A convexity-like structure for polar decomposition with an application to distributed computing

Abstract

We make a full landscape analysis of the (generally non-convex) orthogonal Procrustes problem. This problem is equivalent with computing the polar factor of a square matrix. We reveal a convexity-like structure, which explains the already established tractability of the problem and show that gradient descent in the orthogonal group computes the polar factor of a square matrix with linear convergence rate if the matrix is invertible and with an algebraic one if the matrix is singular. These results are similar to the ones of Alimisis and Vandereycken (2024) for the symmetric eigenvalue problem. We present an instance of a distributed Procrustes problem, which is hard to deal by standard techniques from numerical linear algebra. Our theory though can provide a solution.