B-convergence results for linearly implicit one step methods (Q1090079)

Linearly implicit one-step methods (Rosenbrock-type methods, adaptive Runge-Kutta methods,...) are investigated with respect to their ''B- convergence''. Here B-convergence order q has not the original sense of \textit{R. Frank\textit{, \textit{J. Schneid}}} and \textit{C. W. Ueberhuber} [SIAM J. Numer. Anal. 18, 753-780 (1981; Zbl 0467.65032)]. It means that for semi- linear problems \(y'=Ty+g(t,y)\) the global error can be bounded by \(Ch^ q\), where C may depend only on the logarithmic norm of T, on the Lipschitz constant of g and on derivatives of y(t) and g(t,y(t)). The authors derive sufficient conditions for B-convergence order q and present some particular methods with \(q\leq 3\). Numerical experiments on an artificial test problem confirm the theoretical results.