Magnesium–zinc (Mg–Zn) alloys derive much of their mechanical strength from the formation and evolution of intermetallic compounds. However, the Mg-rich and intermediate compositional regimes remain insufficiently resolved structurally, hindering a comprehensive and phase-consistent understanding of phase stability and its role in microstructural strengthening. Here, we develop a high-accuracy neuroevolution potential (NEP) for the Mg–Zn system by integrating an active-learning strategy with the CALYPSO structure-search method, enabling systematic exploration of complex configurational phase space with near-first-principles accuracy. Extensive NEP-assisted structure searches reproduce all experimentally established Mg–Zn intermetallic phases and predict a previously unreported thermodynamically stable Mg3Zn2 phase with P42/mnm symmetry, together with a metastable I4/mcm-Mg2Zn phase. Mg3Zn2 is a plausible candidate constituent phase, as a selected simulated X-ray diffraction feature is consistent with an unassigned peak previously reported in multiphase Mg–Zn alloys. Additionally, Mg3Zn2 exhibits superior thermal robustness, retaining higher strength and hardness at elevated temperatures compared with the widely studied MgZn2. These results demonstrate the effectiveness of NEP-assisted structure prediction in resolving hidden intermetallic phases and establishing systematic structure–property relationships in chemically complex alloy systems.
Li et al. (Fri,) studied this question.