Typical tensorial rank (Q1064383)

Given finite dimensional vector spaces U, V, and W over some field k and \(t\in U\otimes V\otimes W\), the problem of finding optimal computations for bilinear forms is one of decomposition of t into triads, \(t=\sum^{r}_{\rho =1}u_{\rho}\otimes v_{\rho}\otimes w_{\rho}\), \((u_{\rho}\in U\), \(v_{\rho}\in V\), \(w_{\rho}\in W)\) with minimal possible r. The least r for which such a representation exists is called the rank of t, rk t. Let \(R(n,m,l)=\max \{rk t:\) \(t\in U\otimes V\otimes W\}\) where \(n=\dim U\), \(m=\dim V\) and \(l=\dim W\). Then the typical rank is defined by Ṟ(n,m,l)\(=\min \{r: X_ r=U\otimes V\otimes W\}\) where \(X_ r=\{t: \underline{rk} t\leq r\}\) and \underbar{rk} is the border rank. For the typical rank it is shown that Ṟ(n,m,l)\(=nml/(n+m+l- 2)+O(l)\) (n\(\leq m\leq l)\) and for cubic shape tensors Ṟ(n,n,n)\(=n^ 3/(3n-2)\), \(n\neq 3\) and dim \(X_ r=\min \{r(3n-2),n^ 3\}\), \(n\neq 3\). Upper bounds on the typical rank are also derived applied to cubic shape tensor.