Benchmarking Quantum Chemical Methods for the Calculation of Molecular Dipole Moments and Polarizabilities

Abstract

The calculation of molecular electric moments, polarizabilities, and electrostatic potentials is a widespread application of quantum chemistry. Although a range of wave function and density functional theory (DFT) methods have been applied in these calculations, combined with a variety of basis sets, there has not been a comprehensive evaluation of how accurate these methods are. To benchmark the accuracy of these methods, the dipole moments and polarizabilities of a set of 46 molecules were calculated using a broad set of quantum chemical methods and basis sets. Wave function methods Hartree-Fock (HF), second-order Møller-Plesset (MP2), and coupled cluster-singles and doubles (CCSD) were evaluated, along with the PBE, TPSS, TPSSh, PBE0, B3LYP, M06, and B2PLYP DFT functionals. The cc-pVDZ, cc-pVTZ, aug-cc-pVDZ, aug-cc-pVTZ, and Sadlej cc-pVTZ basis sets were tested. The aug-cc-pVDZ, Sadlej cc-pVTZ, and aug-cc-pVTZ basis sets all yield results with comparable accuracy, with the aug-cc-pVTZ calculations being the most accurate. CCSD, MP2, or hybrid DFT methods using the aug-cc-pVTZ basis set are all able to predict dipole moments with RMSD errors in the 0.12-0.13 D range and polarizabilities with RMSD errors in the 0.30-0.38 Å(3) range. Calculations using Hartree-Fock theory systematically overestimated dipole moments and underestimate polarizabilities. The pure DFT functionals included in this study (PBE and TPSS) slightly underestimate dipole moments and overestimate polarizability. Polarization anisotropy and implications for charge fitting are discussed.

Genetic Optimization of Training Sets for Improved Machine Learning Models of Molecular Properties

“…For isotropic polarizability and norm of dipole moment, target accuracies of 0.1a 3 0 and 0.1D were used. Again, these values are within the predictive uncertainty using CC level of theory [27]. Figure 2 displays learning curves of randomly generated and GA-optimized training sets for out-of-sample predictions of B3LYP level enthalpies of atomization, H, using both direct learning as well as the ∆H B3LYP PM7 -ML model [3].…”

Molecular electric moments calculated by using natural orbital functional theory

“…Thus, it contains sufficient diffuse and polarization functions in order to give an accurate description of the outervalence region. It has been shown [40,41] that the Sadlej basis set has effectively the same accuracy as the aug-ccpVTZ basis set.…”

Assessment of DFT for Computing Sum Frequency Generation Spectra of an Epoxydiol and a Deuterated Isotopologue at Fused Silica/Vapor Interfaces

“…42,43,91 Hybrid functionals also tend to perform better in predicting IR intensities. 92 In fact, PBE has been recently shown by Rowley and coworkers 43 to systematically overestimate polarizabilities in their benchmarking study, which may result in large differences in SFG intensities. However, we do observe improvements in the spectral matches when pairing the pure functionals with the Dunning basis sets, as seen in Table 1.…”