A high order discontinuous Galerkin method for the symmetric form of the anisotropic viscoelastic wave equation

Khemraj Shukla, Jesse Chan, Maarten V. de Hoop

Abstract

Accurately simulating wave propagation in the Earth's subsurface is critical for seismic imaging and reservoir characterization. However, real-world geological media are characterized by complex physical phenomena, including anisotropy (direction-dependent velocity) and viscoelasticity (energy dissipation). Traditional numerical models often simplify these effects, leading to inaccurate amplitude predictions and blurred seismic images. This paper introduces a high-order discontinuous Galerkin (DG) finite element method designed for inhomogeneous, anisotropic, and dissipative media. By reformulating the wave equation into a symmetric hyperbolic system and utilizing a memory variable treatment for stress-strain convolutions, the proposed method ensures mathematical stability and high-order accuracy. Validation against analytical solutions and large-scale 3D simulations demonstrates the method's effectiveness in handling complex topography and heterogeneous material properties.

Introduction

In geophysics, the ability to predict how seismic waves traverse the Earth is fundamental to understanding the planet's interior and locating natural resources. Most standard seismic models rely on two major simplifications: they assume the medium is purely elastic (no energy loss) and isotropic (waves travel at the same speed in all directions).

In reality, materials such as cracked limestone or porous sandstone are both viscoelastic and anisotropic. Viscoelasticity causes the medium to "soak up" wave energy, leading to attenuation and phase shifts, while anisotropy causes waves to travel faster along certain structural orientations (like cracks) than others. Ignoring these factors results in incorrect seismic amplitudes and structural misinterpretations. While various methods exist to model these phenomena, they frequently suffer from numerical dispersion, difficulty in handling irregular geometries (such as mountains or jagged layers), or mathematical instabilities that cause simulations to diverge over time.

Methodology

Symmetric Hyperbolic Formulation

The core innovation of this work is the reformulation of the anisotropic viscoelastic wave equations into a symmetric hyperbolic system. This symmetry is not merely a mathematical convenience; it is a fundamental requirement for proving energy stability. By ensuring the system is symmetric, the researchers can guarantee that the numerical energy within the simulation will not grow artificially over time, preventing the "blow-up" common in complex wave simulations.

Memory Variable Treatment

Viscoelasticity is traditionally modeled using convolution integrals, which require storing the entire history of the wave—a massive computational and memory burden. To circumvent this, the authors employ auxiliary "memory variables" based on the Zener model (standard linear solids). This approach transforms the convolution integrals into a set of first-order differential equations, significantly reducing memory overhead while maintaining physical accuracy.

Discontinuous Galerkin (DG) Framework

The method utilizes a high-order DG finite element approach on unstructured tetrahedral meshes. Unlike finite-difference methods that rely on rigid rectangular grids—which often result in "staircase" errors when approximating sloped surfaces—the DG method uses flexible triangles (2D) and tetrahedra (3D). This allows the mesh to conform precisely to complex geological boundaries and surface topography. Information is communicated between these discontinuous elements via an energy-stable penalty flux, ensuring high-order accuracy without the computational bottlenecks associated with traditional Riemann solvers.

Validation and Results

Benchmarking and Convergence

The method's precision was verified using physical properties from real-world materials, including clay shale (highly anisotropic) and phenolic materials. The researchers compared numerical results against exact mathematical solutions, such as Carcione’s analytical viscoelastic solution. The method demonstrated $O(h^{N+1})$ convergence rates (where $N$ is the polynomial degree), meaning that as the mesh is refined, the error decreases significantly faster than in lower-order methods.

Capturing Complex Wave Phenomena

A key success of the method is its ability to resolve "shear wave cusps." These are sharp, pointed wave patterns that occur specifically in highly anisotropic materials and are notoriously difficult for standard numerical schemes to capture. The DG method successfully resolved these features in both phase and amplitude.

Large-Scale 3D Scalability

To test industrial applicability, the authors performed a simulation of a large-scale 3D reservoir measuring 22.8 km x 17.4 km. This model included undulated surface topography and complex internal layering. The results clearly illustrated the effects of attenuation, showing how viscoelasticity reduces wave amplitude and shifts the phase compared to purely elastic models. This confirms that accounting for these physics is essential for high-fidelity seismic imaging.

Conclusion and Future Work

The high-order DG method presented provides a robust and stable framework for simulating seismic waves in the Earth's most complex materials. By factoring material properties into the mass matrix rather than the fluxes, the scheme remains stable even across abrupt material interfaces.

While the method is computationally intensive due to the high number of variables per grid point (15+ in 3D), it is well-suited for modern high-performance computing architectures. Future research will focus on extending this framework to poroelastic media—where rocks are filled with moving fluids—and developing more advanced absorbing boundary conditions to better simulate infinite domains.

Summary updated

Abstract

Wave propagation in real media is affected by various non-trivial physical phenomena, e.g., anisotropy, an-elasticity and dissipation. Assumptions on the stress-strain relationship are an integral part of seismic modeling and determine the deformation and relaxation of the medium. Stress-strain relationships based on simplified rheologies will incorrectly predict seismic amplitudes, which are used for quantitative reservoir characterization. Constitutive equations for the rheological model include the generalized Hooke's law and Boltzmann's superposition principal with dissipation models based on standard linear solids or a Zener approximation. In this work, we introduce a high-order discontinuous Galerkin finite element method for wave equation in inhomogeneous and anisotropic dissipative medium. This method is based on a new symmetric treatment of the anisotropic viscoelastic terms, as well as an appropriate memory variable treatment of the stress-strain convolution terms. Together, these result in a symmetric system of first order linear hyperbolic partial differential equations. The accuracy of the proposed numerical scheme is proven and verified using convergence studies against analytical plane wave solutions and analytical solutions of viscoelastic wave equation. Computational experiments are shown for various combinations of homogeneous and heterogeneous viscoelastic media in two and three dimensions.
BibTeX
@inproceedings{shukla2021high,
  author = {Khemraj Shukla and Jesse Chan and Maarten V. de Hoop},
  title = {{A high order discontinuous Galerkin method for the symmetric form of the anisotropic viscoelastic wave equation}},
  booktitle = {Comput. Math. Appl},
  year = {2021},
  url = {https://openreview.net/forum?id=5oktZbStdV},
}
Figure