The search for the Holy Grail, or: Predictor-corrector methods for solving ODEIVPs (Q686548)

To motivate some potential for efficiency in treating initial value problems using a parallel implementation, the author develops a general formulation for predictor-corrector methods. The main aspects explored are the effects of errors from low-order predictors, and the stability of the algorithms. It is shown that algorithms with an unrestricted region of stability for linear constant coefficient problems with a connectedly diagonally dominant matrix can be constructed using a trivial predictor. However, such low order predictors introduce errors which persist for a substantial interval of the solution. In contrast, corresponding errors introduced by order 1 predictors are not as large, although the maximum step -- step of stable algorithms is finite. Nevertheless, carefully constructed algorithms of the latter type show substantial improvement on the stability of the standard implementations of the predictor-corrector formulas on which they are based. Other alternatives are mentioned, and the author indicates future work will use \(s\)-stage Runge-Kutta methods as correctors. Such improvements may help to reduce the global errors occurring within the interval \([0,1]\) for the \(m\times m\) matrix \(Q\) of (24) which are considerably larger than the endpoint global errors reported in Table 2.