The Latent Ewald Summation (LES) library is a plug-in to add long-range interactions to short-ranged machine learning interatomic potentials.

• Python 3.6 or higher
• NumPy
• PyTorch

Please refer to the setup.py file for installation instructions.

les can be installed using pip

git clone https://github.com/ChengUCB/les.git
pip install -e . 

We present LES (Latent Ewald Summation) (https://github.com/ChengUCB/les) as a plug-in library designed to add long-range interactions to short-range machine learning interatomic potentials (MLIPs).

Here we demonstrate its integration with MLIPs such as MACE, NequIP, Allegro, CACE, and CHGNet, and provide training scripts and trained models. In particular, we provide MACELES-OFF trained on the SPICE dataset.

Here you can find MLIP packages with LES implementation presented in A Universal Augmentation Framework for Long-Range Electrostatics in Machine Learning Interatomic Potentials.

Example training scripts for these LES-augmented MLIPs can be found in [https://github.com/ChengUCB/les_fit].

Hyperparameters selection: The default parameters (i.e. without setting anything) usually work well. One thing that can be changed is 'remove_self_interaction'. Setting 'remove_self_interaction=True' is the default and is the most robust choice. 'remove_self_interaction=False' can sometimes yield a bit better training accuracy, but is less robust when training on finite systems and then extrapolate to periodic systems.

[2025-10] The MACELES model has been implemented in the main MACE repository. Example training and evaluation scripts are available in les_fit.

This project is licensed under the CC BY-NC 4.0 License.

@article{cheng2025latent,
  title={Latent Ewald summation for machine learning of long-range interactions},
  author={Cheng, Bingqing},
  journal={npj Computational Materials},
  volume={11},
  number={1},
  pages={80},
  year={2025},
  publisher={Nature Publishing Group UK London}
}


@article{King2025Machine,
  title = {Machine Learning of Charges and Long-Range Interactions from Energies and Forces},
  author = {King, Daniel S. and Kim, Dongjin and Zhong, Peichen and Cheng, Bingqing},
  year = 2025,
  journal = {Nature Communications},
  volume = {16},
  number = {1},
  pages = {8763},
  publisher = {Nature Publishing Group}
}



@article{zhong2025machine,
  title={Machine learning interatomic potential can infer electrical response},
  author={Zhong, Peichen and Kim, Dongjin and King, Daniel S and Cheng, Bingqing},
  journal={arXiv preprint arXiv:2504.05169},
  year={2025}
}

@article{Kim2025Universalb,
  title = {A Universal Augmentation Framework for Long-Range Electrostatics in Machine Learning Interatomic Potentials},
  author = {Kim, Dongjin and Wang, Xiaoyu and Vargas, Santiago and Zhong, Peichen and King, Daniel S. and Inizan, Theo Jaffrelot and Cheng, Bingqing},
  year = 2025,
  journal = {Journal of Chemical Theory and Computation},
  publisher = {American Chemical Society},
  doi = {10.1021/acs.jctc.5c01400}
}

For any queries regarding LES, please contact Bingqing Cheng at tonicbq@gmail.com.