Critical Assessment of the Performance of Density Functional Methods for Several Atomic and Molecular Properties

• Published in: Journal of chemical theory and computation Vol. 3; no. 2; pp. 407 - 433
• Main Authors: Riley, Kevin E, Op't Holt, Bryan T, Merz, Kenneth M
• Format: Journal Article
• Language: English
• Published: United States American Chemical Society 01-03-2007
• ISSN: 1549-9618; 1549-9626
• DOI: 10.1021/ct600185a

The reliable prediction of molecular properties is a vital task of computational chemistry. In recent years, density functional theory (DFT) has become a popular method for calculating molecular properties for a vast array of systems varying in size from small organic molecules to large biological compounds such as proteins. In this work, we assess the ability of many DFT methods to accurately determine atomic and molecular properties for small molecules containing elements commonly found in proteins, DNA, and RNA. These properties include bond lengths, bond angles, ground-state vibrational frequencies, electron affinities, ionization potentials, heats of formation, hydrogen-bond interaction energies, conformational energies, and reaction barrier heights. Calculations are carried out with the 3-21G*, 6-31G*, 3-21+G*, 6-31+G*, 6-31++G*, cc-pVxZ, and aug-cc-pVxZ (x = D and T) basis sets, while bond-distance and bond-angle calculations are also done using the cc-pVQZ and aug-cc-pVQZ basis sets. Members of the popular functional classes, namely, local spin density approximation, generalized gradient approximation (GGA), meta-GGA, hybrid-GGA, and hybrid-meta-GGA, are considered in this work. For the purpose of comparison, Hartree−Fock and second-order many-body perturbation methods are also assessed in terms of their ability to determine these physical properties. Ultimately, it is observed that the split valence bases of the 6-31G variety provide accuracies similar to those of the more computationally expensive Dunning-type basis sets. Another conclusion from this survey is that the hybrid-meta-GGA functionals are typically among the most accurate functionals for all of the properties examined in this work.