On the Hessian of cubic hypersurfaces (Q6624331)

In this paper, the Hessian locus associated to a general cubic hypersurface is analysed by describing its singular locus and its desingularization for every dimension. The strategy is based on strong connections between the Hessian and the quadrics defined as partial derivatives of the cubic polynomial.\N\NLet \(S=\mathbb K[x_0, \dots, x_n]\) with \(\mathbb K\) algebraically closed field of characteristic 0; write \(S^d\) for the homogeneous polynomials of degree \(d\). For \(f\in S^d\) let \(X=V(f)\subset \mathbb P^n\) be the associated hypersurface, \(H_f= (\frac{\partial^2 f}{\partial x_i \partial x_j})\) the \(n+1\times n+1\) Hessian matrix of \(f \), and \(h_f=\det(H_f)\). By the classical Gordan-Noether (GN) theorem, if \(n\leq 3\) then \(h_f=0\) iff \(V(f)\) is a cone. In the case \(h_f\neq 0\), \(\mathcal H_f:= V(h_f)\) is said to be the Hessian hypersurface associated to \(X\). The authors are, as many of their predecessors, particularly interested in the cubic case \((d=3)\) but extend various known facts, in particular work of Appendix 4 of Adler in [\textit{A. Adler} and \textit{S. Ramanan}, Moduli of abelian varieties. Berlin: Springer (1997; Zbl 0863.14024)] to dimensions \(n\geq 4\).\N\NFour main results are indicated in the Introduction as Theorems A, B, C, D and are in this review formulated in the summaries to sections 2 (A, B), 3(B), and 5(D), where they are proved.\N\NIn Section 1 after efforts to explain the role of the graded algebra of \(D= K[y_0, \dots, y_n] \) with \(y_i=\frac{\partial}{\partial x_i}\) differential operators, and pointing out the inclusions \(\mathcal C_{cone}=\{[f]: V(f) \text{ is cone}\} \subseteq \mathcal C_{GN}=\{[f]: h_f \text{ is singular}\} \subseteq \mathcal C_{ sing}=\{[f]: V(f) \text{ is singular}\}\subset \mathbb P(S^d)\), for \(d\geq 2\), basic results on the Hessian are presented. Thus e.g. elementary but important is that for all \(v,w\in \mathbb K^{n+1},\) there holds \(H_f(v)\cdot w = H_f(w)\cdot v\).\N\NAuthors recall now that for \(f\in S^d,\) the Jacobian ideal is the graded ideal \(J_f=\oplus_{k\geq d-1} J_f^k \) spanned by the partial derivatives of \(f\), while the apolar ring \(A_f=D/\text{Ann}_D(f)= \oplus_{i=0}^{\deg(g)} A^i\) is a graded Artinian algebra and deal from here onwards only with the cubic case, i.e. \(d=3\). They considers the loci \(\mathcal D_k(f)=\{[x]\in \mathbb P, \text{rank}(H_f(x))\leq k \}\), give them the alternative description via the multiplication operator \(y\cdot\), \(\mathcal D_k(f)=\{[y]\in \mathbb P(A^1): \textsf{Rank}(y\cdot : A^1\rightarrow A^2)\leq k \},\) and find similar alternative descriptions for \(X, \textsf{Sing}(X)\) and \(\mathcal H_f\) in terms of classes in \(\mathbb P(A^1)\). The identification \(\mathcal D_k(f)= P(J_f^2)\cap \mathcal Q_k\) is justified and the expected dimension is determined. Formulas for dimension, codimension in \(\mathbb P(S^2) \) and degree of \(\mathcal Q_k\) are recalled.\N\NSection 2 is dedicated to singular loci and desingularizations. The set parametrizing smooth cubics in \(\mathbb P^n\) is \(U= \mathbb P(S^3)\setminus \mathcal C_{\mathrm{sing}}\). A central concept defined for \(f\in \mathbb P(S^3) \) is the variety \(\Gamma_f=\{([x],[y])\in \mathbb P^n \times \mathbb P^n: H_f(x)\cdot y=0 \}=\{([x],[y])\in \mathbb P(A^1)\times \mathbb P(A^1):xy=0 \}\). On \(\Gamma_f\) acts a natural involution \(\tau([x],[y])= ([y],[x])\) which is fixed-point free iff \([f]\not \in \mathcal C_{sing}\).\N\NVariety \(\Gamma_f\) was used in Adler' s work [loc. cit.] to desingularize the Hessian locus by using the Klein cubic \(f_0=x_0x_4^2+x_1x_0^2+x_2x_1^2+ x_3x_2^2+x_4x_3^2 \) and proving \(\Gamma_f\) to be smooth. This seems not possible for larger \(n\), so in Theorem B below another method is used, but still the methods can be used to prove at this point\N\NTheorem A (Singularities of the Hessian). For any \([f]\in \mathbb P(S^3)\) for which \(X=V(f)\) is smooth, the Hessian variety is reduced and \(\textsf{Sing}(\mathcal H_f)\) coincides with \(\mathcal D_{n-1}(f)\).\N\NAn example shows that the smoothness hypothesis is necessary. Next in a small note the technique of Adler is generalized to any value of \(n\) characterizing cubics \([f]\in U\) for which \(\Gamma_f \) is singular and in Subsection 2.2 we find the proof of\N\NTheorem B (Correspondence and desingularisation). For the general smooth cubic hypersurface \(V(f)\), \(\Gamma_f\) is smooth and the natural projection \(\pi_1:\Gamma_f \rightarrow \mathcal H_f\) is a desingularization.\N\NSection 3 determines in Theorem 3.1 with the notation \(\mathcal{\tilde Q}_k= \{q\in S^2\setminus \{0\}: \textsf{Rank}(q)\leq k\} \) e.g. the dimension and other properties of \(\mathcal J_k=\{(q,v,f)\in \mathcal {\tilde Q}_k\times D^1 \times S^3: v,f \neq 0 \text{ and } v(f)=q \} \), as an auxiliary for the proof of\N\NTheorem C (Singularities of the general cubic). For the general smooth cubic hypersurface \(V(f)\), if \(\mathcal D_k(f)\setminus \mathcal D_{k-1}(f)\) is nonempty, then \(\textsf{Sing}(\mathcal D_k(f)) = \mathcal D_{k-1}(f))\).\N\NSection 4 studies the loci \(\mathcal D_k(f)\) from the perspective of being degeneracy loci of a specific vector bundle map. First for a smooth variety \(X\) of dimension \(n\) and a vector- respectively a line bundle \(E\) and \(L\) on \(X\) and a symmetric morphism \(\varphi: E\rightarrow E^*\otimes L, \) the degeneracy loci at order \(k\) are defined as \(\mathcal D'_k(\varphi)= \{x\in X: \textsf{Rank}(\varphi_x)\leq k\} \) and some general results for these are cited. For a symmetric matrix \(M\) of order \(n+1\) with coefficients in \(S^1\) the symmetric homomorphism of vector bundles \(\varphi_M: \mathcal O_{\mathbb P^n}^{n+1} \stackrel{M\cdot}{\rightarrow} \mathcal O_{\mathbb P^n}^{n+1}(1)\) is considered. In case \(M=H_f\) one gets the equality \(D_k(f)=\mathcal D'_k(\varphi_{H_f})\) so that the general results give information on nonemptyness, connectedness, and smoothness of \(\textsf{Sing}(\mathcal H_f) \) for various dimensions. For general symmetric matrix \(M\) and smooth \(Y= \mathcal D' _k(\varphi_M)\), the canonical divisor \(K_Y\) is determined. In Section 4.1 a general construction that allows to describe the existence of 2:1 covers for suitable degeneracy loci of symmetric maps between vector bundles is presented. Section 4.2 applies the theory to the degeneracy loci of \(\mathcal D'_4(\varphi_M)\subseteq \mathbb P^5\). The case \(n=5\) is interesting because it is the last one where the singular locus of \(\mathcal H\) is smooth generically. Section 4 is probably the conceptually most demanding part of an itself quite demanding paper: Chern numbers, relative Grassmannians, the Koszul sequences, and Cohomology groups, Bott' s vanishing theorem, and the Kuenneth theorem all make their appearance. Some of the notation will only be found explained in papers or books known to experts; others will have to guess it.\N\NIn Section 5, the singular variety \(Y=\mathcal D_4(f)= \textsf{Sing}(\mathcal H_f) \), in case \(V(f)\subseteq \mathbb P^5\) is a general smooth cubic four-fold, is described. Results are summarized in\N\NTheorem D (Hessian of cubic four-folds: geometric invariants of the singular locus). Let Y as stated. Then: \(Y\) is a smooth surface; it is irreducible and minimal surface of general type of degree 35 and canonical divisor \(K_Y =3 H|_Y + \eta\) with \(H\) the hyperplane class in \(\mathbb P^5\) and \(\eta\) being a nontrivial 2-torsion element in \(\textsf{Pic}^0(Y)\). \(Y\) has topological Euler characteristic \(e(Y)=357 \), irregularity \(q=0 \) and \(\chi(Y)=56\).\N\NAmong the books that will be helpful for the reader who wants to look deeper into the matter the reviewer has the impression, these will be [\textit{I. V. Dolgachev}, Classical algebraic geometry. A modern view. Cambridge: Cambridge University Press (2012; Zbl 1252.14001)], [\textit{F. Russo}, On the geometry of some special projective varieties. Cham: Springer (2016; Zbl 1337.14001)], [\textit{R. Lazarsfeld}, Positivity in algebraic geometry. II. Positivity for vector bundles, and multiplier ideals. Berlin: Springer (2004; Zbl 1093.14500)]; many of the cited articles also will be helpful.