Bounds on the tensor rank

From MaRDI portal
Publication:1756466



Abstract: We give a sufficient criterion for a lower bound of the cactus rank of a tensor. Then we refine that criterion in order to be able to give an explicit sufficient condition for a non-redundant decomposition of a tensor to be minimal and unique.


Given a tensor $T\in \mathbb C^{n_1+1}\times \dots\times \mathbb C^{n_k+1}$, we can view (modulo scalar multiplication) $T$ as a point in $Y = \nu(\mathbb P^{n_1}\times\dots \times \mathbb P^{n_k})\subset \mathbb P^{M}$, with $M=-1+\Pi_{i=1}^k(n_i+1)$, where $v$ is the Segre embedding. The tensor rank $r$ of $T$ can be expressed as the minimum number $r$ for which there exixt a set $A$ of $r$ simple points in $Y$ such that $T\in < A >$ (in fact $Y$ parameterizes decomposable tensors). In this paper the rank of $T$ is studied via the behavior of projections of $0$-dimensional schemes $S\subset Y$ under projections on ``flattenings of $Y$, i.e. chosing a partition $\{a_1,\dots,a_h\}\cup \{b_1,\dots,b_{k-h}\}=\{1,\dots,k\}$ and projecting $S$ into the Segre embeddings of $ \mathbb P^{a_1}\times\dots\times \mathbb P^{a_h}$ and $\mathbb P^{b_1}\times \dots\times \mathbb P^{b_{h-k}}$ (see Thm. 3.1). Using this lind of analisys, a critherion is given for determining if some $S\subset Y$ with $T\in < S >$ gives a minimal decomposition of $T$ (i.e. if $S$ computes $r(T)$ or not); moreover, also a critherion to determine if $S$ is the unique minimal decomposition (identifiability ) is given.




Cited in
(30)








This page was built for publication: Bounds on the tensor rank

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1756466)