Machine learning interatomic potentials for solid-state precipitation

Machine learning interatomic potentials (MLIPs) are routinely used to model diverse atomistic phenomena, yet parameterizing them to accurately capture solid-state phase transformations remains difficult. We present error metrics and data-generation schemes designed to streamline the parameterization of MLIPs for modeling precipitation in multi-component alloys. We developed an algorithm that enumerates symmetrically distinct transformation pathways connecting chemical decorations on different parent crystal structures. Additionally, we introduce the weighted Kendall-$\tau$ coefficient and its semi-grand canonical generalization as metrics for quantifying MLIP accuracy in predicting low-temperature thermodynamics. We apply these approaches to parameterize an MLIP for a dilute Mg-Nd alloy. The resulting potential reproduces the complex early-stage precipitation behavior observed in experiment. Large-scale atomistic simulations reveal competition between order-disorder and structural transformations. Furthermore, these results suggest a continuous transition between high-symmetry hcp and bcc crystal structures during aging heat treatments.