In this book Professor Lusztig solves an interesting problem by entirely new methods: specifically, the use of cohomolog
156 3 5MB
English Pages 104 [106] Year 2016
Table of contents :
TABLE OF CONTENTS
INTRODUCTION
CHAPTER 1 Partially Ordered Sets and Homology
CHAPTER 2 The Affine Steinberg Module
CHAPTER 3 The Distinguished Discrete Series Module
CHAPTER 4 The Character of D(V) and the Eigenvalue A(V)
CHAPTER 5 The Brauer Lifting
INDEX
Annals of Mathematics Studies N um ber 81
T H E D IS C R E T E SERIES OF G L nO V E R A F IN IT E FIE LD
BY
G E O R G E L U S Z T IG
PR IN C E T O N
U N IV E R SIT Y
PRESS
AND U N IV E R SIT Y
OF
P R IN C E T O N ,
TOKYO
NEW 1974
PRESS
JERSEY
C opyright ©
1974 by P r in c e t o n U n i v e r s i t y Press ALL
RIGHTS
RESERVED
Pub lis he d in Japan e x c l u s i v e l y by U n i v e r s i t y o f T o k y o Press; In o t h e r parts o f the w o r l d by Pr i n c e t o n U n i v e r s i t y Press
Pr in t ed in the U n i t e d States o f A m e r i c a by P r i n c e t o n U n i v e r s i t y Press, P r i n c e t o n , N e w Jersey
L i b r a r y o f C o n g r e s s C a t a l o g i n g in P u b li c a t i o n da ta w il l be f o u n d on the last pr in te d pa g e o f this b o o k
T A B L E OF C O N T E N T S
I N T R O D U C T IO N ............................................................................................
3
CHAPTER 1 Partially Ordered Sets and Homology ..................................................
5
CHAPTER 2 The Affine Steinberg M odule...................................................................
30
CHAPTER 3 The Distinguished Discrete Series Module ........................................... 43 CHAPTER 4 The Character of D ( V ) and the Eigenvalue A (V ) ............................... 66 CHAPTER 5 The Brauer L i f t i n g .................................................................................... 80 I N D E X ................................................................................................................ 99
v
T he Discrete Series of GL„ Over A Finite Field
T H E D I S C R E T E SERIES O F G L n O V E R A F I N I T E F I E L D George L usztig
IN T R O D U C T I O N Since the fundamental work of Green [5] it has become clear that the central role in the ordinary representation theory of the general linear group
G L n(Fq)
sentations.
over a finite field, is played by the discrete series repre
In this work we give an explicit construction of one distin
guished member,
D (V ),
of the discrete series of G L n(Fq)
(here
the n-dimensional F-vector space on which
G L n(F^)
p-adic representation; more precisely
is a free module of rank
D (V )
(q —1 ) (q 2—1) ••• (q 11 1—1) over the ring of Witt vectors To construct cia l complex
X
D (V ),
mand (over Wp )
X
of
In fact
D (V )
T : A ( V ) -> A ( V )
additional homogeneity condition.
is
This is a
F. of a simpli-
is made out of affine flags in V
It turns out that
in A (V ) .
a certain endomorphism
WF
we consider the top-homology A ( V )
associated to V;
which are away from 0.
acts).
V
D (V )
is naturally a direct sum
is defined as an eigenspace of defined geometrically, with an
The reduction modulo p of D ( V )
is
a modular representation of G L ( V )
which can be described as the top
homology of the Tits complex of
with values in a certain non-constant
coefficient system.
V
This can be used to determine the character of D ( V )
on semisimple elements of
G L (V ).
To deal with the non-semisimple e l e
ments we show that the restriction of
D (V )
to any proper parabolic sub
group splits naturally in a tensor product of a module dim V,
D ( V ') , dim V '
3.
This can be proved by a method of Folkman ([2], Theorem 4.1) as follows. Consider the nerve
Nn
of the maximal covering (se e 1.5) of X(Sn ).
is enough to prove that H ^ N j j j A ) = 0 , Note that the k-simplices of N n hyperplanes in V,
0 < i < n—2 and
H q( N j p A ) = A.
are the sets consisting of k+1
whose intersection is non-zero.
no restriction, hence the ( n - 2 ) - s k e l e t o n of
It
linear
If k < n—2 this is
N jj is the same as the
( n_2)-skeleton of a certain standard simplex and the result follows. We wish to prove analogous results for the other sets (or direct s y s tems) associated to V.
1.9
TH EO REM .
H 0(Sj; A ) = A
H i(SI ; A ) - 0
for all
where I - dim E > 2.
i such that 0 < i < £ — 1 and
P A R T IA L L Y ORDERED SETS AND HOMOLOGY
Proof.
13
Note that the homology is not changed if we change temporarily
the partial order in Sj constant).
to the opposite one (the coefficient system is
U sin g the acyclic covering lemma for the maximal covering
with respect to the opposite order we find that it is enough to prove that H i(N j; A ) = 0 ,
u < i < £ - 1.
H0(Nj; A ) s A,
where
complex whose k-simplices are those subsets of
Nj
E
is the simplicial
which have
elements and which span an affine subspace different from E. this is always the case, hence
Nj
has the same
(k+1)
If
k < £— 1
(£—l)-skeleton as a
certain standard simplex and the theorem is proved.
1.10 T HEO REM . H 0(Sm ; A ) ^ A,
H^(Sm ; A ) = 0 for all
i such that
0 < i < n—1 and
n = dim V > 2.
where
Just as in the previous proof it is enough to prove the following.
P R O P O S IT IO N .
Nni
Let
are those subsets of V
be the simplicial complex whose k-simplices
which have
affine subspace not containing
0.
(k+1)
elements and which span an
Then H - ( N i n ; A ) =- 0 for all
i such
that 0 < i < n—1 and H 0( N III; A ) ^ A, ( n > 2 ) .
I am indebted to M. Kervaire for supplying for me the following.
First observe that
Proof of the proposition.
follows from the fact that given we can find
v3 € V
are not collinear.
such that
Let ?
v x £ v2 in V 0,
and
Let
(vq, Vj , - • , v^)
?£ C ?
[
]
0 ,v 2,v 3
which span an a ^ xec^ ^a s is
as follows:
?£ con
such that < vQ, V j , •••, v ^ > ^ 0
[ v 0 , v l f ••*,vk] n ( [ Q 1 , Q 2 , - ^ Q ^ l + Qf) - 0.
the affine span, and
V,
Q i > Q 2 ' * * ’ ’ Qn
of V
This
such that v 1 ^ 0 , v 2 ^ 0 ,
, v3 are not collinear and
For any £, 1 < £ < n define a subset
sists of all subsets
is connected.
be the set of all subsets of
affine subspace not containing 0. for V.
N jn
denotes the linear span.
Here
denotes
THE D ISC R E TE SERIES OF G L n OVER A F IN IT E F IE L D
14
We shall consider alternate simplicial chains on N j n . Let C 'k ^ III'A )
be the set of k-chains.
Given a k-chain
u,
define its sup
port as
q. supp u - l ( v 0 , vx , •••, vk) 6 J |u(v0, v j , •**,vk) £ 0| .
Define a map
T£ : ®u6Ck^NIII; A ) I SUPP u C
^ Ck+1^NIII; A )
(1 < I < n) by the formula: (T£u)(Q(?, v0 , v l f •••, v k) -
if (vq, v lt -
, v k) e ?£.
On sequences which are permutations of
(Q £, v 0 , vi , ’ “ » vk^ define other sequences define
T^u
Tgu
using the alternacy conditions; on all
to be zero.
Note that ( v Q, V i , •••, v k) e ?£ implies to check
(Qg, v Q, vx , •••, vk) € 3\
It is easy
that if k > 1, d(Tj?u) + T ^ 0 u ) = u where u is inthe source of
Tj? (note that supp u C
Lem m a.
u(v0, v 1? •••, vk)
=>
U f C k( N m ; A )
Le t
supp u Cfl 5g+1 0 ...H
supp du C ?£).
be such that du = 0 and
Then there exist ux e C k( N l n ; A )
u2 f C k+i < N m ; A.) such that supp Uj C ?f>+ l U % 2 U ■ " U
and and
u — Uj = du2 (k > 1).
Proof.
We can write
u = u' — u",
where
supp u'C fPg, supp u" C 3£+ i U •••U supp z C
H
1 U ?2 U •••U ?n there exists
such that u = du2 (k > 1).
u2 f
with
k < n—2 lies in
f?1 U 3^ U •••U f?n. Otherwise, we could find a subset
( v 0, v 1 , •••, v^)
such that [ v 0 , v t , •••, v^] H ( [ Q x , •••, Q ^ ] + Q{?) ^ 0
for all
It follows by induction that Q j , Q 2 , *•*, Q n ^ tvQ, v i » "* » [v 0,v i,
( N l n ;A )
Hence to conclude the proof of the proposition
it is enough to prove that any k-simplex of N m
V
such
This contradicts the assumption that
of
£, l < 2 < n .
hence
V =
k < n—2 and the
proposition is proved.
1.11
TH EO R EM .
for all
Given a linear subspace
VC V
H^(SjV ; A ) = 0
we have
i such that 0 < i < m—1 and H q (Sj v ; A ) = A,
where
n = dim V
> dim V = m > 2.
Proo f. U sin g the
acyclic covering lemma applied to the maximal covering
we see that it is enough to prove that H 0( N i V ; A )
s
A,
where N IV
H^(NJV; A ) = 0, 0 < i < m—1 and
is the simplicial complex
defined as
follows:
The k-simplices of N IV
are precisely the sets
of
linear hyperplanes in V
such that
(k+1) N jV
(H Q fl Hj H ••• fl H ^) + V = V.
can be a ls o described as the simplicial complex whose
k-simplices are the sets
( L Q, L j ,
linear subspaces of V *
such that ( L Q+ L j + ••• + L ^ ) H
V
C V
(H qjH j ,--*,H^)
L^)
of (k+1)
is the orthogonal complement of V.
one-dimensional = 0 where
Note that N JV
nected (the proof similar to the one in 1.10 uses the fact that codim
> 2).
is con
16
THE DISCRETE SERIES OF G L n O VER A F IN IT E F IE L D
Let
be the set of all sets
S
linear subspaces of V * Q 'l, Q 2 » ' " ^ a basis.
of one dimensional
such that ( L Q + L j + ••• + L k) fl
m ^e a set
vectors in
For any I, 1 < I < m,
consists of all elements
(L q , L j , •••, L ^ )
( L Q, L j , •••, L k)
T g : SueCk( N '; A )| s u p p
Let
whose images in V * / V
define a subset
^ ([Q \ >Q 2 > *“ » Q ^ - i l + Q £+ V 1 ) --, 0.
= 0.
form
ffg C J C ' as follows:
of ?
Jg
such that (Lq-hLj f •••+ L k)
Define a map
UC
?g! -> C k+1( N ' ; A )
by the formula 0 > ) ( [ Q g ] , L 0 , L 1 ( - , L k) =
if ( L 0 , L j ,
e fF£.
Tj>u using the alternacy condition; on other
T £ u to be zero.
Note that
(L q , L j , •••, L k) e fPg = >
([Q g ], L q , L j , •••, L k) e S’ . One verifies that for T g(d u ) = u,
LEMMA.
where
Let
L j , •••, L k)
On sequences which are permutations of
([Q ^ ], L 0 , L 1, •••, L k) define sequences define
u( L q ,
k > l
we have
1).
Proof.
Write
u = u '-u ',
where
u '.u 'f C ^ jA )
and
supp u " C 9'i+l U •••U 9'm. Then z = du = du" satisfies
9\ n (?£+1 U •••U 9 m).
Put
and it is enough to prove that (T g
z) ( L q,
L j , ••*, L k) ^ 0.
L 0 = [Q g], (L 1+ -
and
Uj = T g z — u", u2 = T'f u'. supp T g z C
supp z C Then du2 = u - u ^
U ■••U 9 m. Assume that
Then we must have (up to a permutation)
( L j , •••, L k) f supp z C 3^+1 U •••U 9 m- It follows that
+ L k) n ([Q '1, - , Q g +h_ 1T Q g + h + V i ) = 0
Hence also
supp u'C 9^,
for some h, l < h < m - £ .
P A R T IA L L Y ORDERED SETS AND HOMOLOGY
([Q {>] + L j +
for some
f L k) fl ( [ Q ' j , — , Q
h, 1 < h < m—I,
17
j-j_ x ] + Q(!+h+ V -*") = 0
which implies that
L 1, L 2, . . . ,L k) f ^ + 1 U... Ui Fm
([Q j],
and the lemma is proved. Applying the lemma repeatedly we find that given that du = 0 and (N IV ; A)
u e C^CNjy; A )
supp u C fP^ U ? 2 U ••• U 3>m, there exists
such that
u = du2 (k > 1).
such
u2 €
To conclude the proof of the theorem
it is enough to observe that every k-simplex of N JV with k < m—2 lies in
U J 2 U - * * U f?m. Otherwise, we could find
k+1
lines in V *
such
that
(L0+ L 1 + - + L k) n ( [ Q i . Q a . - . Q ^ l + Qg+V1) ^ 0 for all
£, I = 1, 2, •••, m.
L q + L 1 + ••• + follows that
It follows by induction that Q i> Q 2 > * “ ’ Q m £
+ V"k Since clearly
V*
C L Q + L 1 + •••+
= L q + L 1 + •••+ L^ + V which contradicts thehypothesis
k < m—2 (m = codim V^-)
and
the theorem is proved.
Remark. This theorem contains Theorem 1.9 as a special case. V
by
V «F
and take
SI V ( V © F , V).
it
V = V C V © F (codim V = 1).
Then
Replace
S j( V ) =
The reason we have proved Theorem 1.9 separately is that
its proof is easier than that of the present theorem.
1.12
THE ORE M .
= 0
Ho(Sn ; ^ n ) s V,
where
n
for all i such that
0 < i < n—2
and
= dimV > 3.
Proof.
U sin g the acylic covering lemma applied to the maximal covering
of Sn
(see 1.5) we see that it is enough to prove that
0 < i < n-2
and
H 0( N n ; £ n ) = V
defined in 1.8 and follows.
where
Nn
is the simplicial complex
is a coefficient system over
Given a k-simplex
(H 0> H x ,
H k)
= 0,
Nn
defined as
in N n ( H i linear hyper-
18
TH E D ISCR E TE SERIES OF G L n O VER A F IN IT E F IE L D
planes in V
such that H 0 H U 1 H . •• n H k ^ 0) we have
= H 0 fl H 1 fl ••• fl H^. natural inclusions. are the sets
The connecting homomorphisms are given by the
Let
N
(H Q, H x ,
be the simplicial complex whose k-simplices
H k)
of
k linear hyperplanes in V
section is arbitrary (possibly zero). a standard simplex.
over
N
over
with value
N
( i > 0) and
i > 0.
Let
V
V
is clearly
Hj ( Nn ;
Njj
^
denote the constant coefficient system
since
N
Then clearly
is a standard simplex.
natural embedding of coefficient systems # C V is defined.
N
on the simplices of
We have then clearly
at every simplex of N.
H 0( N ; V ) = V
and
can be extended to a
such that ^ = 9u
and ^j = 0 on the other simplices. for all
Nn C N
The coefficient system
coefficient system ^
H -(N ;5 )
We have
whose inter
H ^ (N ;V )= 0
There is a
and the quotient S = V/ 3 ).
o p o s it io n
Proof. ? set
.
can be
This w ill again be similar to the proof of Proposition 1.10.
be the set of all simplices of N. C J
as the set of simplices
For any Z, 1 < I < n, ( L 0,
L k) e
Let
define a s u b
such that
P A R T IA L L Y ORDERED SETS AND HOMOLOGY
19
Lo + L i + ••■ + L k) n ([Q i ,Q 2, ••*,Q g_1] + Qg) = 0. Here Q i > Q 2 > " > Q n is a fixed basis of V * . C k( N ; S )
There is a natural notion of support for chains in
(we use alternating simplicial chains). Tg : { u e C ^ (N ; S) |supp u C 5gi
Define a map
C k+1(N; S)
by the formula ( T g u ) ( [ Q g ] , L 0 , L l f . . . , L k) = n C u C L ^ L ^ - . . , ^ ) )
L ^ ) e ffg.
if ( L q , L j ,
Onsequences which are permutations of ([Qg],
L q , L x , •••,L jc) define quences define FI : ( L
q
Tgu
Tgu using the alternacy condition; on other s e
to be zero.
+ ••• + L ^ ) * -> ([Qg] +
projection
L q
-f ••• + L ^ ) *
is direct since
We shall verify that for all d(T g u ) + Tg(du)
In fact, let
IT is the natural linear map defined as the dual of the natural
[Qg] © ( L Q + L x + ••• + L k) -> L Q + L j + ••• + L k; note that the sum
[Qg] + (L q + L x + ••• + L k)
(2)
(Here
( L Q, L t ,
c9(Tgu)(L0, L x ,
( L Q, L j , •• •, L k) e 3g.)
k > 0 we have u e C k(N; S) and
= u, where
L k) e 5g; we have
L k) = 0 ( ( T g u ) ( [ Q g ] , L Q, L x , •••, L k)) =
0 II
u
( L
q
, L ^ , • • •, L k ) =
where 0 : ([Qg] + L Q+ ••• + L k) * -> ( L Q + ••• + L k) * linear functions.
supp u C 5g .
We als o have in this c ase
u
( L
q
, L ^ , •• *, L k )
is given by restricting
T g ( d u ) ( L 0 , L 1, • • •,L k) = 0,
hence the equality (2) is checked on simplices
(L q , Lj^ , ••*, L k) e 3g.
other simplices, the right hand side of (2) is zero.
If
and [Qg] ^ L Q, L 1, •• *,L k , all terms of (2) are zero.
( L Q, L-^, ••*, L k) i $g Hence it is enough
to check that the left hand side of (2) is zero on simplices of the form ([Qg], L j ,
> ' " >L ^ -
For
20
TH E DISCR E TE SERIES OF G L n OV E R A F IN IT E F IE L D
We have i n x u(L, L t , •••, L k)
L where the sum is over all
L
such that
L ^ L 1, L 2 , •••, L ^ , ( L , L 1, •••, L ^ )
€ Jp£, and Tg(6»u)([Qg], L j , L 2 , •••, L k) = 2
112 0 2 u a . L p - . L ^
L
where
L
runs over the same set as above.
Here
2:
( q - l ) ( q 2- l )
■ (q ‘- l )
/3(1) = y ( l ) - 1.
In order to prove that j8(n) = q 1 ' 2 ' ’ ' ' h(n_1} to prove that for any
it is clearly sufficient
n > 2 we have the identity: n—1
X = q 1+ 2 + " - + ( n - l ) +
( _ l ) n—1q 1'l“2'*'"' *'Cl
1)
i= 1 X
(q n- i+1- l ) ( q n- i , 2 - l ) ; - ( q n- l ) + ( _ 1}n = 0 ( q - l ) ( q 2- i )
( q 1—l )
26
THE D ISCRETE SERIES OF G L n O VER A F IN IT E F IE L D
Now substitute: ( q n - H 1_ l ) ( q n - i » 2 _ 1 ) . , , ( q n _ 1)
(q n - i . 1_ 1} ( q n-i> 2 _ 1} ... m + 1 > m > 1.
O bse rv e that q i ( n _ m- l ) ( q m—i + 2 _ 1 ) ( q m - i i
3 _ 1 } ...
( q m + l _ 1)
- q i(n—m) ( q m ii 1 _ f ) ( q m—if2 _ f ) •.. ( q m—1) = q i ( n- m- 1) ( q m- i+2- l ) ( q m- i43- l ) - - ( q m- l ) ( q i - l ) We have
m+1
e"(n,nu 1) - £"(n,m) = ^ 1=1
( —l ) i+1 (q—1 ) (q2—1)
(q 11 * X- l )
.
29
P A R T IA L L Y ORDERED SETS AND HOMOLOGY
x q i(n—m - l ) ( q m—i 42- l ) ( g m- i 43—l ) - - - ( q m- l ) ( q - l ) ( q 2- l ) - - - ( q i- l )
,
(_ ! ) • " « l ( q_ i ) ( q 2
( q n—m—2 _ 1 ) q n—m - l
= q n - m - l ( ( q _ 1 ) ( q 2 _ 1 ) . . . ( q n - 2 _ 1) _ e ' (n_ l j m ) ) .
This implies clearly the desired formula by induction on formula is obvious for
m - 1 .)
m m.
(The
CHAPTER 2 TH E A F F I N E S T E IN B E R G M O D U LE
2.1.
Let
E
be an affine space of dimension I > 1 over a finite field
with q elements.
The affine Steinberg module associated to E
definition the free A-module A a ( E ) E
(see 1.14).
Let
in
E (dim E- i).
fied with the set of all functions
is by
defined by using the affine flags in
F l a g ( E ) be the set of all complete
E( E q C E 1 C •- - C E p _ j )
F
affine flags
Then A a ( E )
u : F l a g ( E ) -> A,
can be
identi
satisfying the “ cycle
condition” ^ ( E E .
o
CE
j
C - C E ^ C E
jCE.
,
C -C E ^ ) = 0
1
for any given
i (0 < i < f —1)
of dimension
0,1, •••, i—1, i+1, •••, I —1 (this sum has
and
q + 1 terms for
and given
E QC E x C ••• C E - ^ C E^+1 C ••• C Ej?
q terms for
i= 0
0 < i < t—1).
Let
A ff (E )
be the groupof all affine
acts on
A A (E )
by the formula
isomorphisms t : E ? E.
A ff (E )
( t u ) ( E 0 C E 1 C . . . C E £ _ 1) = u ( t - 1( E 0) c r 1( E 1) C - - - C t - 1( E f _ 1))
where
t e A f f (E )
and
u € A a (E ).
Given a flag e ^ ( E Q C E x C ••• C E g _ x) we denote by all
t € A f f (E )
B e the group
such that t E Q = E Q, tEi = E i , •••, tE n- l = ^ n - l *
^ave
the following
THEOREM.
Assume that A
a complete affine flag in E. scalar) function
is a field of characteristic zero.
L et
E be
Then there exists a unique (up to a non-zero
u : F l a g ( E ) -> A
such that 30
THE A F F IN E ST E IN B E R G M ODULE
(i)
u£ 0
(ii)
tu = u for all
(iii)
u£ A a (E ).
31
t e B£
The proof will be given in 2.3.
COROLLARY.
If
A
is a field of characteristic zero,
A a (E )
is an
irreducible Aff (E )-m o d u le .
This follows by applying Frobenius duality to the Aff(E)-module
A a (E )
and the unit representation of B £ .
2.2
Bruhat decomposition in the affine case Let
F lag(V )
V
be an n-dimensional vector space over
be the set of all complete linear flags
in V(dim V -= i). Given ^ F l a g V , •••, tVn_ i = Vn_ i i. The group the formula
a? =
(Vi C V2 C •• -C Vn_ i )
let B^ = { t e GL(V)| tVj = Vj , tV2= V2 ,
G L (V )
acts transitively on F l a g ( V ) by
t(Vj C V2 C •••C V'n_ 1) = (tV, C tV2 C •••C tVn_ 1).
not transitive when restricted to B^,. The orbits of described by Bruhat’s theorem. Given
V 2 n Vi92 l ^ V 2 0 Vi? 2 (i2 Here we use the convention:
n)
B^
This action is on F l a g ( V )
are
Let us recall the content of this theorem. F la g (V )
= ( i j , i2 , •••, in) of (1, 2,
that
F(n > 2) and let
we define a permutation b ( V )
by the formulae: V n n Vi n- i
VQ = 0, VR =
H
^
H Vj ,
^ V n 0 Vi n M H’V V i Bruhat’s theorem states
= V.
are in the same B^,-orbit if and only if b ( V ) = b ( V ' )
the map b : {B^-orbitsi -» Spermutations of 1,2, ■••.n!
and that
is bijective.
We need an extension of this result to the affine case.
Such an exten
sion has been found by Solomon [12] but we shall need a somewhat different approach.
The affine space
E
can be regarded as an affine hyperplane
not containing the origin in an (£fl)-dimensional vector space
V.
be the unique linear hyperplane in V
Note that
which is parallel to E.
Let
H
THE DISCRETE SERIES OF GL n OVER A FINITE FIELD
32
A ff(E )
can be regarded as the set of all
Given e = ( E g C E j C ■• ■ C E g _ j ) B £ on F l a g ( E ) .
(Note that
t e G L (V )
such that t (E ) = E.
our problem is to classify the orbits of
A ff(E )
acts transitively on F l a g ( E ) . )
£ gives rise to a complete linear flag [e ] = ( [ E Q[ C f E j l C - ’ -C [ E £ _ 1] ) in V and to a complete linear flag [ s ] H = ( [ E j ] fl HC [ E 2] H HC- - C [ E g _ 1] fl H) in H.
(Here
[E j]
is the linear span of E j
( E q C E ' C •••CE/(. _ 1) € F l a g ( E ) .
in V . )
Let
s' =
s' gives rise similarly to complete linear
[s'], [ e ' l H in V and H. Define a permutation (iQ, ij , •■■, ip) of
flags
(0 ,1 , •••,(’) by the formulae [ e ' 0] n [ E ] i [ e ' 0] n [ e = -j, o 10 U i 0- i [E '^ n i E j ] t [E 'jin iE j
j ] ( i ^ i g ) , •••,
[ E p ] n [ E ^ ] ^ [ E p l n t E ^ j ] , ( i p ^ i 0 , i 1( •••,ip_1) •
Here we use the convention [ E
1] = 0, [Ep] - lE'p] - V.
a permutation ( j j , j 2 ,
(1 ,2 ,...,()
jp) of
[E ^in tE - ] n H ^ [ E ', ] n [ E j
by the formulae
,]n H ,
[ e '2] n [E. 1n H / [ e 2] n [ E j2 _ t] n H ( j 2 ^ tE g ]n [E j ] _ Let
g h
0 , l , - - 4, f
and one of
), ••■,
^ [ E p ] n [ E j^ _ 1] n H ( j p ^ j 1, j 2 , •■ ■ ,!£ _!).
L , i , , i9 , •■•, ip b (s ') = ( . . • ); h ’h ’
Similarly define
this is a pair of permutations:
one of
t 1 ,2 ,
•••,?.
This invariant depends clearly only on the B £-orbit of £', not difficult to see that it distinguishes e\ e " are in the
B £-orbits (i.e.,
In fact, the entries of b(e )
satisfy
(4)
b (s ') = b (e ')
same B E-orbit). Note however that the invariant
b (£ ') cannot take arbitrary values.
(3)
and it is
j j f I iQ, i1 1, j2 f Si o ’ *1 ’ *2 ji >ix , j2 > i2 -
" ’ ’ if7 f ®*0’ *1 ’ *2 ’ j£ ^ l' l ■
must
TH E A F F IN E S T E IN B E R G M ODULE
33
To prove this, let d
= dim ( [ E ' J n [E - 1 O H ) / [ E ' J n [E ,] n H d Ja a Ja - 1
d a =■- dim ( [ E ' J n [E - ] ) / [ E ' J n [ E . ,] . Ja d Ja It is clear that 0 < dg < d'g 1. We shall prove the lemma by
b to be minimal with the above property.
TH E A F F IN E ST E IN B E R G M ODULE
Then we have: [ E a_ i l n
[ E 'a ] H [ E b] = 0.
35
L (line), L C H, [E 'a ] fl [ E b _ 1] - 0 and
Keeping fixed
[ E q ] , [ E j ] , •••, [E 'a _ 2L [ E 'g], •**, [ E ' ^ j ]
we consider all ( a —l)-dimensional affine subspaces E Q_ 1 of E which lie between and q
E'a_ 2 and
if a = 1.)
E a . (There are
There are exactly
[ E a _ 1] U [ E b] - 0;
(q+1)
such subspaces if a > 1
q values of
moreover the flags
E a _ 1 such that
( E q C E ^ C ••• C E a _ 2 C E Q_ 1 C
E'a C • • • C E f _ 1) corresponding to these values of
E a _ x are all in the
same B £-orbit (cf. Corollary 2.2), and they include the original flag In the case
a > 1 there is a unique value for
[ E a _ 1] fl [ E b] ^ 0 this is precisely for u and the fact that
u is B £-invariant we get:
if a > 1 if
a = 1
s ' = (E qC E'j C ••• C E a _ 2 C E a 2 © L C E a C ••• C E
a ( e ' ) - a ( e ' ) — 1. that q
such that
E'a 2 © L . Using the cycle condition
( -u (e ") q • u ( e ') = \ I 0 where
Ea l
x).
Hence
From this the lemma follows by induction on a.
correspondence with the permutations since the invariants
( i Q,
Note that these are in 1-1 , •••, ij?) of (0,1, •••,£),
j 1, •••,]£ are uniquely determined.
function as in the Theorem 2.1.
Since
Let
u be a
u is constant on B £-orbits and
since it vanishes on flags of type II (by the lemma) we can regard a map u : jpermutations of
E
(Note
is not a zero divisor in A . )
We now concentrate on flags of type I.
Let
s'.
u as
0, 1, •••,£! -» A.
( E q C E ' j C ,,T E a _ 2 C E /a C • • • C E ^ _ 1) be a subcomplete flag in
(the ( a —l)-dimensional subspace is missing).
permutation ( i Q, i j , •• •, i(?) of (0 ,1, •••,£) with
A sso c iate to this a iQ_ 1 > iQ by the formulae:
[E '0] u n [ E j1 0] A [ e '0]n u [E -10- 1J , - , [ E ' a - z ]n [ E .ia _ 2] ^ [ e ' a - z ] n [ E ,ia_ 2 _ t1 ] [E g ln tE j
] i [ E al n [E j _ ! ] , [ E a] n [E j ] ^ [ E a] n [ E j a —1 a—1 a a
[ E a+ l l n t£ ia+1l * [ E 'a+1] n
,
- , [ E ' £] n [ E if] ^ [E '£] n [ E i r l ],
36
T H E D IS C R E T E S E R IE S O F
(Convention:
O V E R A F IN IT E F IE L D
n
[ E _ 1] = 0, [Eg] = [E g ] - V . ) We assume that [E'^I fl [E j] + 0
= > [EC] H [ E j ] C H (E qC E j C
GL
■C E '
for all
i ^ a —1 and all
2 C EgC
with the missing
E'a
C a s e 1.
There are q
(a = 1).
l ’s
C E g _j)
j.
into a complete f lag by filling in
in all possible ways. possibilities for
E g = E'j n E j . The remaining (q —1)
complete flags of type I with invaiiant ( i Q, i j ,
follows that
(q —1) u(i0 , i-j, •••, ig) +
C a se 2.
i
In all
q cases we get
( i j , iQ, i2 , i 3 , •••, ig) for
ig) for the remaining ones.
Next assume a > 2;
E'Q. One of them is
possibilities give rise to complete
flags in the same B £-orbit (cf. Corollary 2.2).
and
We shall make
Ej fl E^
From the cycle condition it
, iQ, i2 , •••, ig) = 0, (i j > i2).
there are now (q+1) possibilities for
j p min ( i Q, i x , •••, ia _ 1)
(a > 2).
E'
^
One possibility is
E'a l
=
< E'
„ E ' H E - > (affine span). The remaining q values of E ' , 9—2. 9 1_ d *■ cl give rise to complete flags in the same B £-orbit (cf. Corollary 2.2). In a ll (q+1) c a se s we get complete flags of type I with invariant
ia ’ ia - l , i a + l , ' " , i f ) ia _ i , ig ,
( i Q, i i , — ,ia 2 »
for the first value of E'a _ 1 and 0 0 > H ’
ig) for the remaining q values.
follows that q u ( i0 , i 1, - - - , i a_ 2 , ia_ j , ia ,
* 8- 2’
From the c y c le condition it ig) 4 u ( i 0 , ^
ia_ 2 , i a ,
ia - l ' ia + l ' - ' if ) = °C a s e 3. [E j
ig l
] fl H
- min (ig, ij ,
is a line not contained in [E 'a 2],
= ( [ E'a _ 2] + L ) fl E which * i » i a_2>
From the cycle condition it follows that
(q—i ) u (i0 , ij , ■■■, i a _ 2 > ia - l ’ *a’ *a+l ’ + u(i0>i 1>- , i a _ 2 , i a , i a _ 1>ia 4 l , - , i g ) = 0 .
THE A F F IN E ST E IN B E R G M ODULE
37
We can collect the three c a se s in a single formula:
(5)
6 ■ u( i0 - il . - - i a- 2 . ia - l ’ ia - 1a+ l ’ “ ' ’ i^) ■+ u( i o ’ h >
ia -2 > *a’ * a - I ’ *at-l ’
if) = 0
where if
ia - 1 7 min(i0 , i2 ,
ia —1)
q —1 otherwise. Given a permutation w = ( i 0 , it , •■■, ip) of (0 ,1, •••,£) we define its length £(w), tions
as usual, as the minimal number of fundamental transposi
(a, a+1), 0 < a < £—1 of which w
Wealso that ig ^ m(w).
can be the product.
define m(w) as the number of indices 0 and ig =
Let
min(i0 , i1(
i ).
Wq = (£,£—1, ■•■,1,0)
length f ( w 0) = 1 + 2 + •••+ £. (0 ,1, •••,?—1,£);
ig(0 < a
be the unique permutation of maximal
It is clear that
m(w0) = £.
we have £(w q ) = 0, m(wg) = 0.
Let
w'0 =
From (5) we get immedi
ately by induction on the length that
u(w)
It follows that where w
=
u (w )= q
( —
q
)
~
m^w ) (1 — q)
u(w q )
.
£(wn)—£(w)t m(w)—m(wn) m(wn) —m(w) 0 0 (q—1) 0 u(w0)
is any permutation of
(0,1, •••,£).
last formula all exponents are positive.) to a scalar.
m(w )
(It is easy to see that in the
This proves the unicity of
The same proof shows’ the existence of u.
u up
The theorem is
proved. Remark. formula
Define a function u': {complete flags of type I in E i - » Z by the u ' (£ ') = number of elements in the B g -orbit of s',
recurrence formula similar to the one satisfied by
u' satisfies a
u:
u ( i 0 , i1; •••, ia _ 2 >^ - l ’ 1a ’ 1a+l ’ = 6 ■ u ( i 0 , ij , ••■, i a _ 2 >ia >ia - l ’ W l ’
^
’
THE D ISCR E TE SERIES OF G L n OVER A F IN IT E F IE L D
38
where 0 is as in (5) and B £-orbits of type I in F l a g ( E ) with permutations of
0,1, ...,£.
It follows that u and
x constant.
u(w )- u '(w ) = ( —1)
are identified
u' are related by
Choosing the constant to be equal to
the order of
B c we finally get the following formula for the function u £(ww ) of Theorem 2.1: u (e ') ■- ( —1) times the number of elements t eB £ such that t e ' = s',
if s ' has type I and
u (s ') = 0 if s' has type II.
2.4 Affine foliations Let
E
be, as above, an F-dimensional affine space over
that I > 2. of
E
An affine foliation O
of E
F.
is by definition a decomposition
into affine subspaces (called leav es) of fixed dimension
in such a way that any point in E
Assume
m (0 0 and
H (-)
denotes reduced homology, i.e.,
H j(-) =
H 0( - ) = k e r(H 0(-)-> A ). )
We now compare
S j(E )
and
S (E ,0 ).
Define a map a : S j (E ) -> S by
the formula ( E'
if
a(E' H / < E ' , £ > Here £
is any leaf of O
The map that
= a (E ") Let
£
first that
be a leaf meeting
hence a ( E ' ) < a ( E " ) .
E\
order.
In fact, let
E', E " eS j (E )
such
E', E " e SI V ( E , 0 ) . Then a ( E ' ) =E ' C E "
Assume next that
E', E " e S j (E ) \ SI V (E, $ ) .
We have a ( E ' ) = < E', £ > C < E", £ > = a ( E " ) ,
Finally, assume that E ' e S j(E ) \ S I V ( E , 0 ) , E " e
Then clearly a ( E ' ) e S j ( E / 0 )
hence a ( E ' ) < a ( E " ) = E " by
the definition of the partial order in S ( E , 0 ) . SI V (E,), E " e S j (E ) \ S I V ( E , 0 ) that a
.
meeting E'.
hence a ( E ' ) < a ( E " ) .
SI V ( E , 0 ) .
if E 'f S j(E )\ S IV (E,4>)
arespects the partial
E C E". Assume
E 'e St v ( E , 0 )
Note that the c ase
cannot occur when
E ' C E".
E' e
This proves
respects the partial order.
It follows that a
induces a simplicial map X (S j (E ) ) -> X(S)
hence als o a map a A : H £ _ 1(SI(E ); A ) -> H£_1(S; A).
and
We shall prove that
40
THE D ISC R E TE SERIES OF G L n OVER A F IN IT E F IE L D
Since a A ~ a g ® 1A
«A
is an isomorphism.
«£
is an isomorphism, and this would follow if we can prove that a A
an isomorphism whenever A
it is sufficient to prove that
Assume now that A
is a field.
is
is a field.
We know that dimA H|?_1(Sj(E); A) - ( q - 1 ) ( q 2- 1 ) ••• (q^—1) (cf. 1.14) and
dimA H£_1(S; A ) = dimA H g_m__1(SI(E/0); A).
dimA Hm_ 1(SIV(E, S>); A ) =
( q - l ) ( q 2 - l ) - - - ( q f - m- l ) - ( q £- m+1- l ) ( q f - m+2- l ) . - ( q £- l )
follows that a A
(cf. 1.14).
It
is a map between A-vector spaces of equal dimension,
hence it is sufficient to prove that a A
is injective.
To see this we first
observe that given any (2—1) simplex o = ( E mC E m+1 C •••C E j ^ j < E j _ mC E f _ m+1 C •••CE^_1) m < i < 0—1 and E'unique (0—1) simplex
in X(S) (where
is transversal to leaves, t
0—m < i < 0—1) there is a
= ( E 0 C E x C •■•C E £_1) in X (S j(E )) which under
a maps isomorphically onto o. (O < i< 0 -m -l)
E- contains some leaf,
In fact we have
and E- = E'j(0—m< i < 0-1).
H E\ _ m
E^ -
A ll other (0—1) simplices of
X (S j(E )) are mapped by a onto simplices of lower dimension. that given a chain u € Cg_j_ (X(S j(E)); A),
It follows
the image « A u e C^ _j(X (S ); A)
is given by the formula: («AuXEmC E m l C - - - C E p_ 1< E ^ _ mC E '£_ m+1C . . . C E ' E_ 1)
= u(EmnE'e_ mc Em+1nE 'F_ mc ...c
e^
de'
^
c
E £_mc E g _ m+1c - c E'g.
This shows that « A U = 0 if and only if u vanishes on all complete flags
( E 0 C E 1 C . . . C E £_ 1)
0 < i < 0-1
(£
in E
such that dim (E- H where
We say that d(e )
£ by the formula
d (e ) = (dQ, d x , •••, dg).
is
A (s ) =
Note that A ( e )
is
minimal if and only if A( e ) = 1 + 2 f •••+ m or if and only if d (s ) = (0, 0, •••, 0,1, 2, 3, level of £.
m).
We shall prove the lemma by induction on the
We know that u (s) = 0 when A( e )
that £ has non-minimal level.
Then there exists some
such that d j _ j < d- = d - f l . Consider E j+1
such that
E -_1
C
is minimal.
Assume now
i, 0 < i < f —1
all i-dimensional subspaces
E- of
E ^ . We have
2 u (E x C -C
k i_ 1 C E j C E i+1 C - . . C E ^ )
-
0 .
Ei In the above sum all flags except for the original flag (d Q, d1 ,
8,
have type
d - _ 1, d- —1, d-+ 1 , •••, dj?) hence have level equal to A(£) — 1,
so we can assume, by the induction hypothesis that u vanishes on them. It follows that
u(£) = 0 and the lemma is proved.
It follows from the lemma that a A
is actually an isomorphism.
Com
bining this with the isomorphism (6) and identifying SI V ( E , 0 ) - SI V (V, V), we have the following THEOREM.
Let
V
b e a n (£4 1ydimensional vector space over
F,
V
an
m-dimensional linear subspace of V (f > m > 0) and E
an affine hyper
plane in V
Let
E/V
be the
Then
E/V
is an
image of E
such that 0 i E
and V
is parallel to E.
under the canonical projection
affine hyperplane in V / V
V ^ V/V.
not containing zero and there exists a canoni
cal isomorphism i/fE : A a (E / V ) ® PA ( V , V ) *
uniquely characterized by the property that
A a (E )
THE D ISCR E TE SERIES OF G L n O VER A F IN IT E F IE L D
42
= 0
Ei "k+1
(E o C E iC -C E ^ C E ^ C -C E ;,!) E 0 f E n - l \ E n - 2 - E i H [ E 0 ^ - - E kl i [ E k
_ l l , E kCEk+1CEk+2
E k+ 2 l l [ E k+ l ] ' - - ' E ; - l H [ E n - 2 ]
(cf. (7k+1) for f)
and (7k) for
Tf
is proved for
Next we prove (7n_ 2) for
0 < k < n—2. Tf
(here we assume
n > 3).
Let
( E 0 C E 1 C - - - C E n_ 3 C E n_ 1) be an incomplete flag in V ( 0 / E p .
We have
(q + D 2
( T f ) ( E 0 C E 1 C •- C E n—3 C E n—2 C E n—1 >
En —2
(qn- q n- 2 ) 2
=
f(E o C E j C - ’- C E ^ - j )
( E qC E j C- •-C E ^ _ j ) e Y E 0fE n_ 1\En_ 2(E 'l|[E 0L - - , E n_ 2 !!fEn_ 3]
(qn - l _ q n- 3) = ( - l ) n- X
(q)
2
2
(EoCE ; o - - C E n_ 2) E 0 f E n - l \ En - 2 - E i l l [E0 ] - - ” ’E n - 2 l l [En - 3 ]
e
; _
f ( E oC E i C-
CEn -l)= °
j
E n - 2 CEn - l
^
e
;_ i ( c f . ( 7 jj_ j ) f o r
and (7 n_ 2) for T f
is proved.
f)
46
THE DISCRETE SERIES OF GL n OVER A FINITE FIELD Finally we prove (7n_ 1) for
complete flag in V (0 / E p .
Tf.
Let
( E QC E x C •• • C E n_ 2) be an in
We have
(q) 2
( T f ) ( E o c E i c ••• c
e
„ _ 2 c E n—i )
K-1
(qn- q n~ 1) 2
= ( - 1 )" - 1
f C E g C E j C •• • C
)
( E qCE^C- ••C E ^ _1) 6 Y E 0f V ' E n - 2 ' E i ,!tEt)1' - - E n - l l l t Eh - 2 ]
(qn - 1 - q n _ 2 ) = ( - l ) n_1
(q) 2
2
f ( E o C E i C - - C E n _ 1) =
(E jCE^C- •'C E ^ _ j )
E„
E 'll|[E0],E '/ E n_ 2
E ^ E;
E 2 ||[Ei],-->E n_il|[En_ 2]
0
(Cf. (70) f o r
f)
and the lemma is proved. It follows from the lemma that T : ? A -» 3 ^ into itself; the restriction of T letter:
3.2.
w ill be denoted by the same
T : A a ( V ) - A a (V ).
In this section we shall prove the following
P ro p o sitio n .
Let
(8 )
q
{(E
whenever Then
to A a (V )
takes the subspace A a (V )
C E
f e A a (V )
j
be such that
C •••€ E n_ j )
=
f ( E o C E ; c . . . C E n_ x)
E Q = Eq, [E.J = [E ^], •••, [ E n_ 1] = [ E /n_ 1L ( 0 ^ E ^ 0/ E p .
Tf = f.
T H E D ISTIN G U ISH E D D IS C R E T E SERIES M ODULE
Proof.
We shall first consider the case
ing (8).
n = 2.
Let
47
f e A a (V )
be satisfy
We have (q -D (T f)(E 0 C E j ) = —
^ (E E
q
^
e
i
qC
f(Eo C E j ) E')£Y
\ E
q
, E j || [ E
q
]
(q-1 ) X
f(E 0 C E i>
( c f . (8 ))
K E 0 C E x)
(cf. (70))
E 0f E l X E o
Let now
n>3
E n_ i ) € (9k)
and let
f e A a (V )
be satisfying (8).
Let
(E q C E jO -C
prove by induction on k the following statement.
^e
( T f ) ( E QC E 1 C ••• C E n_ 1)
= ( - l ) n~ k - 1 £
f ( E QC E j C ■••C E k _ 1CE j {C < E ^ , E k> C . " C < E ^ , E n _ 1> )
Ek E k - l C E kC E n - l
( 0 < k < n —1)
E k ^E n —2
Note that (9
is precisely the statement
from the hypothesis (8).
= ( _ i ) n- k - l
and that (9Q)
Assume that (9k) holds for some
we shall prove the statement (9k+1). (Tf)(E0 C El C
T f = f,
follows
k, 0 < k < n—1;
We have
E n_ 1 )
£
f ( E 0C E 1C . . . C E k _ 1C E kC < E k E k> C - C < E j ( , E n_ 1 >)
Ek E k - l C E kC E n - l E k ^ E n—2
48
THE DISCR E TE SERIES OF G L n OVER A FIN IT E F IE L D
= ( - 1}
“
f ( E 0 C E j C - ••CE k _ j C EjjC E ^ + I c
2
E k+ 1
Ek
E k+l^En_ 2
Ek^Ek
< E k+ l - E k+ l > C -
= (_ l)n -k -2
C < E k+ l ’ E n - l > )
f ( E 0C E 1C . . . C E kC E ^ 1C< E ' k + 1 , E k + 1> C . . . C < E i { + 1 , E n _ 1
2
E k+ 1 EkC E L l C E n - l E k+ l ^ E n - 2 (cf. (7^)) and the proposition is proved by induction. the convention
(Here we have used
E _ 1 = 0 .)
Note that in case
q
is invertible in A,
any element
f e A a (V )
satisfy
ing ( 8 ) must be zero.
3.3
In this section we shall assume that dim V = 2.
the set of homogeneous polynomial functions is clear that
P o l_ 9 (V ) q z
V -» F
L et
P o l q_ 2 (V )
of degree
is an F-vector space of dimension
be
(q —2). It
(q —1).
We
shall describe now two F-vector spaces which are canonically isomorphic to P o l n 9 (V ) 4
A
but whose definitions are somewhat more geometric.
First we recall (see 1.13) that ® ( V ) functions of V
is the F-vector s.pace of all
f which associate to any one-dimensional linear subspace
a vector
f(V j) e
such that
f( V j) = 0 (sum in V).
It is
VjCv
clear that dimF ® ( V ) = number of lines — 2 = (q+1) —2 = q — 1. ® ( _ 1 ) (V )
be the F-vector space of all functions
f(Av) = A- 1 f(v), A e F *
and such that
^
f : V \0 -> F
Let such that
f(v) = 0 for any affine hyper-
V
veEj^ plane
E j C V, 0 / E j .
Let
f 6 P o l q _ 2 (V).
f defines a function
THE DISTINGUISHED D ISCR E TE SERIES M ODULE
f:V \ 0 -> F
so that
f(Av) = A - 1 f(v ),A e F *.
Let
We can choose a basis
e x , e2
which clearly satisfies
an affine hyperplane in V ^ O ^ E j ) . E 1 = \e1 + (ie2 , (i e F L
49
Ej
be
in V
We have
fO i e l + ^ 2 e2) =
2
ai
ai f F
0< i< q-2
hence ^
f (ei + iie2) =
]£
ptfF
0
/ 3i ^
° y ( - 1 } (V ) which is clearly injective. In order to prove that this is an isomorphism it is then enough to prove that dimension at most vQ e E 1
be fixed.
( q —1).
To s ee this, let
In order to prove that
cient to prove that if f e identically zero.
^(V)
f(v) = 0.
^^(V) < q —1 it is suffi
is zero on E ^ V q
Consider such a function
f(v Q) = 0 since
0 \ E x and let
E jC V ,
dim
has
f.
then f
must be
We must clearly have
From the homogeneity property of f
it
v£E1 follows that
f(v ) = 0 for any
parallel to E j . Let
v e V Wj
^ e VjXO.
Let
such that ^ 6 E j , 0 | E j . Then than vt . It follows that P o l n_ , ( V ) -» ? ( “ 1 )( V )
4 ^
Let now
f e ® (V ).
by the formula
E^
is the line through 0,
be an affine hyperplane in V,
f(v ) = 0 for all points
f(vx) = 0,
v
of E^ , other
and hence the assertion that
is an isomorphism is proved. We associa te to f a function
f ( v ) v = f[v]
(here
clear that f(Av) - A - 1 f(v),A 6 F * . hyperplane.
where
[v]
is the linear span of v).
Let
We shall show that
E x C V, 0 | E r
f(v ) = 0. veE^
f : V —0 -* F
defined It is
be an affine
In fact, let v *
be the
50
TH E D ISCRETE SERIES OF G L n OVER A F IN IT E F IE L D
unique linear function
VF
f[v] f f(ker v * ) = 0,
such that
v * ( v ) = 1 for v €
since f e ® (V ).
Applying v *
E 1. We have to this equality
veE1 we get
v * ( f [ v ] ) = 0 and hence
f(v )v *(v ) = ^
vfE^
vfE^
follows that the correspondence
f ( v ) 1.
It
vfE^
f -> f defines a map ® ( V ) -»
This is clearly injective, and since the two vector spaces involved have the same dimension, it must be an isomorphism.
We collect these results
in the following
PROP OS I TI ON.
If V
is a 2-dimensional vector space over F,
there
exist canonical isomorphisms of F -vector spaces
P o iq_ 2( v ) *
3.4
fp( - 1}( v ) ? 3)(v) .
In this section we shall assume that A = F.
the set of all A^f(EQ C E j Here
C
f e A F (V ) ••• C E n_ 1)
such that
Ap(V )^
as
f(AEQC A E j C ••• C \ E n_ 1) =
for all A e F *
and all
k is some fixed element in Z / ( q —1 ) Z .
Ap(V) =
Define
©
( E 0 C E 1 C ••• C E n_ 1) e Y. It is obvious that
Ap(V)(k).
k f Z /(q —1) Z
Note that the endomorphism T subspaces
of A p ( V )
leaves invariant each of the
A p (V)^k).
We now prove the following partial converse to Proposition 3.2.
PR OP OS I TI O N .
Let
f £ A F ( v / - 1 ^ be such that T f = f.
Then
f satis
fies condition ( 8 ) of 3.2.
Proof.
Let
8 - ( E 0 C E x C ••• C E n_ 1 ), £ = ( E QC E x C ••• C E n _ 1 ) be two
elements of Y
such that Eq = Eq, [E ^ ] = [E ^ ], ••*, [ E fl_ j ] = [ E n_ 1 l.
We
T H E DISTIN G U ISH E D D IS C R E T E SERIES M ODULE
wish to prove that f (e ) - f(e).
2
Since
T f = f, this is equivalent to
f ( E o C E i C - C E n_ 1) =
( E 0 C " - C E n - l )eY
2
f(E qCE'^C-"CE'^_1).
(E 0 C E i C ' " C E ' n - l )fY
We define a function 0 : V \ [ E n_ 2] - F where
51
by 0(E'O) = fC E o C E i O . - C E ^ j )
E j ( l < i < n—1) is the unique i-dimensional affine subspace of V
such that Eq e E^ and E^||[E^ 1].
Then the equality to be proved can
be written as
This clearly fo llows from the following
LEMMA.
Let
V
be a v ect o r s p a c e of dim ension
a linear hyperplane in
0.
not containing cj> : V \V
j -> F
E ^ V \ V n_ !
Let
V and E^__j, E ^ - l
n > 2 over F, V n_ j
two affine hyperplanes in
E Q e E ^ _ 1 H E/^ _ 1 H V n_ 1. A s s u m e that
is a function such that 0(A E q ) = A_ 1 0 (E q ),
and
A 6 F*,
(q)
2
* (Eo> = 0
for any one-dimensional affine subspace
C V (E^ ||[E0], E j : V — [ E q]
F
( E Q, E^ , E^ , Vx = [ E q]
must be identically zero, cf. 3.3).
belongs to
(see 3.3).
Eq 6 V \ [ E q].
f e 5^
It follows that
^
0 ( E'0) = f ( E 0)
E oeEi ^Eo and similarly
(q -l) 2 E ofE 1XE0
From what we have
1\v)
(q -l)
and the Lemma is proved.
This map is
must be isomorphism.
S by hypothesis.
just proved, it follows that there exists for all
On the other hand,
Since the vector spaces involved have
the map fF^_ 1 ^(V) -> S
Our function
It is obvious
f e 5^- 1 ^(V) which vanishes on V - [ E q]
clearly injective (a function
dimension q - 1
^ ( V ) -> S
satisfying the
fixed).
is an F-vector space of dimension (q —1 ).
there is a natural restriction map 5^
•
E (/E 1 V E 0
be the set of all functions
hypothesis of the Lemma that
2
* < E o> =
* < E 0 > = f( E o)
such that f ( E ^ ) =
TH E DISTINGUISHED D ISCR E TE SERIES M ODULE
3.5. From 3.2 and 3.4 we deduce that
53
1 ^ A F (v /
T : A F (v /
^
is idem-
potent and its image is precisely 3 = { f €A f ( v / - 1 ^ ( f satisfies ( 8 )S . In this section we shall construct a canonical isomorphism 3 — ® ( V ) . We recall (cf. 1.13) that ® ( V )
is the set of all functions
to any complete linear flag
( V i C V 2 C ••• C Vn_ 1)
0 (V 1
C V 2 C ••• C V
in V
0
which associate a vector
t ) € V1 such that the following conditions are satisfied
(q+l)
2
(lo p
^ C ^ C .-C V ^ C V jC ^ C -.C V ^ )
= o
V.
1
(1 < i < n - 1 ) where ( V j C V j C - C V ^ j C V ^ C - C V ^ j ) incomplete flag (the empty fla g if vectors in
if
i > 1
or in V2
as follows.
Given
( E QC E 1
C V2 C •*• C Vn_
C ••• C E n_ 1)
Ei C V2 \0, •••,Efl_ i C V \ 0 .
x)
- f ( E 0 C E 1 C - . C E n_ 1 ) E 0
is any element of
Y
satisfies the conditions (10p
( l < i < n —1).
from the condition (7^_1) satisfied by
f.
we can clearly assume that dim V = 2. f € fp( —1 } ( V )
jp (-l)(V )
such that
E Q e Vj \0,
(This is independent of the choice since
satisfies ( 8 ) and is homogeneous of degree ( —1).)
E 0 6 V1\ 0
f e J , define
by the formula (a f) (Vj
where
Note that this is a sum of
i = 1.
if
We define a map a : 3* -> ® ( V ) , at e ® ( V )
n = 2 , i= 1).
is any given
If
f
We must check that at
i > 2 this follows easily
In order to check ( 1 0 ! ) for af In this case we must prove that if
(se e 3.3) then the function at defined by afO ^ ) = f ( E Q) E Q, lies in J)(V);
but this follows from the isomorphism © ( V ) 55?
constructed in 3.3.
Next, we define a map /3 : ® ( V ) -> 3*. by the formula:
Given 0 e ® ( V )
define /30 6 J
TH E D ISCR E TE SERIES OF G L n OVER A F IN IT E F IE L D
54
(fi)(E0 C E 1 C •••C En __1 ) E 0 = 0 ( [ E o] C [ E 1] C - - . C [ E n _ 2 ]),
( E q c E x C ■••CEn_ 1 ) e Y . It is clear that (dcj) fies ( 8 ).
is homogeneous of degree
(-1 )
and that /30
s atis
We must check now that /3 satisfies the conditions (7-),
(0 < i < n—1).
If
0 < i < n—1 this follows easily from the condition
(10i 1 1 ) satisfied by 2. r —
Then there is canonical isomorphism
of F -vector spaces fi : ® ( V ) “5 I f f A f ( V ) ( _ 1 ) ! T f = f 5 .
/3 rs uniquely characterized hy the property
(fi(f>)(E0 C E 1 C ••■CEn_ l ) E 0 = 4>(\E0K [ E 1] C - - C [ E n_ 2]) for all cf> f 5 ) ( V ) , ( E 0 C E 1 C - . - C E n_ 1) f Y.
3.6
We recall now the definition of the ring of Witt vectors
to the finite field
F. Let
p
mutative ring together with a
WF
be the characteristic of F. Wp
associated is a com
ring homomorphism WF -» F , which is
uniquely characterized by the following properties (see [10], p. 48). a.
p is not a zero divisor in WF
b.
WF
is Hausdorff and complete with respect to the topology deter
mined by the ideals
pmWF ( m > l ) .
c. The sequence 0 -* pWF -* WF
F
0 is exact.
It is well known that Wp must be in fact an integral domain characteristic zero, that there exists a unique map F*-> Wp
of
(denoted
THE DISTINGUISHED D ISCR E TE SERIES M ODULE
A->A)
such that A • p = Ajz(A, p e F * )
the natural inclusion Wp
F*
> F.
and the composition
(A ^ A
55
F * -> Wp ■-> F
is
is known as the Teichmiiller map.)
can be described explicitly as the set of all sequences
(A q ,A 1 ,A 2 , ■••) of elements in F
in which the addition and multiplication
are given by certain universal polynomials discovered by Witt (see [10], P. 49). If F
is the prime field of characteristic
p,
Wp
is canonically isomor
phic to the ring of p-adic integers. If
F
> F ' is an imbedding of a finite field in another, there is a corre
sponding imbedding Wp -> Wp-. naturally on Wp '
3.7
so that Wp
Moreover the Galois group
G a l(F '/ F )
acts
is precisely the ring of invariants.
In this section we shall prove the following
LEMMA.
field). M ® F
Let
L et
M
be a finitely generated free
W p - module
T : M -> M be a W^-linear map such that T
is idempotent.
a finite
(F
®
: M ® F ->
WF
Let
WF
WF
M' = Sx t M |lim ( 1 - T )1 x = 0 l , M" = |x< Ml lim T ^ O l i-->oo i-»oc
.
Then M = M' © M".
Proof.
We shall prove the lemma assuming first the following statement:
there exists
T e Endw (M)
such that T ' 2 ~ T ' f T ' T
TT
and
T ® lp
F
Wp
- T ' ® 1F . Then we have clearly WF mutes with this decomposition. and
T2 e Endw (ker T ' )
on ker T'.
Let
x 2 e ker T'.
Then
x e M;
M = ker (1—T ' ) © ker T ' and
Moreover there exist
such that
T - 1 — pTj
we can write uniquely
T
com-
Tj_ e Endw^ ( k e r ( l —T ))
on ker ( 1 - T ' )
and
T = pT2
x - x 1 + x 2 , Xj e ker (1—T ),
(1—T ) ix = ( p T j ^ X j + ( l - p T 2 ) ix 2 .
56
THE D ISCR E TE SERIES OF G L n OVER A F IN IT E F IE L D
Since
lim ( l - T / x
lim ( l - p T 2 ) ix2
hence x 2 -
0,
i -*CKj
( p T . ^ X j - 0 we must have
0 and lim
0.
This shows that
The reverse inclusion is obvious hence proves that position
M - M'© M".
Y0 - T, X Q
2
M' - ker (1—T')*
Similarly one
IVT ■= ker T ' and it follows that we have the direct sum decom The existence of T ' is a c la s s ic a l fact about
‘ li ft in g idempotents’ \ Define
M'C ker (1—T ').
The idea
is to construct
Y 2 - Y0 , - - , Y i ( 1
(i > 0),
T ' by a limit procedure.
Y, * Xj - 2 ^ ,
X .+ 1 =
. One checks by induction that
is divisible by
ry1 p
r\1 . It follows that Y itl - Y-
X ^ ( l —2Y j) is divisible by
hence '2 and it is easy to see that T T ,
/
Y-
1 converges to some T ' c Endw w p (M)
T T 'T 'T
and
T
®
- T ' ® 1F . The first part of the proof shows
wF
wF
that
T ' is in fact uniquely determined by these properties.
3.8
We use the notations of 3.5.
We shall apply the Corollary 3.7 in the
?(k)' We define A w (V)*
following situation.
p
as the set of all
f e A w (V )
F
F
satisfying the ‘ ‘homogeneity condition” f(AK 0
for all A f
F*
AH,
.
VM F
Let
(V )®
Here ®
k is some fixed A
k fZ /(q-l)Z
( V ) (k) and F
® F. WF
wp
T : A w (V ) -» A A
F
T :A w wp
” F
C E n l ) t Y.
We have A w ( V ) =
...
(V )W *
AEn .,) = A ^ C E j C - . - C E ^ j )
and all ( E QC E , C
element in Z / ( q - l ) Z . ...
•••
(V ),
( V ) ( - 1 ) be the restriction of the map
defined in 3.1 (note that all the subspaces
F
are invariant under
T).
F
It is obvious that we have a commutative diagram
I A f ( V ) 2.
dimp ( V ) - 1. In
58
THE DISCR E TE SERIES OF G L n OVER A F IN IT E F IE L D
I f : V \ 0 .WF i f(AE0) = A " 1f( E „ ), E 0 f V'. 0, A f F*| .
D (V )
This is a free Wp-module of rank 1 and there is an obvious isomorphism P : ® ( V ) -> D ( V )
® F
defined by / 3 (v )(E q ) =
where
WF
and
E q 6 V\0.
If
n y 2,
(q—l ) ( q 2 —1) ••• (q 0” 1—1). for dimF ® ( V ) ,
3.8
Let
E
v e ® (V ) - V
0
D (V )
is a free Wp-module of rank
This follows from the corresponding formula
see 1.14.
be an affine hyperplane in V
In this section we shall prove that Wp-isomorphism p E : D ( V ) * $ A ^
E
(E).
such that
0 \ E, (dim V > 2).
gives rise to a canonical Define pE
by the formula
F
(pE f ) ( E 0 : E 1
E n_ 2) = f ( E 0
for any complete affine flag It is clear that pE isomorphism. -i (q 11 —1)
K,
-
E„
( E QC E^ C ••• C E n_ 2)
in E
is an isomorphism if and only if pE
Since dimF ( D ( V )
E)
f 6 D (V ).
and any ® 1F
is an
® F ) - dimF ( A p ( E ) ) -= (q —l ) ( q 2 —1) •••
Wp it is sufficient to prove that pp.
®
w
lp.
T posing with the isomorphism D ( V ) W ® F = 'J (V ),
is injective. pF
F
map pE : ® ( V ) -> A p (V )
,
® lp
By com-
becomes the
Wp
given by
(Pe 0 ) ( E o C E 1 C "• C E n_ 2 ) E 0 = 0(1 E q] C [ E j l C "• C [ E n_ 2] )
e $>(V), (see 3 .7 ). Let
f ® (V )
i that cf>(V1
1 < i < n;
We shall prove by induction on
V 2 ” *-’ C V n_ 1) - 0 whenever
use the convention V n = V . )
assumption pE flag in V
be such that pE = 0.
0.
such that
Let
For
V T E
* 0.
i - 1,
this is just the
( V j C V 2 C — C V n_ 1)
fl E - O,
D E = 0
We have then by the cycle condition for yS:
(i
fixed,
be a complete linear for some
k, 1 < k < n.
THE DISTINGUISHED D ISCRETE SERIES M ODULE
o (V ,
V,
-
V k ._;
V k. ,
V.
....
Vn.,,
vk .2
vk
,
59
(q) = -
£
o (V ,
V,
•
Vj E ® I p
Hence cf) =--- 0.
V^_1
and
satisfy It follows
V fl H E 1=0 which is always
This proves that p E
is injective hence
*s inje c tive and we can state the following
THEOREM. V
V n ,)
hence the induction hypothesis can be applied.
that qS(V1 C V 2 C •■• C V n_ 1) - 0 whenever satisfied.
. .
V ^ _ 2 C Vj 2, and let
Assume 0 ( E.
E
be an affine hyperplane in
Then there exists a canonical isomorphism of
p E : D ( V ) 5$ A W ( E )
given by the formula
F
E,
for any complete flag
3,9
-
K n . . ) = f ( E 0 C E l C - C E n_ 2 C E )
(Eq C E j C ••• C E
2)
rn ^
an m > 0).
canonical W^-isomorphism i> : D ( V / V ) ® *
Wp
n over
F
and let
There exists a
Pw ( V , V ) S D ( V ) . F
Remark. This tensor product decomposition corresponds in terms of dimensions to the factorization
(q ~ l)(q ^ ~ l)
( q n _ 1 — 1)
=
( q - l ) ( q 2 - l ) - - - ( q " - " 1- 1 - ! )
—m—1) n ^( q nr- m i l x ( q,n n—m
3.11
1
. /„n—1
l)■ ■■(q r
Applying Theorem 3.10 repeatedly we get the following
C O RO LLARY.
Let
V
let V- C V; C - “ C V: n l2 \ n > i^ > •••> i 2 > i^ > 0.
be a vector space of dimension
n over
be a linear flag in V (dim^ V- = i_) * 1a a
F
with
There exists a canonical WF -isomorphism
and
64
THE DISCR E TE SERIES OF G L
0-
D (V /V-
) «
lk
/
®
W^f 'F I
OVER A F IN IT E F IE L D
n
Pw ( V / V : F
F
xa - l
, V: /V: )\ ? D (V ) la la - l \
yl < a < k
where we use the convention
V-
0.
This result describes the restriction of the GL(V)-module proper parabolic subgroup of G L ( V )
D (V )
to a
(the stabilizer of
V- C V - C - C V- ). h ~ 2 xk (V j C V 2 C ••• C V fl_ 1 ) in V.
Take for example a complete linear flag Then we have a canonical Wp-isomorphism
^ ; D ( V / V n_ l ) w ® /
Pw pC V /V a-r V ^ a - i A
\ 1 < a < n —1
describing the restriction of D ( V )
-
/
to a Borel subgroup of
G L ( V ).
In
terms of dimensions, this corresponds to the factorization of (q—1 ) (q 2 —1 ) ••• (q n 1 —1 )
3.12 all
Let
VC V
t € G L (V )
into the factors
1 , ( q - 1 ), (q 2 - 1 ), •■■, (q n“ 1 - 1 ).
be as in the Theorem 3.10.
such that
Let
be the subgroup of
1 1V = identity and t jV/V = identity.
clearly an elementary abelian p-group of order
This is
q m^n m^.
We have the following
PR O PO SITIO N .
For any V
C
V, 0 < dim V
V
such that t(V) = V.
(D ( V / V )
and
Pw ( V , V ) F
can be regarded as
G^-modules in a natural w a y.)
acts as the identity on D (V / V ). acts as zero on R J; ( V , V ) .
Note that any
te
It is then sufficient to prove that
t uU^
THE DISTINGUISHED D ISCR E TE SERIES M ODULE
Let
V
plete flag in V
2
tuV
V >V )
( V n_ mC V n _ m_| 1 I ••• C
and
such that \L + V = V (n—m < i < n—1).
V n -m C V n -m a : -
65
be an incom We have
CVn -l)
trtjY / V
2
u(t_ 1 v „n—m mc r 1v „n-m-: mjl1 c - - c t _ 1 v n— n l, )7
uuY V 2
2
( V n - m + l C' " CVn - l )
' ’n - n w l f
where
(V l _ , v n—m+1
u(s " 1v n- mc v'n_ m+1c . . . c v ' n_ 1) - o
V
C ■••CVln—11 )/
(t_ 1 Vn_ m f l C •*• C t- 1 Vn_ i )
runs over all distinct flags of the form
^or some
from the cycle condition satisfied by
1~ u.
The last equality follows The proposition is proved.
More generally, let v --- (V- C V- C •••CV- ) be a flag in V as in 3.11. h x2 Let G ^ be the parabolic subgroup of G L ( V ) consisting of all t e G L ( V ) such that t(V: ) = V- (1 < a < k). Let U ^ be the subgroup of G ^ con1a a ^ ^ sisting of all t r G ^ such that tlV- /Videntity, (1 < a < k+1) ^ a a —1 where V- - 0, V: = V. We have the following *0 *k+l
t acts as zero on D(V ).
C O RO LLARY.
teu y
Proo f. This follows easily by induction on k from the proposition. Remark. The conclusion of this corollary means that as a GL(V)-module, D (V )
satisfies the cusp-conditions hence it belongs to the discrete
series (s ee [ 6 ]).
CHAPTER 4 T H E C H A R A C T E R O F D (V ) A N D T H E E IG E N V A L U E A (V )
4.1 Let Let
V
be a vector space of dimension
t : V -> V
Tr (t| V ) 6 WF
be an automorphism of V. as follows.
G al(F V F ).
under
G a l(F V F )
under
G a l(F V F ).
see that
Tr(tjV )
F ' be a finite extension field of R e d e fin e
It is clear that the set
It follows that the set (which acts on WF 0 It follows that
subspace of V
invariant under
dim V = 1,
t :V ? V
so
Tr(t|V) = is invariant
is also invariant
+ ^2 + ‘
+ ^n
(see 3.6).
is independent of the choice of F'.
4.2 We say that an automorphism
any
hence
F
£Aj, A2, •••, A )
( A j , A2 , •••, A )
T r (t / V ) ^ 2 ' ’ ” '^ n
that the eigenvalues
Aj + A2 + ••• + An 6 Wp/. under
Let
n > 1 over a finite field
is invariant
It is easy to
Note that
Tr(t|V)
V.
t : V -> V
is anisotropic if there is no
t other than 0 and
is anisotropic.
V.
In particular, if
In this section we shall prove
the following
PR O PO SITIO N .
If t : V ^ V
is anisotropic then
Trw (t|DV) -
( - l )n _ 1 Tr(t! V ) .
F
P roof. The proposition is obvious when
Mi = D (V ) © ^
n - 1.
Assume now that
© c v D ( V n - 2) ) ® ( Vn© Cv D ( Vn - 4 ) ) ® -
and
66
n > 2.
THE C H A R A C T E R OF D (V ) AND THE E IG E N V A L U E A (V )
“2 *
and
M2
D^n^ are the eigenvalues of t on V.
set of eigenvalues of t on M- ( i - 1 , 2 ) .
S 1 U (A1 , A 2 ,•••, An) § 2 U (A1, A2 ,
Let
Sj
be the
It follows that
§2
(n even)
•••, An) •:=S x
(n o d d ) .
-
This implies that
V
Trw (t|D(V)) =
( —l )n _ 1 —i Trw (t| ©
F
F
K K n -i
Next we observe that Trw (t| © F
D(Vj)) + ( - I ) " " 1 T r ( t | V ) .
v.cv
1
D ( V - ) ) = 0 for
1 < i < n—1 because
acts on the set of i-dimensional linear subspaces of V point (since
4.3
t
v ic v
t is anisotropic).
without any fixed
The proposition is proved.
We shall prove the following:
PR O PO SIT IO N .
Suppose we have
t : V 5 V, t (V ) = V
such that t|V
= ( _ l ) m~ 1 ( N —1) where such that Vn_ m t V = V
N
V C V (0 < dim V - m < n) and is anisotropic.
Then Trw (t|Pw ( V , V ) )
is the number of linear subspaces
and t (V n_ m) = V n_ m-
Vn__m C V
68
THE D ISCR E TE SERIES OF G L
Proof.
n
OVER A F IN IT E F IE L D
From the exact sequence 1.13(e) we see that
2
TrwF ^ l pwF ( v - V ) ) =
( - 1 ) n _ 1 " i T r wF ( tlv ^ - pwF ( v i ' Vi n V ) ) + ( - l ) n
n —m 1 over F.
Consider the
virtual representation brCV)= ( —l ) n- 1 D ( V ) + ( —I ) " ” 2
© v„ _ l Cv
of
G L(V ).
©
D (V ).
V.CV
1
of
Note that G L ( V ) When
D(Vn_ j ) + — t- ( —1)°
©
D (V j)
v i Cv
operates naturally on the free Wp-module
1 < i < n—1 this can be regarded as the representation
1
G L (V )
induced by a representation of the maximal parabolic subgroup
GX , which is trivial on U y , for some fixed VjC V; s ee 3.12 for the i i definition of G y and U y . br(V ) can be considered as an element in i i Rw ( G L ( V ) ) . Under the natural map Rw ( G L ( V ) ) -> R F ( G L ( V ) ) , br(V) F F* 80
T H E B R A U E R L IF T IN G
goes to V,
81
regarded as a G L(V )-m odule in the natural way.
(T h is follow s
from 3.8 and the exact sequence 1.13(c).) We sh all now study the behavior of b r(V ) Let Let
VC V
W
be an m-dimensional linear subspace of V, (0 < m < n).
M be any free WF -module with a G L (V )-actio n .
^ V
(
under extensions.
f =0!
and
M/M'.
Then
M' and
Let
M' =l f f M|
M" are G^-modules since
v V is a normal subgroup of G ^.
Wp -modules
0
M' -> M -» M"
It is obvious that the exact sequence of
0 is split over Wp
general not compatible with the action of G ^ .
but the splitting is in
For example if M = D (V ),
we have
M ' = M = D (V / V ) ® Pm ( V , V ) , by Proposition 3.12 (cusp condiwF wf tion) and 3.10. It follow s that M" = 0 in this case.
P
r o p o s it io n
.
m=
Let
©
D(\A) (0 < i < n).
Then
vxiv (i )
M' =
©
D(VS/ V ) ® Qi(Vs , V )
r TT
TT
W-,-,
S
vcvs m \s\m + i
and (ii)
M" -;©
D (V j) ® /
VCV where
Q j(V g , V ) ^
©
_D C ^ A
/
Pw (V j,V ^ n v )
if
v.cv 1_ s
m < s < m+i,
F
V. + V = V
1
©
V V jC V / V
s
Q i(V s , V ) = ker /
© ViCV s
\v.+v=vs
Pw (Vj, 0) - Wp\ F
/
and Q^(Vg , V ) = 0,
otherwise.
if
s = m+ i
82
T H E D IS C R E T E SERIES OF G L
Proof . Let
f - (f^
6 ^
O V E R A F IN IT E F IE L D
n
D (Vp, f^a ^ e D(V^a ^) where a
parametrizes
v .1c v the set of i-dimensional linear subspaces of V.
Let
f(a )
(0---0, f(a ), 0---0)
Assum e that
be the element obtained from f by replacing
all the coordinates (except for the one on place a) V| fl V f-0, V,
where
V- = V^a ^. Write
V-
by zero.
Assum e that
as a direct sum U '© U v *pjy
y
1
(note that U ^ p .y = I t e U- |t (V -)- V-!.
Then we have
(cf. Proposition 3.12). Assume now that a ll
C V,
t < U - , hence / V
where
Let ? (V S)
V-
such that V- -t V = Vg.
Vi C
(f ^ ) a( j ( V s ) ’ f' ^
The group
In this case,
(f(a \ f 5 ( v y
D (V -) = D (Vg / V )
f D (V ^a b
) _0
w
for all
is the same as
acts on ? (V S) by permutations (it leaves
fC“ ) - o in D(VS/V).
Vg
invari-
lf and only if
“ ‘ From these remarks it follow s that
aeff(Vs )
M' sc/ [
©
for
e D OJs /' V) -
ant), and it is clear that
^
t f ^ = f ^
be the set of all i-dimensional linear subspaces
Hence a family
a family
Then clearly
t \ l ^ = lU y | T (a i Let now VS C V be such that
s = m 4 i. of Vs
V- = V^a ^.
D C V jA ® /
V iC V
I
J
yv.nvfco.v.
©
D (Vi+m/ V ) ® Q i(V i+m, V ) \
( Vi+m C V
F
)
\ y c v i4m
and M" : / ©
VViC^
D (V i )\ ® /
/
©
Vi+mCV
D (Vi+m/V)\ •
T H E B R A U E R L IF T IN G
Since the subspaces
Vi+m C V
83
such that V C
are in 1 -1 corre
spondence with i-dimensional linear subspaces of V / V
the required
formula for M" follows. Next we observe that if Vj VjflVfO,^
1
1 1
Vs>
C O RO LLARY.
pw wf
( V , v .n v ) =
D(V-) fl
1
1
d (v . + v / v ) < s pw ( v ,v . n v ) 1 Wp F
and collect together terms corresponding
The required formula for M' follows.
Le t
be an algebraically closed field of characteristic
fl
-> Q
zero, and let ©
wF
We put Vg = V- + V,
to the same
such that
then
D(V-) = D(v-/v. n v) 0 (cf. 3.10).
is a linear subspace in V
Then the G L (V)-m odules
be a ring homomorphism.
are irreducible for all
1 < i < n except when we have
W
V.CV
F
1
simultaneously
q=2, n>2, i = l
in which case
0 D(\A) ® fl V l CV WF
has
two irreducible components ( one of which is the unit representation).
This is an easy consequence of the proposition.
5.3
We shall now prove that, given
V C V (0 < dim V = m < dim V = n)
there exists a natural exact sequence of WF -modules (12)
0 -> (D (V )X -» /
0
D(vn_ i j —
r ® v D(Vi ) )
VVn - l CV (the symbol
Mf has been defined in 5.2).
In fact, according to Proposition 5.2 it is sufficient to construct natural exact sequences (13)
0
pw (VS ,V) -* Q S_ 1 (VS ,V ) ->
> Q s _ m+1 (Vs .V) - Q s _ m(Vs ,V ) ^ 0
F
for all
Vg C V
such that V C Vg
tensoring with D(VS/ V )
and
m < s.
((12) follows from (13) by
and taking direct sums over all
Vg .) Now the
84
T H E D IS C R E T E SERIES OF G L n O V E R A F IN IT E F IE L D
natural exact sequence (13) is provided by 1.13(e) and our assertion follows.
It is obvious that the exact sequence (12) is compatible with the
action of G ^ .
(14)
It follows that
( —l ) n - 1 ( D ( V ) ) + ( —l ) n - 2 /
0
V
D (VV l ) \ + •■•+ ( - 1)7 - CV
n —1
©
D (V j)V = 0
Vvicv
)
/
in the Grothendieck group Rw ( G ~ ) . wp *F A V Let now
G be a finite group and assume that G
in the vector space
V.
We shall show that the correspondence
defines a group homomorphism br : R F (G ) -> Rw (G). ~
for any exact sequence G-action we have
is acting (linearly)
~
V -* br(V)
We have to show that
F
0 -> V -> V -> V / V -» 0 of F-vector spaces with
br(V ) = br(V ) + b r(V / V )
in Rw (G).
This follows from
F
(14)
and Proposition 5.2(ii). We can state the following
Th
e o r e m
.
Let
G
be a finite group and let F
be a F -vector space on which
G acts.
sequence of free Wp -modules with
be a finite field.
Let
V
Then there exists a canonical
G action
V^n^,V^n ^ , * * * , V ^ \ n = dimF V )
and a canonical exact sequence of G -modules.
0
->
® F -> \Kn - 1 ) ® F -> — > WF wp wF
® F ^ V -
0 .
Moreover the correspondence v - ( - l ) n _ 1 V (n) + ( - l )n - 2 V ( n _ 1 ) + - - - + ( - 1 ) ° V (1) defines a group homomorphism br : R p ( G ) -> RW^ ( G ) . identity map of R p ( G ) ,
where d : Rw (G )
R p (G )
We have
d © br =
is the decomposition
F
homomorphism induced by the canonical projection WF
5.4
F.
We shall now determine the character of b r(V ) on an arbitrary element
of G (where V is an F-vector space with a G-action).
It is clear that
T H E B R A U E R L IF T IN G
85
Trw (g|br(V)) - Tr(g|V) F
whenever
g e G
has order prime to p.
On the other hand we shall prove
that T t y p C g ’ M b r ( V ) ) - Trw (g ' |b r(V ))
whenever a power of
g', g" e G p and
the subgroup group.
G' of
are such that
g' has order prime to p,
g'- g =- g • g . We restrict rhe action of G
generated by
We can find a G'-invariant flag
spaces of V
such that g
is the
and
g
g";
V = V- + V- /V- 4 h x2 h
G
note that
V; C V- C •••CV; h x2 \
on V
l2
••• + V/V- in R^CG'). b
to
G' is a cyclic
of linear sub-
identity on V- ,V- /V- , ll
have
g" has order
. We
h
Ak
Since
br
is
agroup
homomorphism (cf. 5.3), we have br(V ) -
br(V: ) + br(V- /V- ) + • • • + br(V/V: ) H
l2
h
in Rw ( G ' )
xk
F
hence Trw ( g V ' l b r ( V ) ) = Trw Cg"g"" I br(Vi ) ) + T r w Cg"g""! br(Vi /V - )) F
F
1
F
+-
2
+ Trw ( g V I b r C V / V j ) ) F
= Trw (g'lbrCVj ))+ T rw (g | b r ^ F
H
1
F
2k
/V { 2
1
) ) + - + Tr(g | br(V/V- ))
- T r w (g '|b r(V )). F
This shows that the function
g -> Trw (g| br (V ))
is precisely the Brauer
F
lifting of the modular character of V
as a G-module (se e [5], [9]).
The fact that this is a virtual character of G was first pointed out by Green ([5], Theorem 1) and proved using Brauer’s characterization of characters in terms of elementary subgroups [ l l .
(This has been called
the Brauer lifting by Quillen [9], who showed how it can be applied to homotopy theory.)
5.5
In this section we show how Theorem 5.3 implies the following
k
86
T H E D IS C R E T E SERIES OF G L
(Swan [17], Theorem 3).
THEOREM
O VE R A F IN IT E F IE L D
n
G
Let
be a finite group and let F
Then the natural homomorphism Rw (G ) -* R~
be a finite field.
F
duced by the inclusion
Wp -* Q p
in-
isomorphism.
Note that Swan proves a more general result, where by ( A ,K ) A
(G )
yF
any semilocal Dedekind ring, and
K
(Wp , Q p )
is replaced
its field of fractions,
but this is not needed here. We define a map R n (G ) -* Rw (G )
Proof.
VF
Q p -vector space with a G-action. M e R0
in M and we send
vF
as follows.
F
(G )
Let
We choose a G-invariant Wp -lattice
to L e Rw (G).
to prove that given two G-invariant Wp -lattices we have
L = L'
in Rw (G).
L
We have to check that
F
this is well defined, i.e., independent of the choice of
pL C L ' C L,
M be a
L, L '
L.
It is sufficient
in M such that
Using the isomorphism
F
Rw (G ) -» R ^ F
(G )
(see 5.1) we see that it is enough to prove that
L = V
F
in R ^
(G).
We have an exact sequence
F
enough to prove that
L / L ' = 0 in R L
0 -> L ' -» L
(G).
Since
L / V -* 0. p L / L = 0,
It is then it follows
F
that
is an F-vector space with a G-action.
L /V
V ^n _ 1
can find a sequence
U sin g Theorem 5.3 we
V*-1^ of free Wp -modules with
G-action and an exact sequence of F-vector spaces compatible with the G-action:
0
® F -> V ^n _ 1 ^ ® F -> — > V ^1 ^ ® F -> L / V -* 0 . wp wp wp
In order to prove that L / V that
0 in
® F = 0 in R ^ (G), W
® F = Wp
R w (G ) Wp
it
is then enough toprove
i, 1 < i < n.
F
We have an exact sequence y (i)
for all
( G)
0 ->
= 0 in RL (G).
® F It follows that
0 hence
L = L'
in
WF
hence our map RQ (G ) -» Rw (G ) vF F
is w ell defined.
check that this is just the inverse of R W^ ( G ) -> R q f ( g ) is proved.
It is easy to
anc* the theorem
T H E B R A U E R L IF T IN G
1.
C O RO LLARY
Let
L, L ' e R w (G )
87
be such that
Tru; ( g | L ) = T r w (g|L'),
F
g e G.
for all
Then L - L '
F
F
in R w (G). F
C O RO LLARY
2.
The map br : R p ( G ) -> Rw (G )
is a ring homomorphism.
F
In fact, we must check that
br(V ® V ' ) --= br(V ) b r (V ')
in R w (G )
for any
F
F-vector spaces
V, V ' with G-action.
According to Corollary 1, it is
sufficient to check the equality of the corresponding traces, which is obvious.
5.6 Let
V
be a vector space of dimension
known that the Steinberg module
StF ( V )
The
F [G L
(V )]-m o d u le
F.
It is well
(s ee 1.13) is a projective inde
composable module for the group algebra
THEOREM.
n > 1 over
F [G L (V )].
is projective and indecom-
V ® S t F (V) F r
posable. provided q 4- 2 .
Proof.
We can clearly assume that
n > 2.
The fact that
V StF ( V )
is
F
projective follows from the fact that StF ( V ) there exists a unique projective = V ® S t F (V ).
is projective.
WF [G L (V )]-m od u le
It follows that
M such that M ® F Wp
M has the following character: t 6 G L (V ) .
Trw (t|M) = Tr(t| V ) ■ Trw^(tjStw^ ( V ) ) , We have (15)M = D ( V ) © / ©
Vj p ) \W ® . -. .e©// DCV/Vp ® Stw ( V
\^ 1 CV as
Q F [G L(V )]-m odu les.
WF
1
)
V
0
D (V / V n_ j ) , CV
vv« -
® Stw (Vn. WF
"F
This can be checked by a character computation
(the character of the Steinberg module is known, se e [15]).
For example
the equality of the dimensions of the two sides of (15) is the identity
88
THE DIS CRE T E SERIES OF G L n OVER A FIN IT E F IE L D
n . q U 2 + ...+( n - l )
y
q l +2+ ■■-4 ( i - l ) ( q i+ 1 - 1 ) ( q i+2- l > • ( q n- l )
iTo Replacing q by q _1
n
V
and multiplying with q 1 + 2 " + ( n- 1) this becomes
( l ) n *'i—1 (qi f l - l ) ( q i l 2 - l ) --(qn- l )
i= o
(c' n“ 1- 1>
which is precisely the identity expressing the vanishing of the alternating sum of dimensions of the terms in the exact sequence 1.14(c). now that V ® St~ (V ) F
posable.
is not indecomposable.
Assume
Then M is not indecom-
*
Since the GL(V)-modules
®
D(V/V-) ® St(V;) (0 < i < n - 1 ) w
V .^ V
F
are irreducible when the scalars are extended to any field of characteristic zero
(q
2) we must have then the identity
(16)
1 + 2 , K+(nn -- 1l )K £V mq1+21....
q > - 2 ....... ( i - l ) ( q i+1- l ) ( q i t 2 - l ) - - ( q n- l )
i= 0 where
0 < m < n and cr(0 < i < n—1) are numbers equal to 0 or
such that a Q = 0. tive,
M1 f-0,M
(In fact, we can write M -
and such that
the dimension of IV^
D (V )
©M2 with
IV^
1,
and
projec
occurs in M2 - It is w ell known that
must be of the form
m - q 1+2+
+ ^n 1^).
The
identity (16) is im possible since its two sides have distinct p-adic valu a tions.
The theorem is proved.
Remark.
The case
n = 2 of the theorem is due to V. Jeyakumar [7].
5.7 According to 4.10 the G L(V)-m odule
©
D(V^) considered in 5.2
v.cv can be defined naturally over the discrete valuation ring © F • C Kp • (1 < i < n).
T H E B R A U E R L IF T IN G
Let A ( F n)
KF
be the subfield of Q F
for all
Then 0 p -
n > 2.
We have
Kp = Jim^ Kp
lim^ 0 p n and © F
a unique prime ideal (= p ' f F ) D (V )
F , n 0 F ,n
bre
F
(V ) -
generated by
89
x, x e F * Let
and by
0 p = Kp fl Wp .
is a discrete valuation ring in Kp
with residue field
F.
Let
with
DCV)^
0 F . The formal alternating sum
( - 1 )11- 1 D ( V ) e
F
( - 1)° ©
n- 2 (-i):
VjCV
Vn - l CV can then be regarded as an element in R/c> ( G L ( V ) )
F
(cf. 5.1).
get a natural Brauer-lifting homomorphism R p ( G ) -> R q all finite groups
G.
F
(G )
We thus
defined for
It is easy to check that all the proofs given in the
case of the Witt ring Wp
remain valid when
Wp
is replaced by O p .
There is a natural commutative diagram br,c R F (G )
R«
(O
Rw
(G )
br
which shows that the map br^
5.8
Let
k be an integer.
F
is a refinement of br.
Consider the cla ss function X v ^ : G L ( V ) ^ K p
defined as follows: (17)
where
* v>k(t) = ( - l ) n+j Tr(tk|Van) ( q m- l ) ( q 2 m- l ) - - ( q ( j “ l ) m - l ) t