A convex optimization approach for solving large scale linear systems
Journal Title: Bulletin of Computational Applied Mathematics (Bull CompAMa) - Year 2017, Vol 5, Issue 1
Abstract
The well-known Conjugate Gradient (CG) method minimizes a strictly convex quadratic function for solving large-scale linear system of equations when the coefficient matrix is symmetric and positive definite. In this work we present and analyze a non-quadratic convex function for solving any large-scale linear system of equations regardless of the characteristics of the coefficient matrix. For finding the global minimizers, of this new convex function, any low-cost iterative optimization technique could be applied. In particular, we propose to use the low-cost globally convergent Spectral Projected Gradient (SPG) method, which allow us to extend this optimization approach for solving consistent square and rectangular linear system, as well as linear feasibility problem, with and without convex constraints and with and without preconditioning strategies. Our numerical results indicate that the new scheme outperforms state-of-the-art iterative techniques for solving linear systems when the symmetric part of the coefficient matrix is indefinite, and also for solving linear feasibility problems.
Authors and Affiliations
Debora Cores, Johanna Figueroa
Keywords
Related Articles
- EP ID EP243956
- DOI -
- Views 81
- Downloads 0
Fitting of parameters for a temperature control model by means of continuous derivative-free optimization: a case study in a broiler house
Intensive broiler production requires of accurate control systems aimed to maintain ideal conditions inside the facilities. The achievement of an appropriate environment guarantees good performance and sustainability of...
On solution of a nonlocal problem with dynamic boundary conditions for a loaded linear parabolic equation by straight-line methods
We consider a nonlocal problem with dynamic boundary conditions for a loaded linear parabolic equation. For this problem we prove the unique solvability in Sobolev's spaces and the maximum principle under some natural c...
Stability and square integrability of solutions of nonlinear fourth order differential equations
The aim of the present paper is to establish a new result, which guarantees the asymptotic stability of zero solution and square integrability of solutions and their derivatives to nonlinear differential equations of fou...
L(2,1)-Labeling for Subdivisions of Cycle Dominated Graphs
Let G(V,E) be a simple, finite, connected, undirected graph. Distance two labeling or L(2,1)-labeling of a graph G is an assignment f from the vertex set V(G) to the set of non-negative integers such that |f(x)-f(y)| &ge...
On the Modified Methods for Irreducible Linear Systems with L-Matrices
Milaszewicz, [Milaszewic J.P, Linear Algebra. Appl. 93,1987, 161$-$170] presented new preconditioner for linear system in order to improve the convergence rates of Jacobi and Gauss-Seidel iterative methods. Li et al.,[Li...