9
$\begingroup$

Let $S$ be an integral regular scheme and let $T\to S$ be a finite etale morphism. Let $G$ be a smooth affine finite type group scheme over $S$.

Is the set of $S$-isomorphism classes of $G$-torsors over $S$ which are trivial over $T$ finite?

I guess the set-up is ridiculously general. So what if

1) $S$ is of finite type over $\mathbb C$, or

2) $S$ is of finite type $\mathbb F_p$, or

4) $S$ is of finite type over $\mathbb Z_p$, or

5) $S$ is of finite type over $\mathbb Q_p$, or

6) $S$ is of finite type over $\mathbb Z$?

$\endgroup$

1 Answer 1

6
$\begingroup$

The answer to the general question is NO.

We take $S=\mathrm{Spec}\, K$, where $K$ is a number field. Let $L/K$ be a finite Galois extension. We set $T=\mathrm{Spec}\, L$. We consider the norm homomorphism $$N_{L/K}\colon R_{L/K}\mathbb{G}_{m,L}\to \mathbb{G}_{m,K}\,,$$ where $R_{L/K}$ denotes the Weil restriction of scalars. Set $G=\mathrm{ker}\, N_{L/K}$. Then our torus $G$ splits over $L$, hence $H^1(L,G)=1$ by Theorem 90. From the short exact sequence $$1\to G\to R_{L/K}\mathbb{G}_{m,L}\to \mathbb{G}_{m,K}\to 1$$ and the induced Galois cohomology exact sequence $$ L^*\to K^* \to H^1(K,G)\to H^1(K,R_{L/K}\mathbb{G}_{m,L})=1$$ we obtain that $H^1(K,G)= K^*/ N_{L/K}(L^*)$, and this group is infinite. Thus the kernel $$\mathrm{ker}[H^1(K,G)\to H^1(L,G)]=H^1(K,G)= K^*/ N_{L/K}(L^*)$$ is infinite, as required.

$\endgroup$
1
  • 5
    $\begingroup$ I wanted to compute an explicit example. For instance we might view the circles $x^2+y^2=N$ as torsors over the circle group $x^2+y^2=1$, and then the circles $N$ and $M$ are only isomorphic if $N/M=a^2+b^2$ for $a$ and $b$ rational, so the set where $N$ is a product of distinct primes congruent to $3$ mod $4$ are all distinct torsors. $\endgroup$
    – Will Sawin
    Jul 19, 2015 at 22:21

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.