Geometric Decomposition and Efficient Implementation of High Order Face and Edge Elements

Chunyu Chen, Long Chen, Xuehai Huang, Huayi WeiarXiv 2023 · DOI · arXiv

Abstract

High-order face ($H(\text{div})$) and edge ($H(\text{curl})$) finite elements are essential for solving complex physics problems like electromagnetics and fluid flow, yet their implementation is often hindered by complex bookkeeping and continuity requirements. This paper introduces a unified geometric framework that simplifies these elements through a "nodal" approach. By constructing basis functions based on interpolation points and local vector frames, the authors bridge the gap between abstract mathematical theory and practical computational implementation.

Introduction

In Finite Element Methods (FEM), standard Lagrange elements are relatively straightforward because they are node-based, focusing on function values at specific points. However, vector-valued problems require elements that maintain specific continuity across boundaries: $H(\text{div})$ elements must maintain normal component continuity across faces, while $H(\text{curl})$ elements must maintain tangential component continuity along edges.

Traditionally, constructing these basis functions involves complex Piola transformations from a reference element to the actual mesh. While mathematically elegant, this process is computationally cumbersome and difficult to implement for high-order polynomials. There remains a significant "implementation gap"—a lack of clear documentation on managing the global indexing of the thousands of degrees of freedom (DoFs) generated in high-order 3D simulations.

Methodology

The proposed method constructs nodal bases for the "second family" of face (BDM) and edge (Nédélec) elements using three primary pillars:

  1. Geometric Decomposition: The polynomial space of an element is decomposed into a direct sum of "bubbles" associated with its sub-simplices (vertices, edges, faces, and the interior).
  2. Variable Vector Frames: Instead of a single global coordinate system, local frames (tangential and normal directions) are established at each interpolation point. By fixing specific components at these points, the required $H(\text{div})$ or $H(\text{curl})$ continuity is naturally enforced.
  3. Nodal Basis Functions: Explicit formulas are derived for basis functions that are "dual" to these DoFs, ensuring each basis function corresponds to exactly one physical measurement at a specific point.
Interpolation points on a triangular simplex for high-order finite element construction.

A critical component of this methodology is Indexing Management. The paper provides an algorithmic roadmap for mapping local indices within a single tetrahedron to a unique global index across the entire mesh. This involves sophisticated array sorting (e.g., argsort logic) to handle vertex ordering, ensuring continuity is correctly maintained across the 3D tetrahedral mesh.

Implementation and Validation

The algorithms were implemented within the FEALPy (Finite Element Analysis Library in Python) ecosystem. Validation was performed using standard synthetic benchmarks in a 3D domain:

  • The Mixed Poisson Problem: Used to test BDM (face) elements.
  • Time-Harmonic Maxwell Equations: Used to test second-kind Nédélec (edge) elements.

Numerical experiments confirmed that the proposed construction method is both accurate and efficient. For a polynomial degree $k$, the $L^2$ error for the vector field scaled at the theoretically optimal rate of $O(h^{k+1})$, while the divergence/curl error scaled as $O(h^k)$.

Conclusion and Future Work

The "nodal" shift presented in this work demonstrates that vector elements can be treated with nearly the same ease as scalar Lagrange elements by selecting appropriate local vector frames. This unified framework shows that Lagrange, BDM, and Nédélec elements belong to a single geometric hierarchy.

While the current focus is on the "second family" of elements on tetrahedral meshes, future work could extend these bookkeeping algorithms to the "first family" (incomplete spaces) and other mesh geometries such as hexahedral or polyhedral meshes. Additionally, further optimization may be needed to ensure that the use of variable frames does not introduce significant computational overhead compared to traditional Piola-mapped approaches.

Summary updated

BibTeX
@misc{chen2023geometric,
  author = {Chunyu Chen and Long Chen and Xuehai Huang and Huayi Wei},
  title = {{Geometric Decomposition and Efficient Implementation of High Order Face and Edge Elements}},
  year = {2023},
  eprint = {2309.13843},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2309.13843},
}
Figure