HAL Id: hal-00004156
https://hal.archives-ouvertes.fr/hal-00004156v3
Submitted on 30 Jun 2006
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Two-sided cells in type B (asymptotic case)
Cédric Bonnafé
To cite this version:
Cédric Bonnafé. Two-sided cells in type B (asymptotic case). Journal of Algebra, Elsevier, 2006, 304,
pp.216-236. �hal-00004156v3�
ccsd-00004156, version 3 - 30 Jun 2006
TWO-SIDED CELLS IN TYPE (ASYMPTOTIC CASE)
C´EDRIC BONNAF´E
Abstract. We compute two-sided cells of Weyl groups of typeBfor the “asymptotic”
choice of parameters. We also obtain some partial results concerning Lusztig’s conjectures in this particular case.
Let W
nbe a Weyl group of type B
n. The present paper is a continuation of the work done by L. Iancu and the author [3] concerning Kazhdan-Lusztig theory of W
nfor the asymptotic choice of parameters [3, §6]. To each element w ∈ W
nis associated a pair of standard bi-tableaux (P (w), Q(w)) (see [19] or [3, §3]): this can be viewed as a Robinson-Schensted type correspondence. Our main result [3, Theorem 7.7] was the complete determination of the left cells: two elements w and w
′are in the same left cell if and only if Q(w) = Q(w
′). For the corresponding result for the symmetric group, see [12] and [1]. We have also computed the character afforded by a left cell representation [3, Proposition 7.11] (this character is irreducible).
In this paper, we are concerned with the computation of the two-sided cells. Let us state the result here. If w ∈ W
n, write Q(w) = (Q
+(w), Q
−(w)) and denote by λ
+(w) and λ
−(w) the shape of Q
+(w) and Q
−(w) respectively. Note that (λ
+(w), λ
−(w)) is a bipartition of n.
Theorem (see 3.9). For the choice of parameters as in [3, §6], two elements w and w
′are in the same two-sided cell if and only if (λ
+(w), λ
−(w)) = (λ
+(w
′), λ
−(w
′)).
Lusztig [17, Chapter 14] has proposed fifteen conjectures on Kazhdan-Lusztig theory of Hecke algebras with unequal parameters. In the asymptotic case, Geck and Iancu [9]
use some of our results, namely some informations on the preorder 6
LR(see Theorem 3.5 and Proposition 4.2), to compute the function a and to prove Lusztig’s conjectures P
i, for i ∈ {1, 2, 3, 4, 5, 6, 7, 8, 11, 12, 13, 14}. On the other hand, Geck [8] has shown that Lusztig’s conjectures P
9and P
10hold. More precisely, he proved that the Kazhdan-Lusztig basis is cellular (in the sense of [11]). He also proved a slightly weaker version of P
15(but his version is sufficient for constructing the homomorphism from the Hecke algebra to the asymptotic algebra J ).
The present paper is organized as follows. In Section 1, we study some consequences of Lusztig’s conjectures on the multiplication by T
w0, where w
0is the longest element of
Date: June 30, 2006.
1991 Mathematics Subject Classification. According to the 2000 classification: Primary 20C08; Sec- ondary 20C15.
1
a finite Weyl group. From Section 2 to the end of the paper, we assume that the Weyl group is of type B
nand that the choice of parameters is done as in [3, §6]. In Section 2, we establish some preliminary results concerning the Kazhdan-Lusztig basis. In Section 3, we prove the above theorem by introducing a new basis of the Hecke algebra: this was inspired by the work of Geck on the induction of Kazhdan-Lusztig cells [7]. Section 4 contains some results related to Lusztig’s conjectures. In Section 5, we determine which specializations of the parameters preserve the Kazhdan-Lusztig basis.
1. Generalities
1.A. Notation. We slightly modify the notation used in [3, §5]. Let (W, S) be a Coxeter group with |S| < ∞. We denote by ℓ : W → N = {0, 1, 2, . . . } the length function relative to S. If W is finite, w
0denotes its longest element. Let 6 denote the Bruhat ordering on W . If I ⊂ S, we denote by W
Ithe standard parabolic subgroup of W generated by I.
Let Γ be a totally ordered abelian group which will be denoted additively. The order on Γ will be denoted by 6 . If γ
0∈ Γ, we set
Γ
<γ0= {γ ∈ Γ | γ < γ
0}, Γ
6γ0= {γ ∈ Γ | γ 6 γ
0}, Γ
>γ0= {γ ∈ Γ | γ > γ
0} and Γ
>γ0= {γ ∈ Γ | γ > γ
0}.
Let A be the group algebra of Γ over Z . It will be denoted exponentially: as a Z -module, it is free with basis (v
γ)
γ∈Γand the multiplication rule is given by v
γv
γ′= v
γ+γ′for all γ , γ
′∈ Γ. If a ∈ A, we denote by a
γthe coefficient of a on v
γ, so that a = P
γ∈Γ
a
γv
γ. If a 6= 0, we define the degree and the valuation of a (which we denote respectively by deg a and val a) as the elements of Γ equal to
deg a = max{γ | a
γ6= 0}
and val a = min{γ | a
γ6= 0}.
By convention, we set deg 0 = −∞ and val 0 = +∞. So deg : A → Γ ∪ {−∞} and val : A → Γ ∪ {+∞} satisfy deg ab = deg a + deg b and val ab = val a + val b for all a, b ∈ A.
We denote by A → A, a 7→ a ¯ the automorphism of A induced by the automorphism of Γ sending γ to −γ . Note that deg a = − val ¯ a. If γ
0∈ Γ, we set
A
<γ0= ⊕
γ<γ0
Z v
γ, A
6γ0= ⊕
γ6γ0
Z v
γ, A
>γ0= ⊕
γ>γ0
Z v
γand A
>γ0= ⊕
γ>γ0
Z v
γ.
We fix a weight function L : W → Γ, that is a function satisfying L(ww
′) = L(w)+L(w
′)
whenever ℓ(ww
′) = ℓ(w)+ℓ(w
′). We also assume that L(s) > 0 for every s ∈ S. We denote
by H = H(W, S, L) the Hecke algebra of W associated to the weight function L. It is the
associative A-algebra with A-basis (T
w)
w∈Windexed by W and whose multiplication is determined by the following two conditions:
(a) T
wT
w′= T
ww′if ℓ(ww
′) = ℓ(w) + ℓ(w
′) (b) T
s2= 1 + (v
L(s)− v
−L(s))T
sif s ∈ S.
It is easily seen from the above relations that (T
s)
s∈Sgenerates the A-algebra H and that T
wis invertible for every w ∈ W . If h = P
w∈W
a
wT
w∈ H, we set h = P
w∈W
a
wT
w−−11. Then the map H → H, h 7→ h is a semi-linear involutive automorphism of H. If I ⊂ S, we denote by H(W
I) the sub-A-algebra of H generated by (T
s)
s∈I.
Let w ∈ W . By [17, Theorem 5.2], there exists a unique element C
w∈ H such that (a) C
w= C
w(b) C
w∈ T
w+
⊕
y∈W
A
<0T
y. Write C
w= P
y∈W
p
∗y,wT
ywith p
∗y,w∈ A. Then [17, 5.3]
p
∗w,w= 1
p
∗y,w= 0 if y 66 w.
In particular, (C
w)
w∈Wis an A-basis of H: it is called the Kazhdan-Lusztig basis of H.
Write now p
y,w= v
L(w)−L(y)p
∗y,w. Then
p
y,w∈ A
>0and the coefficient of p
y,won v
0is equal to 1 (see [17, Proposition 5.4 (a)].
We define the relations 6
L, 6
R, 6
LR, ∼
L, ∼
Rand ∼
LRas in [17, §8].
1.B. The function a. Let x, y ∈ W . Write C
xC
y= X
z∈W
h
x,y,zC
z, where h
x,y,z∈ A for z ∈ W . Of course, we have
(1 . 1) h
x,y,z= h
x,y,z.
The following lemma is well-known [17, Lemma 10.4 (c) and formulas 13.1 (a) et (b)].
Lemma 1.2. Let x, y and z be three elements of W . Then deg h
x,y,z6 min(L(x), L(y)).
Conjecture P
0(Lusztig): There exists N ∈ Γ such that deg h
x,y,z6 N for all x, y and z in W .
If W is finite, then W satisfies obviously P
0. If W is an affine Weyl group, then it also satisfies P
0[16, 7.2]. From now on, we assume that W satisfies P
0, so that the next definition is valid. If z ∈ W , we set
a(z) = max
x,y∈W
deg h
x,y,z.
Since h
1,z,z= 1, we have a(z) ∈ Γ
>0. If necessary, we will write a
W(z) for a(z). We denote by γ
x,y,z−1∈ Z the coefficient of v
a(z)in h
x,y,z. The next proposition shows how the function a can be calculated by using different bases.
Proposition 1.3. Let (X
w)
w∈Wand (Y
w)
w∈Wbe two families of elements of H such that, for every w ∈ W , X
w− T
wand Y
w− T
wbelong to ⊕
y<wA
<0T
y. For all x and y in W , write
X
xY
y= X
z∈W
ξ
x,y,zC
z. Then, if x, y, z ∈ W , we have:
(a) deg ξ
x,y,z6 min{L(x), L(y)}.
(b) ξ
x,y,z∈ γ
x,y,z−1v
a(z)+ A
<a(z). In particular,
a(z) = max
x,y∈W
deg ξ
x,y,z. Proof - Clear.
1.C. Lusztig’s conjectures. Let τ : H → A be the A-linear map such that τ (T
w) = δ
1,wif w ∈ W . It is the canonical symmetrizing form on H (recall that τ (T
xT
y) = δ
xy,1). If z ∈ W , let
∆(z) = − deg p
∗1,z= − deg τ (C
z).
Let n
zbe the coefficient of p
∗1,zon v
−∆(z). Finally, let D = {z ∈ W | a(z) = ∆(z)}.
Conjectures (Lusztig): With the above notation, we have:
P
1. If z ∈ W , then a(z) 6 ∆(z).
P
2. If d ∈ D and if x, y ∈ W satisfy γ
x,y,d6= 0, then x = y
−1.
P
3. If y ∈ W , then there exists a unique d ∈ D such that γ
y−1,y,d6= 0.
P
4. If z
′6
LRz, then a(z) 6 a(z
′). Therefore, if z ∼
LRz
′, then a(z) = a(z
′).
P
5. If d ∈ D and y ∈ W satisfy γ
y−1,y,d6= 0, then γ
y−1,y,d= n
d= ±1.
P
6. If d ∈ D, then d
2= 1.
P
7. If x, y, z ∈ W , then γ
x,y,z= γ
y,z,x.
P
8. If x, y, z ∈ W satisfy γ
x,y,z6= 0, then x ∼
Ly
−1, y ∼
Lz
−1and z ∼
Lx
−1. P
9. If z
′6
Lz and a(z
′) = a(z), then z
′∼
Lz.
P
10. If z
′6
Rz and a(z
′) = a(z), then z
′∼
Rz.
P
11. If z
′6
LRz and a(z
′) = a(z), then z
′∼
LRz.
P
12. If I ⊂ S and z ∈ W
I, then a
WI(z) = a
W(z).
P
13. Every left cell C of W contains a unique element d ∈ D. If y ∈ C, then γ
y−1,y,d6= 0.
P
14. If z ∈ W , then z ∼
LRz
−1.
P
15. If x, x
′, y, w ∈ W are such that a(y) = a(w), then X
y′∈W
h
w,x′,y′⊗
Zh
x,y′,y= X
y′∈W
h
y′,x′,y⊗
Zh
x,w,y′in A ⊗
ZA.
Lusztig has shown that these conjectures hold if W is a finite or affine Weyl group and L = ℓ [17, §15], if W is dihedral and L is any weight function [17, §17] and if (W, L) is quasi-split [17, §16].
1.D. Lusztig’s conjectures and multiplication by T
w0
. We assume in this subsection that W is finite. We are interested here in certain properties of the multiplication by T
wn0for n ∈ Z . Some of them are partially known [14, Lemma 1.11 and Remark 1.12]. If y ∈ W and n ∈ Z , we set
T
wn0C
y= X
x∈W
λ
(n)x,yC
x.
Note that λ
(n)x,y= 0 if x 66
Ly.
Proposition 1.4. Assume that W is finite and satisfies Lusztig’s conjectures P
1, P
4and P
8. Let n ∈ Z and let x and y be two elements of W such that x 6
Ly. Then:
(a) If n > 0, then deg λ
(n)x,y6 n(a(x) − a(w
0x)). If moreover x <
Ly, then deg λ
(n)x,y<
n(a(x) − a(w
0x)).
(b) If n 6 0, then deg λ
(n)x,y6 n(a(y) − a(w
0y)). If moreover x <
Ly, then deg λ
(n)x,y<
n(a(y) − a(w
0y)).
(c) If n is even and if x ∼
Ly, then λ
(n)x,y= δ
x,yv
n(a(x)−a(w0x)).
Proof - If n = 0, then (a), (b) and (c) are easily checked. Let us now prove (a) and (b).
By [17, Proposition 11.4],
T
w0= X
u∈W
(−1)
ℓ(w0u)p
∗1,w0uC
u. Consequently,
λ
(1)x,y= X
u∈W x6Ru
(−1)
ℓ(w0u)p
∗1,w0uh
u,y,x.
But, by P
1, we have deg p
∗1,w0u6 −a(w
0u). If moreover x 6
Ru, then w
0u 6
Rw
0x and so
−a(w
0x) > −a(w
0u) by P
4. Therefore
deg λ
(1)x,y6 a(x) − a(w
0x).
On the other hand, if deg λ
(1)x,y= a(x) − a(w
0x), then there exists u ∈ W such that x 6
Ru
and deg h
u,y,x= a(x). So, by P
8, we get that x ∼
Ly. This shows (a) for n = 1.
Now, let ν : H → H denote the A-linear map such that ν(C
w) = v
a(w0w)−a(w)C
wfor all w ∈ W and let µ : H → H, h 7→ T
w0h. Then, if w ∈ W , we have
νµ(C
y) = X
u6Ly
v
a(w0u)−a(u)λ
(1)u,yC
u.
So, by the previous discussion, we have v
a(w0u)−a(u)λ
(1)u,y∈ A
60. Moreover, if u <
Ly, then v
a(w0u)−a(u)λ
(1)u,y∈ A
<0. On the other hand, det µ = ±1 and det ν = 1. Therefore, if we write
µ
−1ν
−1(C
y) = X
u6Ly
β
u,yC
u, then β
u,y∈ A
60and, if u <
Ly, then β
u,y∈ A
<0. Finally,
T
w−01C
y= µ
−1(C
y)
= µ
−1ν
−1ν(C
y)
= v
a(w0y)−a(y)µ
−1ν
−1(C
y)
= X
u6Ly
v
a(w0y)−a(y)β
u,yC
u.
In other words, λ
(x,y−1)= v
a(w0y)−a(y)β
x,y. This shows that (b) holds if n = −1. An elementary induction argument using P
4shows that (a) and (b) hold in full generality.
Let us now prove (c). Let K be the field of fraction of A. Let C be a left cell of W and let c ∈ C. We set
H
6LC= ⊕
w6Lc
AC
wand H
<LC= ⊕
w<Lc
AC
w.
Then H
6LCand H
<LCare left ideals of H. The algebra KH = K ⊗
AH being semisimple, there exists a left ideal I
Cof KH such that KH
6LC= KH
<LC⊕ I
C.
We need to prove that, for all h ∈ I
C,
T
wn0h = v
n(a(c)−a(w0c))h.
For this, we may, and we will, assume that n > 0. Let V
1C, V
2C,. . . , V
nCCbe irreducible sub-KH ⊗
AK-modules of I
Csuch that
I
C= V
1C⊕ · · · ⊕ V
nCC.
Let j ∈ {1, 2, . . . , n
C}. Since T
wn0is central and invertible in H, there exists ε ∈ {1, −1}
and i
Cj∈ Γ such that
T
wn0h = εv
iCjh
for every h ∈ V
jC. By specializing v
γ7→ 1, we get that ε = 1. Moreover, by (a) and (b),
i
Cj6 n(a(c) − a(w
0c)). On the other hand, since det µ = ±1, we have det µ
n= 1. But
det µ
n= v
r, where
r = X
C∈LC(W) nC
X
j=1
i
Cjdim V
jC6 n X
C∈LC(W)
(a(C) − a(w
0C))
nC
X
j=1
dim V
jC= n X
C∈LC(W)
(a(C) − a(w
0C))|C|
= n X
w∈W
(a(w) − a(w
0w))
= 0.
Here, LC(W ) denotes the set of left cells in W and, if C ∈ LC(W ), a(C) denotes the value of a on C (according to P
4). The fact that r = 0 forces the equality i
Cj= n(a(C) −a(w
0C)) for every left cell C and every j ∈ {1, 2, . . . , n
C}.
Remark 1.5 - Assume here that w
0is central in W and keep the notation of the proof of Proposition 1.4 (c). Let j ∈ {1, 2, . . . , n
C}. Then there exists ε
j(C) ∈ {1, −1} et e
j(C) ∈ Γ such that T
w0h = ε
j(C)v
ej(C)h for every h ∈ V
jC.
Question: Let j, j
′∈ {1, 2, . . . , n
C}. Does ε
j(C) = ε
j′(C) ?
A positive answer to this question would allow to generalize Proposition 1.4 (c) to the case where w
0nis central.
Corollary 1.6. Assume that W is finite and satisfies Lusztig’s conjectures P
1, P
2, P
4, P
8, P
9and P
13. Let w ∈ W and let n ∈ N . Then deg τ (T
w−0nC
w) 6 −a(w) +n(a(w
0w) − a(w)).
Moreover, deg τ (T
w−0nC
w) = −a(w) + n(a(w
0w) − a(w)) if and only if w
0nw
−1∈ D.
Proof - Assume first that n is even. In particular, w
n0= 1. By Proposition 1.4, we have τ (T
w−0nC
w) = v
n(a(w0w)−a(w))τ (C
w) + X
x<Lw
λ
(x,w−n)τ (C
x).
But, if x <
Lw, then deg τ (C
x) = −∆(x) 6 −a(x) 6 −a(w) by P
1and P
4. So, by Propo-
sition 1.4 (b), we have that deg λ
(x,w−n)τ (C
x) < −a(w) + n(a(w
0w) −a(w)). Moreover, again
by P
1, we have deg τ (C
w) = −∆(w) 6 −a(w). This shows that deg τ (T
w−0nC
w) 6 −a(w) +
n(a(w
0w) − a(w)) and that equality holds if and only if ∆(w) = a(w), that is, if and only
if w ∈ D, as desired.
Assume now that n = 2k+1 for some natural number k. Recall that, by [17, Proposition 11.4], T
w0= P
u∈W
(−1)
ℓ(w0u)p
∗1,w0uC
u. Therefore T
w−0nC
w= X
u∈W
(−1)
ℓ(w0u)p
∗1,w0uT
w−0n−1C
uC
w= X
u,x∈W x6Lw and x6Ru
(−1)
ℓ(w0u)p
∗1,w0uh
u,w,xT
w−0n−1C
x.
This implies that
τ (T
w−0nC
w) = X
u,x∈W x6Lw andx6Ru
(−1)
ℓ(w0u)p
∗1,w0uh
u,w,xτ (T
w−0n−1C
x),
so deg τ (T
w−n0C
w) 6 max
u,x∈W x6Lw and x6Ru
deg(p
∗1,w0uh
u,w,xτ (T
w−n−0 1C
x)).
Let u and x be two elements of W such that x 6
Lw and x 6
Ru. Since n + 1 is even and by the previous discussion, we have
deg τ (T
w−0n−1C
x) 6 −a(x) + (n + 1)(a(w
0x) − a(x)).
By P
1and P
4, deg p
∗1,w0u6 −a(w
0u) 6 −a(w
0x). Moreover, deg h
u,w,x6 a(x). Conse- quently,
deg(p
∗1,w0uh
u,w,xτ (T
w−0n−1C
x)) 6 −a(x) + n(a(w
0x) − a(x)) 6 −a(w) + n(a(w
0w) − a(w)).
Moreover, equality holds if and only if w
0u ∈ D, deg h
u,w,x= a(x) = a(w) and x ∈ D.
We first deduce that
deg τ (T
w−0nC
w) 6 −a(w) + n(a(w
0w) − a(w)) which is the first assertion of the proposition.
Assume now that deg τ (T
w−0nC
w) = −a(w) + n(a(w
0w) − a(w)). Then there exists u and x in W such that x 6
Lw, x 6
Ru, w
0u ∈ D, deg h
u,w,x= a(x) = a(w) and x ∈ D.
Since deg h
u,w,x= a(x) and x ∈ D, we deduce from P
2that w = u
−1, which shows that w
0w
−1∈ D.
Conversely, assume that w
0w
−1belongs to D. To show that deg τ (T
w−0nC
w) = −a(w) +
n(a(w
0w) −a(w)), it is sufficient to show that there is a unique pair (u, x) of elements of W
such that x 6
Lw, x 6
Ru, w
0u ∈ D, deg h
u,w,x= a(x) = a(w) and x ∈ D. The existence
follows from P
13(take u = w
−1and x be the unique el´ement of D belonging to the left cell
containing w). Let us now show unicity. Let (u, x) be such a pair. Since deg h
u,w,x= a(x)
and x ∈ D, we deduce from P
2that u = w
−1. Moreover, since a(x) = a(w) and x 6
Lw,
we have x ∼
Lw by P
9. But, by P
13, x is the unique element of D belonging to the left
cell containing w.
2. Preliminaries on type B (asymptotic case) From now on, we are working under the following hypothesis.
Hypothesis and notation: We assume now that W = W
nis of type B
n, n > 1. We write S = S
n= {t, s
1, . . . , s
n−1} as in [3, §2.1]: the Dynkin diagram of W
nis given by
i i i
· · ·
it s
1s
2s
n−1We also assume that Γ = Z
2and that Γ is ordered lexicographically:
(a, b) 6(a
′, b
′) ⇐⇒ a < a
′or (a = a
′and b 6 b
′).
We set V = v
(1,0)and v = v
(0,1)so that A = Z [V, V
−1, v, v
−1] is the Laurent polynomial ring in two algebraically independent indeterminates V and v.
If w ∈ W
n, we denote by ℓ
t(w) the number of occurences of t in a reduced expression of w. We set ℓ
s(w) = ℓ(w) − ℓ
t(w). Then ℓ
sand ℓ
tare weight functions and we assume that L = L
n: W
n→ Γ, w 7→ (ℓ
t(w), ℓ
s(w)). So H = H
n= H(W
n, S
n, L
n). We denote by S
nthe subgroup of W generated by {s
1, . . . , s
n−1}: it is isomorphic to the symmetric group of degree n.
We now recall some notation from [3, §2.1 and 4.1]. Let r
1= t
1= t and, if 1 6 i 6 n −1, let r
i+1= s
ir
iand t
i+1= s
it
is
i. If 0 6 l 6 n, let a
l= r
1r
2. . . r
l. We denote by S
l, W
l, S
l,n−l, W
l,n−lthe standard parabolic subgroups of W
ngenerated by {s
1, s
2, . . . , s
l−1}, {t, s
1, s
2, . . . , s
l−1}, S
n\ {t, s
l} and S
n\ {s
l} respectively. The longest element of S
lis denoted by σ
l. Let
Y
l,n−l= {a ∈ S
n| ∀ σ ∈ S
l,n−l, ℓ(aσ) > ℓ(σ)}.
If w ∈ W
nis such that ℓ
t(w) = l, then [3, §4.6] there exist unique a
w, b
w∈ Y
l,n−l, σ
w∈ S
l,n−lsuch that w = a
wa
lσ
wb
−w1. Recall that ℓ(w) = ℓ(a
w) + ℓ(a
l) + ℓ(σ
w) + ℓ(b
w).
2.A. Some submodules of H . If l is a natural number such that 0 6 l 6 n, we set T
l= ⊕
w∈Wn
ℓt(w)=l
AT
w, T
6l= ⊕
w∈Wn
ℓt(w)6l
AT
w, T
>l= ⊕
w∈Wn
ℓt(w)>l
AT
w,
C
l= ⊕
w∈Wn
ℓt(w)=l
AC
w, C
6l= ⊕
w∈Wn
ℓt(w)6l
AC
wand C
>l= ⊕
w∈Wn
ℓt(w)>l
AC
w.
Let Π
T?: H
n→ T
?and Π
C?: H
n→ C
?be the natural projections (for ? ∈ {l, 6 l, > l}).
Proposition 2.1. Let l be a natural number such that 0 6 l 6 n. Then:
(a) T
land C
lare sub-H( S
n)-modules-H( S
n) of H
n. The maps Π
Tland Π
Clare mor- phisms of H( S
n)-modules-H( S
n).
(b) T
6l= C
6l.
(c) C
>lis a two-sided ideal of H
n.
Proof - (a) follows from [3, Theorem 6.3 (b)]. (b) is clear. (c) follows from [3, Corollary 6.7].
The next proposition is a useful characterization of the elements of the two-sided ideal C
n.
Proposition 2.2. Let h ∈ H
n. The following are equivalent:
(1) h ∈ C
n.
(2) ∀x ∈ H
n, (T
t− V )xh = 0.
(3) ∀x ∈ H( S
n), (T
t− V )xh = 0.
Proof - If ℓ
t(w) = n, then tw < w so (T
t− V )C
w= 0 by [17, Theorem 6.6 (b)]. Since C
nis a two-sided ideal of H
n(see Proposition 2.1 (c)), we get that (1) implies (2). It is also obvious that (2) implies (3). It remains to show that (3) implies (1).
Let I = {h ∈ H
n| ∀x ∈ H( S
n), (T
t− V )xh = 0}. Then I is clearly a sub-H( S
n)- module-H
nof H
n. We need to show that I ⊂ C
n. In other words, since C
n⊂ I , we need to show that I ∩ C
6n−1= 0. Let I
′= I ∩ C
6n−1= I ∩ T
6n−1(see Proposition 2.1 (b)).
Let X = {w ∈ W
n| τ (I
′T
w−1) 6= 0}. Showing that I
′= 0 is equivalent to showing that X = ∅ .
Assume X 6= ∅ . Let w be an element of X of maximal length and let h be an element of I
′such that τ (hT
w−1) 6= 0. Since h ∈ T
6n−1, we have ℓ
t(w) 6 n − 1. Moreover, T
th = V h, so τ (T
thT
w−1) = τ (hT
w−1T
t) 6= 0. By the maximality of ℓ(w), we get that tw < w. So, there exists s ∈ S
nsuch that sw > w and s 6= t. Then
τ (T
shT
(sw)−1) = τ (hT
(sw)−1T
s)
= τ (hT
w−1) + (v − v
−1)τ (hT
(sw)−1) 6= 0,
the last inequality following from the maximality of ℓ(w) (which implies that τ (hT
(sw)−1) = 0). But T
sh ∈ I
′and so sw ∈ X. This contradicts the maximality of ℓ(w).
2.B. Some results on the Kazhdan-Lusztig basis. In this subsection, we study the elements of the Kazhdan-Lusztig basis of the form C
alσwhere 0 6 l 6 n and σ ∈ S
n. Proposition 2.3. Let σ ∈ S
nand let 0 6 l 6 n. Then C
alC
σ= C
alσand C
σC
al= C
σal. Proof - Let C = C
alC
σ. Then C = C and
C − T
alσ= X
w<alσ
λ
wT
wwith λ
w∈ A for w < a
lσ. To show that C = C
alσ, it is sufficient to show that λ
w∈ A
<0.
But, C
al= T
al+ P
x<al
V
ℓt(x)−lβ
xT
xwith β
x∈ Z [v, v
−1] (see [3, Theorem 6.3 (a)]).
Hence,
C = T
alσ+ X
τ <σ
p
∗τ,σT
alτ+ X
x<al
V
ℓt(x)−lβ
xT
xC
σ.
But, if x < a
l, then ℓ
t(x) < l. This shows that λ
w∈ A
<0for every w < a
lσ. This shows the first equality. The second one is obtained by a symmetric argument.
Proposition 2.3 shows that it can be useful to compute in different ways the elements C
alto be able to relate the Kazhdan-Lusztig basis of H
nto the Kazhdan-Lusztig basis of H( S
n). Following the work of Dipper-James-Murphy [4], Ariki-Koike [2] and Graham- Lehrer [11, §5], we set
P
l= (T
t1+ V
−1)(T
t2+ V
−1) . . . (T
tl+ V
−1)
= X
06k6l
V
k−lX
16i1<···<ik6l
T
ti1...tik.
Lemma 2.4. P
nis central in H
n.
Proof - First, P
ncommutes with T
t(indeed, tt
i= t
it > t
ifor 1 6 i 6 n). By [2, Lemma 3.3], P
ncommutes with T
sifor 1 6 i 6 n − 1. Since the notation and conventions are somewhat different, we recall here a brief proof. First, if j 6∈ {i, i + 1}, s
it
j= t
js
i> t
jso T
sicommutes with T
tj. Therefore, it is sufficient to show that T
sicommutes with (T
ti+ V
−1)(T
ti+1+ V
−1). This follows from a straightforward computation using the fact that s
it
i> t
i, that t
i+1s
i< t
i+1and that s
it
i= t
i+1s
i.
Proposition 2.5. If 0 6 l 6 n, then C
al= P
lT
σ−1l
= T
σ−1l
P
l.
Proof - The computation may be performed in the subalgebra of H
ngenerated by {T
t, T
s1, . . . , T
sl−1} so we may, and we will, assume that l = n. First, we have T
tP
n= V P
n. Since P
nis central in H
n, it follows from the characterization of C
ngiven by Proposition 2.2 that P
n∈ C
n.
Now, let h = C
an− P
nT
σ−n1. Then, by Proposition 2.1 (a), we have h ∈ C
n. Moreover, it is easily checked that h ∈ T
6n−1= C
6n−1. So h = 0.
Corollary 2.6. If 0 6 l 6 n and σ ∈ S
n, then Π
T0(C
alσ) = V
−lT
σ−1l
C
σ. In particular, τ (C
alσ) = V
−lτ (T
σ−l1C
σ).
Proof - Since Π
T0is a morphism of right H( S
n)-modules (see Proposition 2.1 (a)) and
since C
alσ= P
lT
σ−l1C
σ(see Propositions 2.3 and 2.5), we have Π
T0(C
alσ) = Π
T0(P
l)T
σ−l1C
σ.
But, Π
T0(P
l) = V
−l. This completes the proof of the corollary.
3. Two-sided cells
The aim of this section is to show that, if x and y are two elements of W such that ℓ
t(x) = ℓ
t(y) = l, then x 6
LRy if and only if σ
x6
SLRl,n−lσ
y(see Theorem 3.5). Here, 6
SLRl,n−lis the preorder 6
LRdefined inside the parabolic subgroup S
l,n−l. For this, we adapt an argument of Geck [7] who was considering the preorder 6
L.
We start by defining an order relation 4 on W . Let x and y be two elements of W . Then x ≺ y if the following conditions are fulfilled:
(1) ℓ
t(x) = ℓ
t(y), (2) x 6 y,
(3) a
x< a
yor b
x< b
y, (4) σ
x6
SLRl,n−lσ
y.
We write x 4 y if x ≺ y or x = y. If y ∈ W
n, we set Γ
y= T
ayC
aℓt(y)C
σyT
b−1y
.
Lemma 3.1. Let y ∈ W . Then Γ
y∈ T
y+ ⊕
x<yA
<0T
x.
Proof - First, recall that Γ
yis a linear combination of elements of the form T
ayT
zT
b−1y
with z 6 a
ℓt(y)σ
y, so it is a linear combination of elements of the form T
xwith x 6 y.
Let l = ℓ
t(y) and σ = σ
y. We have Γ
y= T
ayT
alT
σT
b−1y
+ X
τ <σ
p
∗τ,σT
ayT
alT
τT
b−1 y+ X
a<al, τ6σ
p
∗a,alp
∗τ,σT
ayT
aT
τT
b−1 y.
If τ < σ, then T
ayT
alT
τT
b−1y
= T
ayalτ b−y1
by [3, §4.6]. On the other hand, if a < a
l, then ℓ
t(a) < l so T
ayT
aT
τT
b−1y
is a linear combination, with coefficients in Z [v, v
−1] of elements T
wwith ℓ
t(w) = ℓ
t(a) < l (because a
y, τ and b
−y1are elements of S
n). Since V
l−ℓt(a)p
∗a,al∈ Z [v, v
−1] by [3, Theorem 6.3 (a)], this proves the lemma.
Lemma 3.2. If y ∈ W
n, then
Γ
y= Γ
y+ X
x≺y
ρ
x,yΓ
xwhere the ρ
x,y’s belong to Z [v, v
−1].
Proof - Let l = ℓ
t(y). Then T
−1a−1y
= T
ay+ X
a∈Yl,n−l
x∈Sl,n−l ax<ay