A concise proof of Kruskal's theorem on tensor decomposition (Q2267410)

Let \(M_{1},M_{2}\) and \(M_{3}\) be matrices of dimensions \(s_{1}\times r\), \(s_{2}\times r\) and \(s_{3}\times r\), respectively, over an arbitrary field. The matrix triple product \([M_{1},M_{2},M_{3}]\) is defined to be the sum of the \(r\) rank-1 tensors \(m_{i}^{1}\otimes m_{i}^{2}\otimes m_{i}^{3}\) (\(i=1,\dots,r\)) where \(m_{i}^{j}\) is the \(i\)th column of \(M_{j}\). Define the \(K\)-rank of \(M_{j}\) to be the largest number \(k\) such that each set consisting of \(k\) columns of \(M_{j}\) is linear independent. \textit{J. B. Kruskal}'s theorem [ibid. 18, 95--138 (1977; Zbl 0364.15021); Theorem 4a] states the following. Assume that the sum of the \(K\)-ranks of the \(M_{j}\) is at least \(2r+2\) and that \(N_{1},N_{2}\) and \(N_{3}\) are matrices each with \(r\) columns. If \([M_{1},M_{2},M_{3}]=[N_{1},N_{2},N_{3}]\) then there exists an \(r\times r\) permutation matrix \(P\) and \(r\times r\) diagonal matrices \(D_{1},D_{2},D_{3}\) such that \(D_{1}D_{2}D_{3}=I\) and \(N_{i}=M_{i}D_{i}P\) for each \(i\). Kruskal's original proof of the theorem was complicated and various simplifications have been given [see, for example, \textit{A. Stegeman} and \textit{N. D. Sidiropoulos}, ibid. 420, No.~2--3, 540--552 (2007; Zbl 1120.15002)]. The present author uses a different approach to provide a short proof.