On the accuracy of operator splitting for the monodomain model of electrophysiology

Abstract

Nonlinear systems of time dependent partial differential equations of real life phenomena tend to be very complicated to solve numerically. One common approach to solve such problems is by applying operator splitting which introduces subproblems that are easier to handle. The solutions of the subsystems are usually glued together by either the Godunov method (first-order) or the Strang method (second-order). However, the accuracy of such an approach may be very hard to analyse because of the complexity of the equations involved. The purpose of the present paper is to introduce another line of reasoning concerning the accuracy of operator splitting. Let us assume that a fully coupled and implicit discretization of the complete system has been developed. Under appropriate conditions on the continuous problem, such discretizations provide reasonable and convergent approximations. As in the continuous case, operator splitting can be utilized to obtain tractable algebraic subsystems. The problem we address in the present paper is to obtain a bound on the difference between the fully coupled implicit discrete solutions, u h, and the solutions, u h, s, obtained by applying operator splitting to these discrete equations. Suppose we know that u h converges to the analytical solution u as the grid is properly refined and, applying the results from this paper, that u h, s converges toward u h under grid refinement. Then, by the triangle inequality, also the splitting approximation u h, s converges toward the analytical solution u and convergence is thus obtained for an approximation that, from a practical point of view, is easier to compute.