Analysis of a multirate theta-method for stiff ODEs (Q1007391)

This paper is concerned with the numerical analysis of the so called multirate methods for the numerical solution of stiff differential systems in which coexist components with different behavior that are integrated with different time steps with the purpose to improve the uniform accuracy and efficiency of numerical integrators. A multirate procedure with automatic step size partitioning for the large error components was introduced by the authors in a previous paper [BIT 47, No.~1, 137--155 (2007; Zbl 1113.65071)]. In the present paper a theoretical analysis of this procedure is presented when the underlying method is the theta-method with one level of temporal refinement. This meas that for those components in which the error estimation indicates an step size reduction, the computation is redone with a halved step size and linear or quadratic interpolation is employed at the half step. Under these conditions error recursions and sharp error bounds are derived for linear constant coefficient systems \( w' = A w + g(t)\) with the stability condition on the logarithmic matrix norm \( \mu_{\infty} ( A ) \leq 0\). Further, a complete study of stability and contractivity properties of multirate theta-methods is presented, in particular for a fixed block partitioning and non stiff coupling. The paper ends with some numerical experiments for two test problems with the aim to compare the relative efficiency of refined and non refined theta methods and different levels of refinement and interpolation.