AI 中文总结
本研究结合EXAFS谱与原子模拟探究hcp锌的晶格动力学，通过反向蒙特卡洛方法分析MSRD揭示其各向异性，发现微调CHGNet可改善模拟与实验的一致性，证实EXAFS分析能有效验证优化机器学习原子间势。
AI 中文摘要
本研究利用温度依赖的锌K边扩展X射线吸收精细结构（EXAFS）谱结合原子模拟，探究作为典型各向异性金属的六方最密堆积（hcp）锌的晶格动力学。通过反向蒙特卡洛方法提取了8个配位壳层的均方相对位移（MSRD），得到了壳层分辨的热运动描述。利用相关爱因斯坦模型分析MSRD的温度依赖性，得到了有效原子间力常数，并揭示了面内与面外相互作用间显著的各向异性；这种各向异性通过第一和第二配位壳层的MSRD比值进一步量化，与衍射实验得到的各向异性位移参数高度吻合。采用CHGNet通用机器学习原子间势的分子动力学模拟显示，原始模型高估了热无序，而微调后的版本大幅提升了与实验EXAFS谱及径向分布函数的一致性。总体而言，基于EXAFS的分析可有效验证和优化机器学习原子间势。
英文摘要
The lattice dynamics of hexagonal close-packed (hcp) zinc, a prototypical anisotropic metal, is studied using temperature-dependent Zn K-edge extended X-ray absorption fine structure (EXAFS) spectroscopy combined with atomistic simulations. The reverse Monte Carlo method enable the extraction of mean-square relative displacements (MSRDs) for eight coordination shells, providing a shell-resolved description of thermal motion. The MSRD temperature dependence, analysed using the correlated Einstein model, yields effective interatomic force constants and reveals pronounced anisotropy between in-plane and out-of-plane interactions. This anisotropy is further quantified by the ratio of MSRDs for the first and second coordination shells, which closely matches the anisotropic displacement parameters from diffraction experiments. Molecular dynamics simulations using the CHGNet universal machine-learning interatomic potential show that the original model overestimates thermal disorder, while a fine-tuned version substantially improves agreement with experimental EXAFS spectrum and radial distribution function. Overall, EXAFS-informed analysis is effective for validating and refining machine-learning interatomic potentials.
Journal refPhysica B 739 (2026) 418945
DOI：10.1016/j.physb.2026.418945
