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).
This repository contains the data and scripts used in the study A Universal Augmentation Framework for Long-Range Electrostatics in Machine Learning Interatomic Potentials. 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 model trained on the SPICE dataset using the the MACE package developed by the Cheng group, while the converted MACELES-OFF model is compatible with the current MACE main branch.
Here you can find MLIP packages with LES implementation.
| Package | Link |
|---|---|
| CACE | github.com/BingqingCheng/cace |
| MACE | github.com/ChengUCB/mace |
| MACE(updated) | github.com/ACEsuit/mace |
| NequIP | github.com/ChengUCB/NequIP-LES |
| Allegro | github.com/ChengUCB/NequIP-LES |
| MatGL | github.com/ChengUCB/matgl |
[2025-10] The MACELES model has been implemented in the main MACE repository. Example training and evaluation scripts are available in ./MLIPs/MACE-LES-new.
[2025-12] For the MACE main branch, one needs to use the converted MACELES-OFF model (MD5: cc10e937b55e09f05b16dba756e2311b).
Please refer to the specific folder for related scripts and trained MLIPs.
This project is licensed under the CC BY-NC 4.0 License.
@article{Kim2025Universal,
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}
}
@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},
journal = {Nature Communications},
volume = {16},
number = {1},
pages = {8763},
year = {2025},
publisher = {Nature Publishing Group},
doi = {10.1038/s41467-025-63852-x},
}
@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}
}
For any queries regarding LES, please contact Bingqing Cheng at tonicbq@gmail.com.