Sets computing the symmetric tensor rank (Q1949838)

Let \(\nu _d: \mathbb {P}^r \to \mathbb {P}^N\), \(N:= \binom{r+d}{r}-1\), denote the degree \(d\) Veronese embedding of \(\mathbb {P}^r\). For any \(P\in \mathbb {P}^N\), the symmetric tensor rank \(\mathrm{sr}(P)\) is the minimal cardinality of a set \(S\subset \nu _d(\mathbb {P}^r)\) spanning \(P\). Let \(\mathcal {S}(P)\) be the set of all \(A\subset \mathbb {P}^r\) such that \(\nu _d(A)\) computes \(\mathrm{sr}(P)\). In this paper the authors classify all \(P\in \mathbb {P}^n\) such that \(\mathrm{sr}(P) < 3d/2\) and \(\mathrm{sr}(P)\) is computed by at least two subsets of \(\nu _d(\mathbb {P}^r)\). For such tensors \(P\in \mathbb {P}^N\), they prove that \(\mathcal {S}(P)\) has no isolated points. Moreover they can prove that under the same hypothesis if \(\mathrm{sr}(P)\) is computed by at least two different sets, then it is computed by an infinite family without isolated points. This is an important result in terms of identifiability of (symmetric) tensors. Nowadays this is among the most studied problems expecially in the applications.