Stochastic Rayleigh-Ritz Ascent: Global Linear Convergence under a Spectral Gap

DOI10.48550/ARXIV.2610.42071MaRDI QIDQ7414851arXivFDO

Authors Jonas Bresch, Oleh Melnyk

Publication date 30 October 2026

Published in arXiv

Tldr We propose the stochastic Rayleigh-Ritz ascent (SRA) algorithm for finding the top eigenvector of a matrix and prove that it achieves global linear convergence rates under a spectral gap condition. The SRA algorithm is a random-plane iteration method without adjoint computation, improving upon existing methods like power iterations and gradient descent in terms of scalability and efficiency.

Full work available at URL https://doi.org/10.48550/ARXIV.2610.42071

arXiv classification

math.NAmath.OCmath.PR

This page was built for publication: Stochastic Rayleigh-Ritz Ascent: Global Linear Convergence under a Spectral Gap

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7414851)