Special holomorphic tensors on orthogonal modular varieties and applications to the Lang conjecture (Q6046119)

There is a very well established way to interpret modular forms of weight \(nm\) as \(m\)-canonical forms on an open part of an orthogonal modular variety \(\mathcal{F}(\Gamma)\) of dimension \(n\). Here \(\mathcal{F}(\Gamma)\) is the quotient of the symmetric space \(\mathcal{D}\) associated with \({\operatorname{{O}}}(2,n)\) by an arithmetic subgroup \(\Gamma\) of the orthogonal group of a lattice \(L\) of signature \((2,n)\). Also the conditions under which such a form will extend to a smooth projective model of \(\mathcal{F}(\Gamma)\) have been much studied, usually in the context of moduli of \(K3\) surfaces as in [\textit{V. A. Gritsenko} et al., Invent. Math. 169, No. 3, 519--567 (2007; Zbl 1128.14027)]. It is usually convenient to consider separately the obstructions coming from cusps, those coming from branch divisors (i.e.\ reflections in \(\Gamma\)), and those coming from elliptic fixed points in codimension \(>2\). In this paper, the author gives a way to similarly interpret modular forms of weight \(pm\) as sections of \((\Omega^p)^{\otimes m}\). Again there are conditions under which the resulting holomorphic tensors extend. For the cusp and reflective obstruction, the picture is broadly similar to the pluricanonical case, but the elliptic fixed points are more difficult to handle, especially when they occur at the cusps. The connection between modular forms and differential forms arises from the automorphic bundles \(\mathcal{L}=\mathcal{O}_{{\mathbb P}L_{\mathbb C}}(-1)|_{\mathcal{D}}\) and \(\mathcal{E}=\mathcal{L}^\perp/\mathcal{L}\), where \(\mathcal{L}^\perp\) is the orthogonal complement in \(L_{\mathbb C}\otimes\mathcal{O}_D\). Using the Euler sequence on \({\mathbb P}L_{\mathbb C}\) one finds \(\Omega^1_{\mathcal{D}}\simeq\mathcal{L}\otimes\mathcal{E}\) and hence \(\Omega^p_{\mathcal{D}}\simeq\mathcal{L}^{\otimes p}\otimes\bigwedge^p\mathcal{E}\). Sections of \((\Omega^p_{\mathcal{D}})^{\otimes m}\) thus correspond to vector-valued modular forms with values in \(\mathcal{L}^{\otimes pm}\otimes(\bigwedge^p\mathcal{E})^{\otimes m}\). The trick then is to construct some sections from scalar-valued modular forms, i.e.\ sections of some power of \(\mathcal{L}\). This is possible because \((\bigwedge^p\mathcal{E})^{\otimes m}\) can be decomposed according to the decomposition of \((\bigwedge^p{\mathbb C}^n)^{\otimes m}\) under \({\operatorname{{O}}\nolimits}(L_{\mathbb C})\): the symmetric square of this has a trivial summand and this gives, for \(m\) even, an embedding of \(\mathcal{L}^{\otimes pm}\) in \(({\operatorname{Sym}}^2\Omega_{\mathcal{D}}^p)^{\otimes m/2}\subset (\Omega_{\mathcal{D}}^p)^{\otimes m}\). The extension properties at the cusps and branch divisors are examined next, and then the extension to singularities, using a variant of the Reid-Tai criterion for canonical singularities: this makes precise the idea, present in several places in the literature, that if the modular form being considered vanishes ``enough'' along exceptional divisors, then the tensor also extends. With sufficient control over the actual singularities, this can be improved a little further. The author then uses these results to prove general type results for certain subvarieties of orthogonal modular varieties, namely those along which the canonical quadratic form on the tangent space to \(\mathcal{D}\) does not drop rank too much. Comparing this with the slope (weight divided by order of vanishing) of the modular forms along the cusps, branch divisors and and exceptional divisors, one obtains general type results for subvarieties not contained in certain loci of \(\mathcal{F}(\Gamma)\). In many cases, this gives a partial version of Lang's conjecture, in the form of a proper closed subset that is guaranteed to contain all non-general type subvarieties of sufficiently (explicitly) large dimension.