Let $k$ be an algebraically closed field of characteristic $p>2$, let $A/k$ be an abelian variety. Let $\iota\colon A\to A, a\mapsto -a$ be the natural involution. Let $x\in\mathrm{Br}(A)[p]$ be a Brauer class on $A$ of order $p$. Is it necessarily true that $\iota^*x=x$?
-
$\begingroup$ I think yes, and it should follow from the Künneth formula (the involution acting as $-1$ on $H^1$, and so as $(-1)^2 = 1$ on $H^2$). I realize I am not giving enough details to be 100% persuasive, but I am very far from being an expert on this. I think the paper by Skorobogatov and Zarhin on the Brauer group of Kummer surfaces must contain enough details to construct a proof though. $\endgroup$– R.P.Commented Jun 24, 2020 at 15:48
-
$\begingroup$ wwwf.imperial.ac.uk/~anskor/crelle.2011.121%5B1%5D.pdf (section 2) $\endgroup$– R.P.Commented Jun 24, 2020 at 15:53
1 Answer
$\newcommand{\bG}{\mathbb{G}}$Let $X$ be any smooth scheme over an algebraically closed field $k$ of characteristic $p$. From the short exact sequence $0\to\mu_p\to \bG_m\to\bG_m\to 0$ of sheaves on the flat site of $X$ we get $0\to (\mathrm{Pic}\,X)/p\to H^2_{fl}(X,\mu_p)\to H^2_{fl}(X,\bG_m)[p]\to 0$.
Let now $X$ be an abelian variety. Ogus showed in Proposition 1.2 of "Supersingular K3 crystals" that under certain assumptions on a smooth proper variety (that are satisfied for abelian varieties) there is a canonical inclusion $$H^2_{fl}(X,\mu_p)\hookrightarrow H^2_{\mathrm{dR}}(X/k) $$
The map is induced by the identification $H^2_{fl}(X,\mu_p)\simeq H^1_{et}(X,\bG_m/(\bG_m)^p)$ and the logarithmic derivative $\bG_m/(\bG_m)^p\xrightarrow{dlog}Z^1\to \Omega^{\bullet}[1]$. Since there is a canonical isomorphism $H^2_{\mathrm{dR}}(X/k)\simeq\Lambda^2 H^1_{\mathrm{dR}}(X/k)$ the multiplication by an integer $[n]:X\to X$ induces multiplication by $n^2$ on $H^2_{\mathrm{dR}}(X/k)$. In particular, the involution $\iota$ acts by identity and hence it does so on $H^2_{fl}(X,\mu_p)$ and $H^2_{fl}(X,\bG_m)[p]=(\mathrm{Br}\,X)[p]$ where the last identification comes from Gabber's theorem.
-
$\begingroup$ Thanks! Is there a similar $\mathrm{dlog}$ sequence for $\mathbb{G}_m/\mathbb{G}_m^{p^n}$, if we care about the involution action on the full $p$-primary part of Brauer group? $\endgroup$– user39380Commented Jun 25, 2020 at 0:15
-
$\begingroup$ Or we can use the sequence $0\to\mu_p\to\mu_{p^2}\to\mu_p\to 0$ and the induction, just wonder if it is possible to do it directly? $\endgroup$– user39380Commented Jun 25, 2020 at 0:52
-
2$\begingroup$ @Qixiao There is a injective dlog map from this sheaf to the de Rham-Witt forms $W_n\Omega^1$ and the analogous argument can perhaps be made to work using e.g. Corollary 3.29 in Illusie's paper on de Rham-Witt complex, but I haven't thought this through carefully. The inductive argument indeed shows this fact for free for $p\neq 2$. $\endgroup$– SashaPCommented Jun 25, 2020 at 5:38