Back to Search Start Over

A Bregman forward-backward linesearch algorithm for nonconvex composite optimization: superlinear convergence to nonisolated local minima

Authors :
Ahookhosh, Masoud
Themelis, Andreas
Patrinos, Panagiotis
Source :
SIAM J Optim 31(1):653-685 (2021)
Publication Year :
2019

Abstract

We introduce Bella, a locally superlinearly convergent Bregman forward backward splitting method for minimizing the sum of two nonconvex functions, one of which satisfying a relative smoothness condition and the other one possibly nonsmooth. A key tool of our methodology is the Bregman forward-backward envelope (BFBE), an exact and continuous penalty function with favorable first- and second-order properties, and enjoying a nonlinear error bound when the objective function satisfies a Lojasiewicz-type property. The proposed algorithm is of linesearch type over the BFBE along candidate update directions, and converges subsequentially to stationary points, globally under a KL condition, and owing to the given nonlinear error bound can attain superlinear convergence rates even when the limit point is a nonisolated minimum, provided the directions are suitably selected.

Details

Database :
arXiv
Journal :
SIAM J Optim 31(1):653-685 (2021)
Publication Type :
Report
Accession number :
edsarx.1905.11904
Document Type :
Working Paper
Full Text :
https://doi.org/10.1137/19M1264783