Let $S$ be a scheme over a field $k$, and let $G$ be a reductive group scheme over $S$. Let us call it trivial, if it is a pullback of a group scheme over $k$ via the structure morphism $S\to k$. Is it always true that $G$ becomes trivial after a certain etale base change $S'\to S$? I am willing to assume that $S$ is smooth if needed.
9

$\begingroup$ The question looks natural, but it would be interesting to know a little more about the background or the implications. $\endgroup$ – Jim Humphreys Apr 23 '11 at 14:56

2$\begingroup$ If you look at SGA 3, Prop XIX.6.1, it is essentially shown that any reductive group over any base is split after a finite etale base change. So if you allow extensions of scalars of $k$ as well, then what you need follows. $\endgroup$ – Keerthi Madapusi Pera Apr 23 '11 at 15:11

$\begingroup$ Sorry, it appears as if the Proposition does not guarantee you a finite etale base change, though I would be surprised if this weren't true. $\endgroup$ – Keerthi Madapusi Pera Apr 23 '11 at 15:13

6$\begingroup$ I have been informed by the esteemed BCnrd that my previous comment was too optimistic. In fact, in his paper 'Groups over $\mathbb{Z}$', B. Gross has shown the existence of nonsplit simplyconnected semisimple groups over $\mathbb{Q}$ that have reductive models over $\mathbb{Z}$. In particular, these models can only split after a nonfinite etale basechange over $\mathbb{Z}$. $\endgroup$ – Keerthi Madapusi Pera Apr 23 '11 at 21:54

$\begingroup$ Answering the question of Jim, if the statement is correct, then reductive group schemes over $S$ are classified by $H^1_{et}(S, Aut_G)$, right? $\endgroup$ – Roman Fedorov Apr 25 '11 at 17:26
6
$\begingroup$
$\endgroup$
Reductive groups schemes over $S$ are classified by $H^1_{fpqc}(S,Aut_G)$, see SGA 3 Exp. XXIV.

6$\begingroup$ You can indeed replace $fpqc$ by $et$, see SGA 3 Exp. XXIV Cor. 1.18 $\endgroup$ – Victor Petrov Apr 29 '11 at 11:26

4$\begingroup$ Note that SGA3 has its own definition of reductive. The geometric fibers have to be connected. $\endgroup$ – Wilberd van der Kallen Apr 29 '11 at 15:55