Boris Kozinsky1,2,Lixin Sun1,Simon Batzner1,Albert Musaelian1,Jonathan Vandermause1,Yu Xie1,Steven Torrisi1
Harvard University1,Bosch Research2
Boris Kozinsky1,2,Lixin Sun1,Simon Batzner1,Albert Musaelian1,Jonathan Vandermause1,Yu Xie1,Steven Torrisi1
Harvard University1,Bosch Research2
Accurate reaction rate prediction is crucial for designing highly efficient heterogeneous catalysis. Ab initio molecular dynamics can be used because it precisely models chemical reactions. But they often scale poorly and thus are prohibitive for long simulation time to accumulate sufficient statistics. On the other hand, enhanced sampling techniques can accelerate the simulation but require good collective variables, which can be hard to design for complex reactions.<br/>In this work, a data-driven method is used to address these two problems using a multitask learning framework [1] based on Neural Equivariant Interatomic Potentials (NequIP) [2]. The framework trains force fields with quantum chemical accuracy and discovers critical collective variables for highly efficient free energy landscape exploration. This learning framework is demonstrated on estimating the reaction free energy of formate dehydrogenation on a Cu(110) surface.<br/><br/>[1] L. Sun, J. Vandermause, S. Batzner, Y. Xie, D. Clark, W. Chen, B. Kozinsky, “Multitask machine learning of collective variables for enhanced sampling of rare events”, J. Chem. Theory Comput. 18, 2341 (2022)<br/>[2] S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. Mailoa, M. Kornbluth, N. Molinari, T. Smidt, B. Kozinsky, “E (3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials”, Nature Comm. 13, 1 (2022)