Exploring the configuration spaces of surface materials using time-dependent diffraction patterns and unsupervised learning

Abstract

Computational methods for exploring the atomic configuration spaces of surface materials will lead to breakthroughs in nanotechnology and beyond. In order to develop such methods, especially ones utilizing machine learning approaches, descriptors which encode the structural features of the candidate configurations are required. In this paper, we propose the use of time-dependent electron diffraction simulations to create descriptors for the configurations of surface materials. Our proposal utilizes the fact that the sub-femtosecond time-dependence of electron diffraction patterns are highly sensitive to the arrangement of atoms in the surface region of the material, allowing one to distinguish configurations which possess identical symmetry but differ in the locations of the atoms in the unit cell. We demonstrate the effectiveness of this approach by considering the simple cases of copper(111) and an organic self-assembled monolayer system, and use it to search for metastable configurations of these materials.