A framework for deflated and augmented Krylov subspace methods

TL;DR

This paper introduces a unified framework for deflated and augmented Krylov subspace methods, clarifying their differences from preconditioning and demonstrating how augmentation can be implemented implicitly or explicitly to improve convergence.

Contribution

The paper develops a general framework for Krylov methods that includes deflation and augmentation, analyzing their properties and providing strategies to avoid breakdowns, with practical numerical experiments.

Findings

Augmentation can be implemented explicitly or implicitly within the framework.

Conditions for avoiding breakdowns in deflated MINRES are identified.

Numerical experiments demonstrate the effectiveness of different deflated MINRES variants.

Abstract

We consider deflation and augmentation techniques for accelerating the convergence of Krylov subspace methods for the solution of nonsingular linear algebraic systems. Despite some formal similarity, the two techniques are conceptually different from preconditioning. Deflation (in the sense the term is used here) "removes" certain parts from the operator making it singular, while augmentation adds a subspace to the Krylov subspace (often the one that is generated by the singular operator); in contrast, preconditioning changes the spectrum of the operator without making it singular. Deflation and augmentation have been used in a variety of methods and settings. Typically, deflation is combined with augmentation to compensate for the singularity of the operator, but both techniques can be applied separately. We introduce a framework of Krylov subspace methods that satisfy a Galerkin condition. It includes the families of orthogonal residual (OR) and minimal residual (MR) methods. We show that in this framework augmentation can be achieved either explicitly or, equivalently, implicitly by projecting the residuals appropriately and correcting the approximate solutions in a final step. We study conditions for a breakdown of the deflated methods, and we show several possibilities to avoid such breakdowns for the deflated MINRES method. Numerical experiments illustrate properties of different variants of deflated MINRES analyzed in this paper.