Stable iterative refinement for solving linear systems with inaccurate computation (Q6964657)

This interesting paper deals with stable iterative refinement for solving linear systems with inaccurate computation. Consider an ill-conditioned linear system \(Ax=b\) with possibly nonsymmetric \(A\). The Iterative Refinement (IR) algorithm is an iterative method where the \(m\)th iteration, corrects the approximation \(x_m\) of \(x\) to the enhanced approximation \(x_{m+1}\). IR is based on an inner-outer iteration scheme, where an outer iteration keeps control of the iterative procedure by calcuating the residual of the approximation whereas the inner iteration performs a linear system solution with the matrix \(A\).\N\NIR, apart from its use in ill-conditioned linear systems and in inner-outer iterative schemes in general have been applied in varied applications for example matrix preconditioning, high-performance computing and many others. While IR can lead to fast convergence towards a highly accurate approximation of \(x\), IR may diverge when \(A\) is ill-conditioned or the inner iteration introduces more error than the outer iteration can sustain.\N\NIn this paper, as a main contribution, the authors produce a new IR scheme which does not result in divergence regardless of the magnitude of the error stemming from the approximate linear system solution or the condition number of the matrix \(A\).\N\NThe paper is well written with a good set of references.