On the nuclear norm and the singular value decomposition of tensors (Q300892): Difference between revisions

Let \(V\) be the tensor product of \(d\) finite-dimensional Hilbert spaces. For some applications one needs to find a decomposition of a given tensor \(T\in V\) as the sum of a number \(r\) of pure tensors. The smallest such \(r\) is called the rank of \(T\). For \(d=2\), the rank of \(T\) corresponds to the rank of a matrix. But already for \(d=3\) it is difficult to determine the rank or even give upper or lower bounds. Convex relaxation is a powerful matrix technique which has been successfully applied to matrix completion problems. The author generalizes this to tensors using the nuclear norm. He also generalizes the singular value decomposition (SVD) for matrices to the so-called diagonal SVD for tensors. In addition, he studies various measures of orthogonality in tensor product spaces.