A data-centric framework for crystal structure identification in atomistic simulations using machine learning

Abstract

Atomic-level modeling performed at large scales enables the investigation of mesoscale materials properties with atom-by-atom resolution. The spatial complexity of such cross-scale simulations renders them unsuitable for simple human visual inspection. Instead, specialized structure characterization techniques are required to aid interpretation. These have historically been challenging to construct, requiring significant intuition and effort. Here we propose an alternative data-centric framework for a fundamental characterization task: classifying atoms according to the crystal structure to which they belong. A group of data-science tools are employed together to make unbiased decisions based on a collection of simple local descriptors of atomic structure. We present a rigorous statistical comparison to the performance of state-of-the-art methods and it is discovered that our data-centric approach performs at least as well as the most popular rule-based methods while being superior in many cases and presenting better generalizability to new structures and chemistries.