The focal-point approach, combining several quantum chemistry computations to estimate a more accurate computation at a lower expense, is effective and commonly used for energies. However, it has not yet been widely adopted for properties such as geometries. Here, we examine several focal-point methods combining Møller–Plesset perturbation theory (MP2 and MP2.5) with coupled-cluster theory through perturbative triples [CCSD(T)] for their effectiveness in geometry optimizations using a new driver for the Psi4 electronic structure program that efficiently automates the computation of composite-energy gradients. The test set consists of 94 closed-shell molecules containing first- and/or second-row elements. The focal-point methods utilized combinations of correlation-consistent basis sets cc-pV(X+d)Z and heavy-aug-cc-pV(X+d)Z (X = D, T, Q, 5, 6). Focal-point geometries were compared to those from conventional CCSD(T) using basis sets up to heavy-aug-cc-pV5Z and to geometries from explicitly correlated CCSD(T)-F12 using the cc-pVXZ-F12 (X = D, T) basis sets. All results were compared to reference geometries reported by Karton et al. [J. Chem. Phys. 145, 104101 (2016)] at the CCSD(T)/heavy-aug-cc-pV6Z level of theory. In general, focal-point methods based on an estimate of the MP2 complete-basis-set limit, with a coupled-cluster correction evaluated in a (heavy-aug-)cc-pVXZ basis, are of superior quality to conventional CCSD(T)/(heavy-aug-)cc-pV(X+1)Z and sometimes approach the errors of CCSD(T)/(heavy-aug-)cc-pV(X+2)Z. However, the focal-point methods are much faster computationally. For the benzene molecule, the gradient of such a focal-point approach requires only 4.5% of the computation time of a conventional CCSD(T)/cc-pVTZ gradient and only 0.4% of the time of a CCSD(T)/cc-pVQZ gradient.

中文翻译：

---

使用大基数耦合簇理论的焦点近似，高效且自动地计算精确的分子几何形状

结合多种量子化学计算以较低成本估算更准确计算的焦点方法是有效且通常用于能源的方法。但是，尚未将其广泛用于诸如几何等特性。在这里，我们研究了几种焦点方法，它们通过微扰三元组[CCSD（T）]将Møller-Plesset扰动理论（MP2和MP2.5）与耦合集群理论相结合，从而在几何优化中使用了新的P si驱动程序4个电子结构程序，可有效地自动计算复合能梯度。测试仪由94个包含第一行和/或第二行元素的闭壳分子组成。焦点方法利用了相关一致基集cc-pV（X + d）Z和重度aug-cc-pV（X + d）Z的组合（X = D，T，Q，5、6）。将焦点几何与使用重载aug-cc-pV5Z的基础集与常规CCSD（T）的几何进行比较，并使用cc-pVXZ-F12将其与显式相关的CCSD（T）-F12的几何进行比较（X = D， T）基础集。将所有结果与Karton等报道的参考几何进行了比较。[J. 化学 物理 145，104101（2016）]在CCSD（T）/ heavy-aug-cc-pV6Z理论水平。通常，基于MP2完整基准集限制的估计的焦点方法，以（重载）cc-pVXZ为基础评估的耦合群集校正，其质量优于常规CCSD（T ）/（重-aug-）cc-pV（X + 1）Z，有时接近CCSD（T）/（重-aug-）cc-pV（X + 2）Z的误差。但是，焦点方法在计算上要快得多。对于苯分子，这种焦点方法的梯度仅需要传统CCSD（T）/ cc-pVTZ梯度的4.5％的计算时间，而仅需要CCSD（T）/ cc-pVTZ梯度的0.4％的计算时间。 pVQZ梯度。