A splitting preconditioned primal-dual algorithm with interpolation and extrapolation for bilinear saddle point problem

Both the popular primal-dual algorithms of Chambolle-Pock and Chang-Yang are full-splitting and efficient for solving bilinear saddle point problem. However, their step sizes depend on the spectral norm of the linear transform or are estimated by using linesearch, which are usually overconservative...

Full description

Saved in:
Bibliographic Details
Published in:Numerical algorithms Vol. 100; no. 3; pp. 937 - 962
Main Authors: Chang, Xiaokai, Xu, Long, Cao, Jianxiong
Format: Journal Article
Language:English
Published: New York Springer US 01.11.2025
Springer Nature B.V
Subjects:
Algebra
Algorithms
Computation
Computer Science
Convergence
Convex analysis
Decomposition
Euclidean space
Extrapolation
Interpolation
Lagrange multiplier
Linear transformations
Numeric Computing
Numerical Analysis
Operators (mathematics)
Preconditioning
Saddle points
Splitting
Theory of Computation
Primal-dual algorithm
Bilinear saddle point problem
Splitting preconditioning
Sequence combination
ISSN:1017-1398, 1572-9265
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Both the popular primal-dual algorithms of Chambolle-Pock and Chang-Yang are full-splitting and efficient for solving bilinear saddle point problem. However, their step sizes depend on the spectral norm of the linear transform or are estimated by using linesearch, which are usually overconservative from the spectral norm or require extra computing from linesearch. By using sequence interpolation and extrapolation, a novel primal-dual algorithm with larger stepsizes is proposed. Moreover for this scheme, we introduce a splitting preconditioned strategy by solving matrix inversion problem for the dual variable, where the preconditioning is separable from the evaluations of proximal operators and then avoids the difficulty of computing complicated preconditioned proximal operator in the existing methods. The preconditioned primal-dual algorithm not only does not require the spectral norm of the linear transform or linesearch to estimate stepsizes, but also holds cheap computation on the matrix inversion problem for the cases with primal or dual variable being modest, as the matrix decomposition is involved only once. Global iterative convergence and O ( 1 / N ) ergodic convergence rate results measured by the function value residual and constraint violation are established, where N denotes the iteration counter. Finally, we demonstrate the performance of the proposed primal-dual algorithm and preconditioning strategy via preliminary numerical experiments.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1017-1398
1572-9265
DOI:10.1007/s11075-024-01974-x