Generalized Multiscale Finite-Element Method (GMsFEM) for elastic wave propagation in heterogeneous, anisotropic media

Kai Gao, Shubin Fu, Richard L. Gibson Jr., Eric T. Chung, Yalchin Efendiev

TL;DR

This paper introduces the Generalized Multiscale Finite-Element Method (GMsFEM) for simulating elastic wave propagation in complex, heterogeneous, and anisotropic geological media. GMsFEM significantly reduces computational cost by constructing specialized basis functions that capture fine-scale details on a coarse grid, enabling faster and more accurate seismic modeling for large-scale geological problems.

What Problem Does This Paper Address?

Simulating seismic wave propagation through Earth's complex subsurface is crucial for understanding geological structures and subsurface resources like oil and gas reservoirs. Traditional numerical methods, such as finite-difference and finite-element methods, become computationally prohibitive for large 3D models due to the need for extremely fine grids to resolve fine-scale heterogeneities. While effective medium theories can simplify calculations, they fail when heterogeneities are comparable to or larger than the seismic wavelength, leading to inaccurate results. This paper addresses the challenge of developing a method that is both accurate for complex media and computationally efficient for large-scale simulations.

Methods

The core innovation is the Generalized Multiscale Finite-Element Method (GMsFEM). This approach constructs multiscale basis functions by solving local problems within coarse grid elements. These basis functions are not predefined polynomials but are derived from the specific material properties of the local region, allowing them to capture fine-scale heterogeneity. The paper explores two types of multiscale basis functions:

  • Type I: Solves a local linear elasticity eigenvalue problem to obtain basis functions representing standing modes within a coarse element.
  • Type II: Decomposes the basis function space into "interior" and "boundary" components, each solved from separate local spectral problems. This type aims to better capture wave propagation across element boundaries.

The GMsFEM framework is presented in both Continuous Galerkin (CG) and Discontinuous Galerkin (DG) formulations. The DG formulation is highlighted for its flexibility with non-conforming meshes. The paper also introduces an oversampling technique to mitigate the influence of boundary conditions on local problem solutions and an adaptive strategy to dynamically assign the number of basis functions per coarse element based on local wave velocity.

Datasets

Not discussed in the paper. The paper uses synthetic models designed to test the method's capabilities.

Key Insights and Ideas

  • Multiscale Basis Functions are Key: The fundamental idea is that by solving local problems that reflect the underlying physics and material properties, we can create basis functions that implicitly capture fine-scale details. This allows us to use a coarse grid for the global simulation while retaining accuracy.
  • Type II Basis Functions Offer More Nuance: The decomposition into interior and boundary basis functions (Type II) seems to provide a more refined way to handle wave phenomena, especially at element interfaces, compared to the simpler Type I approach.
  • Oversampling is Crucial for Accuracy: The oversampling technique significantly reduces the error introduced by fixed boundary conditions in the local problems, leading to more robust and accurate multiscale basis functions.
  • Adaptivity is Practical: The adaptive assignment of basis functions based on local wave velocity is a practical approach to optimize computational cost. Slower regions (longer wavelengths) require more basis functions, while faster regions (shorter wavelengths) can use fewer, leading to efficiency gains.
  • GMsFEM Outperforms Standard FEM in Heterogeneous/Anisotropic Media: The results consistently show that GMsFEM achieves comparable or better accuracy than standard FEM with significantly fewer degrees of freedom, especially in complex, heterogeneous, and anisotropic scenarios.

Key Results

  • Significant DOF Reduction: GMsFEM achieved accuracy comparable to fine-grid FEM simulations but with a reduction in degrees of freedom (DOF) by factors of 10 to 100 or more. For example, in one test, a DG-GMSFEM solution with 60x60 coarse elements and 70 basis functions achieved accuracy similar to a DG-FEM solution on a 600x600 fine mesh, but with drastically fewer DOFs.
  • Improved Accuracy and Reduced Dispersion: GMsFEM solutions exhibited significantly less numerical dispersion, particularly for the slower S-waves, compared to standard FEM solutions on comparable coarse meshes.
  • Computational Efficiency: While the initial computation of multiscale basis functions can be time-consuming, the overall simulation time (time stepping) for GMsFEM was often shorter than for standard FEM, especially when considering the DOF reduction. The adaptive approach further optimized this.
  • Effective for Anisotropic Media: The method demonstrated effectiveness in simulating wave propagation in anisotropic media, which is a significant challenge for many numerical techniques.

Limitations and Future Work

  • Basis Function Computation Cost: The initial computation of multiscale basis functions can be expensive, especially for complex models or when many basis functions are used. The paper notes that this is a one-time cost and can be parallelized.
  • Tuning of Penalty Parameters (DG): The DG formulation requires careful selection of penalty parameters for stability, which can be a tuning effort.
  • Accuracy vs. Number of Basis Functions: The accuracy of GMsFEM is directly related to the number of multiscale basis functions used. While this offers control, finding the optimal number can require test evaluations.
  • Future Work: The authors aim to investigate improvements for CG-GMSFEM to reduce computational burden in mass matrix inversion and to develop parallel implementation for DG-GMSFEM. They also plan to provide a rigorous analysis of stability conditions and dispersion relations for anisotropic elastic wave equations.

Summary updated

Abstract

It is important to develop fast yet accurate numerical methods for seismic wave propagation to characterize complex geological structures and oil and gas reservoirs. However, the computational cost of conventional numerical modeling methods, such as finite-difference method and finite-element method, becomes prohibitively expensive when applied to very large models. We propose a Generalized Multiscale Finite-Element Method (GMsFEM) for elastic wave propagation in heterogeneous, anisotropic media, where we construct basis functions from multiple local problems for both the boundaries and interior of a coarse node support or coarse element. The application of multiscale basis functions can capture the fine scale medium property variations, and allows us to greatly reduce the degrees of freedom that are required to implement the modeling compared with conventional finite-element method for wave equation, while restricting the error to low values. We formulate the continuous Galerkin and discontinuous Galerkin formulation of the multiscale method, both of which have pros and cons. Applications of the multiscale method to three heterogeneous models show that our multiscale method can effectively model the elastic wave propagation in anisotropic media with a significant reduction in the degrees of freedom in the modeling system.
BibTeX
@article{gao2015generalized,
  author = {Kai Gao and Shubin Fu and Richard L. Gibson Jr. and Eric T. Chung and Yalchin Efendiev},
  title = {{Generalized Multiscale Finite-Element Method (GMsFEM) for elastic wave propagation in heterogeneous, anisotropic media}},
  journal = {Journal of Computational Physics},
  year = {2015},
  doi = {10.1016/j.jcp.2015.03.068},
}
Figure