Abstract

Predictive molecular simulations require fast, accurate and reactive interatomic potentials. Machine learning offers a promising approach to construct such potentials by fitting energies and forces to high-level quantum-mechanical data, but doing so typically requires considerable human intervention and data volume.

In this talk, I will present our efforts to tackle these challenges by introducing automation and machine-learned potentials to study reactions mechanisms in the condensed phase, We have recently shown that by leveraging hierarchical and active learning, Gaussian Approximation Potential (GAP) models can be developed for diverse chemical systems in an efficient manner, requiring only hundreds to a few thousand energy and gradient evaluations on a reference potential energy surface. The approach uses separate intra- and inter-molecular fits and active learning to maximise a prospective error metric used to quantify accuracy. We have applied this strategy to study a range of molecular systems: from bulk solvents to chemical reactivity, including a bifurcating Diels–Alder reaction in the gas phase and non-equilibrium dynamics (SN2 reaction) in explicit solvent. While promising, many challenges remain which need to be addressed in order to expand the applicability of this strategy to increasingly complex systems.