Abstract
This paper introduces bootstrap two-grid and multigrid finite element approximations to the Laplace-Beltrami (surface Laplacian) eigen-problem on a closed surface. The proposed multigrid method is suitable for recovering eigenvalues having large multiplicity, computing interior eigenvalues, and approximating the shifted indefinite eigen-problem. Convergence analysis is carried out for a simplified two-grid algorithm and numerical experiments are presented to illustrate the basic components and ideas behind the overall bootstrap multigrid approach.
Introduction
The Laplace-Beltrami operator is fundamental in physics and engineering for describing vibrational modes on curved surfaces, such as the resonance of a drumhead or the structural dynamics of complex 3D shapes. Numerically solving the associated eigenvalue problem requires high-resolution meshes, which often leads to prohibitive computational costs. While standard two-grid methods attempt to mitigate this by solving on a coarse mesh and correcting on a fine one, they face significant challenges on curved geometries.
Specifically, two major hurdles arise:
- Geometric Error: In surface finite element methods, refining a mesh involves projecting new vertices onto the true curved surface. This results in non-nested spaces, which violates the assumptions of standard multigrid theory.
- Spectral Loss: Coarse meshes often lack the resolution to capture high-frequency or oscillatory modes. Consequently, these modes are "lost" during grid transitions, leading to inaccurate approximations for complex spectra.
Methodology
The authors employ the Surface Finite Element Method (SFEM) to discretize the problem on piecewise planar approximations of smooth surfaces. To address the limitations of standard solvers, they propose a Bootstrap Multigrid (BMG) framework.
Geometric Projection and Refinement
A critical component of the BMG approach is handling the non-nested nature of refined surface meshes. When a coarse element is refined, the midpoints of its edges are projected onto the continuous surface to form the fine mesh.
To bridge these spaces, the authors define a specific projection operator ($G_{H \to h}$) that accounts for the geometric shift, ensuring that information is accurately transferred between the coarse and fine representations.
The Bootstrap Framework
The BMG method introduces several innovations:
- Space Enrichment: Instead of a static coarse space, the BMG method dynamically enriches it with approximate eigenfunctions computed on the fine grid. This "bootstrapping" allows the coarse grid to resolve oscillatory modes it would otherwise miss.
- Inverted V-Cycle: The algorithm utilizes an "inverted" cycle (coarse-fine-coarse). It employs Rayleigh quotient iterations to move from coarse to fine grids and simple smoothers, such as Gauss-Seidel or Kaczmarz, to return to the coarse level.
- Shifted Indefinite Solver: To target "interior" eigenvalues—those located within the spectrum rather than at the extremes—the framework incorporates a shifted solver, allowing for localized spectral analysis.
Numerical Experiments and Results
The algorithm was evaluated using synthetic benchmarks on a 2-sphere ($S^2$), where exact eigenvalues (spherical harmonics) are known analytically. Meshes were generated using Distmesh, and experiments were conducted within the iFEM MATLAB toolbox.
Key Findings
- Optimal Convergence: Numerical results confirmed the theoretical prediction of $O(h^2)$ convergence for eigenvalues and $O(h)$ for eigenfunctions in the $H^1$ norm.
- Spectral Recovery: The BMG method successfully recovered high-multiplicity eigenvalues that standard two-grid methods failed to resolve. For instance, in cases where standard methods "lost" the 34th eigenvalue, BMG maintained high precision.
- Efficiency of Smoothers: The study demonstrated that direct solves on the fine mesh are unnecessary. A few iterations of a simple smoother (like the Kaczmarz method) are sufficient to maintain optimal convergence, drastically reducing computational overhead.
Conclusion and Future Directions
The Bootstrap Multigrid approach provides a robust framework for solving the Laplace-Beltrami eigenvalue problem on smooth, closed surfaces. By combining space enrichment with geometric projections, it overcomes the traditional pitfalls of spectral loss and non-nestedness.
Future work may extend this framework to surfaces with boundaries, sharp edges, or complex topologies (high genus). Additionally, the authors suggest that the "Bootstrap Algebraic Multigrid" (BAMG) variant could be further developed for applications where the underlying geometry is not explicitly known.
Summary updated