Spectral approximation of elliptic operators by the Hybrid High-Order method

Victor M. Calo, Matteo Cicuttin, Quanling Deng, Alexandre ErnMath. Comput 2019 · OpenReview

TL;DR

This paper introduces and analyzes a novel Hybrid High-Order (HHO) method for accurately approximating the eigenvalues and eigenfunctions of second-order elliptic differential operators. The method achieves optimal convergence rates, improving upon existing techniques and demonstrating strong numerical performance, even for non-smooth solutions.

What Problem Does This Paper Address?

The accurate computation of eigenvalues and eigenfunctions of elliptic operators is fundamental in many scientific and engineering disciplines, including quantum mechanics, structural analysis, and fluid dynamics. Traditional methods like finite elements have well-established convergence rates, but there's ongoing research into developing more efficient and accurate numerical techniques. This paper focuses on the spectral approximation of elliptic operators, specifically addressing the challenge of achieving high-order accuracy in both eigenvalues and eigenfunctions using a relatively new method called the Hybrid High-Order (HHO) method. While HHO has shown promise for source problems, its application to eigenvalue problems and its theoretical error analysis were less developed. The authors aim to fill this gap by providing a rigorous analysis of HHO for spectral approximation and demonstrating its effectiveness.

Methods

The core of this work is the application and analysis of the Hybrid High-Order (HHO) method for eigenvalue problems. The HHO method is characterized by using both cell-based and face-based polynomial unknowns. A key innovation presented here is the formulation of a discrete eigenvalue problem where the mass bilinear form acts only on cell unknowns, while the stiffness bilinear form acts on both. This leads to the introduction of a "purely cell-based operator" obtained by eliminating the face unknowns. This operator is crucial for the theoretical error analysis, allowing the authors to leverage abstract spectral approximation theory for compact operators. They prove convergence rates for eigenvalues and eigenfunctions, and their theoretical findings are supported by numerical experiments.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and numerical experiments on standard mathematical domains (unit interval, unit square, L-shaped domain, unit disk) rather than specific datasets.

Key Insights and Ideas

  • Eliminating Face Unknowns for Spectral Analysis: The clever elimination of face unknowns to create a cell-based operator is a critical step that simplifies the theoretical analysis of the HHO method for eigenvalue problems.
  • Optimal Convergence Rates: The HHO method, under appropriate smoothness assumptions, achieves optimal convergence rates for both eigenvalues ($h^{2t}$) and eigenfunctions ($h^t$ in the $H^1$-seminorm), where $t$ depends on the polynomial degree and solution regularity. This matches or exceeds the performance of many other methods.
  • Superconvergence Phenomenon: For a specific choice of the stabilization parameter ($\eta = 2k+3$), the authors numerically observe a superconvergence of order $h^{2k+4}$ for eigenvalues in one dimension, which is a significant improvement.
  • Robustness on Various Meshes: The method's effectiveness is demonstrated on both standard (triangular) and more general (polygonal/hexagonal) meshes, highlighting its flexibility.

Key Results

  • Theoretical Convergence: The paper establishes theoretical convergence rates of $h^{2t}$ for eigenvalues and $h^t$ for eigenfunctions in the $H^1$-seminorm, where $t \in [s, k+1]$ and $s$ is related to elliptic regularity. For smooth eigenfunctions, this translates to $h^{2k+2}$ for eigenvalues and $h^{k+1}$ for eigenfunctions.
  • Numerical Verification: Extensive numerical experiments on 1D and 2D domains (unit interval, unit square, L-shaped domain, unit disk) confirm the theoretical convergence rates for both smooth and non-smooth eigenfunctions.
  • Superconvergence in 1D: Numerical results show a superconvergence of order $h^{2k+4}$ for eigenvalues in 1D when the stabilization parameter is set to $\eta = 2k+3$.
  • Comparison with FEM: The HHO method, particularly with appropriate stabilization, shows competitive or superior performance compared to standard continuous finite element methods (FEM(1)).

Limitations and Future Work

  • Smoothness Assumptions: The theoretical analysis relies on certain smoothness assumptions for the eigenfunctions. While numerical results are presented for non-smooth cases, the theoretical guarantees are strongest for smoother solutions.
  • Superconvergence Mechanism: The observed superconvergence in 1D for a specific stabilization parameter is noted, but a deeper theoretical understanding of this phenomenon is suggested for future work.
  • Generalization to Other Problems: The authors mention that extensions to biharmonic eigenvalue problems and non-selfadjoint operators are possible avenues for future research.
  • Maxwell Eigenvalue Problems: The paper suggests that the HHO method could be extended to curl-curl problems, which are relevant in electromagnetics.

Summary updated

BibTeX
@inproceedings{calo2019spectral,
  author = {Victor M. Calo and Matteo Cicuttin and Quanling Deng and Alexandre Ern},
  title = {{Spectral approximation of elliptic operators by the Hybrid High-Order method}},
  booktitle = {Math. Comput},
  year = {2019},
  url = {https://openreview.net/forum?id=qTljFzEc4a},
}
Figure