The approximate Runge-Kutta computational process (Q1577725)

It is well known that the choice of the iteration process for the approximate solution of the Runge-Kutta equations, especially when the underlying differential equations, \(B\dot y= f(t,y)\) (thus encompassing differential-algebraic equations), are stiff, may have a significant impact on the performance of the Runge-Kutta method. The authors introduce the concept of \((\Delta, K)\)-approximate stage pairs (where \(\Delta\) bounds the stage error, and \(K\) is related to the Runge-Kutta matrix \(A= [a_{ij}]\)) to account for the iteration errors and to obtain error bounds uniform with respect to stiffness. They also investigate stage derivative reuse for methods having an explicit first stage. This technique may result in significant performance gains, also when such methods are employed as error estimators. Previous computational heuristics can therefore be replaced by a consistent approach supported by theoretical analysis. A number of representative test problems is used to illustrate the various performance aspects of this approach.