The Noether-Lefschetz conjecture and generalizations (Q2356921)

A quasi-polarized \(K3\) surface of genus \(g\geq 2\) is a pair \((S,L)\) where \(S\) is a \(K3\) surface and \(L\) is a line bundle on \(S\) with primitive Chern class \(c_1(L)\in H^2(S,{\mathbb{Z}})\) satisfying \(L\cdot L=\int_S c_1(L)^2=2g-2\) and \(L\cdot C= \int_C c_1(L)\geq 0\) for every curve \(C\subset S\). Let \({\mathcal{K}}_g\) be the moduli space of complex quasi-polarized \(K3\) surfaces of genus \(g\). The Noether-Lefchetz locus in \({\mathcal{K}}_g\) parametrizes \(K3\) surfaces in \({\mathcal{K}}_g\) with Picard number \(\geq 2\); it is a countable union of divisors. Each of them parametrizes the \(K3\) surfaces whose Picard lattice contains a special curve class; these divisors are called \textit{Nother-Lefschetz (NL) divisors} on \({\mathcal{K}}_g\). The question concerned in this paper is the following: Is the Picard group \(\mathrm{Pic}_{\mathbb{Q}}({\mathcal{K}}_g)\) of \({\mathcal{K}}_g\) with rational coefficients spanned by NL divisors? More generally, this question can be extened to higher NL-loci on \({\mathcal{K}}_g\), which parametrize \(K3\) surfaces in \({\mathcal{K}}_g\) with higher Picard number. Call the irreducible compoents of higher NL-loci the \textit{NL cycles} on \({\mathcal{K}}_g\). This paper proves the Noether-Lefschetz conjecture on the moduli space of quasi-polarized \(K3\) sufaces. Theorem 1(=Theorem 1.1). For all \(g\geq 2\) and all \(r\leq 4\), the cohomology group \(H^{2r}( {\mathcal{K}}_g,{\mathbb{Q}})\) is spanned by NL-cycles of codimension \(r\). In particular (taking \(r=1\)), \(\mathrm{Pic}_{\mathbb{Q}}({\mathcal{K}}_g)\cong H^2({\mathcal{K}}_g,\mathbb{Q})\) and the Noether-Lefschetz conjecture holds on \({\mathcal{K}}_g\) for all \(g\geq 2\). Combined with the works of \textit{R. E. Borcherds} [Invent. Math. 132, No. 3, 491--562 (1998; Zbl 0919.11036)] and \textit{J. H. Bruinier} [Compos. Math. 133, No. 1, 49--63 (2002; Zbl 1036.11018)], an explicit formula for the rank of \(\mathrm{Pic}({\mathcal{K}}_g)\) is obtained as a corollary of Theorem 1. Theorem 1 is deduced from the main theorem on arithmetic manifolds. Let \(K\) be any compact open subgroup of \(G({\mathbb{A}}_f)\). The main theorem (Theorem 2.7). Let \(Y_K\) be a connected arithmetic manifold associated to \(\mathrm{SO}(V)\) and let \(r<\min\{(p+q-1)/3, p/2\}\). Then the subspace \(\bar{H}^{r\times q}(Y_K,{\mathbb{C}})\) is spanned by the Poincaré dual of special cycles, i.e., the natural projection \(SC^{rq}(Y_K)\to \bar{H}^{r\times q}(Y_K,{\mathbb{C}})\) is surjective. Let \(Y\) be a connected smooth Shimura variety of orthogonal type. \(Y\) has many special cycles. One can ask if the classes of special cycles of codimension \(r\) exhaust all the cohomology classes in \(H^{2r}(Y,{\mathbb{Q}})\cap H^{r,r}(Y)\) for sufficiently small \(r\). Theorem 2(=Theorem 1.5). Assume that \(Y\) is a connected Shimura variety associated to \(\mathrm{SO}(p,2)\). If \(r<(p+1)/3\), any cohomology class in \(H^{2r}(Y,{\mathbb{Q}})\cap H^{r,r}(Y)\) is a linear combination (with rational coefficients) of classes of special cycles. Theorem 3 (=Theorem 1.7). Let \(Y\) be a smooth non-compact finite volume congruence arithmetic hyperbolic manifold of dimension \(p\). Then for all \(r<p/3\), the \({\mathbb{Q}}\)-vector space \(H^r(Y,{\mathbb{Q}})=\bar{H}^r(Y,{\mathbb{Q}})\) is spanned by classes of totally geodesic submanifolds of codimension \(r\).