A Fast Fourier-Galerkin Method for Solving Boundary Integral Equations on Torus-Shaped Surfaces

Yiying Fang, Ying Jiang, Jiafeng SuarXiv 2024 · DOI · arXiv

Introduction

Boundary Integral Equation (BIE) methods are powerful tools for solving partial differential equations because they require discretizing only the surface of an object rather than its entire volume. However, for non-axisymmetric toroidal surfaces—complex, "lumpy" donut shapes—these methods typically generate dense $O(N^2)$ matrices, creating a significant computational bottleneck. This research introduces a fast Fourier-Galerkin method that leverages the inherent mathematical decay of the integral operator's kernel to sparsify these matrices to $O(N \ln^2 N)$ complexity without sacrificing accuracy.

Methodology: Exploiting Kernel Decay

The core of the proposed method is a strategic truncation of the Fourier-Galerkin matrix. By representing the solution using bi-periodic Fourier bases, the researchers analyzed the interaction between different Fourier modes.

3D visualization of the matrix entries showing rapid exponential decay from the center.

The "Aha" moment came from discovering that while the kernels for 3D BIEs are not square-integrable, they are Lebesgue integrable. By applying a shear transformation and identifying a uniform analytical continuation radius, the authors proved that matrix entries decay exponentially as they move away from the diagonal. This allows for a truncation rule where only entries with a logarithmic relationship to the total degrees of freedom ($N$) are retained, effectively zeroing out negligible interactions.

Key Results: Efficiency and Stability

The method was tested on complex geometries like "cruller" surfaces. The results demonstrate significant compression while maintaining high-order convergence and numerical stability.

The sparsity pattern of the truncated matrix, illustrating the $O(N \ln^2 N)$ structure.

Key findings include:

  • Compression Efficiency: In numerical experiments on a "cruller" surface with $N \approx 14,641$, the compression ratio reached 0.0868. This means the solver required only about 8.7% of the original matrix entries to reach the solution.
  • Accuracy: The method maintained a theoretical convergence order of approximately 3 ($O(N^{-p/2} \ln N)$), matching the results of the much slower, non-truncated version.
  • Stability: The condition numbers of the matrices remained low and stable (around 3.7 to 4.0), ensuring that the numerical solution is not overly sensitive to small errors.

Conclusion

This fast Fourier-Galerkin method provides a scalable path for solving BIEs on complex toroidal geometries, such as those found in stellarator fusion reactors or complex biological structures. By reducing the computational complexity from $O(N^2)$ to $O(N \ln^2 N)$, the approach enables high-precision modeling of non-axisymmetric shapes that were previously computationally prohibitive. Future work will focus on optimizing numerical quadrature for the truncated entries and extending the approach to other topologies.

Summary updated

BibTeX
@misc{fang2024fast,
  author = {Yiying Fang and Ying Jiang and Jiafeng Su},
  title = {{A Fast Fourier-Galerkin Method for Solving Boundary Integral Equations on Torus-Shaped Surfaces}},
  year = {2024},
  eprint = {2408.02199},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2408.02199},
}
Figure