Age hardening is widely used to strengthen commercial engineering alloys.

Direct experimental observation of precipitation is very challenging. Therefore, atomistic simulations with machine-learned interatomic potentials (ML-IPs) are invaluable to elucidating the atomistic mechanisms of solid-state phase transformations, including those occurring during age hardening.

Despite advancements, developing training datasets and validation methodologies for metallurgical systems remains challenging. ML-IPs still struggle to accurately model solid-state precipitation reactions due to their complex energy landscapes and small energy barriers.

This work demonstrates that carefully designed training datasets and novel error metrics beyond standard statistical measures are essential to reproduce solid-state phase transformations accurately.

We apply this approach to model precipitation in the binary magnesium-neodymium alloy. Atomistic simulations using the fitted ML-IP reveal previously inaccessible nucleation mechanisms and energy barriers. The results of this work provide critical insights to enable the tuning of processing conditions of commercial metallic alloys.