Adaptive Finite Element Solution of Multiscale PDE-ODE Systems

TL;DR

This paper develops an adaptive finite element method for multiscale PDE-ODE systems, coupling macroscale PDEs with microscale ODEs via an intermediate scale, and introduces error estimates to guide efficient computation.

Contribution

It presents a mathematically consistent coupling approach with goal-oriented error estimates for multiscale PDE-ODE models, including a Monte Carlo sampling strategy and adaptive solution methodology.

Findings

Effective coupling of macro and micro models demonstrated

Goal-oriented error estimates improve computational efficiency

Adaptive method successfully applied to realistic heart activity model

Abstract

We consider adaptive finite element methods for solving a multiscale system consisting of a macroscale model comprising a system of reaction-diffusion partial differential equations coupled to a microscale model comprising a system of nonlinear ordinary differential equations. A motivating example is modeling the electrical activity of the heart taking into account the chemistry inside cells in the heart. Such multiscale models pose extremely computationally challenging problems due to the multiple scales in time and space that are involved. We describe a mathematically consistent approach to couple the microscale and macroscale models based on introducing an intermediate "coupling scale". Since the ordinary differential equations are defined on a much finer spatial scale than the finite element discretization for the partial differential equation, we introduce a Monte Carlo approach to sampling the fine scale ordinary differential equations. We derive goal-oriented a posteriori error estimates for quantities of interest computed from the solution of the multiscale model using adjoint problems and computable residuals. We distinguish the errors in time and space for the partial differential equation and the ordinary differential equations separately and include errors due to the transfer of the solutions between the equations. The estimate also includes terms reflecting the sampling of the microscale model. Based on the accurate error estimates, we devise an adaptive solution method using a "blockwise" approach. The method and estimates are illustrated using a realistic problem.