Full operator preconditioning and the accuracy of solving linear systems

Journal Article (2024)
Author(s)

Stephan Mohr (Technische Universität München)

Yuji Nakatsukasa (University of Oxford)

Carolina Urzua-Torres (TU Delft - Numerical Analysis)

Research Group
Numerical Analysis
DOI related publication
https://doi.org/10.1093/imanum/drad104
More Info
expand_more
Publication Year
2024
Language
English
Research Group
Numerical Analysis
Issue number
6
Volume number
44
Pages (from-to)
3259-3279
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Unless special conditions apply, the attempt to solve ill-conditioned systems of linear equations with standard numerical methods leads to uncontrollably high numerical error and often slow convergence of an iterative solver. In many cases, such systems arise from the discretization of operator equations with a large number of discrete variables and the ill-conditioning is tackled by means of preconditioning. A key observation in this paper is the sometimes overlooked fact that while traditional preconditioning effectively accelerates convergence of iterative methods, it generally does not improve the accuracy of the solution. Nonetheless, it is sometimes possible to overcome this barrier: accuracy can be improved significantly if the equation is transformed before discretization, a process we refer to as full operator preconditioning (FOP). We highlight that this principle is already used in various areas, including second kind integral equations and Olver–Townsend’s spectral method. We formulate a sufficient condition under which high accuracy can be obtained by FOP. We illustrate this for a fourth order differential equation which is discretized using finite elements.

Files

2105.07963v1.pdf
(pdf | 1.3 Mb)
License info not available