h(Γj ) = α∈Γj Cα ∨
 j
and kj ∈ h(Γ \Γj ) = α∈Γ \Γj Cα ∨ . Note that [h(Γ \Γj ), d1 ] = {0}; hence
[Xj , kj ] = 0. One also has [Xj , H ] = −ωj (H )Xj ; hence (ad Xj )2 (H ) = 0. Since
j j
the pair (Π, Γ ) is C-admissible, we have [Xj , [Xj , hj ]] ∈ [d1 , d1 ] = {0}; hence
(ad Xj )2 (hj ) = 0. So we have (ad Xj )2 (h) = 0 for all h ∈ h. The same proof
shows that (ad Yj )2 (h) = 0.
Let i be ∈ {0, . . . , n}, and consider the corresponding αi ∈ Π , we have ei ∈ ĝαi .

j j j
• If αi ∈ Γj then ei ∈ d0 and (ad Xj )2 (ei ) ∈ [d1 , d1 ] = {0}.
• If αi ∈ Γ \Γj then (ad Xj )(ei ) = 0.
• If αi ∈
/ Γ (i.e. α i = 0) then there exists l such that αi = ωl and, for some
integer N , one has

(ad Xj )N (ei ) ∈ VNωj +ωl = ĝβ .
β=Nωj +ωl (mod Γ )

For N big enough one has ĝβ = {0} and the result follows.
For fi ∈ ĝ−αi let us consider the following cases:

j j
• If αi ∈ Γj then (ad Xj )2 (fi ) ∈ [d1 , d1 ] = {0}.
• If αi ∈ Γ \Γj then (ad Xj )(fi ) = 0.
j
• If αi = ωl (l = j ) then (ad Xj )(fi ) = 0 because [d1 , d−1
l
] = {0}.
j
• If αi = ωj then (ad Xj )(fi ) ∈ d0 ; so (ad Xj )3 (fi ) = 0.
66 J. Nervi / Journal of Algebra 253 (2002) 50–99

The same proof shows that ad Yj is locally nilpotent on ei , fi for i =

0, . . . , n. ✷

Remark 1.5.7. A slight modification of the preceding proof shows that ĝ is still
an integrable ĝΓ -module even if (Π, Γ ) is admissible.

Theorem 1.5.8. The root system ∆Γ of the C-admissible algebra ĝΓ is ∆\{0}.

Proof. Since ĝ is an integrable ĝΓ -module, we can define the automorphisms

of ĝ:
rjad = exp(ad Yj ) exp(− ad Xj ) exp(ad Yj ).
Denote by rj the fundamental reflection corresponding to the simple root ωj ∈ Ψ .
Recall that

rjad (Vα ) = Vrj (α) , rjad ∈ Aut(ĝΓ ), rjad hΓ = rj ()
(see [11, Lemma 3.8]).
Let w be an element of the Weyl group WΓ of ĝΓ generated by the rj . If
w = ri1 . . . rip then put wad = riad
1
. . . riad
p
∈ Aut(ĝ). The fact that w(α) is a weight
of ĝ for any α ∈ ∆ ∪ {0} follows from the first assertion of () and w(α) ∈ ∆ ∪ {0}.
Let α be an element of ∆ such that α = 0 and α is not proportional to some
root of ∆Γ . Consider the R-subspace hΓ,R of hΓ generated by the elements
dΓ , H0 , . . . , Hk . Since ωj (dΓ ) and ωj (Hi ) are integers for all i, j , then α|hΓ,R ∈
hΓ,R . Denote by Lωj (respectively LδΓ , Lα ) the kernel of ωj |hΓ,R (respectively
δΓ |hΓ,R , α|hΓ,R ).
We have obviously Lα = Lωj , Lα = LδΓ and Lα is not contained in the union
of the kernels LδΓ and Lωj (j = 0, . . . , k).
Let H be an element of Lα which is not in this union. We can choose H
such that δΓ (H ) > 0; so H is in the Tits cone [11, Proposition 5.8]. Hence
there exists w ∈ WΓ such that ωj (w(H )) > 0 for any j . Put β = w(α); we have
Lβ = w(Lα ), w(H ) ∈ Lβ , β = 0 and β is not proportional to some root of ∆Γ .

Set β = kj =0 nj ωj where the coefficients nj are integers of the same sign. We

have β(w(H )) = kj =0 nj ωj (w(H )) = 0. It is impossible since β = 0. Therefore
either β = 0 or β = kν with ν ∈ ∆Γ and the same is true for α (α = 0 or α = kµ,
µ ∈ ∆Γ ).

• If µ is imaginary then α ∈ ∆ImΓ .

• If µ is real then there exists w ∈ WΓ and j such that w(α) = kωj . Since
j j
[d1 , d1 ] = {0}, we have k = ±1. Hence α ∈ ∆Re
Γ .

Therefore one has ∆ ⊂ ∆Γ ∪ {0} and Proposition 1.5.5 implies the final
result. ✷
 J. Nervi / Journal of Algebra 253 (2002) 50–99 67

2. Affine algebras graded by affine root systems

2.1. Definition

The definition of a Lie algebra graded by a finite root system has been
introduced by S. Berman and M. Moody (see also the papers by Benkart and
Zelmanov [2], Allison et al. [4], and Nervi [5]). In this work we study more
particularly affine algebras graded by affine root systems. This needs a slight
extension of the definition given in [4].

Definition 2.1.1. Let Σ be an affine root system. An affine Lie algebra ĝ over C
is graded by Σ or Σ-graded if

(i) ĝ contains as a subalgebra an affine algebra a whose root system relative to a

subalgebra ha = a is equal to Σ;
Cartan 0

(ii) ĝ = α̃∈Σ∪{0} Vα̃ where Vα̃ = {x ∈ ĝ: [h, X] = α̃(h)x for all h ∈ ha }.

The affine subalgebra a is called the grading subalgebra relative to Σ.

Remark 2.1.2. For any α̃ ∈ Σ ∪ {0} the root space aα̃ is included in Vα̃ and one
has Vα̃ = {0}.

Remark 2.1.3. As α̃∈Σ ([Vα̃ , V−α̃ ] ⊕ Vα̃ ) is a non-zero ideal of ĝ that contains
[a, a], we have [ĝ, ĝ] = α̃∈Σ ([Vα̃ , V−α̃ ] ⊕ Vα̃ ) [11, Proposition 1.7.b].

Therefore a general definition of an arbitrary Lie algebra graded by an affine

root system should include the fact that the derived algebra is generated by the
spaces Vα̃ (α̃ ∈ Σ). The preceding computation shows that this is always true if ĝ
is affine.

2.2. The grading defined by a C-admissible subalgebra

Proposition 2.2.1. The root system ∆Γ of any C-admissible subalgebra ĝΓ of

an affine algebra ĝ grades ĝ as in Definition 2.1.1 and ĝΓ is the corresponding
grading algebra.

Proof. For any C-admissible pair (Π, Γ ) in ∆ = (ĝ, h) we have defined

hΓ = {h ∈ h: α(h) = 0, ∀α ∈ Γ }
and we have ∆Γ = (ĝΓ , hΓ ) = ∆\{0} where ∆ = {α = α|hΓ for all α ∈ ∆}
(Theorem 1.5.8). Set

Vα = ĝγ for all α ∈ ∆Γ ∪ {0}.
γ |hΓ =α
68 J. Nervi / Journal of Algebra 253 (2002) 50–99

Since ĝ is an integrable gΓ -module whose weights are ∆Γ ∪ {0} (Proposi-

tion 1.5.6), the proof of the proposition is then straightforward. ✷

2.3. Converse problem: first results

In this section we establish a link with the grading affine subalgebras of ĝ and
the C-admissible subalgebras.
Let ĝ be an affine algebra graded by an affine root system Σ. Denote by
B = { ω0 , . . . ,
ωk } (k
1) a root basis of Σ and by a the grading subalgebra.
The elements of the Cartan subalgebra ha of a are semisimple since ĝ is ha -
diagonalizable (condition (ii) of Definition 2.1.1). One deduces by applying the
conjugacy theorem of Kac–Peterson [14] that there exists a Cartan subalgebra h
of ĝ containing ha .
Put ∆ = (ĝ, h) and ρ : h → ha , α → α|ha .

Remark

2.3.1. For any α̃ ∈ Σ ∪ {0}, the subspace

Vα̃ of the Σ-gradation is equal
to ρ(γ )=α̃ ĝγ and one has V0 = Zĝ (ha ) = h ⊕ ρ(γ )=0 ĝγ .

Lemma 2.3.2. The map ρ is a surjection from ∆ ∪ {0} onto Σ ∪ {0}.

Proof. Since ĝ = α̃∈Σ∪{0} Vα̃ , then for any γ ∈ ∆, there exists α̃ ∈ Σ ∪ {0}
such that ĝγ ⊂ Vα̃ and ρ(γ ) = α̃.
Let α̃ be an element of Σ, since Vα̃ = {0} (Remark 2.1.2), there exists γ ∈ ∆
such that ĝγ ⊂ Vα̃ and ρ(γ ) = α̃. ✷

Denote by δ (respectively δa ) the unique imaginary root of ∆ (respectively Σ)

whose coefficients in a given basis Π (respectively B) are positive relatively prime
integers.

Lemma 2.3.3. The restriction to ha of the root δ is proportional to δa and one

can choose a root basis Π of ∆ such that the coefficient is a positive integer.

Proof. Let Π be a basis of ∆. Lemma 2.3.2 implies that ρ(δ) is a root of Σ or is

zero; for any j ∈ Z, one has ρ(j δ) = jρ(δ) and it is an element of Σ ∪ {0} too.
Suppose that ρ(δ) = 0. Recall that any real root can be written as α = j δ + γ
(γ ∈ ∆,˚ j ∈ Z) (possibly α = 1 ((2j − 1)δ + γ if ĝ is of type A(2)). Since
2 2
|ρ(∆)|
˚ < ∞, the hypothesis ρ(δ) = 0 implies |ρ(∆)| < ∞. This conclusion is
not true since ρ(∆ ∪ {0}) = Σ ∪ {0}. Hence ρ(δ) and jρ(δ) for any j are roots
of Σ. We deduce that there exists k ∈ Z such that ρ(δ) = kδa .
Suppose now that k < 0. One can choose the basis Π  = −Π . For this basis

Π and the corresponding element δ, one has ρ(δ) > 0. (Note that this is just
choosing a conjugacy class of basis under the Weyl group of ∆). ✷
 J. Nervi / Journal of Algebra 253 (2002) 50–99 69

From now on we choose a basis Π such that ρ(δ) > 0.

Define the non-empty set

 
P = γ ∈ ∆: ρ(γ ) ∈ Σ+ ∪ {0} .
We have clearly P ∪ (−P) = ∆ and P is a closed set.

Proposition 2.3.4. There exists a basis Π of ∆ such that ∆+ (Π) ⊂ P (where

∆+ (Π) is the set of positive roots relative to Π ).

+ (Π) = {j δ: j ∈ N }. The choice of Π implies that

Proof. Recall that ∆Im 

∆+ (Π) ⊂ P.
Im

Any real root can be written as α = j δ + γ (γ ∈ ∆,

˚ j ∈ Z) (possibly α =
(2)
2 ((2j − 1)δ + γ if ĝ is of type A2 ) and ∆+ (Π) = {α ∈ ∆ (Π): j > 0} ∪
1 Re Re

∆˚ + (Π̊ ) [11, Proposition 6.3.e]. Set

S = P ∩ (−P) = {α ∈ ∆: ρ(α) = 0}.
One has
   
α ∈ ∆+ (Π): ρ(α) < 0 = ∆+ (Π) ∩ (−P)\S .
This set does not contain any positive imaginary root. Let α be a positive real
root of ∆. If α = j δ + γ (γ ∈ ∆, ˚ j ∈ N) then ρ(α) = j kδa + ρ(γ ). Since
|ρ(∆)|
˚ < ∞, then for j big enough one has ρ(α) > 0. Hence |{α = j δ + γ :
j
0, ρ(α) < 0}| < ∞.
If ĝ is of type A(2)
2l , a positive real root could be written as α = 2 ((2j −1)δ +γ )
1

with 2j − 1 > 0, and the conclusion is the same. Hence we always have
|∆+ (Π) ∩ ((−P)\S)| < ∞. If the previous set is empty then one has obviously
∆+ (Π) ⊂ P.
Suppose now that ∆+ (Π) is not a subset of P, then there exists a root
αi ∈ Π such that ρ(αi ) < 0. Consider the reflection s−αi [11, p. 35] and set
Π  = s−αi (Π). The set Π  is a root basis of ∆ [11, Proposition 5.9] and s−αi
preserves the root δ. We clearly have ρ(−αi ) > 0; so −αi ∈ / (P)\S (1). On the
other hand, one has s−αi (αi ) = −αi ∈ Π  hence −αi ∈ ∆+ (Π  ) (2).
Let us prove now that ∆+ (Π  ) ∩ ((−P)\S) is a proper subset of ∆+ (Π) ∩
((−P)\S). Note that ∆+ (Π  ) ∩ ((−P)\S) ⊂ ∆+ (Π  )\{−αi } (cf. (1) and (2)).
We know that the set ∆+ (Π  )\{−αi } is s−αi -invariant [11, Lemme 3.7]. Hence
for any β ∈ ∆+ (Π  )\{−αi }, one has
 
β = (s−αi )2 (β) ∈ s−αi ∆+ (Π  ) .
Since Π = s−αi (Π  ), we have s−αi (∆+ (Π  )) ⊂ ∆+ (Π). Therefore
   
∆+ (Π  ) ∩ (−P)\S ⊂ ∆+ (Π) ∩ (−P)\S .
70 J. Nervi / Journal of Algebra 253 (2002) 50–99

Furthermore one has αi ∈ ∆+ (Π) ∩ ((−P)\S) and αi ∈ / ∆+ (Π  ). Hence the

inclusion is strict.
After a finite number of such steps we get a root basis Θ such that ∆+ (Θ) ∩
((−P)\S) = ∅ and the proposition follows. ✷

Fix once for all a basis Π that verifies ∆+ (Π) ⊂ P.

The following corollary is straightforward.

Corollary 2.3.5. The map ρ is a surjection from ∆+ ∪ {0} onto Σ+ ∪ {0} and
ρ −1 (Σ+ ) ⊂ ∆+ .

Remark 2.3.6. If the set S = P ∩ (−P) is empty (this means that ∆− ∩ P = ∅ or

∆+ ∩ (−P) = ∅) then ρ is a surjection from ∆+ onto Σ+ .

Proposition 2.3.7. If the set S = P ∩ (−P) is not empty then there exists a non-
empty subset Γ of Π , such that ∆+ ∪ Γ − = P. This condition is equivalent to
Γ  = P ∩ (−P).

Proof. The condition S = ∅ implies that there exists a simple root in (−P)
(Remark 2.3.6). Set Γ = Π ∩ (−P) = {γ ∈ Π: ρ(γ ) = 0}. Γ is a proper subset
of Π and Γ  is a finite root system.
Let us prove now that Γ − = ∆− ∩ P. Let α be an element of Γ − , write
α = −β1 − · · · − βn (βj ∈ Γ ). Let us prove by induction on n that Γ − ⊂
∆− ∩ P. For n = 1, one has α = −β1 ; since β1 ∈ Γ ⊂ (−P) then −β1 ∈ ∆− ∩ P.
Let n be a positive integer, suppose that if α = −β1 − · · · − βn , βj ∈ Γ then
α ∈ ∆− ∩ P. Let α be a root such that α = −β1 − · · · − βn − βn+1 (βj ∈ Γ ).
There exists a permutation σ such that α = −βσ (1) − · · · − βσ (n) − βσ (n+1) and
the partial sums are roots. Set γ = −βσ (1) − · · · − βσ (n) . One has γ ∈ ∆− ∩ P
and −βσ (n+1) ∈ ∆− ∩ P. Since α = γ − βσ (n+1) ∈ ∆− and P is closed, then
α ∈ ∆− ∩ P.
Let us prove now the converse inclusion by induction too. Let α be an
element of ∆− ∩ P, one can write α = −β1 − · · · − βn (βj ∈ Π) If n = 1 then
α = −β1 ∈ P and β1 ∈ Π ∩ (−P) = Γ . Let n be a positive integer, suppose
that if α = −β1 − · · · − βn (βj ∈ Π) and if α ∈ ∆− ∩ P, then α ∈ Γ − . Let α
be an element of ∆− ∩ P such that α = −β1 − · · · − βn − βn+1 (βj ∈ Π).
There exists a permutation σ such that α = −βσ (1) − · · · − βσ (n) − βσ (n+1) and
the partial sums are roots. Set γ = −βσ (1) − · · · − βσ (n) . One has γ ∈ ∆− and
γ = α + βσ (n+1) ; since P is closed then γ ∈ P. We deduce that γ ∈ Γ − .
Furthermore, α − γ = −βσ (n+1) and −γ , α ∈ P, we deduce that βσ (n+1) ∈
(−P) ∩ Π = Γ . ✷
 J. Nervi / Journal of Algebra 253 (2002) 50–99 71

Corollary 2.3.8. The space V0 = h ⊕ ĝΓ  is finite-dimensional.

Proposition 2.3.9. The space Vα̃ is finite-dimensional for all α̃ ∈ Σ.

Proof.
Let α̃ be a root of Σ. We can assume α̃ is a positive root.Recall that

Vα̃ = ρ(γ )=α̃ ĝγ ; let us prove that |ρ −1 ({α̃})| < ∞. Put α̃ = kj =0 mj ωj
(mj ∈ N). Denote by ω0 , . . . , ωq the roots of Π\Γ . One has ρ(ωl ) > 0 for all

l = 0, . . . , q. Set ρ(ωl ) = kj =0 pjl
ωj where pjl are non-negative integers. Let
γ be a root of ∆ such that q ρ(γ ) = α̃,we know that γ ∈ ∆+ (Corollary 2.3.5)
and one can write γ = l=0 nl ωl + βt ∈Γ st βt with nl
0 and st
0. One
has
 k   q 
q 
q   k 
ρ(γ ) = nl ρ(ωl ) = nl pj
l
ωj = nl pj
l
ωj
l=0 l=0 j =0 j =0 l=0


k
= mj
ωj .
j =0

Hence one has for any j :


q
nl pjl = mj .
l=0

The set of the q-uplets (n1 , . . . , nq ) which are solutions of the preceding system
of equations is finite. And for each q-uplet (n1 , . . . , nq ), the number of the roots
in ∆ whose coefficients relative to the simple roots in a non-empty subset of Π
(here Π\Γ ) are fixed, is finite. One deduces that |ρ −1 ({α̃})| < ∞. ✷

Recall that since ρ is a surjection from ∆ ∪ {0} onto Σ ∪ {0}, the set Γ =
{α ∈ Π: ρ(α) = 0} is a proper subset (possibly empty) of Π . Hence we can define
as in Section 1 a Z-gradation in ĝ relating Γ . One has

d0 = h ⊕ ĝΓ  = V0 , d1 = ĝγ .
γ =ω (mod Γ )
ω∈Π\Γ

Proposition 2.3.10. Every root in B has a preimage under ρ in Π\Γ and

any preimage α in ∆ of a root of B can be written α = ω (mod Γ ) with
ω ∈ Π\Γ .

Proof. Let ω ∈ B. Let γbe a preimage of ω in ∆, one has γ ∈ ∆+

q
(Corollary 2.3.5). Put γ = l=0 nl ωl + β where β is a linear combination of
elements of Γ . If we denote the height of any α̃ ∈ Σ relative to the basis B by
72 J. Nervi / Journal of Algebra 253 (2002) 50–99

hB (α̃), one has

 
q
  
hB ρ(γ ) = nl hB ρ(ωl ) .
l=0

Since ρ(ωl ) > 0 for all l, one has


  nl0 = 1, hB (ρ(ωl0 )) = 1,
hB ρ(γ ) = 1 ⇒ ∃!l0 :
n = l0 ⇒ nl = 0.
Hence γ = ωl0 (mod Γ ) and ρ(ωl0 ) =
ω. We can see that the root ωl0 depends on
the choice of the preimage γ . ✷

k
Corollary 2.3.11. One has card(Π\Γ )
card(B)
2 and j =0 V
ωj ⊂ d1 .

Proposition 2.3.12. The map ρ is a surjection from Π\Γ onto B.

Proof. From Proposition 2.3.10 it is enough to prove that for any ω ∈ Π\Γ one
has ρ(ω) ∈ B. Suppose that there exists ω ∈ Π\Γ such that ρ(ω) ∈ Σ+ \B. One
can write
ρ(ω) =
ωj1 + · · · +
ωjp , ()
where
ωl ∈ B and p
2. We can also assume that the partial sums are roots.
 Suppose first that ρ(ω) is a real root. Let (Xρ(ω) , ρ(ω)∨ , Yρ(ω) ) be the
corresponding sl2 -triplet. Since ĝ is a-graded, any Z ∈ Vρ(ω) verifies
     
2Z = ρ(ω)∨ , Z = [Xρ(ω) , Yρ(ω) ], Z = Xρ(ω) , [Yρ(ω) , Z] .
We deduce that Vρ(ω) = [V0 , Xρ(ω) ]. From () we deduce that
 
Xρ(ω) = X ωjp . . . [X ωj1 ] . . .
ωj2 , X

ωjm ∈ a m ⊂ V ωjm ⊂ d1 (Corollary 2.3.11). Hence we have Xρ(ω) ∈ dp
with X ωj

(p
2). Furthermore,
ĝω ⊂ [V0 , Xρ(ω) ] ⊂ [d0 , dp ] ⊂ dp .
This is a contradiction to the fact that we have ĝω ⊂ d1 . Hence if ω ∈ Π\Γ and
if ρ(ω) is real, then ρ(ω) ∈ B.
 Suppose now that ρ(ω) is an imaginary root. Since |B|
2, there exists
at least two simple roots in Π\Γ whose restrictions to ha are simple roots of B;
hence one has |(Π\Γ )|
3. Let ωl0 be one of these roots. We set ρ(ωl0 ) = ω ∈ B.
It is possible to choose ωl0 and
ω such that there exists a sequence (α1 , . . . , αs )
in Γ (possibly empty) such that the sum ξ = ω + α1 + · · · + αs + ωl0 is a positive
root of ∆. Since |(Π\Γ )|
3, the support of ξ is a proper subset of Π , hence ξ is
a real root of ∆. One has ρ(ξ ) = p(ω) +
ω ∈ Σ+ and since ρ(ω) is an imaginary
 J. Nervi / Journal of Algebra 253 (2002) 50–99 73

root, ρ(ξ ) is real. Since ρ(ω) is imaginary, supp(ρ(ω)) = B and ρ(ξ ) is the sum
of p simple roots (p
3). As in the first case, we can write
   
ĝω = ĝξ , ĝ−α1 −···−αs −ωl0 ⊂ [Vρ(ξ ), V− ω ⊂ dp−1 .
ω ] = [V0 , Xρ(ξ ) ], V−

Since p − 1
2, we conclude as in the first case. ✷

Let ω ∈

Π\Γ . Let Πω be the connected component of Γ ∪ {ω} containing ω.
Let gω = p∈Z dp (ω) be the Z-gradation of the corresponding finite-dimensional
subalgebra gω corresponding to Πω (see Section 1.1).

Corollary 2.3.13. Let of B. Let ω1 , . . . , ωq be the corresponding

ω an element

q
preimages of
ω in Π\Γ . Then V ω = l=1 d1 (ωl ). Moreover, if B = { ω0 , . . . ,
ωk }

k
then d1 = j =0 V ωj .

Corollary 2.3.14. The algebra ĝ and the grading algebra a have the same one-
dimensional center CK.

Proof. Let us denote by Ka the standard generator of the one-dimensional center

of the affine algebra a. Ka is an element of the orthogonal of 
ω0 , . . . ,
ωk  in ha .
For any α ∈ Γ , one has α(Ka ) = ρ(α)(Ka ) = 0 and for any ωj ∈ Π\Γ , one has
ωj (Ka ) = ρ(ωj )(Ka ) = 0. Hence Ka ∈ CK. ✷

2.4. The irreducible case

In this section we prove that under a condition of irreducibility for some

representations attached to the gradation, the subset Γ defined in the preceding
gives rise to a C-admissible subalgebra which is exactly the grading subalgebra.
Recall that B = { ω0 , . . . ,
ωk } is a fixed basis of Σ. The proof of the following
proposition is straightforward.

Proposition 2.4.1. The following conditions are equivalent:

(i) The map ρ is a bijection from Π\Γ onto B.

(ii) ha = hΓ .
ωj ) is irreducible for any j = 0, . . . , k.
(iii) The representation (V0 , V

Proposition 2.4.2. If the representation (V0 , V ωj ) is irreducible for any j =

0, . . . , k then the pair Π\Γ is C-admissible and the grading subalgebra a is the
associated C-admissible subalgebra.

Proof. Let
ωj ∈ B. Denote by ωj its unique preimage in Π\Γ , one has
   
ωj∨ ∈ a
ωj −
, a ωj ⊂ [V ωj ] ⊂ d1 (ωj ), d−1 (ωj ) ⊂ d0 (ωj ).
ωj ,V−
74 J. Nervi / Journal of Algebra 253 (2002) 50–99

One deduces that ωj∨ ∈ h(Πωj ). Furthermore ωj ( ωj∨ ) = ωj∨ ) = 2 and

ωj (
α( ωj ) = 0 for all α ∈ Πωj \{ωj } because ρ(α) = 0 by definition of Γ . Hence
∨

ωj∨ = Hj the grading element of the connected component of S(A) relative to ωj .

For any j the triple (X
ωj , Hj , Y
ωj ) is a natural sl2 -triple. Hence the pair (Π, Γ )
is admissible.
Suppose that there exist j ∈ {0, . . . , k} and p
2 such that dp (ωj ) = {0}.
Then there exists a root γ of (gj , hj ) which can be written γ = pωj (mod Γj ).
One has ρ(γ ) = p ωj ∈ Σ+ . Since ωj is not an imaginary root, then it is
not possible and the pair (Π, Γ ) is C-admissible. The C-admissible subalgebra
generated by hΓ = ha and the elements X ωj , Y
ωj is obviously the affine
algebra a. ✷

We can summarize the preceding results as follows.

Theorem 2.4.3. Let ĝ be an affine algebra and Σ be an affine root system. Denote
a basis of Σ by B = { ω0 , . . . ,
ωk }. The following conditions are equivalent:

(i) Σ grades ĝ and the representations (V0 , V ωj ) are irreducible.

(ii) There exist a root system ∆ of ĝ, a basis Π of ∆, and a subset Γ of Π such
that |Π\Γ |
2, the pair (Π, Γ ) is C-admissible and the corresponding C-
admissible subalgebra whose root system is Σ is the grading subalgebra.

In the following we give an example of a Σ-gradation such that one at least

among the representations (V0 , V
ωj ) is not irreducible.

Example 2.4.4. Consider the affine algebra of type A(1)

3 ,

 
ĝ = C t, t −1 ⊕ sl4 (C) ⊕ CK ⊕ Cd,

where C[t, t −1 ] is the algebra of Laurent polynomials in t and sl4 (C) the
simple Lie algebra of complex matrices of trace zero. (For details about concrete
constructions of this affine algebra see [11, Chapter 7].) In the underlying
subalgebra g̊ = sl4 (C), consider h̊ the Cartan subalgebra of the diagonal matrices.
Denote by (α1 , α2 , α3 ) the standard basis of ∆˚ = (g̊, h̊). The subalgebra h =
h̊ ⊕ CK ⊕ Cd is a Cartan subalgebra of ĝ. Let ∆ be the root system of the pair
(ĝ, h). Let δ be the root of ∆ defined by δ|h̊ ⊕ CK = 0, δ, d = 1.
Put α0 = δ − α1 − α2 − α3 = δ − θ , the sequence (α0 , α1 , α2 , α3 ) is a basis
of ∆ and the coroot of α0 is α0∨ = K − θ ∨ . Denote by (Xαi , αi∨ , Yαi )i=0,...,3 the
standard sl2 -triples. Consider the subalgebra å of sl4 (C) generated by the two sl2 -
 J. Nervi / Journal of Algebra 253 (2002) 50–99 75

triples (Xα1 + Xα3 , α1∨ + α3∨ , Yα1 + Yα3 ) and (Xα2 , α2∨ , Yα2 ). The algebra å is the
simple subalgebra (of type C2 ) of the matrices
 
a c −d e
 −a + f
 c b d 
  .
−d b a−f c
e  d  c  −a
Denote by hå the Cartan subalgebra generated by α1∨ + α3∨ and α2∨ . Put ω1 =
α1 |hå = α3 |hå and ω2 = α2 |hå ; (
ω1 ,
ω2 ) is a basis of Σ̊ = (å, hå ). (For details
about these calculations see [5, Example 5.5].) Put a = C[t, t −1 ] ⊕ å ⊕ CK ⊕ Cd,
then a is an affine subalgebra of ĝ of type C2(1) and ha = hå ⊕ CK ⊕ Cd is a
Cartan subalgebra of a. Denote by Σ the root system of (a, ha ) and define δa
in ha by the relations δa |h̊å ⊕ CK = 0, δa , d = 1. Put ω0 = δa − (2 ω1 + ω2 );
ω1 ,
( ω1 ,
ω2 ) is a basis of Σ. Define as precedently ρ : h → ha , α → α|ha . We
have ρ(α1 ) = ρ(α3 ) = ω1 , ρ(α2 ) =
ω2 , ρ(δ) = δa , ρ(α0 ) =
ω0 . It is easy to verify
that ρ(∆) = Σ. The algebra ĝ is Σ-graded and for any α̃ ∈ Σ ∪ {0}, we have

Vα̃ = ĝγ .
γ ∈∆∪{0}
ρ(γ )=α̃

Furthermore, the set Γ = {α ∈ ∆: ρ(α) = 0} is empty; hence the C-admissible

subalgebra associated to the C-admissible pair (Π, ∅) is ĝ itself and a is a proper
subalgebra of ĝ. One has V ω1 = d1 (α1 ) ⊕ d1 (α3 ) and the representation (V0 , V
ω1 )
is not irreducible. We can illustrate this example by the following diagrams:

A(1) (1)
3 graded by C2

In this example, as in the sequel of the paper, we have joined by an arrow the
simple roots having the same restriction.

2.5. The general case

Let ĝ be an affine algebra graded by an affine root system Σ. The notations are
the same as precedently. The following definition is useful.

Definition 2.5.1. The two different roots α and β of Π\Γ are said Γ -connected
if (α, β) = 0 or if there exists in S(A) a string whose extremities are α and β and
whose vertices (except α and β) are all in Γ .

Proposition 2.5.2. Let α and β in Π\Γ be two roots which are Γ -connected, one
has ρ(α) = ρ(β) and (ρ(α), ρ(β)) = 0.
76 J. Nervi / Journal of Algebra 253 (2002) 50–99

Proof. There exists a sequence, possibly empty, (α1 , . . . , αl ) in Γ such that

γ = α + α1 + · · · + αl + β is a root of ∆; hence ρ(γ ) = ρ(α) + ρ(β) is a root
of Σ. Since ρ(α) ∈ B, then 2ρ(α) is not a root and the result follows. ✷

Proposition 2.5.3. Let ω an element of B. Denote the preimages of

ω in Π\Γ by
ω1 , . . . , ωq . One has
ω ∨ = H1 + · · · + Hq ,

where Hj is the semisimple element in h(Πωj ) that grades the simple Lie
algebra gj associated to the simple root ωj .

Proof. One has

 
  
q 
q 
q
∨ −

ω ∈ a ,a ω ω
⊂ [V ω] =
ω , V− d1 (ωl ), d−1 (ωl ) ⊂ d0 (ωl )
l=1 l=1 l=1

and the last sum is direct since, from Proposition 2.5.2, the roots ωl are not Γ -
connected in Π\Γ . Hence the decomposition ω∨ = h1 + · · · + hq is unique. One
ω ) = α(hl ) = 0 for all α ∈ Γl , and ωl (
has α( ∨
ω∨ ) = ωl (hl ) = ω( ω∨ ) = 2. We
deduce that hl = Hl for all l = 1, . . . , q. ✷

Corollary 2.5.4. If the set Γ is empty then

ω∨ = ω1∨ + · · · + ωq∨ ,

where ωl∨ is the coroot in h of the simple root ωl .

Theorem 2.5.5. The pair (Π, Γ ) is C-admissible and one can choose the C-
admissible subalgebra ĝΓ such that ĝΓ contains the grading subalgebra a.

ω be an element of B and ω1 , . . . ωq its preimages in Π \Γ . Since

Proof. Let

q 
q
a
ω
⊂ V
ω= d1 (ωl ) and a−
ω
⊂ V−
ω= d−1 (ωl ),
l=1 l=1

we can write
ω = X1 + · · · + Xq
X ω = Y1 + · · · + Yq ,
and Y
where Xl ∈ d1 (ωl ) and Yl ∈ d−1 (ωl ). One has, from Proposition 2.5.3,

q 
q
ω∨ = [X
ω] =
ω , Y [Xl , Yl ] = Hl .
l=1 l=1

The decomposition is unique, so we have [Xl , Yl ] = Hl for all l and the space
(d0 (ωl ), d1 (ωl )) is regular (Section 1.1). Since ρ is a surjection from Π\Γ
 J. Nervi / Journal of Algebra 253 (2002) 50–99 77

onto B, it is clear that the same is true for any j = 0, . . . , card(Π\Γ ); hence
the pair (Π, Γ ) is admissible.
Suppose now that there exist j ∈ {0, . . . , card(Π\Γ )} and p
2 such that
dp (ωj ) = {0}. This means that there exists a root σ ∈ ∆+ that verifies σ =
pωj (mod Γ ), hence ρ(σ ) = pρ(ωj ) ∈ Σ+ . This is impossible since ρ(ωj ) is a
real root of Σ. The pair (Π, Γ ) is therefore C-admissible. Consider now the affine
subalgebra ĝΓ generated by hΓ = {h ∈ h: α(h) = 0 ∀α ∈ Γ } and the sl2 -triples
(Xj , Hj , Yj ). The algebra ĝΓ contains ha and the generators X ω , Y ω ∈ B).
ω (
Hence ĝΓ contains a. ✷

Proposition 2.5.6. The root system Σ grades ĝΓ and the corresponding grading
subalgebra is a. Moreover, if we apply the preceding procedure to this grading,
the associated C-admissible subalgebra is ĝΓ itself.

Proof. One has

a ⊂ ĝΓ ⊂ ĝ and ha ⊂ hΓ ⊂ h.
Let RΓ be the root system of the pair (ĝΓ , hΓ ) and put

ρ : h −→ hΓ
α −→ α|hΓ = α.
One knows that RΓ = ∆\{0} and Π\Γ is a basis of RΓ (Theorem 1.5.9).
Consider

ρΓ : hΓ −→ ha
α −→ α|ha .
One has ρ = ρΓ ◦ ρ. Since ρ is a surjection from ∆+ ∪ {0} onto Σ+ ∪ {0} and
since Π\Γ is a basis of RΓ , one has ρΓ−1 ({0}) ∩ R+ Γ = ∅ and ρΓ is clearly a


surjection from R+ onto Σ + . Put V α̃ = ĝ α for any α̃ ∈ Σ ∪ {0}. Note
Γ
ρΓ (α)=α̃ Γ
that V0 = ĝ0Γ = hΓ and ĝΓ = α̃∈Σ∪{0} Vα̃ . Hence ĝΓ is Σ-graded. In this new
gradation we can construct a C-admissible subalgebra of ĝΓ that contains a by
the procedure of Theorem 2.5.5. Since the set {α ∈ Π\Γ : ρΓ (α) = 0} is empty,
then this subalgebra is ĝΓ itself. ✷

Definition 2.5.7. If the C-admissible subalgebra of ĝ constructed by the procedure

of Theorem 2.5.5 is the algebra ĝ itself (i.e. the set Γ is empty), then the Σ-
gradation is said to be maximal in ĝ.

Proposition 2.5.8. Let ĝ be an affine algebra graded by a root system Σ1 which

is associated to a grading subalgebra ĝ1 . Let Σ2 be a grading affine root system
in ĝ1 . Then Σ2 grades ĝ.
78 J. Nervi / Journal of Algebra 253 (2002) 50–99

Proof. Denote by ĝ2 the grading subalgebra in ĝ1 which is associated to Σ2 . One
has
ĝ2 ⊂ ĝ1 ⊂ ĝ.
Let h2 be a Cartan subalgebra of ĝ2 , by applying the conjugacy theorem of Cartan
subalgebras [14, Theorem 2] one can find a Cartan subalgebra h1 (respectively h)
of ĝ1 (respectively ĝ) such that
h2 ⊂ h1 ⊂ h.
In the corresponding root systems ∆, Σ1 , Σ2 , one can choose basis Π, Π1 , Π2
such that

ρ : h −→ h1
α −→ α|h1 = α
is a surjection from ∆+ ∪ {0} onto Σ1+ ∪ {0} (Corollary 2.3.5) and

ρ̃ : h1 −→ h2
α −→ α|h2 = α̃
is a surjection from Σ1+ ∪ {0} onto Σ2+ ∪ {0} (Corollary

2.3.5). The spaces of the
Σ1 -gradation are denoted by Vα and one has ĝ = α∈Σ1 ∪{0} Vα . For any α̃ ∈ Σ2 ,
let us set:

Uα̃ = Vα .
α|h2 =α̃

It is easy to prove that ĝ is Σ2 -graded and that the spaces of the gradation are the
Uα̃ ’s, α̃ ∈ Σ2 ∪ {0}. ✷

Remark 2.5.9. If Σ2 is a maximal grading root system in ĝ1 , then U0 = V0 =

h ⊕ ĝΓ  where Γ = {α ∈ Π: α|h1 = 0}.

The following theorem is now straightforward.

Theorem 2.5.10. Let ĝ an affine algebra and Σ be an affine root system. The two
following conditions are equivalent:

(i) ĝ is Σ-graded.
(ii) There exist a root system ∆ of ĝ, a basis Π of ∆, a subset Γ of Π such that
– the pair (Π, Γ ) is C-admissible;
– the C-admissible subalgebra ĝΓ , associated to the pair (Π, Γ ) is graded
by Σ and Σ is maximal in ĝΓ .
 J. Nervi / Journal of Algebra 253 (2002) 50–99 79

It is now clear that the study of the Σ-gradations of the affine algebras
can be reduced to the study of the maximal Σ-gradations of its C-admissible
subalgebras. Recall that maximality is equivalent to say that the subset Γ of Π
precedently defined is empty.
We can assume without loss of generality that the homomorphism ρ is not
a bijection from Π onto B = { ω0 , . . . ,
ωk } the basis of Σ that we have fixed once
for all. Recall that it is equivalent to say that the representations (V0 , V ωj ) are
not all irreducible (Proposition 2.4.1). Such maximal gradations will be called
non-trivial (abbreviated n.t.m.).
For the trivial maximal gradations the grading subalgebra is the algebra ĝ itself
(Theorem 2.4.3).

2.6. Investigation of the possible non-trivial maximal gradations

In this section we assume that the gradation of the affine algebra ĝ by the root
system
 Σ is maximal non-trivial. Let Π i = ρ −1 ({
ωi }) ∩ Π and set H i = ωi∨ =
γ ∈Π i γ for all i = 0, . . . , k.
∨

The following lemma plays a crucial role for the classification.

Lemma 2.6.1. Let α and β be two different roots in Π i ; one has

(a) (α, β) = 0;
(b) α(H j ) = β(H j ) for all j = 0, . . . , k;
(c) let γ be a root in Π\{α} such that (α, γ ) = 0, then γ ∈
/ Π i and there exists
a simple root µ such that ρ(µ) = ρ(γ ) and (µ, β) = 0. If (γ , β) = 0 then
µ = γ .

Proof. (a) Suppose that (α, β) = 0 then α + β is a root of ∆, hence ρ(α + β) =

ωi is a root of Σ. This is obviously not true since
2 ωi is a real root.
(b) Since H j ∈ ha , one has α(H j ) = β(H j ) =
ωi (H j ).
(c) Let γ be a root in Π\{α} such that (α, γ ) = 0, then (a) implies that
γ ∈/ Πi . Let us set ρ(γ ) = ωj where j = i. Since Γ = ∅, one has H j =
γ + ξ ∈Π j \{γ } ξ (Corollary 2.5.4) and d1 (α) = ĝα for α ∈ Π . One has by
∨ ∨

Theorem 2.5.5:

X ωi = Xα + Xβ + Xλ ,
λ∈Π i \{α,β}
 j     j
ωi =
H , X ωi = α H X
ωi H j X ωi ,
   
 j 
H , X
ωi = γ∨ + ξ ∨ , Xα + Xβ + Xλ
ξ ∈Π j \{γ } λ∈Π i \{α,β}
80 J. Nervi / Journal of Algebra 253 (2002) 50–99


= α(γ ∨ )Xα + β(γ ∨ )Xβ + α(ξ ∨ )Xα
ξ ∈Π j \{γ }
 
+ β(ξ ∨ )Xβ + λ(ξ ∨ )Xλ .
ξ ∈Π j \{γ } ξ,λ

Note that α(γ ∨ ) < 0 and α(ξ ∨ )   0 for all ξ ∈ Π j . Hence the coefficient of Xα
in the previous sum (i.e. α(γ ) + ξ ∈Π j \{γ } α(ξ ∨ )) is not zero.
∨

If β(γ ∨ ) = 0 (i.e. (β, γ ) = 0) then we can set µ = γ .

If β(γ ∨ ) = 0 (i.e. (β, γ ) = 0) then ad(H j ) acts diagonally on X ωj and there
exists ξ ∈ Π j such that β(ξ ∨ ) = 0. Then we can set µ = ξ . ✷

In the following we give the list of the affine algebras which are not graded by
an n.t.m. root system. For the others we give the possible n.t.m. Σ-gradations. We
give too some representative proofs.

(1) (1) (2) (3)

Proposition 2.6.2. The algebras A1 , A2 , A2 , and D4 , do not admit any
n.t.m. Σ-gradation.

Proof. The result is straightforward for the three first algebras by applying
Lemma 2.6.1(a). For D4(3) , note that the only possibility can be illustrated by the
following arrow diagram:

Put Π 0 = {α0 , α2 } and Π 1 = {α1 }. One has α0 (H 1 ) = α0 (α1∨ ) = −1 and

α2 (H 1 ) = −4. This is not possible by Lemma 2.6.1(b). ✷

Lemma 2.6.3. If the algebra ĝ is of one of types F4(1) , E6(2) , A(2) (2) (1)
2l , Dl+1 , or Cl ,
−1 −1
and if ρ ({ρ(α0 )}) ∩ Π = {α0 } then for all i = 0, . . . , n, ρ ({ρ(αi )}) ∩ Π =
{αi } and the gradation is trivial.

Proof. Note that the Dynkin diagrams of such algebras are linear (i.e. non-
circular) and without branching point. Let j0 the smallest index
1 such
that |ρ −1 ({ρ(αj0 )}) ∩ Π|
2. Denote by αj0 +k a root in this set. One has
k
2 (Lemma 2.6.1(a)) and ρ(αj0 −1 ) = αj0 +k−1 (or αj0 +k+1 if this root exists)
(Lemma 2.6.1(c)). It follows that |ρ −1 ({ρ(αj0 −1 )}) ∩ Π|
2 and this is not
possible from to the choice of j0 . ✷

(2) (2) (1)

Proposition 2.6.4. The algebras E6 , A2l , E8 do not admit any n.t.m. Σ-
gradation.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 81

Proof. Suppose that ĝ (of type E6(2) ) admits an n.t.m. Σ-gradation, then one has
|ρ −1 ({ρ(α0 )}) ∩ Π|
2. If ρ(α0 ) = ρ(α2 ), by applying Lemma 2.6.1(c) one has
ρ(α1 ) = ρ(α3 ) and ρ(α4 ) = ρ(α2 ). Hence Π 0 = {α0 , α2 , α4 }, Π 1 = {α1 , α3 },
α0 (H 1 ) = α0 (α1∨ ) = −1 and α2 (H 1 ) = α2 (α1∨ + α3∨ ) = −2; this is not possible
by Lemma 2.6.1(b). This case is illustrated by the following diagram:

(2)
The other two possibilities for E6 (illustrated by the following diagrams) are
treated similarly.

(2)
The case A2l . The case l = 1 has been studied in Proposition 2.6.2. The case
l = 2 is straightforward by Lemma 2.6.1(b).
Let us assume l
3 and choose notations such that α0 ∈ Π 0 and α1 ∈ Π 1 .
Denote by j0 the smallest non-zero integer such that ρ(α0 ) = ρ(αj0 ). We clearly
have j0
2. If j0 < l then ρ(α1 ) = ρ(αj0 −1 ) = ρ(αj0 +1 ) by Lemma 2.6.1(c), one
has α0 (H 1 ) = −1 and αj0 (H 1 ) = −2. This is not possible by Lemma 2.6.1(b)
then j0 = l. It follows that ρ(α1 ) = ρ(αl−1 ). This is not possible by Lemma
2.6.1(b) and the fact that α1 (H 0 ) = −2 and αl−1 (H 0 ) = −1. The following
diagram illustrates this case:

(1)
The case E8 is straightforward. ✷

Proposition 2.6.5.
– If the algebra G(1)2 admits an n.t.m. Σ-gradation then Σ is necessarily of
(2)
type A2 .
(1)
– If the algebra C2 admits an n.t.m. Σ-gradation then Σ is necessarily of
type A(1)
1 .
– If the algebra F4(1) admits an n.t.m. Σ-gradation then Σ is of type D4(3) .
(2)
– If l + 1 is even then the algebra Dl+1 does not admit any n.t.m. Σ-gradation.
(2)
– If l + 1 is odd and if the algebra Dl+1 admits an n.t.m. Σ-gradation then Σ
(2)
is of type A2p with p = l/2.
82 J. Nervi / Journal of Algebra 253 (2002) 50–99

– If the algebra Cl(1) admits an n.t.m. Σ-gradation then Σ is of type A(1) 1

or Cp(1) with l = 2pm.
(1) (2)
– If the algebra E7 admits an n.t.m. Σ-gradation then Σ is of type E6 .
– If the algebra E6(1) admits an n.t.m. Σ-gradation then Σ is of type D4(3) .
(1) (2)
– If the algebra Bl admits an n.t.m. Σ-gradation then Σ is of type Dl
(2)
or A2p with l = 2p + 1.
– If the algebra Dl(1) admits an n.t.m. Σ-gradation then Σ is of type Bl−1
(1)

or A(2) (2)
2p+1 or A2p with l = 2p + 2 in these two last cases.
(2) (2)
– If the algebra A2l−1 admits an n.t.m. Σ-gradation then Σ is of type A2l−2 .

Proof. For G(1)

2 the only possibility is illustrated by the following diagram:

with α1 (H 0 ) = α1 (α0∨ + α2∨ ) = −4 and α0 (H 1 ) = α2 (H 1 ) = −1. Hence the

generalized Cartan matrix of Σ is of type A(2)2 .
For the case C2(1) the proof is similar.
(1)
For F4 , the only possibility is illustrated by this diagram and the calculus of
(3)
the
ωj (H i )’s gives a generalized Cartan matrix of type D4 :

(2)
The case Dl+1 . Note that l
2. By Lemma 2.6.1(b), we prove that if |Π 0 |
2
then Π 0 = {α0 , αl }. We prove then by induction that for all j = 0, . . . , [l/2], one
has ρ −1 ({ρ(αj )}) = {αj , αl−j }. If l + 1 = 2p then ρ(αp ) = ρ(αp−1 ). This is
impossible by Lemma 2.6.1(a). The case l + 1 = 2p + 1 is illustrated by the
following diagram:

The computation of the ωj (H i )’s gives the type of the Cartan matrix of the
possible grading root system: A(2)
2p . ✷

For the other cases the proofs are very similar and use essentially Lemma 2.6.1.
(1)
In the last proposition we give the result for the circular case Al .
 J. Nervi / Journal of Algebra 253 (2002) 50–99 83

Remark 2.6.6. Because of the circularity of the diagram, one can replace α0 with
any other simple root and change the indices. The result is the same.

Proposition 2.6.7.
– If algebra Al (l
3) admits an n.t.m. Σ-gradation and ρ −1 ({ρ(α0 )}) ∩
(1)

Π = {α0 } then for all j ∈ {0, . . . , l} one has ρ −1 ({ρ(αj )})∩Π = {αj , αl+1−j }
(one sets αl+1 = α0 ). The integer l is odd and if we set l = 2p + 1, then Σ is
(1)
of type Cp+1 .
(1)
– If algebra Al admits an n.t.m. Σ-gradation and satisfies the hypothesis:
() it does not exist any simple root αj such that ρ −1 ({ρ(αj )}) ∩ Π = {αj }
then Σ is of type A(1) (1)
p with l = mp − 1 or Cp with l = 2mp + 1.

All the preceding propositions are summarized in Appendix B.

2.7. The effective non-trivial maximal gradations

We prove in this section that the possible non-trivial maximal gradations given
in Section 2.6 are effective gradations.
Let ĝ be an affine algebra. Let h be a Cartan subalgebra and let ∆ be the
root system of the pair (ĝ, h). We use the standard notations from the preceding
sections. We take now one of the diagrams obtained in Section 2.6 (listed in
Appendix B) and consider
 the corresponding partition of a basis Π of ∆. That it
is to say that Π = ki=0 Π i where Π i is either a simple root or a set of orthogonal
simple roots connected together by arrows. Let us set α0 ∈ Π 0 . Put
  
Hi = α∨ , Xi = Xα , Yi = Yα ,
α∈Π i α∈Π i α∈Π i

where (Xα , α ∨ , Yα ) is a sl2 -triple. The elements H 0 , . . . , H k are linearly

independent. For any i and any α ∈ Π i , one has α(H i ) = 2 and (Xi , H i , Yi ) is
a sl2 -triple. Since the subset Π i verifies Lemma 2.6.1, one has α(H j ) = β(H j )
for all α, β in Π i and for any j .
The matrix M = (mij )i,j =0,...,k where mij = α(H i ) for any α ∈ Π j is a gene-
ralized Cartan matrix of affine type (cf. Section 2.6). Set M̊ = (mij )i,j =1,...,k the
associated invertible matrix.

Denote by K  the element of h defined by K  = kj =0 aj∨ H j where aj∨ are
relatively prime integers such that tM t (a0∨ , . . . , ak∨ ) = 0 [11, 6.1, 6.2, p. 79–80].
The following lemma is straightforward.
84 J. Nervi / Journal of Algebra 253 (2002) 50–99

Lemma 2.7.1. One has α(K  ) = 0 for all α ∈ Π and K  is an element of the
one-dimensional center CK of ĝ.

Remark 2.7.2. Since a0∨ = 1 [11, p. 79], the elements K  , H 1 , . . . , H k are linearly
independent.

Since Å and M̊ are invertible, the following proposition is easy to prove.

Proposition 2.7.3. There exists in h an element d  unique modulo K such that

α(d  ) = 1 for all α ∈ Π 0 and α(d  ) = 0 for all α ∈ Π\Π 0 . The elements
d  , K  , H 1 , . . . , H k are linearly independent.

Consider now the subalgebra c of h generated by d  , H 0 , . . . , H k (or by

d  , K  , H 1 , . . . , H k ).
Denote by ωj the restriction to c of any root in Π j .

Lemma 2.7.4. The elements (ω0 , . . . , ωk ) of c are linearly independent.

k
Proof. The sequence of coefficients (x0 , . . . , xk ) such that j =0 xj ωj = 0 is
solution of the linear system

k 
k
 
xj ωj (d  ) = 0, xj ωj H i = 0 ∀i = 0, . . . , k. ✷
j =0 j =0

The following proposition is now straightforward.

Proposition 2.7.5. The triple (c(ωj )j =0,...,k , (H j )j =0,...,k ) is a realization of the

generalized Cartan matrix M.

Proposition 2.7.6. The subalgebra a, generated by c and the elements Xj , Yj , is

an affine subalgebra whose Cartan matrix is M.

Proof. It is enough to prove that Serre’s relations are satisfied (i.e.

(ad Yi )1−mij (Yj ) = 0 and (ad Xi )1−mij (Xj ) = 0
for i = j ). Let i, j be two different integers.
If the partition of Π contains more than two subsets (i.e. the order of M
3)
then we can consider the finite-dimensional semisimple algebra, l, generated
by the sl2 -triples (Xα , α ∨ , Yα ) for α ∈ Π i and (Xβ , β ∨ , Yβ ) for β ∈ Π j . The
Lie algebra l is a sl2 -module under the action of (Xi , H i , Yi ). The element Yj
is primitive of weight −mij , it generates an irreducible (1 − mij )-dimensional
submodule and one has ad(Yi )1−mij (Yj ) = 0 [13]. The second equality is proved
similarly.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 85

If the partition contains exactly two subsets (i.e. Π = Π 0 ∪ Π 1 ), the previous

argument is not correct. Let us prove that (ad Y0 )1−m01 (Y1 ) = 0. The Lie algebra ĝ
is a sl2 -module under the action of (X0 , H 0 , Y0 ). The element Y1 is primitive
of weight −m01 . One has (ad Y0 )j (Y1 ) = β∈Π 1 (ad Y0 )j (Yβ ) for any integer j .

Each term of this sum is in α1 ,...,αj ∈Π 0 ĝβ+α1 +···+αj . Since Π 0 ∩ Π 1 = ∅, one
has (ad Y0 )j (Yβ ) = 0 for β fixed in Π 1 and j big enough.
Put N = Inf{j, (ad Y0 )j (Y1 ) = 0}, one has N
1. Put vj = 1 j
j ! (ad Y0 ) (Y1 ),
j
0. The following relations are verified:

(1) (ad Y0 )(vj ) = (j + 1)vj +1 ,

(2) (ad X0 )(vj ) = (−m01 − j + 1)vj −1 .

The integer N is such that vN = 0 and vN−1 = 0. From (1), one deduces that
vj = 0 for all j
N and from (2), that −m01 − N + 1 = 0 therefore N = 1 − m01 .
And the result follows.
The others equalities are proved in the same way. ✷

Denote by Σ the root system of the pair (a, c). The type of Σ is given by the
matrix M.

Proposition 2.7.7. The algebra ĝ is an integrable a-module.

Proof.
The algebra ĝ is a c-diagonalizable a-module and the weight-spaces are

Vω = ρ(γ )=ω ĝγ for ω ∈ c. Let us prove that Xj , Yj are locally nilpotent for
any j . It is enough to prove they are locally nilpotent on the generators of ĝ
[11, Lemmas
 a fixed integer in {0, . . . , k}. Recall that
3.4. and 3.5]. Let j be
Xj = α∈Π j Xα (respectively Yj = α∈Π j Yα ), where Xα , Yα are associated to
the simple root α. Let h ∈ h. One has
   
(ad Xj ) (h) =
2
−β(h)[Xα , Xβ ] = − β(h)[Xα , Xβ ].
α∈Π j β∈Π j α,β∈Π j

As roots of Π j are orthogonal, we have (ad Xj )2 (h) = 0.

Let i0 = 0, . . . , n. If αi0 ∈ Π j , one has clearly (ad Xj )(Xαi0 ) = 0. If αi0 ∈
/ Πj ,

for any integer N one has (ad Xj )N (Xαi0 ) ∈ ĝαi0 +β where β is the sum of N
roots of Π j , all different from αi0 . For N big enough, αi0 + β is not a root of ∆;
hence (ad Xj )N (Xαi0 ) = 0.
Let us prove now that (ad Xj )3 (Yαi0 ) = 0.

If αi0 ∈
/ Π j then (ad Xj )(Yαi0 ) = α∈Π j [Xα , Yαi0 ] = 0.
86 J. Nervi / Journal of Algebra 253 (2002) 50–99


If αi0 ∈ Π j then (ad Xj )(Yαi0 ) = α∈Π j [Xα , Xαi0 ] = αi∨0 . One deduces that
 
(ad Xj )2 (Yαi0 ) = Xα , αi∨0 = −2Xαi0 ;
α∈Π j

hence (ad Xj )3 (Yαi0 ) = 0. ✷

Proposition 2.7.8. The root system Σ of the pair (a, c) is the set ρ(∆). The
algebra ĝ is maximally graded by Σ and the corresponding algebra is a.
Moreover, all non-trivial maximal gradings are obtained this way. The list of the
non-trivial maximal gradings of affine algebras is given in Appendix B.

Proof. We know that Σ ⊂ ρ(∆), let us prove the converse inclusion. Using
Proposition 2.7.7 we show as in Theorem 1.5.8 that for any α ∈ ∆, there exist
an integer n and a root ω in Σ such that ρ(α) = nω. If ω is an imaginary root
then ρ(α) ∈ Σ. If ω is a real root then up to conjugation by an element of the
Weyl group of a, we can assume that ω is a simple root. Assume that n is a
positive integer. This means that α is the sum of n roots of Π whose restrictions
to c are ω. Since such simple roots are orthogonal, this is impossible if n
2.
Hence ρ(α) = ω and ρ(∆) ⊂ Σ.
Let us prove now that ĝ is Σ-graded. Let α ∈ Σ ∪ {0} and set

Vα = ĝγ .
ρ(γ )=α

One has clearly V

0 = h and Vα = {x ∈ ĝ: [h, x] = α(h)x ∀h ∈ c}. Since ρ(∆) =
Σ, we have ĝ = α∈Σ∪{0} Vα . ✷

Let us summarize our main result.

Theorem 2.7.9. Let ĝ be an affine algebra, let h be a Cartan subalgebra and

let Π be a set of simple roots for (ĝ, h). Let also Γ ⊂ Π be a C-admissible subset
and let ĝΓ the corresponding C-admissible subalgebra of ĝ. Let Σ be an affine
root system such that ĝΓ is maximally graded by Σ. Then ĝ is graded by Σ and,
conversely, all gradings of ĝ are obtained this way.

The list of C-admissible subalgebras is given as a subtable of Table A, and the

list of non-trivial maximal gradings is given in Table B.

Appendix A. Admissible subalgebras

In the left column of Table A is the Dynkin diagram of the affine algebra ĝ
and in the right column is the Dynkin diagram of the corresponding admissible
(eventually C-admissible) a.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 87

Table A.1
The algebra ĝ is of type (Aff1)
(1) (1)
A(k+1)(p+1)−1 C-admissible Ak

(1) (2)
Bn C-admissible if p = 1 Dk if k
2

(1)
A1 if k = 1

(1) (2)
Bn C-admissible A2k

(1) (1)
Ckp Ck

(1) (1)
Bn C-admissible Bk

(1) (1)
B3 C-admissible A1
88 J. Nervi / Journal of Algebra 253 (2002) 50–99

Table A.1 (continued)

(1) (1)
Dn C-admissible Bk

(1) (2)
Cn A2k

(1) (2)
Cn Dk

(1) (2)
Dn C-admissible if p = 1 Dk

(1) (2)
Dn A2k

(1) (1)
Dkp C-admissible Ck

(1) (1)
E6 C-admissible A2
 J. Nervi / Journal of Algebra 253 (2002) 50–99 89

Table A.1 (continued)

(1) (2)
E6 A2

(1) (1)
E6 C-admissible A2

(1) (1)
E6 C-admissible G2

(1) (2)
E7 C-admissible A2

(1) (2)
E7 A2

(1) (1)
E7 A1

(1) (2)
E7 A2
90 J. Nervi / Journal of Algebra 253 (2002) 50–99

Table A.1 (continued)

(1) (1)
E7 C-admissible A2

(1) (2)
E7 A2

(1) (2)
E7 A2

(1) (2)
E7 A4

(1) (1)
E7 G2

(1) (1)
E7 C2

(1) (2)
E7 A5

(1) (1)
E7 C-admissible C3
 J. Nervi / Journal of Algebra 253 (2002) 50–99 91

Table A.1 (continued)

(1) (1)
E7 C-admissible F4

(1) (2)
E8 C-admissible A2

(1) (2)
E8 A2

(1) (1)
E8 A1

(1) (2)
E8 A2

(1) (2)
E8 A2

(1) (2)
E8 A2

(1) (1)
E8 A1
92 J. Nervi / Journal of Algebra 253 (2002) 50–99

Table A.1 (continued)

(1) (2)
E8 A2

(1) (2)
E8 C-admissible A4

(1) (1)
E8 C2

(1) (2)
E8 A4

(1) (3)
E8 C-admissible D4

(1) (1)
E8 G2

(1) (1)
E8 C-admissible G2

(1) (2)
E8 C-admissible E6
 J. Nervi / Journal of Algebra 253 (2002) 50–99 93

Table A.1 (continued)

(1) (1)
E8 C-admissible F4

(1) (1)
F4 C-admissible G2

(1) (2)
F4 A2

(1) (2)
F4 A2

(1) (2)
G2 A2

(2) (2)
A2l Dk+1

(2) (2)
A2l A2k

(2) (2)
A2l−1 Dk+1
94 J. Nervi / Journal of Algebra 253 (2002) 50–99

Table A.2
The algebra ĝ is of type (Aff2)

(2) (2)
A2l−1 A2k

(2) (2)
A2l−1 A2k

(2) (1)
A2kp−1 C-admissible Ck

(2) (2)
Dl+1 C-admissible if p = 1 Dk+1

(2) (2)
E6 A2

(2) (2)
E6 A2

(2) (2)
E6 A2
 J. Nervi / Journal of Algebra 253 (2002) 50–99 95

Table A.2 (continued)

(2) (2)
E6 C-admissible A2

(2) (2)
E6 C-admissible A4

Appendix B. Non-trivial maximal gradings

The Dynkin diagram of ĝ takes place in the left column (with arrows joining
roots which have the same restriction) while the right column contains the Dynkin
diagram of the maximal grading subalgebra a.

Table B
(1) (2)
G2 A2 ρ(δ) = 2δa

(1) (3)
F4 D4 ρ(δ) = 3δa

(2) (2)
Dp+1 A2p+2 ρ(δ) = δa

(1) (2)
E7 E6 ρ(δ) = 2δa
96 J. Nervi / Journal of Algebra 253 (2002) 50–99

Table B (continued)

(1) (1)
E6 F4 ρ(δ) = δa

(1) (3)
E6 D4 ρ(δ) = 3δa

(1) (1)
B3 G2 ρ(δ) = δa

(1) (2)
Bl Dl ρ(δ) = 2δa

(1) (2)
B2p+1 A2p ρ(δ) = 2δa

(1) (1)
Dl Bl−1 ρ(δ) = δa
 J. Nervi / Journal of Algebra 253 (2002) 50–99 97

Table B (continued)
(1) (2)
Dl Dl−1 ρ(δ) = 2δa

(1) (2)
D2p+2 A2p+1 ρ(δ) = 2δa

(1) (2)
D2p+2 A2p+2 ρ(δ) = 2δa

(2) (2)
A2l−1 A2l−2 ρ(δ) = δa

(1) (1)
Cl A1 ρ(δ) = (l − 1)δa

(1) (1)
C2mp Cp ρ(δ) = 2mδa
98 J. Nervi / Journal of Algebra 253 (2002) 50–99

Table B (continued)
(1) (1)
A2p+1 Cp+1 ρ(δ) = 2δa

(1) (1)
Amj −1 Aj −1 ρ(δ) = (m − 1)δa
0 0

(1) (1)
A2mp−1 Cp ρ(δ) = (m − 1)δa

References

[1] S. Berman, R. Moody, Lie algebras graded by finite root systems, Invent. Math. 108 (1992) 323–
347.
[2] G. Benkart, E. Zelmanov, Lie algebras graded by finite root systems and intersection matrix
algebras, Invent. Math. 126 (1996) 1–45.
[3] E. Neher, Lie algebras graded by J-graded root systems and Jordan pairs covered by grids, Amer.
J. Math. 118 (2) (1996) 439.
[4] B. Allison, G. Benkart, Y. Gao, Central extensions of Lie algebras graded by finite root systems,
Math. Ann. 316 (2000) 499–527.
 J. Nervi / Journal of Algebra 253 (2002) 50–99 99

[5] J. Nervi, Algèbres de Lie simples graduées par un système de racines et sous-algèbres C-
admissibles, J. Algebra 223 (2000) 307–343.
[6] H. Rubenthaler, Construction de certaines sous-algèbres remarquables dans les algèbres de Lie
semi-simples, J. Algebra 81 (1983) 268–278.
[7] H. Rubenthaler, Les paires duales dans les algèbres de Lie réductives, Astérisque 219 (1994).
[8] M. Sato, T. Kimura, A classification of irreducible prehomogeneous vector spaces and their
relative invariants, Nagoya Math. J. 65 (1977) 1–155.
[9] H. Rubenthaler, Espaces préhomogènes de type parabolique, Thèse, Université de Strasbourg,
1982.
[10] E.B. Vinberg, On the classification of the nilpotent elements of graded Lie algebras, Soviet Math.
Dokl. 16 (1975) 1517–1520.
[11] V.G. Kac, Infinite Dimensional Lie Algebras, third edition, Cambridge Univ. Press, 1990.
[12] N. Bourbaki, Groupes et algèbres de Lie, Masson, Paris, 1981, Chapters 4, 5, 6.
[13] J.P. Serre, Algèbres de Lie semi-simples complexes, Benjamin, 1966.
[14] V.G. Kac, D.H. Peterson, Infinite flag varieties and conjugacy theorems, Proc. Natl. Acad. Sci.
USA 80 (1983) 1778–1782.