Billions of Atoms with Machine Learning Interatomic Potentials: Application to Direct Heterogeneous Reactive Dynamics

Machine learning interatomic potentials (MLIPs) have become a prevalent approach to bridging the gap between slow-but-accurate ab initio calculations and fast-but-inaccurate empirical potentials for molecular dynamics. Among MLIPs, there is a pareto front of models with different tradeoffs between accuracy and speed. The FLARE interatomic potential aims to push the boundary of scalability and performance, while maintaining sufficient accuracy to study complex, reactive systems.

FLARE combines the atomic cluster expansion with a sparse Gaussian process. Bayesian uncertainties enable efficient training with active learning and uncertainty-aware, large-scale molecular dynamics simulations. We implement FLARE in LAMMPS with the Kokkos performance portability library, enabling efficient molecular dynamics simulations on GPUs across a wide range of system sizes. Using 27336 GPUs, we demonstrate state-of-the-art scaling and performance in micrometer-scale heterogeneous catalysis simulations with up to half a trillion atoms [1].

[1] arXiv:2204.12573