Nonconvex Factorization and Manifold Formulations are Almost Equivalent in Low-rank Matrix Optimization

TL;DR

This paper establishes a geometric landscape connection between manifold and factorization formulations in low-rank matrix optimization, explaining their similar empirical performance and enabling transfer of geometric properties.

Contribution

It provides the first geometric landscape equivalence between manifold and factorization approaches for low-rank matrix optimization, including PSD and general cases.

Findings

Established spectrum-based relations between Riemannian and Euclidean Hessians.

Proved equivalence of stationary points and saddle points across formulations.

Applied geometric insights to phase retrieval and low-rank matrix applications.

Abstract

In this paper, we consider the geometric landscape connection of the widely studied manifold and factorization formulations in low-rank positive semidefinite (PSD) and general matrix optimization. We establish a sandwich relation on the spectrum of Riemannian and Euclidean Hessians at first-order stationary points (FOSPs). As a result of that, we obtain an equivalence on the set of FOSPs, second-order stationary points (SOSPs) and strict saddles between the manifold and the factorization formulations. In addition, we show the sandwich relation can be used to transfer more quantitative geometric properties from one formulation to another. Similarities and differences in the landscape connection under the PSD case and the general case are discussed. To the best of our knowledge, this is the first geometric landscape connection between the manifold and the factorization formulations for handling rank constraints, and it provides a geometric explanation for the similar empirical performance of factorization and manifold approaches in low-rank matrix optimization observed in the literature. In the general low-rank matrix optimization, the landscape connection of two factorization formulations (unregularized and regularized ones) is also provided. By applying these geometric landscape connections, in particular, the sandwich relation, we are able to solve unanswered questions in literature and establish stronger results in the applications on geometric analysis of phase retrieval, well-conditioned low-rank matrix optimization, and the role of regularization in factorization arising from machine learning and signal processing.