Finite-Difference Solution of the Helmholtz Equation Based on Two Domain Decomposition Algorithms

Wensheng Zhang, Yunyin DaiJournal of Applied Mathematics and Physics 2013 · DOI

Abstract

In this paper, wave simulation with the finite difference method for the Helmholtz equation based on the domain decomposition method is investigated. The method solves the problem by iteratively solving subproblems defined on smaller subdomains. Two domain decomposition algorithms both for nonoverlapping and overlapping methods are described. More numerical computations including the benchmark Marmousi model show the effectiveness of the proposed algorithms. This method can be expected to be used in the full-waveform inversion in the future.

Introduction

Simulating how acoustic waves travel through the Earth is essential for oil and gas exploration and earthquake monitoring. Mathematically, this is governed by the Helmholtz equation—the frequency-domain version of the wave equation. However, researchers face a significant "high-frequency hurdle": to maintain accuracy, simulations typically require at least ten grid points per wavelength. For high-frequency waves or large geographic areas, this results in massive linear systems with millions of unknowns. These systems are notoriously difficult for standard computer solvers to handle because they are "indefinite" and "non-Hermitian," often leading to convergence failure or memory exhaustion.

Methodology

This paper proposes a "divide and conquer" strategy to tackle computational complexity by breaking large areas into smaller, manageable subdomains.

Finite Difference Scheme and Boundary Conditions

The authors utilize a second-order finite difference method (FDM) to discretize the 2D acoustic wave equation. To prevent waves from artificially reflecting off the edges of the computational domain, paraxial approximation-based Absorbing Boundary Conditions (ABCs) are implemented.

Domain Decomposition Methods (DDM)

The domain is split into $N$ subdomains using two primary algorithms:

  1. Non-overlapping DDM: Subdomains meet at interfaces. The algorithm iteratively adjusts boundary conditions between neighbors using "transmission conditions" until the solutions match seamlessly.
  2. Overlapping DDM: Subdomains share common regions. An auxiliary equation is constructed to maintain the symmetry of the sub-problems, which is critical for the stability and efficiency of numerical solvers.

Preconditioning

To speed up the iterative solver (Bi-CGSTAB), the authors employ a Shifted-Laplacian preconditioner. This technique effectively "massages" the difficult Helmholtz matrix to make it behave more like a simpler Laplacian matrix, significantly accelerating convergence.

Experimental Results and Validation

The study validates these algorithms using several synthetic models, including homogeneous, layered, and complex geological structures.

Wavefield Accuracy

Simulations for varying wave numbers ($k=20, 30, 40$) demonstrate high-resolution wave propagation. As the wave number increases, the complexity of the vibration pattern rises, matching theoretical expectations for high-frequency behavior.

Figure 2: Simulated wavefield for wave number k=40 on a 128x128 grid, demonstrating high-resolution propagation.

Complex Model Performance

The ultimate test for the proposed algorithms is the Marmousi model, a famous geophysical benchmark mimicking the complex geology of the North Sea. The algorithm successfully handled the steep faults and strong velocity gradients of this model. The resulting wavefield contours were virtually identical to those produced by global solvers, proving that domain decomposition does not sacrifice accuracy for efficiency.

Figure 5: Wavefield contours for f=30 Hz on a 232x386 grid, illustrating the algorithm's ability to handle complex propagation patterns.

Conclusion and Future Directions

The research demonstrates that combining FDM with domain decomposition provides a scalable and robust approach for wave simulation. By splitting the global problem into smaller sub-problems, the method becomes viable for complex environments like the Marmousi model.

While currently limited to 2D simulations, the authors note that the next step is extending this work to 3D environments and integrating it into Full-Waveform Inversion (FWI), where the goal is to work backward from recorded seismic data to map the Earth's interior in high detail.

Summary updated

Abstract

In this paper, wave simulation with the finite difference method for the Helmholtz equation based on the domain decomposition method is investigated. The method solves the problem by iteratively solving subproblems defined on smaller subdomains. Two domain decomposition algorithms both for nonoverlapping and overlapping methods are described. More numerical computations including the benchmark Marmousi model show the effectiveness of the proposed algorithms. This method can be expected to be used in the full-waveform inversion in the future.
BibTeX
@article{zhang2013finite,
  author = {Wensheng Zhang and Yunyin Dai},
  title = {{Finite-Difference Solution of the Helmholtz Equation Based on Two Domain Decomposition Algorithms}},
  journal = {Journal of Applied Mathematics and Physics},
  year = {2013},
  doi = {10.4236/jamp.2013.14004},
}
Figure