Matrix-oriented FEM formulation for stationary and time-dependent PDEs on x-normal domains

TL;DR

This paper introduces a matrix-oriented finite element method (MO-FEM) for solving elliptic and parabolic PDEs on structured and normal domains, reformulating the problem as Sylvester matrix equations to improve computational efficiency.

Contribution

The work develops a novel MO-FEM approach that reformulates PDE discretizations as Sylvester matrix equations, enabling efficient spectral solution methods on complex domains.

Findings

Significant reduction in computational time compared to classical methods.

Memory savings when solving large-scale PDE problems.

Successful application to high-resolution Turing pattern simulations.

Abstract

When numerical solution of elliptic and parabolic partial differential equations is required to be highly accurate in space, the discrete problem usually takes the form of large-scale and sparse linear systems. In this work, as an alternative, for spatial discretization we provide a Matrix-Oriented formulation of the classical Finite Element Method, called MO-FEM, of arbitrary order $k\in\mathbb{N}$. On structured 2D domains (e.g. squares or rectangles) the discrete problem is then reformulated as a Sylvester matrix equation, that we solve by the reduced approach in the associated spectral space. On a quite general class of domains, namely normal domains, and even on special surfaces, the MO-FEM yields a multiterm Sylvester matrix equation where the additional terms account for the geometric contribution of the domain shape. In particular, we obtain a sequence of these equations after time discretization of parabolic problems by the IMEX Euler method. We apply the matrix-oriented form of the Preconditioned Conjugate Gradient (MO-PCG) method to solve each multiterm Sylvester equation for MO-FEM of degree $k=1,\dots,4$ and for the lumped $\mathbb{P}_1$ case. We choose a matrix-oriented preconditioner with a single-term form that captures the spectral properties of the whole multiterm Sylvester operator. For several numerical examples, we show a gain in computational time and memory occupation wrt the classical vector approach solving large sparse linear systems by a direct method or by the vector PCG with same preconditioning. As an application, we show the advantages of the MO-FEM-PCG to approximate Turing patterns with high spatial resolution in a reaction-diffusion PDE system for battery modeling.