boss_bench

The BoSS: a benchmark of stellarator simulations

The BoSS: a benchmark of stellarator simulations#

Download Data Download Models Data Explorer

The Benchmark of Stellarator Simulations (BoSS) is a database of stellarator configurations paired with physics simulations. It holds about 3 TB of data from ideal MHD and collisionless guiding-center simulations of roughly 180,000 stellarators. The database keeps growing: new configurations are added over time, and results from other kinds of simulation will be added as new sections. It can be used to develop machine-learning methods for fusion applications, or to improve our understanding of the underlying physics.

A burst of 400 alpha particles followed along their guiding-center orbits in one stellarator configuration. Orange marks show where lost particles strike the plasma boundary. The panel on the right counts the particles lost so far: the loss fraction is the share of the 400 particles that have been lost, and the loss rate is the number of particles lost per millisecond at each moment.

Installation#

The BoSS comes with a Python package for working with the data: reading the files, computing derived quantities such as loss fractions and heat fluxes, and evaluating models with the benchmark’s metrics. It needs Python 3.9 or newer. Install it with:

pip install boss-bench

From source#

git clone https://github.com/mishapadidar/boss_bench.git
cd boss_bench
pip install .

See the package documentation for how to use it.

Citing the BoSS#

If you find this project useful for your research, please consider citing:

  • M. Padidar, T. Huang, M. Goldstein, M. Czekanski, and J. Soules. Generative modeling for fusion: a large-scale benchmark of stellarator fast-ion simulations. 2026. arXiv:2610.37910.

The BibTeX entry is below.

@misc{padidar2026generativemodelingfusionlargescale,
      title={Generative modeling for fusion: a large-scale benchmark of stellarator fast-ion simulations},
      author={Misha Padidar and Teresa Huang and Mark Goldstein and Michael Czekanski and Jeffrey Soules},
      year={2026},
      eprint={2610.37910},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2610.37910},
}

License#

The data is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. The boss-bench package and the trained models are released under the Apache License 2.0. Copyright 2026 The Simons Foundation.

Contact#

For questions regarding this project, please contact Misha Padidar.

Acknowledgments#

We thank the Flatiron Institute’s Scientific Computing Core and the Simons Foundation.