An inexact regularized proximal Newton method for nonconvex and nonsmooth optimization

DOI10.1007/S10589-024-00560-0arXiv2209.09119MaRDI QIDQ6411230FDOQ6411230

Authors: Ruyu Liu, Shaohua Pan, Yuqia Wu, Xiao Qi Yang 

Publication date: 19 September 2022

Abstract: This paper focuses on the minimization of a sum of a twice continuously differentiable function $f$ and a nonsmooth convex function. We propose an inexact regularized proximal Newton method by an approximation of the Hessian $abl{a}^2!f(x)$ involving the $varrho$th power of the KKT residual. For $varrho=0$, we demonstrate the global convergence of the iterate sequence for the KL objective function and its $R$-linear convergence rate for the KL objective function of exponent $1/2$. For $varrhoin(0,1)$, we establish the global convergence of the iterate sequence and its superlinear convergence rate of order $q(1!+!varrho)$ under an assumption that cluster points satisfy a local H"{o}lderian local error bound of order $qin(max(varrho,frac11+varrho),1]$ on the strong stationary point set; and when cluster points satisfy a local error bound of order $q>1+varrho$ on the common stationary point set, we also obtain the global convergence of the iterate sequence, and its superlinear convergence rate of order $frac(q-varrho{)}^{2}q$ if $q>frac2varrho+1+sqrt4varrho+12$. A dual semismooth Newton augmented Lagrangian method is developed for seeking an inexact minimizer of subproblem. Numerical comparisons with two state-of-the-art methods on $el{l}_1$-regularized Student's $t$-regression, group penalized Student's $t$-regression, and nonconvex image restoration confirm the efficiency of the proposed method.

Recommendations

• A globally convergent proximal Newton-type method in nonsmooth convex optimization
• A proximal quasi-Newton trust-region method for nonsmooth regularized optimization
• An inexact regularized proximal Newton-type method for nonconvex composite optimization problems
• An inexact regularized proximal Newton method without line search
• An inexact proximal gradient algorithm with extrapolation for a class of nonconvex nonsmooth optimization problems

Mathematics Subject Classification ID

Methods of successive quadratic programming type (90C55) Nonconvex programming, global optimization (90C26) Newton-type methods (49M15) Mathematical programming (90Cxx)

Cited In (3)

• A VMiPG method for composite optimization with nonsmooth term having no closed-form proximal mapping
• An inexact regularized proximal Newton method without line search
• Generalized Newton Method with Positive Definite Regularization for Nonsmooth Optimization Problems with Nonisolated Solutions

This page was built for publication: An inexact regularized proximal Newton method for nonconvex and nonsmooth optimization

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6411230)