Optimizing the architecture of Behler-Parrinello neural network potentials

Author(s)
Lukáš Kývala, Christoph Dellago
Abstract

The architecture of neural network potentials is typically optimized at the beginning of the training process and remains unchanged throughout. Here, we investigate the accuracy of Behler-Parrinello neural network potentials for varying training set sizes. Using the QM9 and 3BPA datasets, we show that adjusting the network architecture according to the training set size improves the accuracy significantly. We demonstrate that both an insufficient and an excessive number of fitting parameters can have a detrimental impact on the accuracy of the neural network potential. Furthermore, we investigate the influences of descriptor complexity, neural network depth, and activation function on the model’s performance. We find that for the neural network potentials studied here, two hidden layers yield the best accuracy and that unbounded activation functions outperform bounded ones.

Organisation(s)
Computational and Soft Matter Physics
Journal
Journal of Chemical Physics
Volume
159
No. of pages
8
ISSN
0021-9606
DOI
https://doi.org/10.1063/5.0167260
Publication date
09-2023
Peer reviewed
Yes
Austrian Fields of Science 2012
102019 Machine learning, 103043 Computational physics, 103029 Statistical physics
ASJC Scopus subject areas
General Physics and Astronomy, Physical and Theoretical Chemistry
Portal url
https://ucrisportal.univie.ac.at/en/publications/a14cd157-f5fc-48bb-9314-543b34703f41