Stability and error bounds for numerical methods for stiff initial value problems

Abstract: This article is based on a plenary lecture presented at the 1993 Conference on Scientific Computing and Differential Equations (SCADE 93), in Auckland, New Zealand, in honor of the 60th birthday of John Butcher. The article is a survey of the theory of stability and error bounds for numerical methods for stiff initial value problems. We begin with problems in singular perturbation form, where the distinction between ``fast'' and ``slow'' variables is clearest. Next we explain Kreiss's characterization of stiff problems that includes singular perturbation problems as a special case. The requirement of uniform stability for a class of problems provides an effective characterization of linear multistep methods appropriate for stiff problems. We describe the recent extension of these results to Runge-Kutta methods, and conclude with a brief discussion of error bounds.

Available by anonymous ftp from ftp.math.iastate.edu in the directory pub/alex/preprints/scade93