AuCol is a python package for learning collective variables and latent representations of chemical systems employing pre-trained or fixed atomic representations.
The package currently builds on top of the Schnetpack package. Two possible approaches to generate atomic representations are currently:
- Atomic centered symmetric functions (ACSF) for smaller systems with no NNPs [1]
- Pre-trained Schnet network if they are available [2]
The metadynamics package is currently linked to a modified version of the PLUMED library using a modified version of the ASE PLUMED calculator.
The main uses of the package are the following:
- Create an automatic collective variable learning workflow, possibly employing iterative learning procedure
- Use collective variables defined as neural network in PLUMED library. Send CV value and gradient using PLUMED cmd interface.
Example use of the library is demonstrated on a set of examples. For some a neural network potential is needed, for the main simple case ACSF are employed. The example can be downloaded from Google drive
If you use this library please cite:
@misc{https://doi.org/10.48550/arxiv.2203.08097,
doi = {10.48550/ARXIV.2203.08097},
url = {https://arxiv.org/abs/2203.08097},
author = {Šípka, Martin and Erlebach, Andreas and Grajciar, Lukáš}
keywords = {Chemical Physics (physics.chem-ph), FOS: Physical sciences, FOS: Physical sciences},
title = {Understanding chemical reactions via variational autoencoder and atomic representations},
publisher = {arXiv},
[1] Behler, J. Atom-centered symmetry functions for construct- ing high-dimensional neural network potentials. The Journal of Chemical Physics, 134(7):074106, 2011. doi: 10.1063/1.3553717. URL https://doi.org/10. 1063/1.3553717.
[2] Schütt, K. T., Kessel, P., Gastegger, M., Nicoli, K. A., Tkatchenko, A., and Müller, K.-R. Schnetpack: A deep learning toolbox for atomistic systems. Journal of Chemical Theory and Computation, 15(1):448–455,