An apocalypse-free first-order low-rank optimization algorithm with at most one rank reduction attempt per iteration

TL;DR

This paper introduces a first-order optimization algorithm for minimizing functions over the determinantal variety that avoids the 'apocalypse' problem, guarantees convergence to stationary points, and limits rank reduction attempts to one per iteration.

Contribution

The proposed algorithm is the first to prevent 'apocalypse' failures while requiring at most one rank reduction attempt per iteration, improving efficiency over existing methods.

Findings

Proves convergence to stationary points avoiding 'apocalypse' scenarios.

Limits rank reduction attempts to at most one per iteration.

Demonstrates improved efficiency over previous algorithms.

Abstract

We consider the problem of minimizing a differentiable function with locally Lipschitz continuous gradient over the real determinantal variety, and present a first-order algorithm designed to find stationary points of that problem. This algorithm applies steps of a retraction-free descent method proposed by Schneider and Uschmajew (2015), while taking the numerical rank into account to attempt rank reductions. We prove that this algorithm produces a sequence of iterates the accumulation points of which are stationary, and therefore does not follow the so-called apocalypses described by Levin, Kileel, and Boumal (2022). Moreover, the rank reduction mechanism of this algorithm requires at most one rank reduction attempt per iteration, in contrast with the one of the $\mathrm{P}^2\mathrm{GDR}$ algorithm introduced by Olikier, Gallivan, and Absil (2022) which can require a number of rank reduction attempts equal to the rank of the iterate in the worst-case scenario.