Mixed Generalized Multiscale Finite Element Method for a Simplified Magnetohydrodynamics Problem in Perforated Domains

Valentin Alekseev, Qili Tang, Maria V. Vasilyeva, Eric T. Chung, Yalchin EfendievComput 2020 · DOI

TL;DR

This paper introduces a novel multiscale finite element method (Mixed GMSFEM) to efficiently solve simplified magnetohydrodynamics (MHD) problems in complex domains with perforations. By constructing specialized multiscale basis functions, the method significantly reduces computational cost while maintaining good accuracy, making it a promising tool for simulating challenging fluid dynamics scenarios.

What Problem Does This Paper Address?

Magnetohydrodynamics (MHD) describes the intricate interplay between electrically conducting fluids and magnetic fields, crucial for understanding phenomena like liquid metal cooling in nuclear reactors and astrophysical processes. However, simulating MHD flows in domains with complex geometries, such as those containing perforations (holes), poses a significant computational challenge. Traditional finite element methods require extremely fine grids to accurately capture the geometry of these perforations, leading to massive systems of equations that are computationally prohibitive. This paper aims to bridge this gap by developing a multiscale approach that can effectively handle these perforated domains without the need for prohibitively fine meshes.

Methods

The core of this work lies in the development and application of a Mixed Generalized Multiscale Finite Element Method (Mixed GMSFEM). Here's a breakdown of their approach:

  • Problem Formulation: They start with a simplified 2D MHD problem, focusing on the coupling of incompressible Navier-Stokes and Maxwell's equations. The system is linearized using a Picard iteration to decouple the equations.
  • Fine Grid Approximation: For the detailed resolution of perforations, a fine grid is constructed. On this fine grid, a mixed finite element method is employed for both the flow and magnetic field components.
  • Multiscale Basis Functions: The key innovation is the construction of multiscale basis functions. This involves:
  • Local Snapshot Spaces: For each coarse grid element (or face), local problems are solved with various boundary conditions on the fine grid within that local domain. The solutions to these local problems form a "snapshot space" that captures the essential multiscale features of the solution.
  • Local Spectral Decomposition: A spectral problem is then solved within this snapshot space to identify dominant modes. These dominant modes are selected as the multiscale basis functions. This process effectively reduces the dimensionality of the problem while retaining crucial information about the sub-grid geometry (perforations).
  • Coarse Grid Approximation: The Mixed GMSFEM then uses these multiscale basis functions to construct a reduced-order model on a coarser grid. This significantly reduces the number of degrees of freedom compared to a full fine-grid solution.
  • Discontinuous Galerkin (DG) for Flow: For the flow problem, a Discontinuous Galerkin (DG) method is employed on the fine grid, which offers advantages in handling discontinuities and is then adapted for the multiscale setting.

Datasets

Not discussed in the paper. The paper focuses on a model problem with defined parameters and geometries rather than using pre-existing datasets.

Key Insights and Ideas

  • Effective Representation of Perforations: The GMSFEM approach, by constructing local multiscale basis functions, is shown to effectively capture the influence of perforations on the overall flow and magnetic fields, even on a coarse grid.
  • Dimensionality Reduction: The core idea of using local spectral decomposition on snapshot spaces is a powerful technique for reducing the dimensionality of complex problems, leading to significant computational savings.
  • Accuracy with Fewer Basis Functions: A crucial finding is that the proposed method achieves good accuracy with a surprisingly small number of multiscale basis functions, highlighting the efficiency of the constructed basis.
  • Unified Framework: The paper demonstrates the applicability of the GMSFEM framework to a coupled MHD system, extending its utility beyond simpler elliptic or parabolic problems.

Key Results

The paper presents numerical results for a 2D perforated domain. Key findings include:

  • Magnetic Field (B): The multiscale solution for the magnetic field achieved a DOF of 980, which is only 0.48% of the fine-grid DOF (155,580), with a relative error of approximately 0.83% for the flux and around 1.57% for the magnetic field.
  • Velocity (u) and Pressure (p): For the velocity, the multiscale solution used a DOF of 4100 (0.9% of the fine-grid DOF of 433,384). The relative velocity error was reduced from around 60% to 8% with an increasing number of multiscale basis functions. The pressure error was also significantly reduced, from about 25% to around 1%.
  • Influence of Basis Functions: The results clearly show that increasing the number of multiscale basis functions generally leads to improved accuracy, but the method achieves good results even with a relatively small number of basis functions.

Limitations and Future Work

  • Simplified 2D Problem: The current study focuses on a simplified 2D MHD problem. The authors plan to extend the method to the full 3D problem in future work.
  • Oversampling Techniques: For further error reduction in the flow approximation, the authors suggest applying oversampling techniques in future research.
  • Generalization: While the paper demonstrates the method's effectiveness, further exploration of its performance on different perforated geometries and more complex MHD scenarios would be valuable.

Summary updated

BibTeX
@article{alekseev2020mixed,
  author = {Valentin Alekseev and Qili Tang and Maria V. Vasilyeva and Eric T. Chung and Yalchin Efendiev},
  title = {{Mixed Generalized Multiscale Finite Element Method for a Simplified Magnetohydrodynamics Problem in Perforated Domains}},
  journal = {comput.},
  year = {2020},
  doi = {10.3390/computation8020058},
}
Figure