Parallel block predictor-corrector methods of Runge-Kutta type (Q1308565)

This paper discusses block predictor-corrector (PC) methods with Runge- Kutta (RK) correctors. After a discussion of effective stability boundaries, the authors present a brief introduction to block parallel implicit RK methods (BPIRK), based on \(s\)-stage implicit RK correctors applied on a block of size \(r\) from the current point. These \(r\) block approximations can be computed in parallel. High-order predictor formulae are obtained by Lagrange or Hermite interpolation. It is found that by close spacing of the \(r\) block points, accurate predictor estimates can be made. The parallel implementation ensures that costs may be significantly better than those of conventional sequential computation. Application of these methods to a few well-known test problems indicates that the overall cost reduction can be as high as a factor of 11.