A new iterative refinement for ill-conditioned linear systems based on discrete gradient

Abstract

In this paper, a new iterative refinement for ill-conditioned linear systems is derived based on discrete gradient methods for gradient systems. It is proved that the new method is convergent for any initial values irrespective of the choice of the stepsize h. Some properties of the new iterative refinement are presented. It is shown that the condition number of the coefficient matrix in the linear system to be solved in every step can be improved significantly compared with Wilkinson’s iterative refinement. The numerical experiments illustrate that the new method is more effective and efficient than Wilkinson’s iterative refinement when dealing with ill-conditioned linear systems.