12
$\begingroup$

This question is a sequel to Pointless groups, to which @DanielLitt produced an elegant and easy-to-understand counter-example, and Pointless groups II, where @R.vanDobbendeBruyn pointed out that my implicit claim about Zariski density of rational points of groups over infinite fields is only guaranteed for reductive groups (or perfect fields), and gave a counterexample (still elegant, though less easy for me to understand than a split torus over $\mathbb F_2$!) for an infinite but imperfect field.

At the suggestion of @R.vanDobbendeBruyn and @YCor, I am engaging in the extreme bad manners of once again modifying my search for counterexamples to rule out the counterexamples already provided. The revised question is @DanielLitt's wording.

So, 3rd time's a charm, hopefully … fix a finite field $k$. Then "group" means "smooth, connected, affine algebraic group scheme of finite type over $k$".

If $G$ is a (smooth, connected) group, and $G(k)$ is trivial, then does that imply that $k = \mathbb F_2$ and $G$ is a split torus?

$\endgroup$
4
  • 1
    $\begingroup$ By Corollary 1.5 here we can reduce to the case of tori. Some version of Weil's bound on number of points on a variety might give the required result. (Edit: not quite correct, I forgot the unipotent case. YCor's comment below is correct though) $\endgroup$
    – Wojowu
    Jun 11, 2022 at 14:01
  • 3
    $\begingroup$ By a theorem of Grothendieck, $G$ has a maximal torus that is defined over $k$. Hence, if, for given $k$, $G$ is a counterexample of minimal dimension, then $G$ is either a torus or unipotent. In Borel's book(16.5(ii)) it is proved that quotient maps are surjective on $k$-points for finite fields. Hence one can suppose that $G$ is abelian. For perfect fields, the unipotent abelian groups are $k$-isomorphic to the standard one. So a minimal counterexample should be a torus. $\endgroup$
    – YCor
    Jun 11, 2022 at 14:02
  • $\begingroup$ In turn, since a non-trivial split torus over a field $k \ne \mathbb F_2$ has non-trivial $k$-points, a minimal counterexample when $k \ne \mathbb F_2$ is an anisotropic torus. A minimal counterexample when $k = \mathbb F_2$ is not split (by the way the question was stated), so has a non-trivial anisotropic torus, which necessarily has no non-trivial $k$-points (since the full torus has none); so, even for $k = \mathbb F_2$, a minimal counterexample is anisotropic. $\endgroup$
    – LSpice
    Jun 11, 2022 at 15:11
  • 3
    $\begingroup$ I think this finishes it. The number of rational points of a torus of dimension $d$ over $\mathbb{F}_q$ is computed by $(-1)^d\det(\Phi-q\cdot \text{Id})$, where $\Phi$ is the action of Frobenius on the characters. But for $q>2$, or $\Phi$ not equal to the identity, $\Phi-q\cdot \text{Id}$ has all eigenvalues of size at least $1$, and at least one eigenvalue greater than one. So $(-1)^d\det(\Phi-q\cdot \text{Id})=1$ implies $q=2$ and $\Phi=\text{Id}$, which is what we were trying to prove. $\endgroup$ Jun 11, 2022 at 15:16

1 Answer 1

13
$\begingroup$

Yes.

Let $|k| = q$. Let $T$ be an $n$-dimensional torus defined over $k$ and let $\overline{T}$ be the base change of $T$ to $\overline{k}$. Then the character group $\text{Hom}(\overline{T}, \mathbb{G}_m)$ is isomorphic to $\mathbb{Z}^n$ and the Frobenius acts on $\text{Hom}(\overline{T}, \mathbb{G}_m)$ by a matrix $F$. If $T$ splits over $\mathbb{F}_{q^n}$, then $F$ obeys $F^n = q^n \text{Id}$. The isomorphism class of the torus $T$ is determined by the $\text{GL}_n(\mathbb{Z})$ conjugacy class of $F$, and we have $\# T(\mathbb{F}_q) = \det(F - \text{Id})$. So we are reduced to the following linear algebra problem:

If $F$ is an integer matrix obeying $F^n = q^n \text{Id}$ and $\det(F - \text{Id}) = 1$, can we deduce that $q=2$ and $F = 2 \text{Id}$?

The answer is "yes". The condition that $F^n = q^n \text{Id}$ means that the eigenvalues of $F$ are of the form $q \zeta$ for $\zeta$ a root of unity. So we have $\det(F - \text{Id}) = \prod_i (q \zeta_i - 1) = 1$ where the $\zeta_i$ are various roots of unity.

But $|q \zeta - 1| \geq 1$ for any prime power $q$ and any root of unity $\zeta$, and strictly greater unless $q=2$ and $\zeta=1$, so the only solution is $q=2$ and $F = \text{Id}$. $\square$

$\endgroup$
9
  • 1
    $\begingroup$ The point-counting approach was also mentioned by @DanielLitt. $\endgroup$
    – LSpice
    Jun 11, 2022 at 15:58
  • 2
    $\begingroup$ The proof above implies the result for every affine group scheme $G$ (since we are counting points over a finite field $k$, we can replace $G$ by its associated reduced scheme, which is a smooth affine group scheme). There is a smooth projective variety over $k$ that parameterizes Borel subgroup schemes, and this variety is $\mathcal{O}$-acyclic, hence has a point. Thus $G$ contains a Borel $B$. This is a torsor for $R_u(B)$ over $B/R_u(B)$, and all $R_u(B)$-torsors over $k$ have $|k|^d$ points, where $d=\text{dim}R_u(B)$. So $R_u(B)$ is trivial, and $B/R_u(B)$ has a unique point . . . $\endgroup$ Jun 11, 2022 at 17:08
  • 2
    $\begingroup$ . . . Thus $B/R_u(B)$ is a split torus and $|k|=2$ by the proof above. Since $d$ equals $0$, the semisimple part of $G$ is trivial. So $G$ equals $B$ equals $B/R_u(B)$ equals a split torus. $\endgroup$ Jun 11, 2022 at 17:09
  • 1
    $\begingroup$ Here $\mathcal{O}$-acyclic is as in the work of Katz on the Ax-Grothendieck theorem: every smooth, projective, geometrically connected variety over a finite field such that the higher cohomology of the structure sheaf vanishes has number of rational points congruent to $1$ modulo the cardinality of the field. $\endgroup$ Jun 11, 2022 at 21:31
  • 1
    $\begingroup$ Just a correction. That theorem is the "Ax-Katz theorem". Katz's paper does follow earlier results of Ax and Grothendieck, but the "Ax-Grothendieck theorem" is a different result (using point counts of varieties over finite fields). $\endgroup$ Jun 11, 2022 at 22:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.