Line search acceleration of iterative methods (Q915366)

Relaxation methods are applied to solve real systems \(Ax=b\) where A is an \(m\times n\) matrix, large and sparse. An iterated method of solution is considered as a minimization algorithm for an objective function F(x); it is supposed that convergence is accelerated by minimizing F(x) along a line that connects successive iterants. This known technique is proved to be advantageous if A is unstructured (i.e. there is no simple rule which specifies the location of its nonzero elements), the computation of optimal parameters is avoided. Column relaxation and row relaxation are detailed. Numerical comparisons with a successive overrelaxation scheme is given.