Moduli spaces of sheaves on \(K3\) surfaces and Galois representations (Q2286220)

Suppose the zeta functions of two \(K3\) surfaces defined over a finite field agree. Then the zeta functions of any two smooth proper moduli spaces of stable sheaves over those surfaces also agree, provided only that both moduli spaces are of the same dimension. This rather striking result is proved here as a corollary of somewhat more general statement: now the two \(K3\) surfaces \(S_1\) and \(S_2\) are defined over an arbitrary field \(k\) and the hypothesis is that the Galois representations \(H^2_{\mathop{\mathrm {\acute{e}}t}}(\overline{S}_j,{{\mathbb Q}_\ell})\) agree. The conclusion is then that for moduli spaces \(M_j\) on \(S_j\) as above, there are isomorphisms \[ H^i_{\mathop{\mathrm {\acute{e}}t}}(\overline{S}_j,{{\mathbb Q}_\ell})\cong H^i_{\mathop{\mathrm {\acute{e}}t}}(\overline{S}_j,{{\mathbb Q}_\ell}) \] as Galois representations, for any \(i\ge 0\). The statement about zeta functions follows by the Lefschetz trace formula. It is striking because there is no apparent geometric map underlying this, as the moduli spaces need not be birationally equivalent. On the other hand, if \(S_1=S_2\) (when the result is already known for fine moduli spaces) then it is at least plausible that the moduli spaces are always derived equivalent, and in that case a conjecture of Orlov states that they should have the same rational motive. So this result is, perhaps, evidence for those conjectures in these cases. If the moduli spaces are fine then the results follow easily from [\textit{F. Charles}, Ann. Math. (2) 184, No. 2, 487--526 (2016; Zbl 1387.14102)], but the general case is somewhat delicate. If \(v\) is a Mukai vector then over \({\mathbb C}\) the Hodge isometry between \(v^\perp\) (or \(v^\perp/v\) if \(v^2=0\)) and \(H^2(M(v),{\mathbb Z})\) is well-known, and the idea of replacing that with a Galois-equivariant isometry in étale cohomology is not new, but here it is essential to do that in the greatest possible generality. Thereafter, the idea is that \(v^\perp\) is also (for \(v^2>0\)) isometric to \(H^2_{\mathop{\mathrm {\acute{e}}t}}(\overline{S},{\mathbb Q}_\ell(1))\oplus {\mathbb Q}_\ell\) and that the cohomology of \(M\) is the same as that of the Hilbert scheme \(\overline{S}^{[n]}\). Analogous statements over \(\mathbb C\) are known but considerable care is needed in transferring them to étale cohomology. Note that the isomorphisms claimed are not, in general, ring isomorphisms: indeed, this seems to be clearly the case only when \(S_1=S_2\).