Molecular dynamics simulations of collision cascades in niobium: comparing interatomic potentials

Radiation damage in structural materials is a critical concern for advanced nuclear energy systems, and niobium is of interest due to its high melting point, mechanical strength, and corrosion resistance. In this work, large-scale molecular dynamics simulations of collision cascades in pure niobium were performed at 300 K over a primary knock-on atom (PKA) energy range of 1–75 keV using four interatomic potentials: an embedded atom method (EAM), two Finnis–Sinclair models (FS-1 and FS-2), and a machine-learning-based spectral neighbour analysis potential (SNAP) developed in our previous work. Previous studies have shown that SNAP and FS-2 correctly reproduce the stable 12⟨111⟩ self-interstitial atom (SIA) configuration, whereas EAM and FS-1 incorrectly predict the ⟨110⟩ configuration as the lowest-energy SIA state rather than the physically correct 12⟨111⟩ orientation. Building on this understanding, we examine how these differences influence radiation damage predictions. All potentials capture the general features of cascade formation, but clear differences are observed in defect production, clustering, and morphology. At low PKA energies, defect behaviour is mainly governed by the stiffness-to-range ratio (|S/R|), threshold displacement energy, and defect formation energies, while subcascade formation dominates at higher energies. EAM and FS-1 predict higher defect survival, whereas SNAP and FS-2 show similar and lower defect retention. Vacancy clustering dominates in all cases, with EAM exhibiting the highest clustering fraction. Morphological analysis reveals distinct defect populations across the potentials, highlighting the improved reliability of SNAP and FS-2 for radiation damage simulations in niobium.