A necessary condition for B-convergence of Runge-Kutta methods (Q911235)

The author considers initial value problems of the form (*) \(y'=f(t,y)\), \(y(0)=y_ 0\) with f: [0,1]\(\times {\mathbb{R}}^ n\to {\mathbb{R}}^ n\) and \(y_ 0\in {\mathbb{R}}^ n\). The set of functions f satisfying \(<f(t,y)- f(t,z),\quad y-z>\leq m\| y-z\|^ 2\) with \(m\in {\mathbb{R}}\) is denoted by \(J_ m\). Let \(h>0\) and \(t_ k=kh\) \((k=0,1,2,...)\) then the Runge- Kutta method is given by \[ Y^ k_ i=\eta_{k- 1}+h\sum^{s}_{j=1}a_{ij}f(t_{k-1}+c_ jh,\quad Y^ k_ j),\quad 1\leq i\leq s \] \[ \eta_ k=\eta_{k-1}+\sum^{s}_{j=1}b_ jf(t_{k-1}+c_ jh,\quad Y^ k_ j) \] where \(s\in {\mathbb{N}}\) and \(a_{ij}\), \(b_ i\), \(c_ i\) are real parameters. The Runge-Kutta method is called B-convergent of order \(p>0\) on \(J_ m\) if for any initial value problem (*) with \(f\in J_ m\) there is an error bound \(\| y(t_ k)-\eta_ k\| \leq Ch^ p\) for \(h\leq H\), \(0\leq t_ k\leq 1\), where y(t) denotes the solution of (*) and C is a suitable constant. The author proves necessary conditions for B-convergence on \(J_ m\). As an application two examples are given which are not B- convergent.