基于EOS的格林艾森函数构建有限温度体积模量
Finite-temperature bulk moduli from an EOS-based Grüneisen function
浏览论文内容
中文总结 AI 辅助
本文开发了基于EOS的格林艾森函数构建方法，无需拟合热数据即可捕捉体积模量的主要热软化行为，可用于检验多种材料并为通用机器学习原子间势的训练验证提供参考。
中文摘要 AI 辅助
本文开发了一种基于状态方程（EOS）的格林艾森函数构建方法，并通过预测有限温度体积模量对其进行评估。该方法中，格林艾森函数的体积依赖性以静态EOS信息为基础解析表达，锚定了从平衡附近采样的弹性数据获得的德拜温度，并受无限压缩极限约束。所得形式仅需静态能量-体积数据和近平衡弹性质作为材料特定输入，无需针对热数据调整任何参数。在米-格林艾森-德拜框架内，针对金刚石、氧化镁、硅和氯化钠这四种涵盖宽刚度范围的材料，采用来自原子通用模型（UMA）和通用点边变换器（UPET）系列的机器学习原子间势，以及UMA的色散修正变体对该方法进行检验。计算得到的体积模量重现了预期的软化趋势，并捕捉了体积模量温度导数的整体量级，尽管定量吻合程度取决于材料、底层原子间势及所选解析EOS形式。这些结果表明，基于EOS的格林艾森函数构建方法无需拟合热数据即可捕捉体积模量的主要热软化行为。观察到的对底层静态描述的敏感性进一步表明，该框架或有助于识别与热弹性可迁移性相关的缺陷，从而为通用机器学习原子间势的未来训练和验证策略提供信息。
英文摘要
An equation-of-state (EOS)-based construction of the Grüneisen function is developed and assessed through predictions of finite-temperature bulk moduli. In this approach, the volume dependence of the Grüneisen function is expressed analytically in terms of static EOS information, anchored by Debye temperatures obtained from elastic data sampled near equilibrium, and constrained by the infinite-compression limit. The resulting form requires as material-specific input only static energy-volume data and near-equilibrium elastic properties, with no parameters adjusted to thermal data. Within a Mie--Grüneisen--Debye framework, the approach is examined for diamond, magnesium oxide, silicon, and sodium chloride, chosen to span a broad range of stiffness, using machine-learning interatomic potentials from the Universal Models for Atoms (UMA) and Universal Point Edge Transformer (UPET) families, together with a dispersion-corrected variant of UMA. The calculated bulk moduli reproduce the expected experimental softening trends and capture the overall scale of the bulk-modulus temperature derivatives, although the level of quantitative agreement depends on the material, the underlying interatomic potential, and the selected analytic EOS form. These results show that an EOS-based construction of the Grüneisen function can capture the leading thermal-softening behavior of bulk moduli without fitting to thermal data. The observed sensitivity to the underlying static description further suggests that the framework may help identify deficiencies relevant to thermoelastic transferability and thereby inform future training and validation strategies for universal machine-learning interatomic potentials.
