On SOR Waveform Relaxation Methods

From MaRDI portal





Consider the system of linear ordinary differential equations (ODEs) \(B\dot u(t)+Au(t)=f(t)\), \(u(0)=u_0\), where the matrices \(B, A\) belong to \({\mathbb{C}}^{d\times d}\) and \(B\) is supposed to be nonsingular. The paper deals with the acceleration of the standard waveform relaxation (WR) method by successive overrelaxation (SOR) techniques when solving these general ODE systems. The WR is a numerical method convenient for solving large scale systems of ODEs on parallel computers. The splitting \(B=M_B-N_B\) and \(A=M_A-N_A\) leads to an iteration \[ \Biggl( M_B{d\over dt}+M_A \Biggr)u^{\nu }(t)=\Biggl(N_B{d\over dt}+N_A\Biggr) u^{(\nu -1)}(t)+f(t),\;u^{\nu}(0)=u_0. \] Different SOR acceleration schemes based on multiplication with a scalar parameter or convolution with a time-dependent function are analyzed. The nature of the different continuous-time and discrete-time SOR iteration operators is identified. The corresponding spectral radii and norm expressions are derived. Optimal pointwise overrelaxation in the \(B=I\) case is studied. Application of these results leads to explicit formulae for the optimal convergence factors as a function of the mesh size \(h\) for given model problems. Particularly, one- and two-dimensional heat equation for point relaxation and line relaxation are considered. The paper also illustrates by numerical experiments that the convolution SOR algorithm can attain excellent convergence rates, even in cases where the current theory does not apply.




Cited in
(27)








This page was built for publication: On SOR Waveform Relaxation Methods

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4377562)