SpookyNet: Learning force fields with electronic degrees of freedom and nonlocal effects


Por: Unke O.T., Chmiela S., Gastegger M., Schütt K.T., Sauceda H.E., Müller K.-R.

Publicada: 1 ene 2021
Resumen:
Machine-learned force fields combine the accuracy of ab initio methods with the efficiency of conventional force fields. However, current machine-learned force fields typically ignore electronic degrees of freedom, such as the total charge or spin state, and assume chemical locality, which is problematic when molecules have inconsistent electronic states, or when nonlocal effects play a significant role. This work introduces SpookyNet, a deep neural network for constructing machine-learned force fields with explicit treatment of electronic degrees of freedom and nonlocality, modeled via self-attention in a transformer architecture. Chemically meaningful inductive biases and analytical corrections built into the network architecture allow it to properly model physical limits. SpookyNet improves upon the current state-of-the-art (or achieves similar performance) on popular quantum chemistry data sets. Notably, it is able to generalize across chemical and conformational space and can leverage the learned chemical insights, e.g. by predicting unknown spin states, thus helping to close a further important remaining gap for today’s machine learning models in quantum chemistry. © 2021, The Author(s).

Filiaciones:
Unke O.T.:
 Machine Learning Group, Technische Universität Berlin, Berlin, 10587, Germany

 DFG Cluster of Excellence “Unifying Systems in Catalysis” (UniSysCat), Technische Universität Berlin, Berlin, 10623, Germany

Chmiela S.:
 Machine Learning Group, Technische Universität Berlin, Berlin, 10587, Germany

Gastegger M.:
 Machine Learning Group, Technische Universität Berlin, Berlin, 10587, Germany

 DFG Cluster of Excellence “Unifying Systems in Catalysis” (UniSysCat), Technische Universität Berlin, Berlin, 10623, Germany

Schütt K.T.:
 Machine Learning Group, Technische Universität Berlin, Berlin, 10587, Germany

Sauceda H.E.:
 Machine Learning Group, Technische Universität Berlin, Berlin, 10587, Germany

 Baslearn, BASF-TU Joint Lab, Technische Universität Berlin, Berlin, 10587, Germany

Müller K.-R.:
 Machine Learning Group, Technische Universität Berlin, Berlin, 10587, Germany

 Department of Artificial Intelligence, Korea University, Anam-dong, Seongbuk-gu, Seoul, 02841, South Korea

 Max Planck Institute for Informatics, Stuhlsatzenhausweg, Saarbrücken, 66123, Germany

 BIFOLD—Berlin Institute for the Foundations of Learning and Data, Berlin, Germany

 Google Research, Brain team, Berlin, Germany
ISSN: 20411723
Editorial
NATURE PUBLISHING GROUP, MACMILLAN BUILDING, 4 CRINAN ST, LONDON N1 9XW, ENGLAND, Reino Unido
Tipo de documento: Article
Volumen: 12 Número: 1
Páginas:
WOS Id: 000730391400003
ID de PubMed: 34907176
imagen Gold