1
$\begingroup$

Let $k$ be a field with separable closure $\overline k$ and absolute Galois group $G$. We know that sheaves of abelian groups on the etale site of $k$ correspond to abelian groups with a $G$ action (obtained by taking stalks of the sheaf at a geometric point).

Let $l/k$ be a finite separable extension of fields of dimension $n$ and let $\pi: \operatorname{Spec} l \to \operatorname{Spec} k$ be the map on schemes. Given a sheaf $\mathscr A$ on the etale site over $l$, we can define it's pushforward $\pi_*\mathscr A$.

Is there a simple description of this in terms of the module $\mathscr A$ along with what the Galois action of $G$ looks like in terms of the Galois action of $\operatorname{Gal}(\overline k/l)?$

At least when $\mathscr A$ is the constant sheaf, I think the pushforward is $\mathscr A\times {\operatorname{Gal}(l/k)}$ and the action is trivial on the first factor and on the left on the second factor. I suspect that the answer in general is something like $\mathscr A\times {\operatorname{Gal}(l/k)}$ but I am not sure what the action on the first factor should be...

$\endgroup$
2
$\begingroup$

Your description $\mathscr A\times\operatorname{Gal}(l/k)$ does not works, in fact $\mathscr A\times\operatorname{Gal}(l/k)$ might not be an abelian group.

You know that the pull-back corresponds to forgetting the action. More precisely, if $\mathcal{F}$ is an étale sheaf on $k$, corresponding to a Galois representation on $\mathcal{F}_{\overline{k}}$, then the pull-back $\pi^*\mathcal{F}$ correspond to the restriction of the representation to the group $\operatorname{Gal}(\overline{k}/l)\subset\operatorname{Gal}(\overline{k}/k)$.

The push-forward is the right-adjoint of this functor. You can explicity compute this right-adjoint using the adjunction isomorphism with $\pi^*\mathbb{Z}[\operatorname{Gal}(\overline{k}/k)]$. You will find that this corresponds to $\mathscr{A}\mapsto \operatorname{Hom}_{\operatorname{Gal}(\overline{k}/l)}(\mathbb{Z}[\operatorname{Gal}(\overline{k}/k)],\mathscr{A})$ with the action $(\sigma.f)(x)=f(x\sigma)$.

When $\mathscr{A}$ is constant and if $l/k$ is Galois, another description is as follow : $\pi_*\mathscr{A}=\prod_{\sigma\in\operatorname{Gal}(l/k)}\mathscr{A}$ where the action on the right-hand side is given by $\tau(a_\sigma)_\sigma=(a_{\sigma\tau})_\sigma$. (If $l/k$ is not Galois, you will have to replace $\operatorname{Gal}(l/k)$ by $G_l\backslash G_k$ the set of right-cosets). Using representative of right-cosets, you can have a similar description when $\mathscr{A}$ is not constant.

Note that this also works if $l/k$ is not-finite but you will need to be careful with the topology as $\pi_*\mathscr{A}$ will not be discrete anymore.

$\endgroup$
  • $\begingroup$ Thanks for answering - the adjoint perspective clears everything up immediately and I really should have seen that. Also, I was using non standard notation, by $\mathscr A\times G$, I meant $\prod_{g\in G}\mathscr A$ and the action is by permuting the factors. Sorry for that. $\endgroup$ – Asvin Jun 20 '17 at 15:15
  • $\begingroup$ Yes I was confident that it was what you meant though it is not clear in the context of groups ! (And just to be pedantic, $\mathscr{A}\times G=\coprod_{g\in G}\mathscr{A}$ in categories where $\coprod\neq\prod$, this distinction is also relevant in the case of infinite extension). $\endgroup$ – Roland Jun 20 '17 at 15:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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