GIT semistability of Hilbert points of Milnor algebras (Q514355)

Let \(V\) be a an \(n\)-dimensional vector space over \(\mathbb C\), let \(S=\text{Sym}V\). By GIT it is defined a projective moduli space of isomorphism classes of degree \(d+1\) hypersurfaces in \(\mathbb P V^\vee\). A homogeneous form \(F\) of degree \(d+1\) on \(V^\vee\) is \textit{semistable} with respect to the standard \(\text{SL}(V)\)-action if (and only if) the closure of the orbit does not contain \(0\), and the GIT quotient \(\mathbb P(\text{Sym}^{d+1}V)^{ss}\slash\text{SL}(V)\) parametrizes orbits of semistable hypersurfaces that are closed in the semistable locus. Only for certain small values of \(n\) or \(d\), the Hilbert-Mumford numerical criterion gives a method for geometric description of the resulting quotient. Because the \(\text{SL}(V)\)-action on \(\mathbb P(\text{Sym}^{d+1}V)\) gives a unique linearisation, there exists only one notion of \(\text{SL}(V)\)-stability. However, the author considers the Milnor algebra associated to a homogeneous form and the Hilbert points of this algebra. Then different variants of stability is obtained. This article explore two variants of stability appearing in work of \textit{J. Alper} and \textit{A. Isaev} [Math. Ann. 360, No. 3--4, 799--823 (2014; Zbl 1308.14048); ``Associated forms and hypersurface singularities: the binary case. to appear in J. Reine. Angew. Math., \url{arXiv:1407.6838}] on an invariant-theoretic approach to the reconstruction problem arising from the Mather-Yau theorem. For a point \([W]\in\text{Grass}(n,\text{Sym}^dV)\) its ideal is \(I_W=(W)\subset S\), and its coordinate ring is \(S_W=S/I_W\). The \(m\)th Hilbert point of \(S_W\) is defined to be the short exact sequences \(0\rightarrow (I_S)_m\rightarrow\text{Sym}^mV\rightarrow (S_W)_m\rightarrow 0\), and this is regarded as a point in \(\text{Grass}(\dim(I_W)_m,\text{Sym}^mV).\) The author gives conditions on \(S_W\), and assume these holds, such that \(S_W\) is a graded local Artinian Gorenstein algebra with socle in degree \(n(d-1).\) There exists an \(\text{SL}(V)\)-invariant divisor \(\text{Res}\subset\text{Grass}(n,\text{Sym}^dV)\) parametrizing subspaces that don't generate a complete intersection ideal. The affine complement of \(\text{Res}\) is denoted \(\text{Grass}(n,\text{Sym}^dV)_{\text{Res}}\), \(\iota(m)=\dim(I_W)_m\), where \([W]\in\text{Grass}(n,\text{Sym}^dV)_{\text{Res}}.\) There is a rational map \(H_m:\text{Grass}(n,\text{Sym}^dV)\rightarrow\text{Grass}(\iota(m),\text{Sym}^mV)\) sending \([W]\) to the \(m\)th Hilbert point of \(S_W\). This map is a morphism on \(\text{Sym}^dV)_{\text{Res}}\), and it is equivariant with respect to the natural actions of \(\text{SL}(V)\) on both sides. Following [loc. cit.], \(H_{n(d-1)}\) is denoted \(\mathbf A\), for \([W]\in\text{Sym}^dV)_{\text{Res}}\), \(\mathbf{A}(W)=[\text{Sym}^{n(d-1)}V\rightarrow (S_W)_{n(d-1)}\rightarrow 0]\in\mathbb P((\text{Sym}^{n(d-1)}V)^\vee).\) Here \(\dim(S_W)_{n(d-1)}=0\) is used to identify \(\mathbf{A}(W)\) with the corresponding point in the space of lines in \((\text{Sym}^{n(d-1)}V)^\vee.\) From the natural isomorphism \(\mathbb P((\text{Sym}^{n(d-1)}V)^\vee)\simeq\mathbb P(\text{Sym}^{n(d-1)}V^\vee)\), \(\mathbf{A}(W)\) is identified with an element in \(\mathbb P(\text{Sym}^{n(d-1)}V^\vee)\), giving an element in \(\text{Sym}^{n(d-1)}V^\vee\) defined up to non-zero scalar. This element is called \textit{the associated form of} \(g_1,\dots,g_n\) by Alper and Isaev [loc. cit.]. The associated form of \(\mathbf{A}(W)\) is called the \textit{homogeneous Macaulay inverse system} of \(S_W\) with respect to the presentation \(S_W=S/I_W\). This set-up gives the possibility to compare \(\text{SL}(V)\)-actions on the two sides, and as \(\text{Res}\) is an \(\text{SL}(V)\)-invariant divisor, every point of \(\text{Grass}(n,\text{Sym}^dV)_{\text{Res}}\) is \(\text{SL}(V)\) stable. Given \(F\in\text{Sym}^{d+1}V\), the gradient point of \(F\) is defined as the subspace \(F\subseteq\text{Sym}^dV\) spanned by the first partial derivatives of \(F\). Then \(\dim\nabla F=n\) if and only if \(F\notin\text{Sym}W\), \(W\subset V\). If \(\dim\nabla F=n\) , \(\nabla(F)\) denotes the corresponding point of \(\text{Grass}(n,\text{Sym}^dV)\). Letting \(J_F:=I_{\nabla F}=(\partial F|\partial\in V^\vee),\) the \textit{Milnor algebra} of \(F\) is defined as \(M_F=S/J_F.\) If \(F\) is smooth, then \(M_F\) is a graded local, Artinian, Gorenstein, \(\mathbb C\)-algebra with socle in degree \(\nu:=n(d-1)\), and the interesting Hilbert points of \(M_F\) occur for \(d\leq m\leq\nu\). These prerequisites give the possibility to define the \textit{gradient morphism}, then the GIT quotient of this morphism can be studied, and the following theorems stated verbatim can be proved: Theorem 1.1. Let \(F(x_1,\dots,x_n)\) be a homogeneous polynomial of degree \(d+1\) in \(n\) variables. Then \(F\) is semistable with respect to the \(\text{SL}(V)\)-action on \(\mathbb C[x_1,\dots,x_n]_{d+1}\) if and only if \[ \nabla(F)=\text{span}\langle\partial F/\partial x_1,\dots,\partial F/\partial x_n\rangle\in\text{Grass}(n,\mathbb C[x_1,\dots,x_n]_d) \] is semistable with respect to the \(\text{SL}(n)\)-action on the Grassmannian. Furthermore, if \(F\) is stable, then \(\nabla(F)\) is polystable, and stable if and only if \(F\) is not a non-trivial sum of two polynomials in disjoint set of variables. Theorem 1.2. Suppose that \(g_1,\dots,g_n\) is a regular sequence of homogeneous degree \(d\) polynomials in \(\mathbb C[x_1,\dots,x_n]\). Then the associated form \[ \mathbf{A}(g_1,\dots,g_n)\in\mathbb P(\mathbb C[\partial/\partial x_1,\dots,\partial/\partial x_n]_{n(d-1)}) \] is semistable with respect to the \(\text{SL}(n)\)-action. In particular, for every smooth form \(F\), the associated form of the regular sequence \(\partial F/\partial x_1,\dots,\partial F/\partial x_n\) is semistable. The article is a clever study comparing stability and GIT quotients in corresponding spaces. It gives answers to interesting questions using elementary theory.