On product identities and the Chow rings of holomorphic symplectic varieties (Q832410)

The main purpose of the paper is to understand the Chow ring of irreducible holomorphic symplectic varieties. The authors conjecture a series of identities in the Chow rings \(\mathrm{CH}_{\star}(M \times X^{l})\), \(l \geq 1\), for a moduli space \(M\) of stable sheaves on a \(K3\) surface \(X\), inspired by the classical Beauville-Voisin identity for a \(K3\) surface. Assuming the conjectures are true all tautological classes fit in the lowest piece of a natural filtration emerging on \(\mathrm{CH}_{\star}(M)\). In the two dimensional setting [\textit{A. Beauville} and \textit{C. Voisin}, J. Algebr. Geom. 13, No. 3, 417--426 (2004; Zbl 1069.14006)] an essential role in approaching the cycle structure is played by a distinguished zero-cycle \(c_X\). The cycle \(c_X\) has degree one and is the Chow class of any point lying on a rational curve in \(X\). The intersection of any two divisors is a multiple of \(c_X\), while the second Chern class of the tangent bundle satisfies \(c_2(T X) = 24c_X\). In higher dimensions, they consider a moduli space \(M\) of stable sheaves on a \(K3\) surface with respect to a Mukai vector \(v\), which is a smooth projective irreducible holomorphic symplectic variety ([\textit{D. Huybrechts}, Lectures on \(K3\) surfaces. Cambridge: Cambridge University Press (2016; Zbl 1360.14099)]) of dimension \(v^{2}+2\). Also in this case there is a distinguished zero-cycle \(c_M \in \mathrm{CH}_0(M)\) of degree one: this is the class of any stable sheaf \(\mathcal{F}\) such that \(c_2(\mathcal{F})=k c_X\) in \(\mathrm{CH}_0(X)\), where \(k\) is the degree of the second Chern class specified by the Mukai vector \(v\). As in the two dimensional case, one expects ([\textit{J. Shen} et al., Compos. Math. 156, No. 1, 179--197 (2020; Zbl 1436.14071),\textit{C. Voisin}, Prog. Math. 315, 365--399 (2016; Zbl 1352.32010)]) that the special cycle corresponds to the largest rational equivalence orbit of points on \(M\), but the intersection properties of \(c_M\) are not understood as well as those of the cycle in the two-dimensional context. Fix the notation for the projections: \(\pi:M \times X \to M\) and \(\rho: M \times X \to X\). The authors consider a universal sheaf \(\mathcal{F} \to M \times X\) and they study its geometry and the geometry of the special cycles \(c_X\) and \(c_M\) in two steps: the first step describes the tautological subring as \(R_{\star}(X)=\mathrm{CH}_{2}(X) + \mathrm{CH}_{1}(X) + \mathbb{Z} c_{X}\subset \mathrm{CH}_{\star}(X)\), in the second step they focus on the rank zero virtual sheaf \(\overline{\mathcal{F}}=\mathcal{F}-\rho^{*}(\mathcal{F})\) with \(\mathcal{F}\in M\) such that \([\mathcal{F}]=c_{M}\in \mathrm{CH}_{0}(M)\). These two steps bring to Conjecture 1 that states that for every \(i_1,\ldots,i_l \geq 0\), it holds that \[ \alpha \cdot \ \mathrm{ch}_{i_1}(\overline{\mathcal{F}_1})\cdots \mathrm{ch}_{i_l}(\overline{\mathcal{F}_l})=0 \in \mathrm{CH}_{\star}(M\times X^l) \] where \( \alpha \in R_{\star}(M)\) is a tautological class of codimension \(d\), and \(\mathcal{F}_i\) denote the pullback to \(M\times X^l\) of the virtual universal sheaf on \(M\) and the i-th factor of \(X\). Conjecture 1 brings to a generalization of identities that hold for \(K3\) surfaces, in a more general setup of moduli spaces. Moreover the identities of Conjecture 1 bring to a large collection of conjectural Chow vanishings in the self-products \(M \times M \times \cdots \times M\). They set \(\overline{\Delta}=\Delta - M\times c_M \in \mathrm{CH}_m(M \times M)\). In Theorem 1 they show that the system of identities of Conjecture 1 is equivalent to the vanishing \(\alpha \cdot \ \overline{\Delta}_{0,1} \cdots \overline{\Delta}_{0,l}=0 \in \mathrm{CH}_{\star}(M \times M^{l})\), for any tautological class \(\alpha \in R_{\star}(M)\) of codimension \(d\) and integer \(l\) satisfying \(d + l > \dim M\). The theorem has many interesting consequences: the tautological ring has rank 1 in dimension 0, i.e. \(R_{\star}(M)=\mathbb{Q}\cdot c_M\), the modified diagonal cycle \(\Gamma^{m+1}(M,c_M)=\Delta - \Delta_{c} + \Delta_{c,c}- \cdots + \Delta_{c,c,\ldots,c}\) vanishes in \(\mathrm{CH}_{m}(M^{m+1})\). Finally for every codimension \(0 \leq d \leq m\) they consider the increasing filtration \[ S_{0} \subset S_1 \ldots \subset S_i \subset \ldots \subset \mathrm{CH}^{d}(M), \] where \(S_i(\mathrm{CH}^{d}(M))=\{ \alpha \ with \ \alpha \cdot \overline{\Delta}_{0,1} \ldots \overline{\Delta}_{0,m-d+i+1}=0 \subset \mathrm{CH}_{\star}(M\times M^{m-d+i+1})\}\). The filtration terminates [\textit{A. Beauville} and \textit{C. Voisin}, J. Algebr. Geom. 13, No. 3, 417--426 (2004; Zbl 1069.14006)] and the authors can prove that \(S_d(\mathrm{CH}^{d}(M))=\mathrm{CH}^{d}(M)\). Finally they remark that a stronger conjecture can be formulate, i.e Conjecture 2, that says that the restriction of the cycle class map to the tautological subring is injective. Anyway Conjecture 1 is easier to study since concerns in explicit relations in the Chow ring. As evidence for Conjecture 1 they prove that it holds true for Hilbert schemes of \(n\) points on a smooth projective \(K3\) surface.