Constructing grids for molecular quantum dynamics using an autoencoder

TL;DR

This paper introduces a machine learning method using an autoencoder to reduce the dimensionality of molecular configurations, enabling efficient quantum dynamics calculations on a low-dimensional grid.

Contribution

It presents a novel autoencoder-based approach to identify relevant coordinates for molecular quantum dynamics, simplifying grid construction for complex systems.

Findings

Successfully applied to a test system

Generated a low-dimensional potential energy surface

Facilitated quantum dynamics calculations

Abstract

A challenge for molecular quantum dynamics (QD) calculations is the curse of dimensionality with respect to the nuclear degrees of freedom. A common approach that works especially well for fast reactive processes is to reduce the dimensionality of the system to a few most relevant coordinates. Identifying these can become a very difficult task, since they often are highly unintuitive. We present a machine learning approach that utilizes an autoencoder that is trained to find a low-dimensional representation of a set of molecular configurations. These configurations are generated by trajectory calculations performed on the reactive molecular systems of interest. The resulting low-dimensional representation can be used to generate a potential energy surface grid in the desired subspace. Using the G-matrix formalism to calculate the kinetic energy operator, QD calculations can be carried out on this grid. In addition to step-by-step instructions for the grid construction, we present the application to a test system.