LES-BEC

This repository contains the data and scripts used in the study Machine learning interatomic potential can infer electrical response.

Specifically, it provides the training datasets, training scripts, BEC inference scripts, molecular dynamics (MD) scripts with and without external electric fields, and trained CACE potentials.

Implementation

In this work, the LES method was implemented in the CACE code by adding new modules: ewald, polarization, grad, and dephase. Please check the cace/modules/ for detailed implementations on CACE.

For a detailed explanation of the LES method, please refer to Latent Ewald summation for machine learning of long-range interactions, and A Universal Augmentation Framework for Long-Range Electrostatics in Machine Learning Interatomic Potentials, which describes the LES library. This library is already integrated with a broad range of MLIPs, including MACE, NequIP, Allegro, CACE, and CHGNet.

Example scripts

Each example folder - water, superionic_water, and PbTiO3 - includes its own README that explains the scripts available in that directory.

• Training scripts can be found in each data repository (e.g., water/fit).
• BEC inference scripts are available at superionic_water/eval_bec.
• Example MD scripts, with or without an external electric field, can be found in each dataset folder (e.g., ferro_PbTiO3/md).
• Scripts for calculating the polarization current J(t) are provided in superionic_water/md and water/md.

License

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

Citation

If you use this data in your academic work, please cite:

@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}
}

And also consider citing:

1. Latent Ewald summation for machine learning of long-range interactions

2. Machine Learning of Charges and Long-Range Interactions from Energies and Forces

3. A Universal Augmentation Framework for Long-Range Electrostatics in Machine Learning Interatomic Potentials