This package implements the Allegro E(3)-equivariant machine learning interatomic potential.
In particular, allegro implements the Allegro model as an extension package for the NequIP framework.
- Installation
- Usage
- LAMMPS Integration
- References & citing
- Community, contact, questions, and contributing
Important
A major backwards-incompatible update to the nequip framework was released on April 23rd 2025 as version v0.7.0. The corresponding allegro version is v0.4.0. Previous versions of Allegro remain available if needed in the GitHub Releases and must be used with older versions of nequip.
allegro requires the nequip package. Details on nequip and its required PyTorch versions can be found in the nequip docs.
allegro can be installed from PyPI (note that it is known as nequip-allegro on PyPI):
pip install nequip-allegroInstalling allegro in this way will also install the nequip package.
The allegro package provides the Allegro model for use within the NequIP framework. The framework's documentation describes how to train, test, and use models.
A minimal example of a config file for training an Allegro model is provided at configs/tutorial.yaml.
We offer a LAMMPS plugin pair_allegro to use Allegro models in LAMMPS simulations, including support for Kokkos acceleration, MPI, and parallel multi-GPU simulations.
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:
-
The Allegro paper
Albert Musaelian, Simon Batzner, Anders Johansson, Lixin Sun, Cameron J. Owen, Mordechai Kornbluth, and Boris Kozinsky.
"Learning local equivariant representations for large-scale atomistic dynamics."
Nature Communications 14, no. 1 (2023): 579 -
The preprint describing the NequIP software framework and Allegro's performance within it
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."
https://doi.org/10.48550/arXiv.2504.16068 -
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.
And also consider citing:
-
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 -
The
e3nnequivariant neural network package used by NequIP, through its preprint and/or code
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 Disucssions.
If your post is related to the NequIP software framework in general, please post in the issues or discussions on that repository. Discussions on this repository should be specific to the allegro package and Allegro model.
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. Register for the NequIP community here.
If you want to contribute to the code, please read CONTRIBUTING.md from the nequip repository; this repository follows the same processes.
We can also be reached by email at [email protected].