Title:On the convergence of higher-order orthogonality iteration

Abstract:The higher-order orthogonality iteration (HOOI) has been popularly used for finding a best low-multilinear-rank approximation of a tensor. However, its iterate sequence convergence is still an open question. In this paper, we propose to implement HOOI in a greedy way. We update each factor matrix by selecting from the best candidates one that is closest to the current iterate. With this greedy selection, we establish the subsequence convergence of HOOI without any additional assumptions, i.e., any limit point of the iterates generated by HOOI is a critical point. Assuming the existence of a nondegenerate limit point, we further show that the entire iterate sequence converges to a critical point through the so-called Kurdyka-Łojasiewicz (KL) property. In addition, we show that if the starting point is sufficiently close to any nondegenerate globally optimal solution, HOOI produces a sequence convergent to a globally optimal solution.