Search papers, labs, and topics across Lattice.
The paper explores the use of machine learning potentials (MLPs), specifically leveraging transfer learning, to reduce the computational cost of high-accuracy molecular simulations. These MLPs, built with minimal training data, are then used in conjunction with advanced semiclassical approximations to quantum dynamics, such as perturbatively corrected instanton theory. The key result is a computationally efficient approach for studying chemical reactions that accurately captures tunneling and anharmonicity effects.
Accurate molecular simulations, previously limited by computational cost, are now achievable thanks to machine-learning potentials that enable advanced semiclassical approximations.
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.
