NequIP

NequIP is an open-source code for building E(3)-equivariant interatomic potentials.

• Installation and usage
• Tutorial
• Foundation potentials & pre-trained models
• Highlighted Features
• Extension Packages
• References & citing
• Authors
• Community, contact, questions, and contributing

[!IMPORTANT] A major backwards-incompatible update to the nequip package was released on April 23rd 2025 as version v0.7.0. The previous version v0.6.2 can still be found for use with existing config files in the GitHub Releases and on PyPI.

Installation and usage

Installation instructions and user guides can be found in our docs.

Tutorial

The best way to learn how to use NequIP is through the tutorial notebook. This will run entirely on Google Colab's cloud virtual machine; you do not need to install or run anything locally.

Foundation potentials & pre-trained models

The NequIP framework provides a family of foundation potentials — pre-trained, wide-purpose interatomic potentials covering most of the periodic table — which can be used directly for production simulations or fine-tuned on your own data. These are hosted at nequip.net and described in "Fast and Accurate Foundation Models for Equivariant Machine-Learned Interatomic Potentials".

See the foundation potentials docs for how to download, compile, and run them, and the fine-tuning docs for adapting them to your own dataset, with the citation information below.

Highlighted Features

The following are some notable features, with quick links for more details:

• Compiled training and compiled inference
• Multi-GPU training
• GPU kernel accelerations with OpenEquivariance and CuEquivariance (alpha)
• ASE calculator integration and LAMMPS integrations through the pair styles in pair_nequip_allegro and our LAMMPS ML-IAP integration.

Extension Packages

The NequIP software framework is designed to be flexible and extensible: you can build custom architectures, implement new training techniques, and develop additional methods on top of it through extension packages. If you're interested in developing your own extension package, please refer to the extension package docs and consider joining our Zulip for developer-focused discussions and collaborations.

A notable example of a NequIP framework extension package is the allegro package that implements the strictly local equivariant interatomic potential architecture, Allegro. More extension packages can be found at https://www.nequip.net/extensions.

References & citing

Any and all use of this software, in whole or in part, should clearly acknowledge and link to this repository.

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

1. The paper describing the NequIP software framework:

   Chuin Wei Tan, Marc L. Descoteaux, Mit Kotak, Gabriel de Miranda Nascimento, Seán R. Kavanagh, Laura Zichi, Menghang Wang, Aadit Saluja, Yizhong R. Hu, Tess Smidt, Anders Johansson, William C. Witt, Boris Kozinsky, Albert Musaelian.
   "High-performance training and inference for deep equivariant interatomic potentials."
   Digital Discovery, 2026, Advance Article.
   https://doi.org/10.1039/D5DD00423C

2. The NequIP/Allegro foundation potentials paper, if you use any of the foundation potentials (from nequip.net, matbench-discovery or otherwise):

   Seán R. Kavanagh, Chuin Wei Tan, Menghang Wang, Marc L. Descoteaux, Gabriel de Miranda Nascimento, Ulrik Unneberg, Laura Zichi, Francesco Libbi, Norma Rivano, Austin Glover, Vivek Bharadwaj, Anders Johansson, William C. Witt, Albert Musaelian, Boris Kozinsky.
   "Fast and Accurate Foundation Models for Equivariant Machine-Learned Interatomic Potentials."
   arXiv:2607.28461 (2026).
   https://doi.org/10.48550/arXiv.2607.28461

And also consider citing:

3. The original NequIP paper

   Simon Batzner, Albert Musaelian, Lixin Sun, Mario Geiger, Jonathan P. Mailoa, Mordechai Kornbluth, Nicola Molinari, Tess E. Smidt, and Boris Kozinsky.
   "E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials."
   Nature Communications 13, no. 1 (2022): 2453

4. The computational scaling paper that discusses optimized LAMMPS MD

   Albert Musaelian, Anders Johansson, Simon Batzner, and Boris Kozinsky.
   "Scaling the leading accuracy of deep equivariant models to biomolecular simulations of realistic size."
   In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, pp. 1-12. 2023.

5. The e3nn equivariant neural network package used by NequIP, through its preprint and/or code

Extension packages like Allegro have their own additional relevant citations.

BibTeX entries for a number of the relevant papers are provided for convenience in CITATION.bib.

Authors

Please see AUTHORS.md.

Community, contact, questions, and contributing

If you find a bug or have a proposal for a feature, please post it in the Issues. If you have a self-contained question or other discussion topic, try our GitHub Discussions.

Active users and interested developers are invited to join us on the NequIP community chat server, which is hosted on the excellent Zulip software. Zulip is organized a little bit differently than chat software like Slack or Discord that you may be familiar with: please review their introduction before posting. Fill out the interest form for the NequIP community here.

If you want to contribute to the code, please read "Contributing to NequIP".

We can also be reached by email at allegro-nequip@g.harvard.edu.