VSPs of cubic fourfolds and the Gorenstein locus of the Hilbert scheme of 14 points on \(\mathbb{A}^6\) (Q1790470)

The author studies the apolar algebra of a complex cubic form \(f\) in 6 variables, which corresponds to a cubic hypersurface \(F\) in a projective space \(\mathbb P^5\). Assuming that \(F\) is not a cone, then the apolar algebra of \(f\) determines a Gorenstein subscheme \(G\) of length \(14\), supported at the origin of an affine space \(\mathbb A^6\). The author studies the location of subschemes \(G\) arising from cubic forms in the scheme \(\mathcal{H}\) parametering Gorenstein subschemes of length \(14\). It is known that \(\mathcal{H}\) consists of two components: the component \(\mathcal{H}_{\mathrm{gen}}\) containing smooth schemes (i.e, sets of distinct points), which has dimension \(84\), and another \(78\)-dimensional component \(\mathcal{H}_{1661}\), corresponding to schemes concentrated at the origin. The author denotes with \(\mathcal{H}_{1661}^{\mathrm{gr}}\) the closed subvariety of \(\mathcal{H}_{1661}\) corresponding to schemes invariant under scalar multiplication. The main result of the paper shows a connection between the geometry of the decompositions of cubic forms \(f\) in 6 variables and the locus \(\mathcal{H}_{1661}^{\mathrm{gr}}\). Indeed, the author shows that \(\mathcal{H}_{1661}^{\mathrm{gr}}\) is canonically isomorphic to the open scheme parametrizing cubic forms in \(6\) variables which do not correspond to cones. In other words, the apolar subschemes of forms \(f\) not corresponding to cones are in \(1:1\) correspondence with points of \(\mathcal{H}_{1661}^{\mathrm{gr}}\). The intersection \(\mathcal{H}_{\mathrm{gen}}\cap \mathcal{H}_{1661}^{\mathrm{gr}}\) is a divisor of \(\mathcal{H}_{1661}^{gr}\) which, in the isomorphism, corresponds to the \textit{Iliev-Ranestad divisor} \(D\) defined in [\textit{A. Iliev} and \textit{K. Ranestad}, Trans. Am. Math. Soc. 353, No. 4, 1455--1468 (2001; Zbl 0966.14027)]: \(D\) is the divisor of cubic forms \(f\) such that the variety \(\mathrm{VSP}(f,10)\), which parameterizes the Waring decompositions of \(f\) with \(10\) summands, corresponds to the Hilbert scheme of two points on a \(K3\) surface. The author points out several consequences, both for the study of the Waring decompositions of cubic forms and for the geometry of the Gorenstein locus \(\mathcal{H}\). For instance, the author proves that the apolar scheme of forms of border rank at most \(9\) is smoothable, i.e. it lies in the intersection \(\mathcal{H}_{\mathrm{gen}}\cap \mathcal{H}_{1661}^{gr}\). This presents naturally the \(9\)-secant variety as a divisor of \(D\). The author also finds an effective description of the equation of \(D\), as the determinant of a \(120\times 120\) matrix. Another consequence of the description of \(\mathcal{H}_{1661}^{gr}\) is the proof that both the components of the Gorenstein locus \(\mathcal{H}\) are rational.