Algebraic theory of multiplicative Schwarz methods (Q5952126)

The authors present results about the convergence of the multiplicative Schwarz methods for solving systems of linear equations with coefficient matrices which are either non-singular \(M\)-matrices or symmetric positive definite matrices. The convergence is studied using the theory of splitting. Nevertheless, the authors concentrate on the case of the algebraic multiplicative Schwarz method, some new results on the additive Schwarz method and hybrid methods are included. The approach used in this paper is suitable for problems where the coefficient matrix does not come from a differential equation (e.g. electrical power network, Leontief models in economics, etc.) as well as where the coefficient matrix comes from a discretization of a differential equation but no geometric information about the underlying mesh is available. The authors emphasize the use of Schwarz methods as solvers rather than preconditioners. The convergence bounds are presented in terms of matrix norms as well as spectral radii. Both of them are used to compare the convergence of different versions. The effect on the convergence of the algorithmic parameters such as the number of blocks or subdomains, the amount of the overlap, the inexact local solvers and the effect of adding coarse grid corrections are analysed.