Irreducible modular representations of a reductive p-adic group and simple modules for Hecke algebras.
Marie-France Vign´eras
Let R be an algebraically closed field of characteristic l / =p, let F be a local non archimedean field of residual characteristic p, and let G be the group of rational points of a connected reductive group defined over F. The two main points in the search for a classification of the irreducible R-representations of G is to try to prove that any irreducible cuspidal representation is induced from an open compact subgroup and that the irreducible representations with a given inertial cuspidal support are classified by simple modules for the Hecke algebra of a type. Over a field R which is not the complex field new serious difficulties arise and the purpose of this article is to indicate a way to avoid them. The mirabolic trick used when the group is GL(n, F ) does not generalize but our new method is general and we can extend from the complex case to R the results of Morris and Moy-Prasad for level 0 representations.
Summary Introduction
1. Almost projectivity
2. A simple criterium for almost projectivity.
3. A simple criterium for irreducibility.
4. Finite reductive group 5. Morris types of level 0.
6. Cuspidal representations of level 0.
7. Bushnell-Kutzko types.
Bibliography
Introduction
Let (K, σ) be an irreducible simple cuspidal R-type in GL(n, F ) as defined by Bushnell and Kutzko, or a cuspidal R-type of level 0 in G as defined by Morris. We consider also an irreducible extended maximal type (N, Λ). We denote by ind the compact induction. We will prove :
Theorem 1) The irreducible R-representations of G which contain (K, σ) are in bijection with the simple modules for the Hecke algebra of (K, σ) in G.
2) ind
GNΛ is irreducible.
It is known that the Hecke algebra of (K, σ) in G is very closed to products of affine Hecke R-algebras hence the construction of simple modules for affine Hecke R-algebras is strongly related to the construction of irreducible R-representations of G if 1) is true.
The purpose of this article is to give a simple proof of the theorem. As a consequence one can extend to R-representations the complex theory of representations of level 0 by Morris or by Moy and Prasad.
The theorem known for GL(n, F ) lead to a complete classification of the irreducible R-representations of GL(n, F ) [Vigneras1,Vigneras2]. But the proof did not transfer to a general reductive group G and the proof that we give here is simpler and general.
When the characteristic of R is banal there is nothing to do. Indeed, if the pro-order of the profinite
group K is invertible in R, the representation ind
GKσ of G is projective and 1) results from an easy lemma in
algebra. Let us denote by Z the center of G. The group N contains Z and N/Z is profinite. If the pro-order
of N/Z is invertible in R, the category of R-representations of N where Z acts by a given character is
semi-simple, and End
RGind
GNΛ R implies the irreducibility of ind
GNΛ.
We explain now the proof of the theorem. In the general case the representation ind
GKσ is not projective, but we know that 1) will be true if ind
GKσ is almost projective [Vigneras2]. One knows that End
RGind
GNΛ R, because the proof [Morris] [Moy-Prasad] [Bushnell-Kutzko] for complex types transfers to R-types. We denote by Irr
RG the set of irreducible R-representations of G, modulo isomorphism.
The properties
1’) ind
GKσ is almost projective,
2’) if Λ is contained in π|
Nthen Λ is a quotient of π|
N, for any π ∈ Irr
RG.
are - a priori - stronger property than 1) and 2). This is explained in the paragraph 2 and 3.
We denote by U the pro-p-radical of K. Suppose that we are in the level 0 case. The group N is the G-normalizer N
G(K) of a maximal K and Λ is an irreducible representation of N
G(K) which contains σ.
The main point of the article is the remark that the action of K (resp. N ) on th e U -invariant vectors of ind
GKσ (resp. ind
GNΛ) can be computed, and this allows to prove 1’) and 2’). The precise result given in the paragraphs 5,6 implies:
Proposition (Level 0) The action of K on the U -invariant vectors of ind
GKσ is a direct sum of irreducible representations conjugate to σ.
When K is maximal, the action of N
G(K) on the U -invariant vectors of ind
GNG(K)Λ is isomorphic to Λ.
The properties 1’) and 2’) and the fact that End
RGind
GNG(K)Λ = R follow from the proposition.
In the GL(n, F )-case, similar results are obtained with the non trivial Bushnell-Kutzko representations η ∈ Irr
R(U ), κ ∈ Irr
R(K) (see the definition in the paragraph 7). When K is maximal there exists an extension E/F with E
∗⊂ GL(n, F ); we have N = KE
∗and Λ is an extension of σ to KE
∗. One computes the action of K (resp. N ) on th e η-isotypic part of ind
GKσ (resp. ind
GNΛ) in the paragraph 7. The properties 1’) and 2’) and the fact that End
RGind
GKE∗Λ = R follow from:
Proposition (GL(n, F )) The action of K on the η-isotypic part of ind
GKσ is of the form κ ⊗ W where W is semi-simple with constituents isomorphic to conjugates of σ.
When (KE
∗, Λ) is an extended maximal Bushnell-Kutzko type, the action of KE
∗on the η-isotypic part of ind
GKE∗Λ is isomorphic to Λ.
When G is a reductive finite group, our method applies and gives a new proof of the almost projectivity of any R-representation parabolically induced from a cuspidal irreducible representation [Geck-Hiss-Malle].
This is done in the paragraph 4.
In bothcases, H = K/U is a finite reductive group. The propositions follow from the computation of the functor
σ → σ
: Mod
RH → Mod
RH
such that the η-isotypic part of ind
GKκ ⊗ σ isomorphic to κ ⊗ σ
(in the level 0 case, η (resp. κ) is the trivial representation of U (resp. K)). This functor is the analogue of a well known functor : the parabolic induction followed by the parabolic restriction in H or in G. Parabolic groups of G are replaced by parahoric subgroups or by Bushnell-Kutzko groups and the unipotent radical by the pro-p-radical with a trick : the representation of the pro-p-radical is not trivial in the Bushnell-Kutzko case. The intersection of two Bushnell-Kutzko groups or of two parahoric subgroups have similar properties than the intersection of two parabolic subgroups and the representations κ have a coherent behaviour. The classical formula for parabolics originally due to Harish-Chandra for finite reductive groups and generalised by Casselman for p-adic groups has an analogue for open compact parahoric or Bushnell Kutzko groups.
It is pleasure to thank Alberto Arabia for his work [Arabia] which lead to the simple criterium of almost
projectivity of ind
GKσ, which is basic in our proof.
1 Almost-projectivity
The notion of almost projectivity was introduced by Dipper for finite groups, and generalised by Arabia to the more general notion of quasi-projectivity for modules over an algebra.
Let R be a field and let A be an R-algebra. We consider the category Mod
tf(A) (resp. Mod
otf(A)) of unital finite type left (resp. right) A-modules.
1.1 Definition A finite type unital A-module Q is called quasi-projective when for any two morphisms
Q
π
//
α
// V
in Mod
tf(A) with π surjective, there exists β ∈ End
AQ such that α = π ◦ β.
It is called almost projective when there exists a surjective morphism π : P → Q from a projective finite type unital A-module P to Q, such that for any morphism α ∈ End
RQ there exists β ∈ Hom
A(Q, P ) such that α = π ◦ β.
An almost projective module is quasi-projective. The fundamental property of quasi-projective modules is the following :
1.2 Main property (Arabia [Vigneras2, appendice th. 10]) When Q is quasi-projective, the functor Hom
A(Q, −) : Mod
tf(A) → Mod
of t(End
AQ)
induces a bijection between
a) the isomorphism classes of simple A-modules V such that Hom
A(Q, V )/ =0 and b) the isomorphism classes of simple right End
AQ-modules.
This functor is not in general an equivalence of category.
2 A simple criterium ofalmost projectivity
Let G be a locally profinite group, K an open compact subgroup of G, R a field and σ an irreducible smooth R-representation of K.
We denote by
Mod
R(G) the category of smooth R-representations of G.
V
σthe σ-isotypic part of V ∈ Mod
R(G), i.e. the biggest RK -submodule of V which is isomorphic to a direct sum ⊕
Iσ.
The functor of compact induction ind
GK: Mod
R(K) → Mod
R(G) is exact when G has an R-Haar measure, has a right ajoint (the restriction from G to K denoted by π → π |
K), respects projectivity and the property of beeing of finite type [Vigneras1 I.5.10, I.5.7, I.5.9].
2.1 Lemma Suppose that the R-representation of G Q = ind
GKσ
admits a K-equivariant direct decomposition Q = Q
σ⊕ Q
σand no subquotient of Q
σisomorphic to σ. Then Q is almost-projective.
We will show in the paragraphs 4,5 that the property of the lemma is verified when R is an algebraically
closed field of characteristic =p / and
- K is a parabolic subgroup of a reductive finite group over a field of characteristic p, or a parahoric subgroup of a reductive p-adic group,
- σ is trivial on the pro-p-radical of K and cuspidal,
We will show in the paragraph 7 that the property of the lemma is verified when (K, σ) is a simple Bushnell- Kutzko R-type.
This simple lemma which is basic for us was found by Arabia when G is a finite group. In this case he proved a better result [Arabia 2.4.2]: when G is a finite group, then Q is quasi-projective if and only if the σ-isotypic part of Q/Q
σis zero.
2.2 Exercise Proof of the lemma.
Let f : P
σ→ σ be a projective cover in Mod
RK (we are reduced to the case of finite groups because σ is trivial on a subgroup of finite index). We take
P := ind
GKP
σ, π = ind
GKf.
The isomorphism of adjunction,
Hom
RG(ind
GKP
σ, ind
GKσ) Hom
RK(P
σ, ind
GKσ).
is given by restriction to the RK-submodule P
σisomorphic to P
σof functions withsupport in K in the canonical model of ind
GKP
σ. The image of ind
GKf under the isomorphism of adjunction is f : P
σ→ σ (where σ σ is the space of functions with support in K in the canonical model of ind
GKσ).
The hypothesis on ind
GKσ and the definition of a projective cover imply that the map γ → γ ◦ f : Hom
RK(σ, ind
GKσ) → Hom
RK(P
σ, ind
GKσ)
is an isomorphism. Hence the map
β → β ◦ π : End
Rind
GKσ → Hom
RG(ind
GKP
σ, ind
GKσ) is surjective (it is clearly injective). We deduce that ind
GKσ is almost projective.
2.3 Exercise If Q = ind
GKσ satisfies the simple criterium (2.1), then Q is quasi-projective.
Let α, π : ind
GKσ → V be two morphisms in Mod
RG with π surjective. We look for β ∈ End
RGind
GKσ suchthat α = π ◦ β . The criterium implies that there exists a simple RK-submodule W
of Q
σsuchthat π(W
) = α(σ). Let β
: σ → ind
GKσ be the RK -equivariant morphism with image W
with π ◦ β
= α|
σ, and β ∈ End
RGind
GKσ the image of β
by adjunction. Then α = π ◦ β.
3 A simple criterium for irreducibility
We replace the property “compact” for K by “compact mod center”, we denote by Ind
GK: Mod
RK → Mod
RG the induction without condition on the support. This functor has a left adjoint (the restriction π → π|
Kfrom G to K) [Vign´eras1 I.5.7]. By the Mackey decomposition and by adjunction one sees that :
3.1 Lemma Let Λ ∈ Irr
RK. When the space End
RG(ind
GKΛ) is finite dimensional, it is equal to Hom
RG(ind
GKΛ, Ind
GKΛ).
3.2 Lemma The R-representation ind
GKΛ is irreducible when a) End
RG(ind
GKΛ) = R
b) If Λ is contained in π |
Kthen Λ is also a quotient of π |
K, for any π ∈ Irr
RG.
Proof Suppose that a) and b) are true. Let π ∈ Irr
RG be a quotient of ind
GKΛ. By adjunction and b),
Λ is quotient of π|
K. By adjunction π ⊂ Ind
GKΛ. Hence there is a morphism ind
GKΛ → Ind
GKΛ withimage
π. By a) and (3.1) ind
GKΛ = π. Hence ind
GKΛ is irreducible.
We will show in the paragraph 6 and 7 that the properties a) and b) of the lemma can be proved when R is an algebraically closed field of characteristic =p / and
- K is the normalizer of a maximal parahoric subgroup in a reductive p-adic group, or an extended maximal Bushnell-Kutzko group in GL(n, F ),
- the restriction of Λ to the maximal parahoric subgroup is cuspidal, or Λ is an extended maximal Bushnell-Kutzko R-type.
4 Finite reductive group Let
G the group of rational points of a reductive connected group over a finite field of characteristic p, P = M U a parabolic subgroup of G withunipotent radical U , and Levi subgroup M ,
Cusp
R(G) ⊂ Irr
R(G) the subset of irreducible cuspidal R-representations, σ ∈ Cusp
R(M ) identified witha representation of P trivial on U .
We denote by
N
G(M ) th e G-normalizer of M , W (M ) = N
G(M )/M,
g
σ
P(?) = σ
P(g
−1?g) ∈ Cusp
R(gP g
−1) for g ∈ G,
w
σ =
gσ with g ∈ N
G(M ) above w ∈ W (M ), W (M, σ) := {w ∈ W (M ),
wσ σ}
V → V
U: Mod
RG → Mod
RM the functor of U -invariant vectors.
4.1 Proposition [Harish-Chandra] We have
(ind
GPσ)
U⊕
w∈W(M)wσ.
With(2.1) we get a new proof of :
4.2 Corollary [Geck-Hiss-Malle (2.3)] ind
GPσ satisfies the simple criterium of almost-projectivity.
Proof. As U is the p-radical of P , we have a direct P -equivariant decomposition ind
GPσ = (ind
GPσ)
U⊕ W
⊕
W(M,σ)σ ⊕ W
where W
has no non zero U-fixed vector, and W has no subquotient isomorphic to σ.
5 Morris types oflevel 0 Let
F a local non archimedean field of residual field F
qwith q elements and characteristic p, G the group of rational points of a reductive connected group G over F ,
P , Q two parahoric subgroups of G [BT2 5.2.4], withtheir canonical exact sequence.
1 // U // P
fP// P (q) // 1,
1 // V // Q
fQ// Q (q) // 1
The kernels U , V are pro-p groups, the quotients P (q), Q (q) are the groups of rational points of connected reductive finite groups over F
q. We denote by f
P∗: Mod
RP (q) → Mod
RP the inflation along f
P, and f
P∗−1the inverse (on the representations of P trivial on U ).
5.1 Definition We call parahoric induction the functor of compact induction : ind
GP(q)= ind
GP◦f
P∗: Mod
RP(q) → Mod
RG
and parahoric restriction the functor of U -invariants :
res
GP(q)= f
P∗−1◦ res
GP: Mod
RG → Mod
RP (q).
(where res
GPV = V
U∈ Mod
RP ).
The composite functor
T
QG(q),P(q)= res
GQ(q)◦ ind
GP(q): Mod
RP (q) → Mod
RQ(q) can be computed, as in the parabolic case, using the Mackey decomposition.
We change notations. Let
T a maximal split torus of G (more precisely the group of rational points of this torus), N = N
G(T ) th e G-normalizer or T (the N of the introduction is no more used), A the appartment in the semi-simple Bruhat-Tits building of G defined by T . x, y ∈ A suchthat G
x= P , G
y= Q .
We denote G
1x= U , G
1y= V , G
x(q) = P (q), G
y(q) = Q (q), f
x= f
P, f
x∗= f
P∗.
5.2 Lemma Let z be a point in the building, G
zthe corresponding parahoric sugroup of G and f
x: G
x→ G
x(q) the canonical surjection. Then f
x(G
x∩G
z) is a parabolic subgroup of G
x(q) with unipotent radical f
x(G
x∩ G
1z).
Proof [Morris]. We reduce easily to the case treated by Morris where x, x
∈ A are in the closure of a chamber and z = nx
, using the following properties :
a) gG
xg
−1= G
gxfor g ∈ G [T 2.1], [BT2 5.2.4], [BT1 6.2.10], b) the Bruhat decomposition
G = G
oN G
owhere G
ois the Iwahori subgroup fixing a point o in a chamber C of A [BT1 7.2.6, 7.3.4] [SS page 105], c) G
x⊃ G
zwhen x is contained in the closure of the facet containing z [BT 5.2.4],
d) given a point a and a chamber C of closure C in the building, there exists g ∈ G suchthat ga ∈ C [T 1.7, 2.1]
By d) we can suppose z = g
x
where x, x
belong to the closure of a chamber C of A and g
∈ G.
By b) we can write g
= b
nb where b
, b ∈ G
owhere o ∈ C and n ∈ N . By c) G
o⊂ G
x∩ G
xhence bx
= x
, b
−1x = x. Witha) we get G
x∩ G
z= b
(G
x∩ G
nx)b
−1. The lemma for G
x∩ G
nx[Morris] implies the lemma for G
x∩ G
z.
The quotient of the parabolic subgroup by its unipotent radical in the lemma (5.2) is denoted by (G
x∩ G
z)
x(q) = f
x(G
x∩ G
z)/f
x(G
x∩ G
1z)
Withthis notation (G
x)
x(q) = G
x(q). The groups G
x(q) are isomorphic when x belongs to an orbit of G in the buiding (property a)) and the conjugation by g ∈ G induces an equivalence of categories
σ →
gσ : Mod
RG
x(q) → Mod
RG
gx(q)
and more generally an equivalence of categories
Mod
R(G
x∩ G
z)
x(q) → Mod
R(G
gx∩ G
gz)
gz(q)
because Mod
R(G
x∩ G
z)
x(q) is identified via f
xto the category of R-representations of G
x∩ G
ztrivial on G
x∩ G
1zand on G
1x∩ G
z, and Mod
R(G
gx∩ G
gz)
gz(q) is identified via f
gzto the category of R-representations of G
gx∩ G
gztrivial on G
gx∩ G
1gzand on G
1gx∩ G
gz.
Withthe notations P , Q we denote (G
x∩ G
y)
x(q) = ( P ∩ Q )
P(q).
5.3 Definition The functor
F
Q(q)gP(q)G: Mod
RP (q) → Mod
R( P ∩ g
−1Q g)
P(q) → Mod
R(g P g
−1∩ Q )
Q(q) → Mod
RQ (q) is the composite of
a) the f
P(P ∩ g
−1Vg)-invariants b) the conjugation by g
c) the parabolic induction.
5.4 Proposition T
Q(q),PG (q)⊕
g∈Q\G/PF
Q(q)gP(q)G.
Proof. By the Mackey formula, we have a Q -equivariant decomposition ind
GP(q)σ ⊕
g∈Q\G/Pind
QQ∩gPg−1(f
P∗σ)(g
−1?g).
Taking the V invariants we get the proposition.
When the functor F
Q(q)gP(q)Gdoes not vanishon Cusp
RP (q), then f
P(P ∩ g
−1Vg) = {1} i.e
(5.4.1) f
P(P ∩ g
−1Qg) = P (q).
When there exists σ ∈ Mod
RP (q) suchthat F
QG(q)gP(q)(σ) admits a cuspidal non zero representation as a sub-module or as a quotient, then f
Q(gU g
−1∩ Q) = {1} i.e.
(5.4.2) f
Q(g P g
−1∩ Q ) = Q (q).
The equations (5.4.1) and (5.4.1) are not independant.
5.5 Lemma Let x, z be two points in the building with corresponding parahoric subgroups G
x, G
z. If f
x(G
x∩ G
z) = G
x(q) then the three following properties are equivalent :
(5.5.i) f
z(G
x∩ G
z) = G
z(q),
(5.5.ii) the order of G
z(q) is less or equal to the order of G
x(q), (5.5.iii) G
x(q) G
z(q) (G
x∩ G
z)/(G
1x∩ G
1z).
If x is a vertex, then f
x(G
x∩ G
z) = G
x(q) is equivalent to z = x.
Proof. f
x(G
x∩ G
z) = G
x(q) is equivalent to G
x∩ G
1z= G
1x∩ G
1z. When this holds, we have (G
x∩ G
z)/(G
1x∩ G
1z) f
z(G
x∩ G
z) ⊂ G
z(q)
(G
x∩ G
z)/(G
1x∩ G
1z) → f
x(G
x∩ G
z) = G
x(q)
where the map is the canonical surjection. We denote by a, b, c the number of elements of G
x(q), (G
x∩ G
z)/(G
1x∩ G
1z), G
z(q). Then a ≤ b ≤ c. They are equal if and only if a = c. This proves that (5.5.ii) and (5.5.iii) are equivalent. If b = c, then G
x(q) is a quotient of G
z(q). As the relation x ∼ z given by
f
x(G
x∩ G
z) = G
x(q), f
z(G
x∩ G
z) = G
z(q)
is symetric, we deduce that G
z(q) is a quotient of G
x(q). As the groups are finite, they are isomorphic.
Hence a = b = c. This proves that (5.5.i) and (5.5.iii) are equivalent.
The last part the lemma follows from the following facts :
Let C be a chamber of A and let ∆ = {α
o, . . . , α
n} be the basis of the affine roots of G associated to C [T 1.8]. For x in the closure C of C we denote by ∆
x⊂ ∆ the proper subset of α
iwith α
i(x) = 0 and by W
xthe group generated by all reflexions s
αfor α ∈ ∆
x.
- Let x, y ∈ C and n ∈ N. Suppose that the image w of n in the affine Weyl group W is of minimal lengthin W
xwW
y. Then f
x(G
x∩ G
ny) = G
x(q) implies w∆
y⊃ ∆
x[Morris].
- x ∈ C is a vertex if and only if ∆
xhas n elements. Then y ∈ C is equal to x is and only if ∆
x= ∆
y. If y / =x then there is no element w ∈ W suchthat w∆
y⊃ ∆
x. Otherwise s
α∆
y= ∆
xwhere ∆ = ∆
y∪ { α } but α ∈ ∆
xand α/ ∈ s
α∆.
On an orbit of G the relation x ∼ z defined above is equivalent to f
x(G
x∩ G
z) = G
x(q)
because the groups G
z(q) G
x(q) are isomorphic on an orbit and (5.5). The set of g ∈ G with x ∼ gx is a disjoint union of double classes G
xnG
xfor n in some set of representatives N (x) ⊂ N. For n ∈ N (x) withimage w ∈ W and σ ∈ Mod
RG
x(q) the isomorphism class of
nσ depends only on w and we denote by
n
σ =
wσ and W (x) the image of N (x) in N . We define W (x, σ) := {w ∈ W (x),
wσ σ}. When P = G
xwe replace x by P in the notation. We deduce from (5.5) and (5.4) :
5.6 Corollary Let σ ∈ Cusp
RP (q). The U -invariants vectors of ind
GPσ is isomorphic to
⊕
w∈W(P)wσ.
As in (4.2) this implies the existence of a P -equivariant decomposition ind
GPσ = ⊕
W(P,σ)σ ⊕ W
where W has no subquotient isomorphic to σ. From (2.1) we deduce that ind
GPσ satisfies the simple criterium of almost projectivity.
6 Cuspidal representations oflevel 0
We suppose that x is a vertex (not necessarily special). The parahoric subgroup G
xis maximal , th e G-normalizer P
xwhich is the fixator of x, is an open compact mod center subgroup of G, and is the set of g ∈ G such f
x(G
x∩ G
gx) = G
x(q) by (5.5). Let Λ ∈ Irr
RP
xtrivial on G
1xand restriction to G
xidentified by f
x∗to a cuspidal representation of G
x(q).
6.1 Proposition (ind
GPxΛ)
G1xis an R-representation of P
xisomorphic to Λ.
Proof. The functor
Λ → (ind
GPxΛ)
G1x: Mod
RP
x→ Mod
RP
x/G
1xis a direct sum
(ind
GPxΛ)
G1x= ⊕
g∈Px\G/PxF
gG(Λ) of functors F
gG: Mod
RP
x→ Mod
RP
x/G
1xcomposite of
- the invariants by P
x∩ g
−1G
1xg
−1- the conjugation by g
- the induction from (P
x∩ gP
xg
−1)/(G
1x∩ gP
xg
−1) to P
x/G
1x.
The cuspidality of Λ implies that if the P
x∩g
−1G
1xg
−1-invariants vectors are not 0 then G
x∩g
−1G
1xg
−1= G
1x∩ g
−1G
1xg
−1, i.e. g ∈ P
x. Then (ind
GPxΛ)
G1x= F
1G(Λ) = Λ and the proposition is proved.
The simple criterium for irreduciblity (3.2) is easily deduced from this proposition. By adjunction (6.1) implies
End
RG(ind
GPxΛ) = R and if π ∈ Irr
RG is a quotient of ind
GPxΛ (6.1) implies
π
G1xΛ.
In Mod
RP
x, π
G1xis a direct factor of π|
Pxand the simple criterium for irreduciblity is satisfied. Hence ind
GPxΛ is irreducible.
7 Bushnell-Kutzko simple types in GL(n, F )
We suppose G = GL
F(V ) GL(n, F ) for an F -vector space V of dimension n.
We fix β ∈ GL
F(V ) suchthat - the algebra E = F(β) is a field
- k
F(β) < 0 [Bushnell-Kutzko 1.4.5, 1.4.13, 1.4.15] [Bushnell-Kutzko2 4.1]. We will not use k
F(β) and we do not recall its definition.
We denote by q
Ethe number of elements of the residual field of E and by B
∗= GL
E(V ) the centralizer of E
∗in GL
F(V ). If d[E : F] = n then B
∗GL(d, E).
We consider the Bushnell Kutzko group associated associated to a “defining sequence” for β and a parahoric subgroup P in B
∗[Bushnell-Kutzko 2.4.2, 3.1.8, 3.1.14]. The group does not depend on the defining sequence [Bushnell-Kutzko 3.1.9 (v)] and is denoted J = J (β, P ). The Bushnell-Kutzko groups, called the BK-groups, have the following properties [Bushnell-Kutzko 1.6.1]:
7.1 Lemma Let P be a parahoric subgroup of B
∗with the canonical exact sequence
1 // U // P
fP// P (q
E) // 1.
The BK-group J = J(β, P ) is an open compact subgroup of G normalized by E
∗with pro-p radical J
1, with
(1) J = J
1P, J ∩ B
∗= P , J
1∩ B
∗= U .
The canonical surjection f
Jgiven by the lemma
1 // J
1// J
fJ// P(q
E) // 1.
has the property
(2) f
J(J ∩ K) = f
P( P ∩ Q )
because f
J(J ∩ K) = f
P(J ∩ K ∩B
∗) and J ∩ K ∩ B
∗= P ∩Q. By (2), the properties of parahoric subgroups
of B
∗seen in the paragraphs 5 and 6 transfer to the Bushnell-Kutzko groups associated to β.
7.2 We consider irreducible Bushnell Kutzko (or BK) representations η
J∈ Irr
RJ
1, κ
J∈ Irr
RJ attached to a fixed endo-class Θ [Bushnell-Kutzko 5.1.8, 5.2.1 and 5.2.2] [Bushnell-Kutzko2 4.3]. We do not recall the definitions. The BK-representations satisfy :
(3) The restriction of κ
Jto J
1is η
Jand κ
Jis normalised by E
∗.
When η
Jis fixed there is some choice for κ
Jbut only by multiplication by a character trivial on J
1and normalized by E
∗. We consider different functors. We use the definition of the paragraph 5 for the parahoric subgroup P of B
∗.
7.3 Definition a) The κ
J-inflation functor
σ → κ
J⊗ f
P∗σ : Mod
RP (q
E) → Mod
RJ
which induces an equivalence of categories between M od
RP (q
E) and the η
J-isotypic R-representations of J . b) The compact κ
J-induction functor :
ind
GκJ: Mod
RP (q
E) → Mod
RG
composite of the κ
J-inflation Mod
RP(q
E) → Mod
RJ by the compact induction ind
GJ: Mod
RJ → Mod
RG.
c) The κ
J-restriction functor
res
GκJ: Mod
RG → Mod
RP (q
E) π → σ
given by the η
J-invariants π → π
ηJ= κ
J⊗ f
P∗σ : Mod
RG → Mod
RJ followed by the inverse of the κ
J-inflation.
d) When K is the BK-group J(β,Q) attached to another parahoric Q of B
∗, the functor T
κGK,κJ= res
GκK◦ ind
GκJ: Mod
RP (q
E) → Mod
RQ(q
E).
We will prove (7.7) that the functor T
κGK,κJis equal to the functor T
Q(qB∗E),P(qE)
associated to the parahoric subgroups P , Q of B
∗(5.1) and already described (5.3) (5.4).
7.4 Remark For g ∈ G, we have J (gβg
−1, g P g
−1) = gJ(β, P )g
−1.
This is not immediately apparent that the elaborate definition of J is G-equivariant.
Let L = (L
i)
i∈Zbe a strictly decreasing periodic lattice chain of O
E-modules in V suchthat P is the set of f ∈ GL
E(V ) with f (L
i) ⊂ L
ifor all i ∈ Z. We consider
- the period e of L (the smallest integer n suchthat L
i+n= p
EL
ifor all i ∈ Z), - th e L -valuation v of β (the biggest integer n suchthat βL
i⊂ L
i+nfor all i ∈ Z).
- the herditary order End
oOFL of End
F(V ) associated to L seen as a chain of O
F-modules (the F - endomorphisms f suchthat f (L
i) ⊂ L
ifor all i ∈ Z).
- s : End
FV → End
EV the tame corestriction map relative to E/F [Bushnell-Kutzko 1.3.3].
We take g ∈ G and we replace (β, L) by (gβg
−1, gL). It is easy to see what happens to the various
objects and numbers introduced above. First we see that (P , End
oFL) is replaced by gP g
−1, g(End
oFL)g
−1,
that the period e and the valuation v do not change. Then looking at the definition of k
F(β ) [Bushnell-
Kutzko 1.3.5], we see that it does not change, and finally if c
g: End
E(V ) → End
gEg−1(V ) is the natural
isomorphism f (x) → gf(g
−1xg)g
−1, we see on the definition [Bushnell-Kutzko 1.3.3] that c
g◦ s is a tame
corestriction map relative to gEg
−1/F .
These remarks imply that a defining sequence ( A
i, n, r
i, γ
i) for ( A , n, r, β) gives a defining sequence (g A
ig
−1, n, r
i, gγ
ig
−1) for (g A g
−1, n, r, gβg
−1) with the definitions [Bushnell-Kutzko 2.4.2]. We deduce from [Bushnell-Kutzko 3.1.8] that J (gβg
−1, g P g
−1) = gJ(β, P )g
−1.
It seems to me that the construction of η
Jand κ
Jis G-equivariant [Bushnell-Kutzko 3.5, 5.7] but I didnt check all the details.
7.5 The three κ
J-functors (7.3) do not determine the group J neither the representation κ
J. In fact we will not keep (J, κ
J).
Let P
maxbe a maximal parahoric subgroup of B
∗and let P
minbe a minimal parahoric subgroup of B
∗contained in P
max. We suppose P
min⊂ P ⊂ P
max. Let η
max∈ Irr
RJ
max1, κ
max∈ Irr
RJ
maxbe the BK-representations associated to (β, Θ). We consider
J
= J
max1P , J
1= J
max1U , η
J= κ
max|
J1, κ
J= κ
max|
JIt is clear that (1) hence (2), (3) are satisfied for (J
, J
1, η
J, κ
J).
Let L
max= (L
op
ZE) be a O
E-lattice chain in End
EV suchthat P
max= GL
oE(L
max) is the set of g ∈ GL
E(V ) with gL
o⊂ L
o. We denote GL
1F(L
max) the set of g ∈ GL
F(V ) with gL
op
iE⊂ L
op
i+1Efor all i ∈ Z. The open compact subgroup A = GL
1F(L
max)P of G has a pro-p-radical A
1= GL
1F(L
max)U and satisfies (1). By construction J, J
are contained in A and J
1= A
1∩ J, J
1= A
1∩ J
. Moreover [Bushnell-Kutzko 5.2.5] proved that
η
A:= ind
AJ11η
Jind
AJ1η
Jare isomorphic and irreducible, and one may suppose (or we twist κ
Jby a character normalised by E
∗) that κ
A:= ind
AJκ
Jind
AJκ
Jare isomorphic and irreducible. This implies :
7.6 Lemma The κ
A, κ
J, κ
J-functors are equal, the compact BK-induction and BK-restriction functors associated to κ
A, κ
J, κ
Jare equal.
Hence the functor T
κGJ,κK= T
κGJ,κKcan be computed using κ
J, κ
K. By the Mackey formula, we have a K
-equivariant decomposition (where we forget f
P∗hence σ instead of f
P∗σ)
ind
GJκ
J⊗ σ ⊕
g∈K\G/Jind
J1 maxQ
Jmax1 Q∩gJmax1 Pg−1
(κ
max⊗ σ)(g
−1?g).
We compute the η
K-isotypic part, i.e the κ
max|
Jmax1 V-isotypic part. We recall [Bushnell-Kutzko 5.1.8 page 160, 5.2.7 page 170] :
dim
RHom
J1max∩gJmax1 g−1
(η
max,
gη
max) = dim
RHom
Jmax∩gJmaxg−1(κ
max,
gκ
max) is equal to 1 for g ∈ J
max1B
∗J
max1, and is equal to 0 when g / ∈J
max1B
∗J
max1.
The terms in g / ∈ J
max1B
∗J
max1give no contribution to the η
K-isotypic part of ind
GJκ
J⊗ σ and Q\ B
∗/ P = J
max1Q\ J
max1B
∗J
max1/J
max1P
Using the property (2) for f
J(J
∩ gK
g
−1) when g ∈ B
∗we deduce withthe same proof and notations of (5.3) and (5.4) :
7.7 Proposition T
κGK,κJ⊕
g∈Q\B∗/PF
QE(q∗E)gP(qE)
T
QB(q∗E),P(qE)