Parallel `peer' two-step W-methods and their application to MOL-systems. (Q1426336)

Numerical methods for solving large system of ordinary differential equations obtained by some space discretization of time dependent partial differential equations are investigated. Among many methods so called W-methods are used very frequently. For higher dimensional partial differential equations special efforts have to be made to solve the linear equations efficiently in the linearly implicit W-methods. So combination of Krylov techniques with stiff integrators has become popular. In order to overcome the low stage order of classical one-step W-methods, the class of two-step W-methods were introduced. A weakness of these methods is some critical dependence on the stepsize ratio in the stiff case. This makes the construction of robust methods for a large number \(s\) of stages very difficult. To avoid this problem a new class -- `peer' two step W-methods -- of parallel methods was introduced recently. In these methods \(s\) numerical solutions of equal importance are computed in parallel at each time step. Some aspects of the stability and convergence of the method are investigated and different types of such methods are considered with order and stage order \(s-1\) or \(s\). These new methods are more robust with respect to stepsize changes, since they have optimal damping at infinity. Four numerical examples for testing the efficiency of presented methods are included. The conclusion of numerical comparison confirm that new methods in combination with Krylov-techniques for method of lines (MOL)-systems are well suited for parallel implementation. Moreover numerical tests reflect the theoretical results on stability.