5
$\begingroup$

Given two three order tensors $T$ and $S$ in $F^{m\times n\times p}$ and $F^{a\times b\times c}$. Clearly $\operatorname{rk}(T\otimes S)\le \operatorname{rk}(T)\operatorname{rk}(S)$. Does the equality holds for generic $T$ in $F^{m\times n\times p}$ and generic $S$ in $F^{a\times b\times c}$ (tensors are viewed as projective varieties via Segre embedding).

$\endgroup$

1 Answer 1

9
$\begingroup$

We don't know.

There are formats in which the equality is false for generic tensors: take $F^{n \times n \times 1}$ and $F^{n \times n \times n^2}$. Generic tensors in both formats are isomorphic to matrix multiplication tensors $\left<n,1,1\right>$ and $\left<1,n,n\right>$ with ranks $n$ and $n^2$ respectively, but the rank of their tensor product $\left<n,n,n\right>$ is less than $n^3$ for $n \geq 2$. If you don't like dimension $1$, the same happens for $F^{n \times n^2 \times n}$ and $F^{n \times n \times n^2}$.

For formats in which the subgeneric rank tensors form a hypersurface, the techniques of the paper "Border Rank Is Not Multiplicative under the Tensor Product" by Christandl, Gesmundo and Jensen, I believe, can be used to prove that for a generic tensor $R(T^{\otimes 2}) < R(T)^2$, but this does not extend to products of different tensors.

On the other hand, for generic $2 \times 2 \times 2$ tensors multiplicativity is true.

My conjecture would be that rank multiplicativity does not hold in most cases.

$\endgroup$
1
  • 2
    $\begingroup$ Thanks a lot for your answer and counterexample $\endgroup$
    – Nick Chen
    Dec 16, 2023 at 11:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.