Abstract
Quantum mechanical methods like Density Functional Theory (DFT) are used withgreat success alongside efficient search algorithms for studying kinetics ofreactive systems. However, DFT is prohibitively expensive for large scaleexploration. Machine Learning (ML) models have turned out to be excellentemulators of small molecule DFT calculations and could possibly replace DFT insuch tasks. For kinetics, success relies primarily on the models capability toaccurately predict the Potential Energy Surface (PES) around transition-statesand Minimal Energy Paths (MEPs). Previously this has not been possible due toscarcity of relevant data in the literature. In this paper we train state ofthe art equivariant Graph Neural Network (GNN)-based models on around 10.000elementary reactions from the Transition1x dataset. We apply the models aspotentials for the Nudged Elastic Band (NEB) algorithm and achieve a MeanAverage Error (MAE) of 0.13+/-0.03 eV on barrier energies on unseen reactions.We compare the results against equivalent models trained on QM9 and ANI1x. Wealso compare with and outperform Density Functional based Tight Binding (DFTB)on both accuracy and computational resource. The implication is that ML models,given relevant data, are now at a level where they can be applied fordownstream tasks in quantum chemistry transcending prediction of simplemolecular features.
Original language | English |
---|---|
Article number | 045022 |
Journal | Machine Learning: Science and Technology |
Volume | 3 |
Issue number | 4 |
Number of pages | 13 |
ISSN | 2632-2153 |
DOIs | |
Publication status | Published - 2022 |
Keywords
- Reaction barriers
- Computational chemistry
- Transition state
- Reaction kinetics
- Density function theory (DFT)
- Nudged elastic band (NEB)
- Neural networks (NN)