TY - JOUR
T1 - How robust are modern graph neural network potentials in long and hot molecular dynamics simulations?
AU - Stocker, Sina
AU - Gasteiger, Johannes
AU - Becker, Florian
AU - Günnemann, Stephan
AU - Margraf, Johannes T.
N1 - Publisher Copyright:
© 2022 The Author(s). Published by IOP Publishing Ltd.
PY - 2022/12/1
Y1 - 2022/12/1
N2 - Graph neural networks (GNNs) have emerged as a powerful machine learning approach for the prediction of molecular properties. In particular, recently proposed advanced GNN models promise quantum chemical accuracy at a fraction of the computational cost. While the capabilities of such advanced GNNs have been extensively demonstrated on benchmark datasets, there have been few applications in real atomistic simulations. Here, we therefore put the robustness of GNN interatomic potentials to the test, using the recently proposed GemNet architecture as a testbed. Models are trained on the QM7-x database of organic molecules and used to perform extensive molecular dynamics simulations. We find that low test set errors are not sufficient for obtaining stable dynamics and that severe pathologies sometimes only become apparent after hundreds of ps of dynamics. Nonetheless, highly stable and transferable GemNet potentials can be obtained with sufficiently large training sets.
AB - Graph neural networks (GNNs) have emerged as a powerful machine learning approach for the prediction of molecular properties. In particular, recently proposed advanced GNN models promise quantum chemical accuracy at a fraction of the computational cost. While the capabilities of such advanced GNNs have been extensively demonstrated on benchmark datasets, there have been few applications in real atomistic simulations. Here, we therefore put the robustness of GNN interatomic potentials to the test, using the recently proposed GemNet architecture as a testbed. Models are trained on the QM7-x database of organic molecules and used to perform extensive molecular dynamics simulations. We find that low test set errors are not sufficient for obtaining stable dynamics and that severe pathologies sometimes only become apparent after hundreds of ps of dynamics. Nonetheless, highly stable and transferable GemNet potentials can be obtained with sufficiently large training sets.
KW - graph neural networks
KW - interatomic potentials
KW - molecular dynamics
UR - http://www.scopus.com/inward/record.url?scp=85141712545&partnerID=8YFLogxK
U2 - 10.1088/2632-2153/ac9955
DO - 10.1088/2632-2153/ac9955
M3 - Article
AN - SCOPUS:85141712545
SN - 2632-2153
VL - 3
JO - Machine Learning: Science and Technology
JF - Machine Learning: Science and Technology
IS - 4
M1 - 045010
ER -