Multirate extrapolation methods for differential equations with different time scales (Q678115)

The authors develop and analyze a multirate extrapolation method for the numerical integration of problems where a small number of rapidly changing components restrict the stepsize of the integrator and where, in contrast to stiff problems, both fast and slow components are present throughout the interval of integration. The method is based on Richardson extrapolation: the fact that local error estimates are available at all extrapolation levels leads to an inexpensive strategy for the dynamic partitioning into several classes of slow to fast components. Also, slow and fast components are integrated simultaneously over a macro-step, and the refinement of the solution by high-order Richardson extrapolation is restricted to the faster components (with slow components being inactivated at earlier levels). The analysis is centered around the explicit Euler method, and detailed numerical experiments illustrate the effectiveness of the method (for which a FORTRAN code is available).