Resolving vacancy clustering mechanisms in tungsten via machine- learning-potential-based simulations at experimental time scales

G Wei and ZC Song and J Hou and JQ Wang and A Singh and L Li, ACTA MATERIALIA, 301, 121529 (2025).

DOI: 10.1016/j.actamat.2025.121529

Vacancy clustering drives swelling and embrittlement in metals, particularly under extreme environments. In tungsten (W), traditional cluster dynamics theory primarily attributes clustering to mono-vacancy (V1) migration, often overlooking the role of larger clusters. This oversight stems from the accuracy and timescale limitations of conventional computational methods. Here, we employ a machine-learning- potential-based off-lattice kinetic Monte Carlo method to model vacancy clustering in W with first-principles-level accuracy, achieving time scales on the order of seconds. Our simulations uncover the high mobility of tri-vacancy (V3) clusters, which diffuse at rates approximately three orders of magnitude higher than other clusters. This high mobility enables a V3-mediated clustering mechanism, Vn + V3 -> Vn+3, positioning V3 as an efficient vacancy carrier at elevated temperature. Additional object kinetic Monte Carlo simulations reveal the critical interplay between thermodynamics and kinetics during the early stages of stable vacancy cluster formation, highlighting the critical role of the mobility of V3 clusters. This mechanism becomes increasingly significant at dilute vacancy concentrations and reconciles discrepancies between theoretical models and experimental observations. These findings refine the cluster dynamics theory and offer critical insights into defect evolution in metals under extreme conditions.

Return to Publications page