Propagation of round-off errors and the role of stability in numerical methods for linear and nonlinear PDEs (Q1091088)

The author points out that the usual stability-convergence-consistency theory of the numerical solution of PDE is based on a linearity assumption on round-off errors. This is not realistic and leads to the paradoxical conclusion that better results can be achieved for PDE than for ODE. The paper presents a method based on an essentially nonlinear approach to the propagation of round-off errors. This allows for a separation between stability under the effect of truncation errors from stability under the effect of the propagation of round-off errors. As a consequence, for smooth initial data, stability conditions can be weakened and convergence properties improved.