Accelerated iterative method for \(Z\)-matrices (Q2564284)

The authors consider Gauss-Seidel-type methods for solving a linear system \(Ax=b\), where \(A\) is supposed to be a \(Z\)-matrix, with \(a_{ii}=1\). The standard Gauss-Seidel method is applied to the preconditioned system \[ PAx =Pb, \] where the preconditioner \(P=I +\beta U\) is derived from the Gauss-Seidel decomposition \(A=I- L-U\), with some positive parameter \(\beta\), the identity matrix \(I\), the strictly lower triangular matrix \(L\), and the strictly upper triangular matrix \(U\). The authors study the convergence and the choice of the parameter \(\beta\). The numerical results show that the proposed method converges faster than the standard Gauss-Seidel method, the adaptive Gauss-Seidel method corresponding to the case \(\beta=1\), and even the successive overrelaxation method.