Recommended by

Indexed by

Learning local equivariant representations for large-scale atomistic dynamics

Albert Musaelian1*, Simon Batzner1*, Anders Johansson1*, Lixin Sun1*, Cameron J. Owen1*, Mordechai Kornbluth2*, Boris Kozinsky1,2*

1 John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138, USA

2 Robert Bosch Research and Technology Center, Cambridge, MA 02139, USA

* Corresponding authors emails: albym@seas.harvard.edu, batzner@g.harvard.edu, andersjohansson@g.harvard.edu, lixinsun@g.harvard.edu, cowen@g.harvard.edu, Mordechai.Kornbluth@us.bosch.com, bkoz@seas.harvard.edu
DOI10.24435/materialscloud:fr-ts [version v1]

Publication date: Oct 17, 2022

How to cite this record

Albert Musaelian, Simon Batzner, Anders Johansson, Lixin Sun, Cameron J. Owen, Mordechai Kornbluth, Boris Kozinsky, Learning local equivariant representations for large-scale atomistic dynamics, Materials Cloud Archive 2022.128 (2022), doi: 10.24435/materialscloud:fr-ts.


A simultaneously accurate and computationally efficient parametrization of the energy and atomic forces of molecules and materials is a long-standing goal in the natural sciences. In pursuit of this goal, neural message passing has lead to a paradigm shift by describing many-body correlations of atoms through iteratively passing messages along an atomistic graph. This propagation of information, however, makes parallel computation difficult and limits the length scales that can be studied. Strictly local descriptor-based methods, on the other hand, can scale to large systems but do not currently match the high accuracy observed with message passing approaches. This work introduces Allegro, a strictly local equivariant deep learning interatomic potential that simultaneously exhibits excellent accuracy and scalability of parallel computation. Allegro learns many-body functions of atomic coordinates using a series of tensor products of learned equivariant representations, but without relying on message passing. Allegro obtains improvements over state-of-the-art methods on the QM9 and revised MD-17 data sets. A single tensor product layer is shown to outperform existing deep message passing neural networks and transformers on the QM9 benchmark. Furthermore, Allegro displays remarkable generalization to out-of-distribution data. Molecular dynamics simulations based on Allegro recover structural and kinetic properties of an amorphous phosphate electrolyte in excellent agreement with first principles calculations. Finally, we demonstrate the parallel scaling of Allegro with a dynamics simulation of 100 million atoms.

Materials Cloud sections using this data

No Explore or Discover sections associated with this archive record.


File name Size Description
8.4 MiB Ag data used
549 Bytes README
261.4 MiB zip file of Li3PO4 extxyz file, first 25k frames / 50ps is melt, 2nd 25k frames / 50ps is quench


Files and data are licensed under the terms of the following license: Creative Commons Attribution 4.0 International.
Metadata, except for email addresses, are licensed under the Creative Commons Attribution Share-Alike 4.0 International license.


machine learning molecular dynamics density-functional theory

Version history:

2022.128 (version v1) [This version] Oct 17, 2022 DOI10.24435/materialscloud:fr-ts