On secant dimensions and identifiability of flag varieties (Q2069822)

Problems about dimension of secant varieties of many kind of algebraic varieties have been quite a subject of study in the last twenty years or so, mainly related to problems of tensor decompositions, by studying varieties which parameterize tensors of given kind (``structured tensors''), main examples are Segre Varieties (parameterizing all tensors of a given format), Veronese Varieties (parameterizing symmetric tensors, i.e. polynomials), Segre-Veronese Varieties (and partially symmetric tensors), Grassmannians (skew symmetric tensors). Moreover the problem of \(h\)-identifiability of a tensor (i.e. unicity of its decomposition as sum of \(h\) decomposable tensors of the given kind) as been quite studied, also for its relation with applications. In this paper the author study both problems (dimension of \(h\)-Secant varieties and identifiability) for flag varieties \(\mathbb{ F}(k_1,\dots, k_r; n)\) which are set of flags, that is the set of nested subspaces, \( V_{1} \subset\dots\subset V_{r}\) in a vector space \( V\), where \(V \cong \mathbb{C}^{n+1}\), and \(\dim V_i=k_i\), for \(i=1,\dots,r\) and \(k_1 \leq\cdots \leq k_r\). If \(\mathbb{G}(k_i, n)\) is the Grassmannian of \(k_i\)-dimensional linear subspaces of \(\mathbb{P}(V )\), viewed in its Plucker embedding into \(\mathbb{P}^{N_i}\), where \(N_i =\binom{n+1}{k_i+1}-1\), we have an embedding of the product of these Grassmannians: \[\mathbb{G}(k_1, n)\times\cdots\times\mathbb{G}(k_r, n) \subset \mathbb{P}^{N_1} \times\cdots\times{P}^{N_r} \subset \mathbb{P}^N,\] where \(N = \binom{n+1}{k_1+1}\dots\binom{n+1}{k_r+1}- 1\). So the flag variety \(\mathbb{ F}(k_1,\dots, k_r; n)\), via Plucker and Segre embeddings, can be viewed as a subvariety of \(\Pi_{i=1}^r\mathbb{G}(k_i, n) \subset \mathbb{P}^N\). The \(h\)-secant varieties of \(\mathbb{ F}(k_1, \dots, k_r; n)\), i.e. the Zariski closure of the union of all the linear spaces generated by \(h\) points of \(\mathbb{ F}(k_1,\dots, k_r; n)\), have expected dimension \(\min\{nh + h-1, N\}\), and if their actual dimension is less that that, they are said to be \(h\)-defective. The main result in the paper enstablishes a bound on \(h\)-defectivity; namely it is proved that if \(n \leq 2k_j + 1\) for some index \(j\) and \(\ell\) is the maximum among these \(j\). Then, for \[h \leq\binom{n + 1}{k_l + 1}^{\log 2(\sum^\ell_{j=1} k_j+\ell-1)}\] \(\mathbb{ F}(k_1, \dots , k_r; n)\) is not \((h +1)\)-defective. Moreover, under the same bound, the generic point of \(\mathbb{ F}(k_1,\dots, k_r; n)\) is identifiable. In the last section of the paper the case of the flag varieties \(\mathbb{ F}(0,k; n)\) is considered and it is proved that their chordal variety (2-secant variety) is non-defective, except for \(k=n-1\). One main tool of the proofs is the study of obsulating spaces to products of Grassmannians and projections from them.