High-Accuracy Molecular Simulations with Machine-Learning Potentials and Semiclassical Approximations to Quantum Dynamics

TL;DR

This paper presents a method combining machine-learning potentials with semiclassical quantum dynamics approximations to enable accurate, low-cost molecular simulations of chemical reactions, capturing tunnelling and anharmonic effects.

Contribution

It introduces transfer learning for efficient potential energy surface construction and demonstrates the integration with advanced semiclassical methods for improved quantum dynamics modeling.

Findings

Machine-learning potentials reduce computational cost significantly.

Transfer learning requires minimal high-level training data.

Semiclassical approximations accurately capture tunnelling and anharmonicity.

Abstract

Accurate simulations of molecules require high-level electronic-structure theory in combination with rigorous methods for approximating the quantum dynamics. Machine-learning approaches can significantly reduce the computational expense of this workflow without any loss of accuracy. We discuss various methods for constructing potential energy surfaces including transfer learning, which requires a minimal number of expensive training points. In this way, we can study chemical reactions at a high level but a low cost. In particular, as the potentials are smooth and differentiable, they enable the use of more advanced semiclassical approximations to quantum dynamics, such as perturbatively corrected instanton theory, which can capture both tunnelling and anharmonicity.