symmetrix — a package for functions equivariant under:
translation, rotation, inversion, and exchange of particles.
See the symmetrix README to build and use the Python package.
See the pair_symmetrix README for use from LAMMPS.
The earliest symmetrix results are reported in:
- D. P. Kovács, J. H. Moore, N. J. Browning, I. Batatia, J. T. Horton, Y. Pu, V. Kapil, W. C. Witt, I.-B. Magdău, D. J. Cole, G. Csányi, "MACE-OFF: Short-Range Transferable Machine Learning Force Fields for Organic Molecules", Journal of the American Chemical Society 147, 17598 (2025). [arxiv] [journal]
Subsequent performance improvements are reported in:
- I. Batatia, P. Benner, Y. Chiang, A. M. Elena, D. P. Kovács, J. Riebesell, ...+78 others..., W. C. Witt, T. Wolf, F. Zills, G. Csányi, "A foundation model for atomistic materials chemistry," Journal of Chemical Physics 163, 184110 (2025). [arxiv] [journal]
Please cite these papers when using symmetrix.
The default license for this project is the MIT License.
The pair_symmetrix subdirectory, which enables integration with LAMMPS,
is licensed under the GNU General Public License (GPLv2)
to maintain consistency with LAMMPS.
An early phase of this project, leading to the Kokkos-based MACE implementation, was supported by the Schmidt Sciences Virtual Institute for Scientific Software (VISS). This engagment involved key contributions from Dave Brownell and Ketan Bhardwaj of the Center for Scientific and Software Engineering at Georgia Tech.