Representations of algebraic groups and their Lie algebras

Jens Carsten Jantzen

Lecture I
Set-up. Let K be an algebraically closed eld. By convention all our algebraic groups
will be linear algebraic groups over K. For such a group G let K[G] denote the algebra
of regular functions on G. For the background on algebraic groups I refer to the books
Linear Algebraic Groups by J. E. Humphreys, T. A. Springer, and A. Borel.
Modules. Let G be an algebraic group. A Gmodule is by denition a module V over
the group algebra KG of G as an abstract group such that V satises two additional
conditions:
V is locally nite: each element is contained in a nite dimensional KGsubmodule.
For each v V and V

the matrix coecient c

,v
given by c
,v
(g) = (g v) for
all g G is a regular function: c
,v
K[G].
For example, any nite dmensional vector space V is a module under the natural action
for the general linear group GL(V ) as well as for any closed subgroup of GL(V ). For any
algebraic group G the algebra of regular functions K[G] is a Gmodule under the (left)
regular representation given by (gf) (g

) = f(g
1
g

).
It is the easy to chech that the category of all Gmodules is closed under:
submodules
factor modules
direct sums
tensor products
symmetric and exterior powers.
If a Gmodule V is nite dimensional, then also its dual space V

is a Gmodule.
Frobenius twist. Assume char(K) = p > 0. In this case we have another method to
construct new Gmodules from known ones. If V is a vector space over K we denote by
V
(1)
the vector space over K that coincides with V as an additive group and where the
scalar multiplication is given by
a v =
p

a v for all a K, v V
where the left hand side is the new multiplication and the right hand side the old one.
(Since K is algebraically closed, the Frobenius morphism a a
p
is a bijection K K, so
any element in K has a unique pth root.)
1
If V is a Gmodule, then V
(1)
is a Gmodule using the given action of any g G on
the additive group V
(1)
= V . The action is again Klinear. A nite dimensional G
stable subspace of V is also a nite dimensional Gstable subspace of V
(1)
(of the same
dimension); so also V
(1)
is locally nite. For any V

the map

(1)
: V
(1)
K, v (v)
p
is Klinear. One checks that (V
(1)
)

=
(1)
[ V

and gets nally for all g G

c

(1)
,v
(g) =
(1)
(gv) = (gv)
p
= c
,v
(g)
p
hence c

(1)
,v
= c
p
,v
K[G].
One denes higher Frobenius twists inductively: Set V
(r+1)
= (V
(r)
)
(1)
.
Exercise: Consider K[G] as a Gmodule under the (left) regular representation. Show
that f f
p
is a homomorphism K[G]
(1)
K[G] of Gmodules.
Induction. Let H be a closed subgroup of an algebraic group G. Any Gmodule V is
by restriction an Hmodule since the restriction to H of any function in K[G] belongs
to K[H]. This restriction of modules is clearly a functor from Gmodules to Hmodules.
We can dene an induction functor ind
G
H
in the opposite direction, from Hmodules to
Gmodules. For any Hmodule M set
ind
G
H
M = f: G M regular [ f(gh) = h
1
f(g) for all g G, h H.
Here a function f: G M is called regular if there exists m
1
, m
2
, . . . , m
r
M and
f
1
, f
2
, . . . , f
r
K[G] such that f(g) =

r
i=1
f
i
(g) m
i
for all g G. All regular functions
G M form a vector space over K and ind
G
H
M is then a subspace. We get a Gaction
on this subspace by setting (gf)(g

) = f(g
1
g

). In order to show that this KGmodule

is a Gmodule one uses the fact that K[G] is a Gmodule.
A trivial example for an induced module is ind
G
{1}
= K[G]. Well soon see more interesting
examples for G = SL
2
(K).
The functor ind
G
H
is right adjoint to the restriction functor from Gmodules to Hmodules.
This means that we have functorial isomorphisms
Hom
G
(V, ind
G
H
M)

Hom
H
(V, M)
Here any Ghomomorphism : V ind
G
H
M is sent to : V M with (v) = (v) (1).
Conversely any Hhomomorphism : V M goes to

: V ind
G
H
M such that

(v) (g) =
(g
1
v).
Example: SL
2
(K). Consider G = SL
2
(K) acting on its natural module V = K
2
with its
standard basis e
1
, e
2
. Then G acts also on V

and on the symmetric algebra S(V

) that
we can identify with the algebra K[X
1
, X
2
] of polynomial functions on V . Here X
1
, X
2
is
the basis of V

dual to e
1
, e
2
.
2
The action of G on S(V

) is given by (gf) (v) = f(g

1
v). Each symmetric power S
n
(V

)
is a submodule. We choose as basis all v
i
= (1)
i
X
i
1
X
ni
2
, 0 i n, and get (e.g.) for
all a K

1 a
0 1

v
i
=
i

j=0

i
j

a
ij
v
j
and

1 0
a 1

v
i
=
n

j=i

n i
n j

a
ji
v
j
(1)
and for all t K

t 0
0 t
1

v
i
= t
n2i
v
i
. (2)
Consider the closed subgroups in G
U =

1 0
a 1

[ a K and B =

t 0
a t
1

[ a K, t K

.
Let us rst observe that we have an isomorphism of Gmodules
S(V

)

ind
G
U
K
where K is the trivial Umodule such that
ind
G
U
K = f K[G] [ f(gu) = f(g) for all g G, u U .
Well, any S(V

) denes

: G K setting

(g) = (ge
2
). One checks that

ind
G
U
K
and that

is a morphism of Gmodules. On the other hand, any f ind
G
U
K denes
f: V 0 K: For any v V , v ,= 0 there exists g G with v = ge
2
; we then set
f(v) = f(g). This is well dened as ge
2
= g

e
2
implies g

gU, hence f(g) = f(g

) since
f ind
G
U
K. One checks that f is a regular function on V 0. Algebraic geometry tells
you that f has a (unique) extension to a regular function on V . So there exists S(V

)
with f(v) = (v) for all v ,= 0, hence with f(g) = (ge
2
) for all g G, i.e., with f =

.
For each n Z let K
n
denote K considered as a Bmodule such that

t 0
a t
1

v = t
n
v for all a, t, v K, t ,= 0.
Since any u U acts trivially on any K
n
, it is clear that ind
G
B
K
n
ind
G
U
K. The inverse
image of ind
G
B
K
n
under the isomorphism S(V

) ind
G
U
K from above consists of all
S(V

) with (tv) = t
n
(v) for all t K

and v V . This shows that we get

isomorphisms
ind
G
B
K
n

S
n
(V

) if n 0,
0 if n < 0.
Simple modules for SL
2
(K). Consider again G = SL
2
(K). If char K = 0, then one can
show that all S
n
(V

) with n N are simple Gmodules. [Using (2) above one checks that
3
each non-zero submodule contains some v
i
. Then (1) shows that the submodule contains
all v
j
. Here one uses that all binomial coecients are non-zero in K.] Furthermore, each
simple Gmodule is isomorphic to S
n
(V

) for a unique n.
If char K = p > 0, then the same argument shows that all S
n
(V

) with 0 n < p are

simple Gmodules. But one can quickly see that S
p
(V

) is not simple: The map

: V

S
p
(V

), f f
p
is semi-linear (i.e., satises (f + f

) = (f) + (f

) and (af) = a
p
(f) for all a K)
and Gequivariant. So its image is a Gsubmodule of S
p
(V

) that is non-zero and not the

whole module.
Going back to the denition of a Frobenius twist one checks that is an isomorphism of
Gmodules
(V

)
(1)

KGX
p
2
onto the submodule of S
p
(V

) generated by X
p
2
.
For general n the characteristic 0 argument shows rst that each non-zero submodule
of S
n
(V

) contains some v
i
. Then the rst formula in (1) implies that the submodule
contains v
0
= X
n
2
. This shows that the submodule
L(n) := KGv
0
= KGX
n
2
of S
n
(V

) generated by X
n
2
is simple, in fact, it is the only simple submodule of S
n
(V

).
It then turns out that each simple Gmodule is isomorphic to L(n) for exactly one n.
One can describe L(n) alternatively as follows. Write n in padic expansion:
n = n
0
+n
1
p +n
2
p
2
+ +n
r
p
r
with 0 n
i
< p for all i.
Then the map
f
0
f
1
f
2
f
r
f
0
f
p
1
f
p
2
2
. . . f
p
r
r
is an isomorphism of Gmodules
S
n
0
(V

) S
n
1
(V

)
(1)
S
n
2
(V

)
(2)
S
n
r
(V

)
(r)

L(n).
Notations. Let G be a connected reductive group over K. Choose a maximal torus T
in G. We want to generalise the SL
2
example. This involves some notation:
X = Hom
alg.gr
(T, K

), the character group of T

X

= Hom
alg.gr
(K

, T), the cocharacter group of T

, ), the pairing X X

Z such that ((t)) = t

,
for all t K

X, the root system of G with respect to T

[ X

the corresponding dual roots

+
a chosen system of positive roots
4
B and B
+
the Borel subgroups of G containing T that correspond to
+
and
+
respectively
U and U
+
the unipotent radicals of these Borel subgroups. So we have semi-direct
decompositions B = T U and B
+
= T U
+
as algebraic groups.
Each X can be extended to a character : B K

such that (tu) = (t) for all t T

and u U. Denote by K

the corresponding one dimensional Bmodule. The general

theory of connected soluble algebraic groups yields:
Proposition: Each simple Bmodule is isomorphic to K

for a unique X.
A corresponding statement holds for B
+
instead of B.
Simple Gmodules. Any Gmodule is by denition locally nite, i.e., a union of nite
dimensional submodules. Therefore any simple Gmodule V is nite dimensional. Consid-
ered as a Bmodule it then has a composition series. This implies in particular that there
is a surjective homomorphism of Bmodules from V onto a simple Bmodule, hence by
the proposition above onto some K

with X. The basic property of induction implies

then
Hom
G
(V, ind
G
B
K

) Hom
B
(V, K

) ,= 0.
Using the simplicity of V we get now:
Claim 1: Each simple Gmodule is isomorphic to a simple submodule of some ind
G
B
K

with X.
In the case of SL
2
(K) this implies that any simple module is (as claimed above) isomorphic
to some L(n).
The next step towards a classication of the simple modules requires more subtle arguments
that I cannot discuss here. They yield:
Claim 2: ind
G
B
K

,= 0 dominant
where the set X
+
of dominant characters is dened by
X
+
= X [ ,

) 0 for all
+
.
As a third step we want to see that each ind
G
B
K

with dominant contains a unique

simple submodule. This is done by looking at xed points for the subgroup U
+
. For any
non-zero U
+
module M the space of xed points
M
U
+
= x M [ ux = x for all u U
+

is non-zero as U
+
is connected and unipotent.
The denition of ind
G
B
K

and of the Gaction on this space shows for any f (ind

G
B
K

)
U
+
that
f(u
1
tu
2
) = (t)
1
f(1) for all u
1
U
+
, t T, u
2
U.
5
So the restriction of f to the subset U
+
TU of G is determined by f(1). As U
+
TU is dense
in G, it follows that f itself is determined by f(1). This implies that
(ind
G
B
K

)
U
+
= K f

where f

K[G] is the unique regular function on G with

f

(u
1
tu
2
) = (t)
1
for all u
1
U
+
, t T, u
2
U.
Now every non-zero Gsubmodule M of ind
G
B
K

satises M
U
+
,= 0, hence f

M and
KGf

M. It follows that
L() := KGf

is the unique simple submodule of ind

G
B
K

.
Theorem: Each simple Gmodule is isomorphic to L() for a unique X
+
.
In order to get uniqueness one notes that L()
U
+
= Kf

and that T acts on this subspace

via .
In case char K = 0 one can show that L() = ind
G
B
K

for all X
+
. The examples
above show that this fails in prime characteristic already for G = SL
2
(K).
Steinbergs tensor product theorem. In order to simplify notation assume now that G
is semi-simple and simply connected. Our choice of positive roots
+
determines a system
of simple roots. The corresponding dual roots form under our assumption a basis for the
free abelian group X

and X has then a dual basis consisting of fundamental weight. For

any integer m > 0 the set
X
m
= X [ 0 ,

) < m for all simple roots X

+
consists of those characters whose coecients with repect to the basis of fundamental
weights belong to the interval from 0 to m1.
Suppose now that char K = p > 0. Any X
+
has a padic expansion
=
0
+p
1
+p
2

2
+ +p
r

r
with all
i
X
p
.
Now Steinbergs tensor product theorem says (generalising the result for SL
2
) that we have
an isomorphism
L(
0
) L(
1
)
(1)
L(
2
)
(2)
L(
r
)
(r)

L().
Note that each L(
i
) ind
G
B
K

i
K[G] is a submodule of K[G]; similarly for L(). The
isomorphism is then simply given by
f
0
f
1
f
2
f
r
f
0
f
p
1
f
p
2
2
. . . f
p
r
r
Application to nite groups of Lie type. Assume again that G is semi-simple and
simply connected. Suppose that char K = p > 0 and that G is dened over a nite subeld
F
q
of K. (So q = p
n
for some n > 0.) The group G(F
q
) of points of G over F
q
is then a
nite group of Lie type. We can describe G(F
q
) as the xed points of a suitable Frobenius
map on G.
6
In this situation a theorem due to Curtis for q = p and to Steinberg in general says:
Theorem: Each L() with X
q
is still simple when considered as a KG(F
q
)module.
Any simple KG(F
q
)module is isomorphic to L() for a unique X
q
.
Note that the Ree and Suzuki groups are nite groups of Lie type, which are not of the
form G(F
q
). However, a modication of this theorem yields also a classication of their
irreducible representations over an algebraically closed eld of their dening characteristic.
References. For a more detailed survey of the material in Lectures I and II see my article
Modular representations of reductive groups in the proceedings of a conference in Beijing
published under the title Group Theory, Beijing 1984 as volume 1185 of the Springer
Lecture Notes in Mathematics.
7