2
$\begingroup$

Let $S=\mathrm{Spec}(R)$, $s=\mathrm{Spec}(k)$ and $\eta=\mathrm{Spec}(K)$, where $R$ is a d.v.r. with fraction field $K$. Let $j:\eta\rightarrow S$

Now how to compute the sheaf $R^1j_*(\mathbb{G}_{m,\eta})$ in the fppf topology?

The case for etale topology is zero by considering the stalks and use the Hilbert 90.

Improve this question
$\endgroup$
1
  • $\begingroup$ is there any general way to compute such staff in flat topology? $\endgroup$
    – Heer
    Commented Jul 16, 2012 at 22:07

2 Answers 2

Reset to default
1
$\begingroup$

It seems to me that (equation) 5.2 and Appendice 11.7 of "Le Groupe de Brauer, III" imply that the cohomology for the \'etale topology equals the cohomology for the fppf topology.

Improve this answer
$\endgroup$
9
  • $\begingroup$ @Jason Starr: thanks a lot for the reference. The condition (L) is satisfied obviously for $j_*\mathbb{G}_m$. the condition (R) is less obvious, in particular what should be the open subfunctor? Is it $\mathbb{G}_m$. $\endgroup$
    – Heer
    Commented Jul 18, 2012 at 23:06
  • $\begingroup$ @Jason Starr: what is the definition of an open subfunctor? $\endgroup$
    – Heer
    Commented Jul 18, 2012 at 23:07
  • $\begingroup$ I am not suggesting to apply Theorem 11.7 to $j_*\mathbb{G}_m$, but rather to $\mathbb{G}_m$, where condition (R) is tautological (the open subfunctor is the entire functor). Grothendieck proves that for any scheme $X$, for every integer $p$, the natural map $H^p(X_{\text{\'et}},\mathbb{G}_m) \to H^p(X_{\text{fppf}},\mathbb{G}_m)$ is an isomorphism. I claim that this implies that for every morphism $f:X\to Y$, also the natural map `$R^p f_*\mathbb{G}_{m,\text{\'et}} \to R^p f_*\mathbb{G}_{m,\text{fppf}}$ is an isomorphism, i.e., the derived etale and fppf pushforwards agree ... $\endgroup$
    – Jason Starr
    Commented Jul 19, 2012 at 13:23
  • $\begingroup$ ... To go from Grothendieck's theorem to the claim, use Prop. 5.1, Exp. V, SGA 4. $\endgroup$
    – Jason Starr
    Commented Jul 19, 2012 at 14:26
  • $\begingroup$ @Jason Starr: I see what you meant. But the natural map should be $R^p f_*\mathbb{G}_{m,\text{et}} \to u_{Y*}R^p f_*\mathbb{G}_{m,\text{fppf}}$, where $u_Y:Y_{\text{fppf}}\rightarrow Y_{\text{et}}$. Then $u_{Y*}R^p f_*\mathbb{G}_{m,\text{fppf}}=0$ provided $R^p f_*\mathbb{G}_{m,\text{et}}$, But this doesn't imply that $R^p f_*\mathbb{G}_{m,\text{fppf}}=0$ $\endgroup$
    – Heer
    Commented Jul 19, 2012 at 20:41
1
$\begingroup$

Here's another way of seeing this. Since $R$ is a dvr, any fppf covering can be refined to a connected cover $p:U\to Spec(R)$ where $U=Spec(A)$ and $A$ is a faithfully flat $R$ algebra of finite type. Also, for our purposes, we might as well replace $R$ with its strict henselization. Now taking generic hyperplane sections of $A$ through a point over $m_R$ refines the covering $p:Spec(A)→Spec(R)$ to a finite, faithfully flat cover $\overline p:Spec(A/(f_1,...,f_n))\to Spec(R)$ which we may assume is connected. But there are no non-trivial line bundles on a connected algebra that is finite over a field.

Improve this answer
$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .