Preconditioning waveform relaxation iterations for differential systems (Q1913584): Difference between revisions

Preconditioning and overlapping of waveform relaxation methods (WRM) are discussed for sparse linear systems \(y'(t)+Qy(t) = g(t)\), \(t\in [t_0,T]\), \(y(t_0) = y_0\). They are based on the splitting \(Q=C+D\), where \(D\) absorbs the stiffness of \(Q\). Setting \(z(t)= \exp ((t-t_0) D)y(t)\), a WRM \((d/dt) z^{(k+1)}(t) + Mz^{(k+1)} (t)= N(t-t_0) z^{(k)}(t) + \exp ((t-t_0) D)g(t)\), \(z^{(k+1)} (0)= y_0\), is applied, where the matrix \(\exp (tD)C \exp (-tD)= M-N(t)\) is splitted so that \(M\) is constant and determines the particular iteration scheme (Gauss-Jacobi, Gauss-Seidel, etc.). Error bounds are derived and the heat equation in one and two space variables is considered as an example.