Retooling the method of block conjugate gradients (Q5942489)

This exploratory paper presents an effective means to generate block implementations of the conjugate gradient method for the solution of large, sparse systems of linear equations with symmetric positive definite matrix and vector right-hand side. The new algorithms presented avoid rank estimation and deflation through the use of changes of bases and algorithmic reformulation that eliminate rank near defects. The transformations implemented include a robust process of nonunitary orthogonalization in the metric of a symmetric positive definite matrix. Sample results of experiments are summarized at the end of the paper.