A family of immersed finite element spaces and applications to three dimensional H(curl) interface problems

Long Chen, Ruchi Guo, Jun ZouarXiv 2022 · arXiv · OpenReview

Introduction

Maxwell interface problems are central to electromagnetic applications, such as simulating fields in motors or complex electronic components. Traditionally, these problems require "fitted" meshes where element boundaries align with material interfaces. In 3D, generating such meshes is computationally expensive and often leads to poor-quality elements (slivers) that degrade accuracy. This paper presents a new family of 3D Immersed Finite Element (IFE) spaces that solve these problems on simple, unfitted Cartesian meshes, achieving optimal convergence without the complexity of traditional fitted methods.

The Challenge of Unfitted Meshes

Unfitted mesh methods allow the interface to cut through elements, simplifying mesh generation significantly. However, standard unfitted methods often struggle to maintain accuracy because they do not naturally account for the physical "jumps" in electromagnetic fields at material boundaries. Many existing solutions rely on complex penalty terms that are difficult to tune and can hinder solver performance.

The concept of an unfitted mesh where the interface Γ cuts through the background grid.

Methodology: Nédélec-type IFE Spaces

The authors propose a framework based on three core innovations:

  1. Nédélec-type IFE Spaces: Special shape functions are constructed for $H(\text{curl})$ and $H(\text{div})$ spaces. These functions "encode" the physical jump conditions directly, making the basis functions themselves aware of the material interface.
  2. Discrete de Rham Complex: The paper proves that these new IFE spaces form a complete de Rham complex. This mathematical structure is essential for ensuring that simulations are physically consistent and free of "spurious solutions."
  3. Petrov-Galerkin (PG) Formulation: By using IFE spaces for trial functions and standard conforming FE spaces for test functions, the authors achieve optimal convergence without needing extra penalty terms. This keeps the resulting linear systems manageable and square.
A tetrahedral element intersected by a curved material interface, illustrating the "immersed" nature of the method.

Fast Solvers and Robustness

To handle the resulting linear systems, which can be non-symmetric near the interface, the authors developed a modified Hiptmair-Xu (HX) preconditioner. This solver includes a "block diagonal smoother" specifically targeting elements near the interface.

Numerical experiments on benchmark problems—including flat, spherical, and torus interfaces—confirm that the method achieves optimal $O(h)$ convergence in the $H(\text{curl})$ norm and $O(h^2)$ in the $L^2$ norm. The solver remains robust even when the interface is extremely close to mesh nodes, a scenario that often causes standard methods to fail. For a problem with over 2.4 million degrees of freedom, the solver converged in just 41-43 iterations, demonstrating its efficiency for large-scale 3D problems.

Conclusion

This work provides a robust, penalty-free approach to 3D Maxwell interface problems. By leveraging the de Rham complex and a Petrov-Galerkin scheme, the authors enable high-accuracy simulations on simple meshes. Future work may extend this framework to higher-order elements and provide more rigorous theoretical proofs for 3D stability conditions.

Summary updated

Abstract

Maxwell interface problems are of great importance in many electromagnetic applications. Unfitted mesh methods are especially attractive in 3D computation as they can circumvent generating complex 3D interface-fitted meshes. However, many unfitted mesh methods rely on non-conforming approximation spaces, which may cause a loss of accuracy for solving Maxwell equations, and the widely-used penalty techniques in the literature may not help in recovering the optimal convergence. In this article, we provide a remedy by developing N\'ed\'elec-type immersed finite element spaces with a Petrov-Galerkin scheme that is able to produce optimal-convergent solutions. To establish a systematic framework, we analyze all the $H^1$, $\mathbf{H}(\text{curl})$ and $\mathbf{H}(\text{div})$ IFE spaces and form a discrete de Rham complex. Based on these fundamental results, we further develop a fast solver using a modified Hiptmair-Xu preconditioner which works for both the GMRES and CG methods.
BibTeX
@misc{chen2022family,
  author = {Long Chen and Ruchi Guo and Jun Zou},
  title = {{A family of immersed finite element spaces and applications to three dimensional H(curl) interface problems}},
  year = {2022},
  eprint = {2205.14127},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.14127},
}
Figure