MultiGrid Preconditioners for Mixed Finite Element Methods of Vector Laplacian

Long Chen, Yongke Wu, Lin Zhong, Jie Zhou

Abstract

Due to the indefiniteness and poor spectral properties, the discretized linear algebraic system of the vector Laplacian by mixed finite element methods is notoriously difficult to solve. While existing block diagonal preconditioners have shown effectiveness, they often require specialized, geometry-dependent smoothers. This paper proposes alternative block diagonal and block triangular preconditioners utilizing a variable V-cycle multigrid method with a standard point-wise Gauss-Seidel smoother. We prove that this approach provides a robust solver for discrete vector Laplacian operators and saddle point systems, including Maxwell equations with divergence-free constraints. The primary advantage of this method is its algebraic nature, allowing the point-wise Gauss-Seidel smoother to be implemented as a "black-box" tool in existing engineering workflows.

Introduction

In computational physics, particularly within electromagnetics and fluid dynamics, Mixed Finite Element Methods (MFEM) are essential for transforming continuous differential equations into discrete systems. However, the resulting linear algebraic systems are "saddle point" problems characterized by indefiniteness—having both positive and negative eigenvalues—and poor conditioning. As the mesh resolution increases, these numerical properties typically deteriorate, making standard solvers inefficient.

Traditional solutions, such as those proposed by Arnold, Falk, and Winther, rely on specialized smoothers (e.g., Schwarz or Hiptmair smoothers) to handle the complexities of $H(curl)$ and $H(div)$ spaces. While mathematically sound, these smoothers require deep integration with the underlying mesh geometry, making them difficult to implement in general-purpose software. This research addresses the need for a more "algebraic" approach that maintains high performance without requiring specialized geometric knowledge.

Methodology

The Schur Complement and Preconditioning

The authors focus on the Schur complement of the system, which encapsulates the most challenging numerical properties of the saddle point problem. To address this, they designed two primary types of preconditioners:

  1. Block Diagonal Preconditioners: Providing a stable, symmetric approach to scaling the system.
  2. Block Triangular Preconditioners: An alternative that often yields faster convergence by utilizing the coupling between variables more effectively.

Variable V-cycle Multigrid

The core of the solver is a variable V-cycle multigrid strategy. Multigrid methods solve problems across multiple levels of mesh resolution to eliminate errors at different scales. A key theoretical hurdle in this work was the use of "non-inherited" operators on coarse grids—operators that are not simple copies of the fine grid due to the presence of inverse mass matrices. The authors utilized the Bramble-Pasciak-Xu (BPX) framework to prove that the method remains stable despite this complexity.

The "Black-Box" Smoother

The most significant innovation is the proof that a standard point-wise Gauss-Seidel smoother is sufficient within this framework. Unlike previous methods that required complex block-smoothing operations, this point-wise approach is purely algebraic. It treats the matrix entries as a collection of numbers without needing to reference the physical coordinates or the connectivity of the mesh elements, facilitating a "black-box" implementation.

Numerical Experiments

To validate the theoretical framework, the authors conducted benchmarks across several 2D and 3D domains designed to test both standard performance and edge cases involving geometric singularities:

  • Standard Geometries: Unit square (2D) and unit cube (3D).
  • Complex Geometries: L-shaped domains and "crack" domains (a square with a slit), which test the solver's robustness in the presence of partial regularity where physics becomes singular at sharp corners.
  • Applications: The tests covered both the Vector Laplacian and Maxwell equations with divergence-free constraints.

Key Results and Discussion

Mesh Independence

A critical metric for any multigrid solver is mesh independence—the property where the number of iterations required for convergence remains constant regardless of the number of degrees of freedom (DoF). The results demonstrated excellent scaling:

  • In 2D unit square tests, the diagonal preconditioner consistently required approximately 28 iterations to converge, whether the system had 4,000 or 263,000 DoF.
  • The triangular preconditioner proved even more efficient, often halving the iteration count (e.g., 14 iterations vs. 28).

Performance in Maxwell Equations

When applied to the Maxwell equations on a 3D L-shaped domain, the triangular preconditioner maintained its efficiency. It solved a system with 241,000 variables in just 19 iterations, demonstrating that the method is robust even for complex 3D physics and non-smooth geometries.

Robustness to Geometry

Despite the rigorous mathematical proofs relying on "full elliptic regularity" (typically found in smooth domains), the numerical experiments showed that the solver performed remarkably well on L-shaped and crack domains. This suggests the method's practical utility exceeds its current theoretical bounds.

Conclusion

This work provides a simplified yet mathematically rigorous path toward solving complex vector Laplacian systems. By proving that a standard Gauss-Seidel smoother can be used effectively within a variable V-cycle multigrid framework, the authors have bridged the gap between high-level numerical analysis and practical engineering implementation. Future work may extend these findings to higher-order finite elements and more complex prolongation operators for different types of Nedelec edge elements.

Summary updated

Abstract

Due to the indefiniteness and poor spectral properties, the discretized linear algebraic system of the vector Laplacian by mixed finite element methods is hard to solve. A block diagonal preconditioner has been developed and shown to be an effective preconditioner by Arnold, Falk, and Winther [Acta Numerica, 15:1--155, 2006]. The purpose of this paper is to propose alternative and effective block diagonal and block triangular preconditioners for solving this saddle point system. A variable V-cycle multigrid method with the standard point-wise Gauss-Seidel smoother is proved to be a good preconditioner for a discrete vector Laplacian operator. This multigrid solver will be further used to build preconditioners for the saddle point systems of the vector Laplacian and the Maxwell equations with divergent free constraint. The major benefit of our approach is that the point-wise Gauss-Seidel smoother is more algebraic and can be easily implemented as a black-box smoother.
BibTeX
@misc{chen2016multigrid,
  author = {Long Chen and Yongke Wu and Lin Zhong and Jie Zhou},
  title = {{MultiGrid Preconditioners for Mixed Finite Element Methods of Vector Laplacian}},
  year = {2016},
  eprint = {1601.04095},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1601.04095},
}
Figure