On the cactus rank of cubic forms (Q1930172)

Fix \(f\in \mathbb {C}[x_0,\dots ,x_n]\), \(f\neq 0\), homogeneous of degree \(d \geq 3\). The rank \(r(f)\) of \(f\) is the minimal integer \(s\) such that \(f = \ell _1^d+\cdots +\ell _s^d\) for some linear forms \(\ell _i\). \(f\) corresponds to a point \(P\) of a projective space \(\mathbb {P}^N\), \(N= \binom{n+d}{n}-1\), containing the order \(d\) Veronese embedding \(X_{n,d}\) of \(\mathbb {P}^n\). The rank of \(f\) is the minimal cardinality \(\mathrm{sr} (P)\) of a set \(S\subset X_{n,d}\) such that \(P\in \langle S\rangle\), where \(\langle \;\;\rangle\) denote the linear span. The cactus rank \(\mathrm{cr} (f)\) of \(f\) is the minimal degree \(\mathrm{cr} (P)\) of a zero-dimensional scheme \(Z\subset X_{n,d}\) such that \(P\in \langle Z\rangle\) (such a minimal \(Z\) is Gorenstein by [\textit{W. Buczynska} and \textit{J. Buczynski}, ``Secant varieties to high degree Veronese reembeddings, catalecticant matrices and smoothable Gorenstein schemes '' , \url{arXiv:1012.3563}; to appear on Journal of Algebraic Geometry]); the cactus rank was introduced (with the name \textit{scheme length}) in [\textit{A. Iarrobino} and \textit{V. Kanev}, Power sums, Gorenstein algebras and determinantal loci. Lect. Notes Math. 1721. Berlin: Springer-Verlag (1999; Zbl 0942.14026)]. The integer \(\mathrm{cr} (f)\) may be defined using the notion of apolarity. See [\textit{K. Ranestad} and \textit{F.-O. Schreyer}, J. Algebra 346, No. 1, 340--342 (2011; Zbl 1277.13016)] for another interesting use of apolarity. The border rank \(\mathrm{br} (f) =\mathrm{br} (P)\) of \(f\) or \(P\) is the minimal integer \(b\) such that \(P\) is contained in the \(b\)-secant variety of \(X_{n,d}\). Now assume that \(f\) is general. Set \(N(d,n):= 2\binom{n+k}{n}\) if \(d=2k+1\) is even and \(N(d,n):= \binom{n+k}{k}+\binom{n+k+1}{k+1}\) if \(d=2k+1\) is odd. A famous theorem by \textit{J. Alexander} and \textit{A. Hirschowitz},[ J. Algeb. Geom. 4, No. 2, 201--222 (1995; Zbl 0829.14002)] says that \(\mathrm{br} (P) = \lceil \binom{n+d}{d}/(n+1)\rceil\), except in \(4\) well-studied cases. The paper under review shows that for general \(f\) we have \(\mathrm{br} (f) \gg \mathrm{cr} (f)\) (a very interesting result). They prove that \(\mathrm{cr} (f)\) is at most the dimension of the linear span of the partial derivatives of all order of \(f\). In particular they prove that \(\mathrm{cr} (f) \leq N(d,n)\). In the case \(d=3\) they prove that \(\mathrm{cr} (f) \leq 2n+2\) for a general \(f\).