Autonomous Learning of Atomistic Structural Transitions via Physics-Inspired Graph Neural Networks

Materials processing often occurs under extreme dynamic conditions leading to a multitude of unique structural environments. These structural environments generally occur at high temperatures and/or high pressures, often under non-equilibrium conditions, which results in drastic changes in the material’s structure over time. Computational techniques such as molecular dynamics simulations can probe the atomic regime under these extreme conditions. However, characterizing the resulting atomistic structures has proved challenging due to the intrinsic levels of disorder present. Here, we introduce SODAS++, a universal and interpretable graph neural network framework that can accurately and intuitively quantify the transition between any two arbitrary phases. The SODAS++ framework also quantifies local atomic environments, providing one with the power to encode global state information at the atomic level. We showcase SODAS++ for both solid-solid and solid-liquid transitions for systems of increasing geometric and chemical complexity such as elemental metals, oxides, and ternary alloys.