Local group theory : from Frobenius to Derived Categories
Michel Brou´e
Universit´e Paris–Diderot Paris 7
September 2012
LOCAL GROUP THEORY
Feit–Thompson, 1963
If G is a non abelian simple finite group, then 2 | | G | .
Cauchy (1789–1857)
If p | | G | , there are non trivial p –subgroups in G . Sylow, 1872
The maximal p–subgroups of G are all conjugate under G .
Brauer–Fowler, 1956
There are only a finite number of isomorphism types of finite simple
groups with a prescribed type of centralizer of an involution.
Assume P ⊂ S and P ⊂ S
′. There is g ∈ G such that S
′= S
g(= g
−1Sg ), hence
P ⊂ S and
gP (= gPg
−1) ⊂ S . This is a fusion.
The Frobenius Category Frob
p(G ) :
Objects : the p–subgroups of G,
Hom(P, Q) := { g ∈ G | (
gP ⊂ Q) } /C
G(P) .
Note that Aut(P ) = N
G(P )/C
G(P ) .
Alperin’s fusion theorem (1967) says essentially that Frob
p(G ) is
known as soon as the automorphisms of some of its objects are
known.
Main question of local group theory
How much is known about G from the knowledge (up to equivalence of categories) of Frob
p(G ) ?
Well, certainly not more than G /O
p′(G ) !
(where O
p′(G ) denotes the largest normal subgroup of G of order not divisible by p)
Indeed, O
p′(G ) disappears in the Frobenius category, since, for P a p–subgroup,
O
p′(G ) ∩ N
G(P ) ⊆ C
G(P ) .
But perhaps all of G /O
p′(G ) ?
Control subgroup
Let H be a subgroup of G . The following conditions are equivalent : (i) The inclusion H ֒ → G induces an equivalence of categories
Frob
p(H) −→
∼Frob
p(G ) ,
(ii) H contains a Sylow p –subgroup of G , and if P is a p–subgroup of H and g is an element of G such that
gP ⊆ H, then there is h ∈ H and z ∈ C
G(P ) such that g = hz.
If the preceding conditions are satisfied, we say that H controls
p–fusion in G , or that H is a control subgroup in G .
The first question may now be reformulated as follows :
If H controls p–fusion in G , does the inclusion H ֒ → G induce an isomorphism
H/O
p′(H) −→
∼G /O
p′(G ) ? In other words, do we have
G = HO
p′(G ) ?
Frobenius theorem, 1905
If a Sylow p–subgroup S of G controls p–fusion in G , then the inclusion induces an isomorphism S ≃ G/O
p′(G ).
p–solvable groups, ?
Assume that G is p–solvable. If H controls p-fusion in G , then the inclusion induces an isomorphism H/O
p′(H) ≃ G /O
p′(G ).
Z
p∗–theorem (Glauberman, 1966 for p = 2, Classification for other primes)
Assume that H = C
G(P ) where P is a p–subgroup of G . If H controls p-fusion in G , then the inclusion induces an
isomorphism H/O
p′(H) ≃ G /O
p′(G ).
But
Burnside (1852–1927)
Assume that a Sylow p –subgroup S of G is abelian. Set
H := N
G(S ). Then H controls p-fusion in G .
Consider the Monster , a finite simple group of order
2
46· 3
20· 5
9· 7
6· 11
2· 13
3· 17 · 19 · 23 · 29 · 31 · 41 · 47 · 59 · 71 ≃ 8.10
53. (predicted in 1973 by Fischer and Griess, constructed in 1980 by Griess, proved to be unique by Thompson)
and the normalizer H of one of its Sylow 11–subgroups, a group of order 72600, isomorphic to (C
11× C
11) ⋊ (C
5× SL
2(5)) (here we denote by C
mthe cyclic group of order m).
Here we have G 6 = HO
11′(G ) since G is simple.
Remark : one may think of more elementary examples...
LOCAL REPRESENTATION THEORY
Let K be a finite extension of the field of p –adic numbers Q
pwhich contains the | G | -th roots of unity. Let O be the ring of integers of K over Z
p, with maximal ideal p and residue field k := O /p.
K 2 R O
dd ■■
■■ // // k = O /p Q
p?
OO
Z
p1 Q
cc ●● ● ?
OO // // F
p= Z
p/pZ
p?
OO
Block decomposition
O G = L
B (indecomposable algebra)
↓ ↓
kG = L
kB (indecomposable algebra)
The augmentation map O G → O factorizes through a unique block of O G called the principal block and denoted by B
p(G ).
O G ● ● ● ● ● ● ● ● // ● B ● ##
p(G )
O
Set e
p′(G ) :=
|O 1p′(G)|
P
s∈Op′(G)
s. Then e
p′(G ) is a central idempotent of
O G and O Ge
p′(G ) is a product of blocks containing the principal block
B
p(G ).
Factorisation and principal block
If H is a subgroup of G , the following assertions are equivalent (i) G = HO
p′(G) .
(ii) The map Res
GHinduces an isomorphism from O Ge
p′(G ) onto O He
p′(H).
In particular, in that case, the map Res
GHinduces an isomorphism from B
p(G ) onto B
p(H).
== Let us re-examine the counterexamples to factorization coming from Burnside’s theorem.
Assume that a Sylow p–subgroup S of G is abelian, let H := N
G(S) be its normalizer.
Even if G 6 = H O
p′(G), it appears that there is some connection
between the (representation theory of) B
p(G ) and B
p(H).
First of all, there are many examples where there is no factorization, but where the algebras are Morita equivalent — but then not through the Res
GHfunctor.
A kind of generic example :
G = GL
n(q )
B H
❙❙❙ ❙❙❙
❙❙❙ ❙❙❙
❙❙❙ ❙❙❙ ❙❙
U
q q q q q q
q q T = S × T
p′❙❙❙ ❙❙❙ ❙❙❙ ❙
q q q q q q q
1
▼▼ ▼▼ ▼▼
▼▼ ▼
❦ ❦
❦ ❦
❦ ❦
❦ ❦
❦ ❦
p ∤ q , p > n, S p-Sylow
H := N
G(T ) = N
G(S)
←− H/T = S
nWe certainly have
G 6 = HO
p′(G ) .
But the principal block algebras of G and H respectively are Morita equivalent.
There exist M and N, respectively a O G –module– O H and a O H–module– O G with the following properties :
(With appropriate cutting by the principal block idempotents) M ⊗
OHN ≃ B
p(G ) as O G –modules– O G N ⊗
OGM ≃ B
p(H) as O H–modules– O H Viewed as an O G –module– O T , we have
M ≃ O (G /U ) ,
i.e., the functor M ⊗
OT? is the Harish–Chandra induction.
SOME NUMERICAL MIRACLES
Let us consider the case G = A
5and p = 2. Then we have H ≃ A
4. Remark : on a larger screen, we might as well consider the above case of the Monster and of the prime p = 11.
Table : Character table of A
5(1) (2) (3) (5) (5’)
1 1 1 1 1 1
χ
44 0 1 -1 -1
χ
55 1 -1 0 0
χ
33 -1 0 (1 + √
5)/2 (1 − √ 5)/2
χ
′33 -1 0 (1 − √
5)/2 (1 + √
5)/2
Table : Character table of B
2(A
5)
(1) (2) (5) (5’) (3)
1 1 1 1 1 1
χ
55 1 0 0 -1
χ
33 -1 (1 + √
5)/2 (1 − √
5)/2 0
χ
′33 -1 (1 − √
5)/2 (1 + √
5)/2 0
Table : Character table of A
4(1) (2) (3) (3’)
1 1 1 1 1
− α
3-3 1 0 0
− α
1-1 -1 (1 + √
− 3)/2 (1 − √
− 3)/2
− α
′1-1 -1 (1 − √
− 3)/2 (1 + √
− 3)/2
ABELIAN SYLOW CONJECTURE
Assume that a Sylow p–subgroup S of G is abelian, let H := N
G(S) be its normalizer.
(ASC) :
The algebras B
p(G ) and B
p(H) are derived equivalent.
Which means :
There exist M and N, respectively a bounded complex of B
p(G )–modules–B
p(H) and a bounded complex of B
p(H)–modules–B
p(G ) with the following properties :
M ⊗
Bp(H)N ≃ B
p(G ) as complexes of B
p(G )–modules–B
p(G )
N ⊗
Bp(G)M ≃ B
p(H) as complexes of B
p(H)–module–B
p(H)
(Strong ASC) :
They are Rickard equivalent, that is, derived equivalent in a way which is compatible with the equivalence of Frobenius
categories
Which means : There is a G –equivariant collection of derived equivalences
{ E (P) : D
b(B
p(C
G(P)) −→ D
∼ b(B
p(C
H(P )) }
P⊆Scompatible with Brauer morphisms.
RICKARD’S EXPLANATION FOR A
5G := A
5H := N
G(S
2), (S
2a Sylow 2–subgroup of G ) View B
2(G ) as acted on as follows
B2(G)
B
2(G )
B2(H)
B2(G)
IB
2(G )
B2(H)
:= kernel of augmentation map B
2(G ) ։ O .
C := a projective cover of the bimodule IB
2(G ).
Thus we have
C
$$ ■ ■ ■ ■ ■
{0}
//
IB2(G)//
B2(G)
//
{0}{0}
set
Γ
2:= { 0 } → C → B
2(G ) → { 0 }
where B
2(G ) is in degree 0 (and C in degree − 1).
We have homotopy equivalences : Γ
2⊗
OH
Γ
∗2≃ B
2(G ) as complexes of (B
2(G ), B
2(G ))–bimodules, Γ
∗2⊗
OG
Γ
2≃ B
2(H) as complexes of (B
2(H), B
2(H))–bimodules.
Case of finite reductive groups
Usual notation
G is a connected reductive algebraic group over F
q, with Weyl group W , endowed with a Frobenius endomorphism F defining a rational structure over F
q.
Here we assume that ( G , F ) is split.
G := G
Fis the corresponding finite reductive group, with order
| G | = q
NY
d>0
Φ
d(q)
a(d)a polynomial which depends only on the reflection representation of W on Q ⊗ Y ( T ).
Indeed, that polynomial is
q
Pidi−1Y
i
(q
di− 1) .
Sylow Φ
d–subgroups, d –cyclotomic Weyl group
There exists a rational torus S
dof G , unique up to G –conjugation, such that
| S
d| = | S
Fd
| = Φ
d(q)
a(d). Set L
d:= C
G( S
d) and N
d:= N
G( S
d) = N
G( L
d)
W
d:= N
d/L
dis a true finite group, a complex reflection group in its action on C ⊗ Y ( S
d).
= This is the d –cyclotomic Weyl group of the finite reductive group G .
Example : For G = GL
n(q ) and n = dm + r (r < d ), then
L
d= GL
1(q
d)
m× GL
r(q) , W
d= µ
d≀ S
mThe Sylow ℓ–subgroups and their normalizers ℓ a prime number, prime to q, ℓ | | G | , ℓ ∤ | W |
= ⇒ There exists one d (a(d ) > 0) such that ℓ | Φ
d(q), and the Sylow ℓ–subgroup S
ℓof S
dis a Sylow of G .
L
d= C
G(S
ℓ) and N
d= N
ℓ= N
G(S
ℓ) : N
ℓ WdL
d1 Since the “local” block is
B
ℓ(Z
ℓN
ℓ) ≃ Z
ℓ[S
ℓ⋊ W
d] our conjecture reduces to
Conjecture
D
b(B
ℓ(Z
ℓG )) ≃ D
b(Z
ℓ[S
ℓ⋊ W
d])
Role of Deligne–Lusztig varieties
Let P be a parabolic subgroup with Levi subgroup L
d, and with unipotent radical U .
Note that P is never rational if d 6 = 1.
The Deligne–Lusztig variety is V
P:=
G
{ g U ∈ G / U | g U ∩ F (g U ) 6 = ∅}
L
dhence defines an object
RΓ
c( V
P, Z
ℓ) ∈ D
b(
ZℓGmod
ZℓLd
) Conjecture
There is a choice of U such that
1
RΓ
c( V
P, Z
ℓ)
0is a Rickard complex between B
ℓ(Z
ℓG ) and its
commuting algebra C ( U ).
The case where d = 1
If d = 1,
S
d= T = L
dand W
d= W
V
B= G/U and RΓ
c( V
P, Z
ℓ) = Z
ℓ(G /U )
Z
ℓG
Z
ℓ(G /U )
C (U) where
1
C (U ) ≃ Z
ℓT .Z
ℓH (W , q)
2
Q
ℓH (W , q) ≃ Q
ℓW
The unipotent part
Extend the scalar to Q
ℓ=: K ⇒ Get into a semisimple situation RΓ
c( V ( U ), Z
ℓ) becomes
H
c•( V ( U ), K ) := M
i
H
ci( V ( U ), K )
Replace V ( U ) by V ( U )
un:= V ( U )/L
d⇒
Only unipotent characters of G are involved Semisimplified unipotent conjecture
1
The different H
ci( V ( U )
un, K ) are disjoint as KG–modules,
2
H ( U ) := End
KGH
c•( V ( U )
un, K ) ≃ KW
dCase where d is regular
L
d=: T
dis a torus ⇐⇒ d is a regular number for W The set of tori L
dis a single orbit of rational maximal tori under G , hence corresponds to a conjugacy class of W .
For w in that class, we have W
d≃ C
W(w ).
The choice of U corresponds to the choice of an element w in that class.
We then have
V ( U
w)
un= X
w:= { B ∈ B | B →
wF ( B ) } B is the variety of all Borel subgroups of G
The orbits of G on B × B are canonically in bijection with W
and we write B →
wB
′if the orbit of ( B , B
′) corresponds to w .
Relevance of the braid groups
Notation
V := C ⊗ Y ( T ) acted on by W , A := set of reflecting hyperplanes of W V
reg:= V − S
H∈A
H B
W:= Π
1(V
reg/W , x
0)
“Section” W → B
W, w 7→ w , since If W =< S | ststs . . .
| {z }
ms,t factors
= tstst . . .
| {z }
ms,t factors
, s
2= t
2= 1 >
then B
W=< S | ststs . . .
| {z }
ms,t factors
= tstst . . .
| {z }
ms,t factors
>
π := t 7→ e
2iπtx
0= ⇒ π ∈ ZB
W2 h
A theorem of Deligne
Theorem (Deligne)
Whenever b ∈ B
W+there is a well defined variety X
(F)b
) such that X
(Fw)= X
(Fw).
For b = w
1w
2· · · w
nwe have X
(F)b
= { ( B
0, B
1, . . . , B
n) | B
0→
w1B
1→ · · ·
w2→
wnB
nand B
n= F ( B
0) }
The variety X
πX
π= { ( B
0, B
1, B
2) | B
0→
w0B
1→
w0B
2and B
2= F ( B
0}
= { ( B
0, B
1, . . . , B
h) | B
0→
cB
1→ · · ·
c→
cB
hand B
h= F ( B
0) } The (opposite) monoid B
W+acts on X
π: For w ∈ B
Wred, and
π = w b = b w ,
if B →
wB
0→
bF (B )
D
w: ( B , B
0, B
1= F ( B )) 7→ ( B
0, F ( B ), F ( B
0)) Hence B
Wacts on H
c•( X
π)
Proposition : The action of B
Won H
c•( X
π) factorizes through the (ordinary) Hecke algebra H (W ).
Conjecture :
End
KGH
c•( X
π) = H (W )
Relevance of roots of π
Proposition
d regular for W ⇐⇒ there exists w ∈ B
W+such that w
d= π.
Application
1
X
(Fw)embeds into X
(Fπd):
X
(Fw)֒ → X
(Fπd)B 7→ ( B , F ( B ), . . . , F
d( B ))
2
Its image is
{ x ∈ X
(Fπd)| D
w( x ) = F ( x ) }
3
C
B+W
( w ) acts on X
(Fw).
Belief
A good choice for U
wis : w a d –th root of π.
Theorem (David Bessis) There is a natural isomorphism
B
CW(w)−→
∼C
BW( w ) From which follows :
Theorem
The braid group B
CW(w)of the complex reflections group C
W(w ) acts on H
c•( X
w).
Conjecture
The braid group B
CW(w)acts on H
c•( X
w) through a d –cyclotomic Hecke
algebra H
W(w ).
d –cyclotomic Hecke algebras
A d –cyclotomic Hecke algebra for C
W(w ) is in particular an image of the group algebra of the braid group B
CW(w),
a deformation in one parameter q of the group algebra of C
W(w ), which specializes to that group algebra when q becomes e
2πi/dExamples :
The ordinary Hecke algebra H (W ) is 1–cyclotomic, Case where W = S
6, d = 3 :
C
W(w ) = B
2(3) = µ
3≀ S
2←→
s
3
t 2
H
W(w ) =
S , T ;
STST = TSTS
(S − 1)(S − q)(S − q
2) = 0 (T − q
3)(T + 1) = 0
W = D
4, d = 4, C
W(w ) = G (4, 2, 2) ←→
s2♥ 2 t
2uH
W(w ) =
S, T , U ;
( STU = TUS = UST (S − q
2)(S − 1) = 0
)
Let us summarize
1
ℓ d , d regular, i.e., L
d= T
w, w
d= π, V ( U
w)/L
d= X
w2
End
KGH
c•( X
w) ≃ H
W(w )
3
Z
ℓH
W(w ) −→
∼Z
ℓC
W(w )
4
End
ZℓGRΓ
c( V ( U
w), Z
ℓ) ≃ Z
ℓ(T
w)
ℓ· End
ZℓGRΓ
c( X
w, Z
ℓ) ≃ B
ℓ(Z
ℓN
ℓ)
What is really proven today
Everything
if d = 1 (Puig),
for G = GL
2(q) (Rouquier), SL
2(q) (cf. a book by Bonnaf´e) for G = GL
n(q) and d = n (Bonnaf´e–Rouquier)
About : End
KGH
c•( X
w) ≃ H
W(w ) ? All H
W(w ) are known, all cases (Malle) Assertion End
KGH
c•( X
w) ≃ H
W(w ) known for
d =h(Lusztig),
d = 2 (Lusztig, Digne–Michel), small rank GL,
d = 4 forD4(q) (Digne–Michel).