Accurate Modeling of Water Clusters with Density-Functional Theory Using Atom-Centered Potentials

The ability of atom-centered potentials
(ACPs) to improve the modeling
of water clusters using density-functional methods is explored. Water-specific
ACPs were developed using accurate ab initio reference
data to correct the deficiencies of the BHandHLYP density functional
in the calculation of absolute and relative binding energies of water
clusters. In conjunction with aug-cc-pVTZ basis sets and with or without
dispersion corrections, it is possible to obtain absolute binding
energies for water clusters containing up to 10 H2O molecules
to within 0.44 kcal/mol or 0.04 kcal/mol per water molecule. In contrast,
dispersion-corrected BHandHLYP/aug-cc-pVTZ predicts binding energies
with errors as large as 6 kcal/mol for (H2O)10 in the absence of ACPs. Therefore, the ACPs improve predicted binding
energies in these clusters by more than an order of magnitude. The
conformers of (H2O)16 and (H2O)17 were used to validate the application of ACPs to larger
clusters. ACP-based approaches are able to predict the binding energies
in (H2O)16,17 within a range of 0.3–2.2
kcal/mol (less than 1.3%) of recently revised ab initio wave function results. ACPs for basis sets smaller than aug-cc-pVTZ
are also presented. However, the ability of the BHandHLYP/ACP approach
to predict accurate binding energies deteriorates as the size of the
basis sets decreases. Nevertheless, ACPs improve predicted binding
energies by as much as a factor of 50 across the range of the basis
sets studied. The BHandHLYP/aug-cc-pVTZ-ACP method was applied to
(H2O)25 in order to identify the minimum-energy
structure of a collection of proposed global minimum-energy structures.
The BHandHLYP/aug-cc-pVTZ-ACP approach is an accurate and computationally
affordable alternative to wave function theory methods for the prediction
of the binding energies and energy ranking of water clusters.

本研究探讨了原子中心势（atom-centered potentials, ACPs）对基于密度泛函方法的水团簇建模的优化能力。研究人员基于高精度从头算参考数据开发了针对水体系的专用ACPs，用以修正BHandHLYP密度泛函在计算水团簇绝对结合能与相对结合能时存在的缺陷。配合aug-cc-pVTZ基组并添加或不添加色散修正时，对于包含至多10个H₂O分子的水团簇，其绝对结合能的计算误差可控制在整体不超过0.44 kcal/mol、单分子误差不超过0.04 kcal/mol的范围内。与之相比，未引入ACPs的色散修正版BHandHLYP/aug-cc-pVTZ方法在计算(H₂O)₁₀的结合能时，最大误差可达6 kcal/mol。由此可见，ACPs可将此类团簇的结合能预测误差降低一个数量级以上。研究使用(H₂O)₁₆与(H₂O)₁₇的同分异构体验证了ACPs在更大尺寸水团簇中的适用性。基于ACPs的方法可将(H₂O)₁₆、₁₇的结合能预测值与近期修订的从头算波函数参考结果之间的误差控制在0.3~2.2 kcal/mol范围内，相对误差小于1.3%。本研究同时提供了适用于尺寸小于aug-cc-pVTZ基组的ACPs。不过，BHandHLYP/ACP方法的结合能预测精度会随基组尺寸缩小而有所下降。尽管如此，在本次研究所涉及的所有基组范围内，ACPs仍可将结合能的预测精度提升至多50倍。研究将BHandHLYP/aug-cc-pVTZ-ACP方法应用于(H₂O)₂₅，以在一系列已提出的全局最低能量结构候选集中筛选出最优的最低能量构型。综上，BHandHLYP/aug-cc-pVTZ-ACP方法是一种兼具高精度与计算经济性的替代方案，可用于水团簇的结合能预测与能量排序，其性能优于传统波函数理论方法。