v1.0.0

skelFMM: A Simplified Kernel-Independent Fast Multipole Method (FMM)

Release Notes

Introducing skelFMM, a novel kernel-independent fast multipole method (FMM) for efficient discrete convolution kernel evaluation. This lightweight implementation simplifies traditional FMMs and is optimized for modern hardware.

Key Features

• Simplified Data Structures: Operates on a near-neighbor list at every level of the tree instead of interaction lists.
• Kernel Independence: Supports the Laplace and low-frequency Helmholtz kernel functions and is extensible to more kernel functions. The methodology relies on low rank compression of far-field interactions and relies on kernel evaluations for the fast convolution.
• Parallel Efficiency: GPU-accelerated for modern hardware.
• Adaptive Tree Compatibility: Handles both uniform and non-uniform point distributions in 2D/3D.
• Precomputation Optimization: Uses tailored skeleton representations for efficient representations on surfaces.

Citation

If skelFMM aids your research, please cite:

@article{yesypenko2024simplified,
  title={A simplified fast multipole method based on strong recursive skeletonization},
  author={Yesypenko, Anna and Chen, Chao and Martinsson, Per-Gunnar},
  journal={Journal of Computational Physics},
  pages={113707},
  year={2024},
  publisher={Elsevier}
}

For more details and the source code, visit the [GitHub repository](https://github.com/annayesy/skelFMM).