A new radial basis function collocation method based on the quasi-uniform nodes for 2D fractional wave equation

Yiran Xu, Jingye Li, Guofei Pang, Zhikai Wang, Xiaohong Chen, Benfeng WangarXiv 2018 · DOI · arXiv

Abstract

This paper presents a novel time-space domain radial basis function (RBF) collocation method designed to solve the decoupled fractional Laplacian wave equation. This equation is essential for describing seismic wave propagation in attenuating media, where energy loss and dispersion occur. By introducing a quasi-uniform node configuration scheme, the method achieves high computational efficiency and flexibility in mesh-free discretization. Comparative analysis with traditional pseudo-spectral methods demonstrates that the quasi-uniform RBF approach provides superior performance, particularly as the computational domain increases in size and complexity. In complex geological models, the proposed method achieves comparable accuracy to standard techniques while requiring only half the number of computational points.

Introduction

In the field of geophysics, accurately modeling seismic wave propagation is fundamental to subsurface imaging and characterization. However, the Earth’s subsurface is not a perfectly elastic medium; it exhibits visco-acoustic properties where waves undergo attenuation and dispersion. These phenomena are mathematically captured using fractional wave equations, which utilize fractional-order derivatives to account for the "memory" and non-local effects inherent in seismic attenuation.

Solving these fractional equations presents significant numerical challenges. Traditional Finite Difference Methods (FDM) are often constrained by regular grids, making them ill-suited for irregular boundaries like the seafloor. Furthermore, the non-local nature of fractional derivatives causes FDMs to lose their typical computational efficiency. While pseudo-spectral methods offer high accuracy, they are prone to Gibbs phenomena at boundaries and face scaling issues in three-dimensional applications. This research addresses these limitations by proposing a mesh-free, RBF-based collocation method utilizing a quasi-uniform node distribution.

Methodology

The proposed framework integrates three key technical components to optimize the simulation of the 2D fractional wave equation:

1. RBF Collocation

The method employs Radial Basis Functions (specifically Multiquadrics) to approximate the wavefield. Unlike grid-based methods, RBF collocation is mesh-free, relying solely on the Euclidean distance between nodes. This allows for seamless discretization of irregular geometries and complex geological structures without the overhead of complex mesh generation.

2. Quasi-Uniform Node Configuration

To enhance computational efficiency, the authors implemented an "advancing front" algorithm to generate a quasi-uniform node layout. Unlike a dense rectangular grid, this scheme places nodes strategically to capture the underlying physics while avoiding redundant points. This distribution maintains the flexibility of mesh-free methods while optimizing the number of degrees of freedom.

3. Directional Fractional Laplacian

To handle the fractional Laplacian operator, the study adopts a directional fractional derivative definition. This approach is more computationally tractable for RBFs than standard Riesz potential definitions, particularly when dealing with boundary conditions in a bounded domain.

Experimental Evaluation

The performance of the new method was validated against three benchmark models:

  • 2D Homogeneous Model: Used to establish baseline accuracy and compare CPU execution times against uniform grid schemes.
  • Partial SEG Hess Model: A complex model featuring a salt body, testing the method's robustness against sharp velocity contrasts.
  • 2D Modified Marmousi Model: A highly complex, multi-layered geological model with numerous faults, representing a realistic seismic imaging scenario.

Results and Discussion

The experimental results highlight significant advantages in both efficiency and accuracy:

  • Computational Efficiency: For a 5.0km x 5.0km domain, the quasi-uniform scheme required only approximately 11,000 nodes, compared to the 251,000 nodes required by a standard uniform grid. This reduction led to a substantial decrease in CPU time, nearly halving the processing requirements in specific test cases (e.g., from 63 seconds to 32 seconds).
  • Point Reduction in Complex Media: In the Marmousi model simulation, the RBF method with quasi-uniform nodes achieved the same level of accuracy as the pseudo-spectral method while utilizing roughly 50% fewer computational points.
  • Accuracy Maintenance: Relative error analysis confirmed that the reduction in node density did not compromise precision. In fact, the quasi-uniform scheme consistently maintained lower relative errors than uniform grid configurations across various model scales.

Conclusion

The RBF collocation method based on quasi-uniform nodes offers a powerful alternative for solving fractional wave equations in geophysics. Its mesh-free nature provides the flexibility needed for complex geological models, while the quasi-uniform node scheme ensures computational efficiency by minimizing redundant calculations.

Future research will focus on extending this framework to handle spatially dependent Quality factors (variable fractional-orders) to better represent heterogeneous rock layers. Additionally, optimizing the inversion of large RBF matrices for 3D scaling and integrating advanced boundary conditions, such as Perfectly Matched Layers (PML), remain key areas for further development.

Summary updated

Abstract

We mainly concerned with a decoupled fractional Laplacian wave equation in this paper. A new time-space domain radial basis function (RBF) collocation method is introduced to solve the fractional wave equation, which describes seismic wave propagation in attenuation media. The directional fractional Laplacian is adopted to cope with the fractional Laplacian of RBFs. In order to increase the computational efficiency, we introduced the quasi-uniform nodes configuration scheme, which is suitable for mesh-free discretization of wave equations. The comparison between the new method and the commonly-used pseudo-spectral method are implemented on square homogeneous models with different model size. The CPU time and relative errors of different methods show that the quasi-uniform configuration scheme provides better performance and the calculation efficiency advantage is significantly prominent as the computation domain increases. The relative errors indicate that the RBF collocation method with quasi-uniform configuration could improve the computational efficiency effectively and provide satisfactory accuracy. This advantage was especially highlighted in complex models, where the new approach achieved the same accuracy with only a half number of points. The implementation on the 2D complex model further demonstrated the accuracy, efficiency, and flexibility of the proposed new method.
BibTeX
@misc{xu2018new,
  author = {Yiran Xu and Jingye Li and Guofei Pang and Zhikai Wang and Xiaohong Chen and Benfeng Wang},
  title = {{A new radial basis function collocation method based on the quasi-uniform nodes for 2D fractional wave equation}},
  year = {2018},
  eprint = {1806.01846},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1806.01846},
}
Figure