Remarks on the Lefschetz standard conjecture and hyperkähler varieties (Q1955666)

Let \(X\) be a smooth projective variety over \({\mathbb C}\) of dimension \(n\) and \(\xi \in H^2 (X, {\mathbb Q})\) be the cohomology class of a hyperplane section of \(X\). Then for any \(0 \leq k \leq n\) the cup product \(\cup \xi ^{n-k} : H^k (X, {\mathbb Q}) \rightarrow H^{2n-k} (X, {\mathbb Q})\) is an isomorphism by the hard Lefschetz theorem. The Lefschetz standard conjecture in degree \(k\) questions the existence of an algebraic cycle \(Z\) of codimension \(k\) in \(X \times X\), such that \([Z]_* : H^{2n-k} (X, {\mathbb Q}) \rightarrow H^k (X, {\mathbb Q})\) is the inverse of \(\cup \xi ^{n-k}\). This conjecture was formulated by Grothendieck in 1968. A cohomology class \(\alpha \in H^k (X, {\mathbb Z})\) is of level \(c\) if it is a push-forward of a cohomology class on a subvariety \(Y \subset X\) of codimension \(c\). The set \(N^c H^k (X, {\mathbb Z})\) of the cohomology classes \(\alpha \in H^k (X, {\mathbb Z})\) of level at least \(c\) is contained in \(H^k (X, {\mathbb Z}) \cap [H^{k-c,c}(X) + \dots + H^{c,k-c}(X)]\). Grothendieck's generalized Hodge conjecture from 1969 asks whether \(N^c H^k (X, {\mathbb Z})\) is the largest sub-Hodge structure of \(H^k (X, {\mathbb Z})\), contained in \(H^{k-c,c}(X) + \dots + H^{c, k-c}(X)\). Let \(S\) be a smooth quasi-projective scheme with tangent sheaf \(\mathcal{T}_S\) and \(Z \in CH^k (X \times S)\) be a family of codimension \(k\) cycles in \(X\) parameterized by \(S\). The cohomology class \([Z] \in H^k (X \times S, \Omega ^k _{X \times S} ) \simeq H^0 (S, \Omega ^k_S) \otimes H^k (X, \mathcal{O}_X)\) induces a morphism \(\phi _Z : \bigwedge ^k \mathcal{T}_S \rightarrow H^k (X, \mathcal{O}_X) \otimes \mathcal{O}_S\) of sheaves on \(S\). Making use of the map \(\phi _{Z,s}\) at a point \(s \in S\), the article under review reduces the Lefschetz conjecture for \(X\) to the existence of local deformations of algebraic cycles in \(X\). More precisely, let us assume that the generalized Hodge conjecture holds for \(H^k (X, {\mathbb Z})\) and the Lefschetz conjecture in degree \(k-2\) is true for all smooth projective varieties. Then \(X\) is shown to be subject to the Lefschetz conjecture in degree \(k\) if and only if there exist a smooth quasi-projective scheme \(S\), a codimension \(k\) cycle \(Z \in CH^k (X \times S)\) and a point \(s \in S\), such that the morphism \(\phi _{Z,s} : \bigwedge ^k \mathcal{T}_S \rightarrow H^k (X, \mathcal{O}_X)\) is surjective. In particular, \(X\) satisfies the Lefschetz conjecture in degree \(2\) exactly when \(\phi _{Z,s} : \bigwedge ^2 \mathcal{T}_S \rightarrow H^2 (X, \mathcal{O}_X)\) is surjective for some smooth quasi-projective variety \(S\), some \(Z \in CH^2 (X \times S)\) and some \(s \in S\). The article provides an explicit formula for the morphism \(\phi _Z\) if the class \(Z \in CH^k (X \times S)\) is given by the codimension \(k\) part \(ch_k(\mathcal{E})\) of the Chern character of a vector bundle \(\mathcal{E}\) over \(X \times S\). More precisely, if \(p: X \times S \rightarrow S\) is the projection on the second factor then the deformation of vector bundles determined by \(\mathcal{E}\) is described by the Kodaira-Spencer morphism of sheaves \(\rho : \mathcal{T}_S \rightarrow \mathcal{E}xt^1_p (\mathcal{E}, \mathcal{E})\). For \(\forall k \in {\mathbb N}\) let \(\phi _k (\mathcal{E}) : \bigwedge ^k \mathcal{T}_S \rightarrow H^k (X, \mathcal{O}_X) \otimes \mathcal{O}_S\) be the composition of \(\bigwedge ^k \rho : \bigwedge ^k \mathcal{T}_S \rightarrow \bigwedge ^k \mathcal{E}xp ^1 _p (\mathcal{E}, \mathcal{E})\), the Yoneda product \(\bigwedge ^k \mathcal{E}xt ^1 _p (\mathcal{E}, \mathcal{E}) \rightarrow \mathcal{E}xt ^k _p ( \mathcal{E}, \mathcal{E})\) and the trace map \(\mathcal{E}xt ^k _p ( \mathcal{E}, \mathcal{E}) \rightarrow H^k (X, \mathcal{O}_X) \otimes \mathcal{O}_S\). If there exist a smooth quasi-projective scheme \(S\), a vector bundle \(\mathcal{E} \rightarrow X \times S\) and a point \(s \in S\), such that \(\phi _k (\mathcal{E}) _s : \bigwedge ^k \mathcal{T}_{S,s} \rightarrow H^k (X, \mathcal{O}_X)\) is surjective then the Lefschetz conjecture in degree \(2\) holds for \(X\). Moreover, the Lefschetz conjecture in degree \(k-2\) for all smooth projective varieties and the generalized Hodge conjecture for \(H^k (X, {\mathbb Z})\) suffice for the Lefschetz conjecture in degree \(k\) for \(X\). The results on the Lefschetz conjecture in degree \(2\) are specified for the hyperkähler varieties \(M\). A simply connected Kähler manifold \(M\) is an irreducible hyperkähler variety if it admits a closed, everywhere non-degenerate \(2\)-form \(\omega\). The presence of such \(\omega\) is equivalent to the existence of three complex structures \(I, J, K\) on \(M\), subject to the quaternionic relations \(I^2 = J^2 = K^2 = - \text{Id}\), \(IJ = - JI = K\). A complex hermitian vector bundle \(\mathcal{E} \rightarrow M\) with a hermitian connection \(\theta\) is hyperholomorphic if \(\theta\) is compatible with \(I, J\) and \(K\). For instance, the tangent bundle of an irreducible hyperkähler variety \(M\) is a hyperholomorphic. The article under review establishes that if a stable hyperholomorphic bundle \(\mathcal{E}\) over a projective irreducible hyperkähler variety \(M\) admits a non-trivial positive-dimensional family of deformations then the Lefschetz conjecture in degree \(2\) holds for \(M\). Let us suppose that for every \(K3\) surface \(S\) there exists a stable hyperholomorphic sheaf \(\mathcal{E}\) with a non-trivial positive-dimensional family of deformations on the Hilbert scheme \(S^{[n]}\) parameterizing the subschemes of \(S\) of length \(n\). Then the Lefschetz conjecture in degree \(2\) is true for any projective deformation of \(S^{[n]}\).