Parallel relaxed multisplitting methods for a symmetric positive definite matrix (Q2493762)

The authors consider the convergence of the relaxed multisplitting method (Method 1) and the relaxed nonstationary two-stage multisplitting method (Method 2) for solving a linear system of the form \(Ax=b\) where \(x,b\in {\mathbb R}^n\) and \(A\in {\mathbb R}^{n\times n}\) is a large sparse symmetric positive definite matrix. The multisplitting method can be thought of as an extension and a generalization of the classical block Jacobi method. Method 2 is implemented in a parallel environment by means of incomplete factorizations, instead of using \(P\)-regular splittings, as inner splittings and applied as a preconditioner of the conjugate gradient (CG) method. Parallel performance results are provided for Methods 1 and 2 and the CG method with a parallel preconditioner.