On singularities of third secant varieties of Veronese embeddings (Q1638012): Difference between revisions

Let \(\sigma_k(V_{d,n})\) be the \(k\)-th secant variety of the Veronese embedding \(V_{d,n}\) of degree \(d\) of \(\mathbb P^n\). This \(k\)-th secant variety corresponds to the set of symmetric tensors of type \((n+1)\times \cdots\times (n+1)\) (\(d\) times), having symmetric border rank \(k\). In order to understand several properties of symmetric tensors (e.g. the identifiability of a tensor rank decomposition), it is relevant to have a description of the singular locus of \(\sigma_k(V_{d,n})\). It is well known that \(\sigma_{k-1}(V_{d,n})\) is contained in the singular locus of \(\sigma_k(V_{d,n})\), and the author addresses the question about the equality of the two loci. By a result of \textit{V. Kanev} [in: Zero-dimensional schemes and applications. Proceedings of the workshop, Naples, Italy, February 9--12, 2000. Kingston: Queen's University. 263 (2002; Zbl 0988.14511)], equality holds for the second secant variety, thus the author studies the case \(k=3\). The main result is that \(\sigma_2(V_{d,n})\) equals the singular locus of \(\sigma_3(V_{d,n})\), unless \(d=4\). When \(d=4\), indeed, Sing(\(\sigma_3(V_{d,n})\)) also contains the locus of forms essentially defined by two variables, i.e. forms that belong to \(\mathbb C[L,M]\), for some choice of two linear forms \(L,M\).