Quantifying disorder one atom at a time using an interpretable graph neural network paradigm.

Abstract in brief

• This research introduces SODAS, a new, interpretable metric using graph neural networks to measure local atomic disorder in materials, capturing how atomic configurations vary from solid to liquid states.
• The method is applied to different examples like grain boundaries, interfaces, microstructures, and fracture, demonstrating its ability to track how disorder evolves over time and space in materials.
• By comparing SODAS with traditional methods, the study shows how it can help predict material behavior and failure, providing a clearer link between atomic-level details and large-scale material properties.]