LINE SEARCH ACCELERATION OF ITERATIVE METHODS

This paper describes and analyzes a simple technique that accelerates the convergence of iterative methods for solving large linear systems. The proposed technique can be used with any convergent method. However, it is especially useful when solving unstructured systems, since in this case traditional methods, as SOR or Chebyshev semiiteration, have difficulty in obtaining optimal parameters. Special attention is given to column relaxation and row relaxation. It is demonstrated experimentally that an accelerated Gauss-Seidel scheme runs faster than an optimally tuned SOR scheme. © 1990.