Geometric description of \(\langle 2\rangle \)-polarised Hilbert squares of generic \(K3\) surfaces (Q6103354)

The main contribution of the paper under review is an explicit description of certain \(K3^{[2]}\)-fourfolds of degree 2 as EPW double sextics. O'Grady proved that the general \(K3^{[2]}\)-fourfold \(X\) of degree \(2\) is an EPW double sextic, but the exact description of the EPW locus in moduli space is unknown. The author's main result is thus strengthening of O'Grady's in the special case that \(X\) is the Hilbert square of a generic \(K3\) surface \(S\) (generic in a sense to be made precise below): whenever \(S^{[2]}\) has a polarization of degree \(2\), then \(S^{[2]}\) is an EPW double sextic. In more detail, let \((S_{2t}, H) \) be a polarized \(K3\) surface of degree \(2t\) over the complex numbers, and suppose that \(S_{2t}\) is \textit{generic}, meaning \(\mathrm{Pic}(S_{2t}) = \mathbb{Z} H\) and \(\mathrm{End}_{\text{Hodge}}(T_{S_{2t}})= \mathbb{Q}\). Suppose further that the Hilbert square \(X=S_{2t} ^{[2]}\) has an ample class \(D\) of degree \(q_X(D) = 2\), where \(q_X\) is the Beauville-Bogomolov-Fujiki form. The first few values of \(t\) for which this happens are \(t=2, 10, 13...\) (the relevant \(t\) are controlled by some Pell equations, as shown by \textit{S. Boissiére} et al. [Prog. Math. 315, 1--15 (2016; Zbl 1375.14015)]). It turns out that the linear system \(|D|\) is basepoint-free. The paper under review provides a geometric description the morphism \(\varphi_{|D|}\) associated to \(|D|\). Previously, it was known from work of Welters and Beauville that for \(t = 2\), the morphism \(\varphi_{|D|}\) is degree 6 cover of the Grassmanian of lines in \(\mathbb{P}^3\). The new result here is that for \(t \ne 2\), the morphism \(\varphi_{|D|} \colon X \to Y \subset \mathbb{P}^5\) is a double cover of a EPW sextic \(Y\) branched over the singular locus, i.e EPW double sextic. The geometry of \(\varphi_{|D|}\) is studied via a careful analysis of a nontrivial involution \(\iota\) that generates \(\Aut(X)\), first studied by Boissiére et al. [loc. cit.]. The strategy is to show that (i) \(\varphi_{|D|}\) factors through the quotient variety \(X/\iota\), and then show that (ii) \(X/\iota\) is an EPW sextic. For (i), one first sees that the vector space \(H^0(X, \mathcal{O}(D)) \) decomposes into a sum \(\iota^*\)-eigenspaces, and each summand can be interpreted as the cohomology of some line bundles on a desingularization of \(X/\iota\). Then, by leveraging a description of the fixed locus of \(\iota\) in terms of the explicit integral Hodge classes on \(S^{[2]}\) from the author's related paper [``Hodge classes of type (2, 2) on Hilbert squares of projective \(K3\) surfaces'', Preprint, \url{arXiv:2112.11306}], it is shown that \(\varphi_|D|\) factors through \(X/\iota\). For (ii), the author the blueprint follows a blueprint out by \textit{K. G. O'Grady} [Commun. Contemp. Math. 10, No. 4, 553--608 (2008; Zbl 1216.14040)]. The key difference here is that while O'Grady only proves deformation equivalence with an EPW double sextic, the author now needs to upgrade deformation-equivalence to isomorphism. To achieve this, the author again uses the nontrivial involution of \(\iota\) together with an \textit{irreducibility criterion} for the intersection of two distinct divisors in \(|D|\). This criterion only holds when \(t\ne 2\), thereby explaining why the \(t =2\) case is special. A related result also appears in the work of \textit{O. Debarre} and \textit{E. Macrì} [Int. Math. Res. Not. 2019, No. 22, 6887--6923 (2019; Zbl 1436.14022)], who prove that for a \(K3\) belonging to some Zariski open subset of the moduli space of degree \(2t\) \(K3\) surfaces, \(t = 10,13,\dots\), the Hilbert square of such a \(K3\) is an EPW double sextic. The proof techniques also have some key differences: Debarre and Macrì make use of a birational map from the moduli space of degree \(2t\) \(K3\) surfaces to a Heegner divisor of Noether-Lefschetz special \(K3^{[2]}\)-fourfolds, while the present work uses explicit analysis of fixed loci of a certain anti-symplectic involution together with the authors description of integral Hodge classes in \(K3^{[2]}\)-fourfolds in [the author, loc. cit., \url{arXiv:2112.11306}].