No label defined (Q4036848)

[For the entire collection see Zbl 0755.00007.] The article studies time parallelism for the numerical solution of nonstiff ordinary differential equations. Many explicit parallel methods for ordinary differential equations do not possess satisfactory stability properties. The paper presents a precise understanding of how stability limits the potential of parallelism in numerical integration of ordinary differential equations and proposes a specific approach to construct good parallel methods. Thus, if an optimally stable sequential method is stability-bounded when applied to a certain problem for a given tolerance, parallelism offers no speedup at all. A strategy to derive good parallel methods is to find zero-stable methods whose stability polynomials are perfect powers of those of simple methods with good stability regions. Based on this approach some new efficient parallel methods are proposed. A family of one-stage block methods and a family of two-stage block methods are introduced. The stability regions of these methods do not change as the order increases and the stability properties are much better than the Adams predictor-corrector scheme of the same order.