A refined derived Torelli theorem for Enriques surfaces (Q2664159)

It has become quite common to study derived categories of coherent sheaves on varieties, with the hope that the derived category encodes interesting geometric information about the varieties. A fundamental question in this area is: Under which circumstances are the derived categories of two varieties equivalent? For some classes of varieties, this can only happen if both varieties are already isomorphic. A statement of this type is usually called a Derived Torelli Theorem. For Enriques surfaces over fields of characteristic different from 2, a derived Torelli theorem holds: If \(X_1\) and \(X_2\) are two Enriques surfaces, we have \[ D^b(X_1)\cong D^b(X_2)\quad\iff\quad X_1\cong X_2\,; \] as proven by \textit{T. Bridgeland} and \textit{A. Maciocia} [Math. Z. 236, No. 4, 677--697 (2001; Zbl 1081.14023)] in characteristic zero and by \textit{K. Honigs} et al. [Math. Res. Lett. 28, No. 1, 65--91 (2021; Zbl 1471.14040)] in positive characteristic. Line bundles \(L\) on Enriques surfaces are always exceptional objects, which means that \(\operatorname*{Ext}^*(L,L)\) is concentrated in degree zero and one-dimensional. Exceptional objects are of interest because they induce decompositions of the derived category with small pieces. For \(X\) a general Enriques surface, there is a collection \(\mathcal L=\{L_1,\dots, L_{10}\}\) of ten pairwise orthogonal line bundles; see [\textit{S.\ Zube}, Math.\ Notes 61, No.\ 6, 693--699 (1997; Zbl 0933.14023)]. Denoting the right-orthogonal complement by \[ \operatorname{Ku}(X,\mathcal L):=\bigl\{E\in D^b(X)\mid \operatorname{Hom}_{D^b(X)}(L_i,E[m])=0\mid \text{ for all \(i=1,\dots, 10\), \(m\in \mathbb Z\)}\bigr\}\,, \] we get a semi-orthogonal decomposition \[ D^b(X)=\bigl\langle \operatorname{Ku}(X,\mathcal L), L_1,\dots , L_{10}\bigr\rangle\,.\tag{\(\star\)} \] The authors of the article prove a refined Torelli theorem stating that, for two general Enriques surfaces \(X_1\) and \(X_2\) together with orthogonal collections of line bundles \(\mathcal L_1\) and \(\mathcal L_2\) of length 10, we have \[ \operatorname{Ku}(X_1,\mathcal L_1)\cong \operatorname{Ku}(X_2,\mathcal L_2)\quad \implies \quad X_1\cong X_2\,. \] Let us give a brief sketch of the proof: In Section 2, the authors prove a useful general criterion for extending equivalences from admissible subcategories to bigger subcategories. Omitting some technical details, the key Proposition 2.5 says the following: Let \(D^b(X)=\langle \mathcal A, \mathcal A'\rangle\) and \(D^b(Y)=\langle \mathcal B, \mathcal B'\rangle\) be two semi-orthogonal decompositions of derived categories of smooth projective varieties, and let \(\Phi\colon \mathcal A\to \mathcal B\) be an equivalence. Then, given exceptional objects \(E\in \mathcal A'\) and \(F\in \mathcal B'\) satisfying \[ \Phi(\alpha^! E)\cong \beta^! F\,, \] where \(\alpha\colon \mathcal A\hookrightarrow D^b(X)\) and \(\beta\colon \mathcal B\hookrightarrow D^b(Y)\) are the embeddings, one can extend \(\Phi\) to an equivalence \(\widehat \Phi\colon \langle \mathcal A,E\rangle \to \langle \mathcal B,F\rangle\). The plan is to apply this Proposition to the decomposition \((\star)\). The main technical work which allows to do this is done in Section 4 (working in a slightly more general setup of Enriques categories). There, it is shown that, given a semi-orthogonal decomposition of the form \((\star)\), and denoting the embedding by \(\alpha\colon \operatorname{Ku}(X,\mathcal L)\hookrightarrow D^b(X)\), all the objects \(\alpha^!L_i\in \operatorname{Ku}(X,\mathcal L)\), for \(i=1,\dots, 10\), are \(3\)-spherical. Conversely, all \(3\)-spherical objects in \(\operatorname{Ku}(X,\mathcal L)\) are of this form, up to degree shifts. This allows the authors to proceed as follows with the proof of their main theorem: Let \(\Phi\colon \operatorname{Ku}(X_1,\mathcal L_1)\to \operatorname{Ku}(X_2,\mathcal L_2)\) be an equivalence, and write \(\mathcal L_j=\{L^j_1,\dots,L^j_{10}\}\). As equivalences map \(3\)-spherical objects to \(3\)-spherical objects, we have \(\Phi(\alpha_1^! L^1_1)\cong \alpha_2^!L^2_i[m]\) for some \(i\in \{1,\dots ,10\}\) and some \(m\in \mathbb Z\). Hence, Proposition 2.5 applies and gives an extended equivalence \[ \widehat \Phi\colon\bigl\langle \operatorname{Ku}(X_1,\mathcal L_1), L^1_1\bigr\rangle\to \bigl\langle\operatorname{Ku}(X_2,\mathcal L_2), L^2_i\bigr\rangle\,. \] Repeating this step ten times (with a small adjustment) gives the desired extended equivalence \(D^b(X_1)\cong D^b(X_2)\). Then, the (non-refined) derived Torelli theorem for Enriques surfaces can be applied to conclude \(X_1\cong X_2\). The authors also use the general results of Sections 2 and 4 to give a simplified proof of a result of \textit{S. Hosono} and \textit{H. Takagi} [Kyoto J. Math. 60, No. 1, 107--177 (2020; Zbl 1475.14033)] relating the derived categories of Enriques surfaces and blow-ups of Artin--Mumford double solids.