Let be the Frobenius map which raises elements to the q-th power, and
let it act entrywise on G. The set of fixed points under F in G
is
of order
Let R be the stabilizer in G of f and S its connected component.
Then is finite, say of size n, so there
are coset representatives
of S in R.
We assume that q has been
chosen so they are fixed by F. Assume
is a complete set of
representatives for the conjugacy classes of
. We prove the following:
Theorem. There are exactly r orbits of H acting on Y.
The orbits are precisely
,
, where
with
. Moreover, if
denotes the subgroup of R consisting of
all elements
for which
, the number
of elements in
is
.
We proceed in five steps.
Notice first that ,
as f has coefficients in
. It follows that
if
, then
,
and so
so
for
and
. Suppose on the other hand that
for some i,
and s in S;
then
. This means that all elements in Y are of the
form gf for such a g.
B. For each i, there is
a
for which
.
Lang's theorem applies to
G acting on itself as G is connected. The fixed points of
the automorphism F are finite and so there is a for which
.
To see the converse suppose satisfies
. Then
and so
.
This means the coset
is exactly the set of elements g of
G for which
. There is one orbit of H in
Y given by
.
D. Suppose that satisfies
with
and
for some
i. Then
.
We will show that there exists
satisfying
. Then
by C.
Notice that
,
and so we want to find
with
or, equivalently,
Let . Note that
since
, the Frobenius map F stabilizes R and so S as well.
By Lang's theorem,
all elements in the coset FG are conjugate in
.
Thus
fixes only finitely many
points (indeed |H| points) on G and so also on S.
By Lang's theorem, applied to
acting on S, there is such an
.
E.
For
, we have
if and only if
and
are conjugate in
.
Suppose and
are conjugate in
by an element
. This means
for
.
Then
. From D.,
it follows that
. Note that
and so the
H-orbits of
and of
are the same.
Suppose on the other hand that and
are in the same
H-orbit. We may assume
for some
and
so
for one of the coset representatives t and
. Now
and so
. But under F, the coset tS is mapped
to tS as S is mapped to S and so
mod S,
and
are conjugate.
This means we need only consider the orbits
.
F. The only remaining part is to show that the number of
elements in the orbit of is as indicated. The stabilizer
in H of
is exactly
. But,
for
, the conjugate
is in H if and only if it is fixed by F. This gives
which is
equivalent to
as given, that is,
.
Corollary. Suppose the stabilizer R splits
as an extension over S (so that we can choose
the coset representatives
of S in R in such a way that they form a
group themselves fixed by F).
Then
As above, let be the set of elements r in R
satisfying
, so that
.
Suppose
. Write
with
. Then
Because
we have chosen the
to be fixed by F,
this means
. But this
implies that
and
commute modulo S and so commute
as the
form a group.
This implies
.
Thus,
where
is the centralizer of
in
.
By the theorem, we now have
.
But
is the size of the conjugacy class and so the result
follows.
Corollary. If R is finite, then
|Y|=|H|.
|S|=1 in the above.
Corollary. If (in particular
if R=S), then
(where
are the just the fixed
points of F on S).
Any induces an inner
automorphism on S. It follows by Lang's theorem,
that
is conjugate to F by an element of S and so has the
same number of fixed points as F on S.
We now add the contributions to
from the orbits of forms
,
.
Note that the stabilizers in G of these forms are given in
.2. Here h=|H| is as in the beginning of this section.
For
,
we shall denote by
the set
.
The stabilizer of is finite and so, by Corollary 6.3,
the number of points in
is h.
The stabilizer of is
where T is a
one-dimensional
torus and
is the cyclic group of order 2
generated by the matrix
. There are two
orbits, each of size
, so
.
The stabilizer of is connected and so the number of orbits is
one and the size is
.
Now adding the contributions for these three orbits and subtracting
from gives a polynomial of degree 13.
We now quote a result by [LW] which shows there are no more orbits of dimension 14 or higher.
Theorem. [LW]. Let X be a d-dimensional
subvariety of an n-dimensional projective space defined
over . There exists a constant c (depending only on
X, d and n - in fact, not on X explicitly, but
on the degree of X in the projective space)
Note this result also works
for subvarieties of affine space.
For one direction, embed the affine variety in
projective space and consider the closure;
then where
is the number
of
-points of the closure and use the upper bound
for that. For the other bound, choose an open subset O
contained in the original variety. Then the number of
-points
on the closure
of O
and the complement of O in
can be estimated, and so also
on O; this gives the appropriate lower bound for
.
Corollary.
Theorem
holds.
If satisfies
, then, by Theorem 6.5, the number of
-points of Gf
(the only dense orbit of
) is at least
(for some constant c), which contradicts
the fact that the number of
-points of
is a polynomial in q of degree 13. This shows
there are no further orbits of dimension 14 or higher. In
particular, all other orbits lie in
. In , any form in
was shown to be listed in Table 1.
Our work is over the algebraic closure of
. However arguments
in [GLMS] show the same applies over any algebraically closed field of
characteristic 3.