A functional fitting Runge-Kutta method with variable coefficients (Q5937418)

Generalized \(s\)-stage Runge-Kutta methods are presented. The approach is based on the exact integration of a set of \(s\) linearly independent functions \(\varphi_i\), \(i= 1,\dots, s\). In the case \(\varphi_i(t)= t^i\), \(i= 0,\dots, s-1\) these methods reduce to Runge-Kutta methods of collocation type. In this approach the coefficients \(b_i\), \(i= 1,\dots, s\), \(a_{ij}\), \(1\leq i,j\leq s\) of the Butcher-tableau depend on the step-size parameter \(h\). It is shown that the stage order of such a variable coefficient Runge-Kutta method (VCRK) is \(s\) and the classical order is at least \(s\). Using certain assumptions about the collocation points the classical order \(2s\) is attainable. This corresponds to the standard Runge-Kutta case. Methods of this type seem to have advantages if there is some a priori knowledge about the solution characteristics of the underlying ordinary differential equation. In this case the functions \(\varphi_i\), \(i= 1,\dots, s\) can be used to mimic this behaviour. This is shown in some examples. Unfortunately, the computational effort is higher, since the \(h\)-dependent coefficients of the VCRK have to be computed in advance and the evaluation at every time step is more complicated.