发表机构
University of Toronto; Acceleration Consortium, University of Toronto; Vector Institute for Artificial Intelligence; Universidad de Oviedo（多伦多大学; 多伦多大学加速联盟; 向量人工智能研究所; 奥维耶多大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出非经验XDM色散模型neXDM，去除实验参数并改进离子系统描述，性能与XDM相当，并在X23晶格能基准中创下GGA泛函新纪录。
AI 中文摘要
在密度泛函理论（DFT）中，包含色散效应对于正确模拟非共价相互作用至关重要，而这一任务在该理论的许多应用中都是必不可少的。过去已经提出了许多色散泛函。交换-空穴偶极矩（XDM）模型将色散能量的阻尼成对渐近表达式的简洁性与基于理论计算色散系数的方法相结合。可以说，XDM是描述分子晶体最准确的色散校正方法，并且已在广泛的化学应用中得到充分测试。在此，我们解决了XDM的两个主要缺点。首先，XDM依赖于使用实验确定的自由原子极化率。其次，由于XDM的分子中原子性质（体积、极化率、交换-空穴偶极矩）使用Hirshfeld划分方法，XDM对具有较大原子部分电荷的系统（如碱金属阳离子或卤化物阴离子）描述不佳。我们提出了neXDM，一种非经验的XDM变体，通过使用Kirkwood极化率公式去除实验参数，从而使neXDM成为纯粹的meta-GGA色散泛函。此外，借鉴Bučko等人先前关于类似的Tkatchenko-Scheffler（TS）方法的工作，我们将Hirshfeld划分替换为其迭代版本。结果表明，neXDM在标准分子和晶体基准测试中的性能与XDM相当，并大大改善了离子系统的建模。新的neXDM方法在X23集合中为分子晶体晶格能设定了最佳色散校正广义梯度近似（GGA）泛函的新纪录（0.700 kcal/mol）。
英文摘要
The inclusion of dispersion effects is important in density-functional theory (DFT) to model non-covalent interactions correctly, a task that is essential in many applications of the theory. Many dispersion functionals have been proposed in the past. The exchange-hole dipole moment (XDM) model combines the simplicity of a damped pairwise asymptotic expression for the dispersion energy with a theory-grounded approach to calculate the dispersion coefficients. XDM is, arguably, the most accurate dispersion correction for the description of molecular crystals and it has been thoroughly tested for other applications across a wide range of chemistries. Here, we address the two main shortcomings of XDM. First, XDM relies on the use of experimentally determined free-atom polarizabilities. Second, because the XDM atom-in-molecule properties (volumes, polarizabilities, exchange-hole dipole moments) use the Hirshfeld partition method, XDM describes systems with large atomic partial charges, like alkali cations or halide anions, poorly. We propose neXDM, a non-empirical variant of XDM that removes the experimental parameters by using the Kirkwood polarizability formula, thereby making neXDM a pure meta-GGA dispersion functional. In addition, following previous work by Bučko et al. on the similar Tkatchenko--Scheffler (TS) method, we replace the Hirshfeld partitioning with its iterative counterpart. The performance of neXDM is shown to be on par with XDM in standard molecular and crystal benchmark sets, and greatly improves the modeling of ionic systems. The new neXDM method sets a new record for the best dispersion-corrected generalized-gradient approximation (GGA) functional for molecular crystal lattice energies in the X23 set (0.700~kcal/mol).
Comments38 pages 2 figs
