DFT calculations for Cu-, Ag-, and Au-containing molecules

The physicochemical properties of a set of molecules containing Cu, Ag, or An atoms were calculated using the GAUSSIAN program suite, with the purpose of investigating the various density functional theory (DFT) approaches for subsequent application in cluster calculations. The test set comprised the copper-based molecules CuH, CUO, CuS, Cu-2, CuCl2-, CuCH3, CuC2H2, Cu-2(HCO2)(4)(H2O)(2), and Cu6H6(PH3)(6) and the silver and gold diatomics AgH, AgO, AgS, Ag-2, AgCu, AuH, AuO, AuS, Au-2, AuCu, and AuAg. The DFT methods used were SVWN, BLYP, and BPW91, together with the DFT hybrids B3LYP and B3PW91. The calculations were carried out with the basis sets LANL2MB, LANL2DZ, 3-21G, and 6-311G (when available). The properties calculated were the molecular geometry, vibrational frequencies, and dissociation energies. It was found that all the DFT-based methods, particularly when allied with the LANL2DZ basis set, produced results which are significantly closer to experimental values than those of the traditional Moller-Plesset (MP2) electron correlation method and which are also applicable to considerably larger molecules. Over the whole test set of molecules, the RMS errors of the four BX methods, in conjunction with LANL2DZ, were typically 3-4% for geometries, 6-8% for frequencies, and 10-16% for dissociation energies, with BPW91 and the popular B3LYP at the lower and upper ends of these ranges, respectively, and with the errors being overestimates and underestimates, respectively. The corresponding values for SVWN and MP2 were 2% and 6%, 12% and 12%, and 33% and 42%, with these errors typically being +/- and +, + and -, and + and -, with + and - representing overestimates and underestimates, respectively. While the best bond lengths are produced by SVWN (a local spin density approximation), which is not uncommmon, this advantage over the gradient corrected (BX) methods is only slight, and the latter are markedly superior for frequencies and especially dissociation energies. Not supported by the present results are the notions that (all) pure DFT methods underestimate metal-ligand bond lengths and overestimate bond strengths and that hybrid methods are superior (and neither that DFT methods are overcorrelated). Testing on a subset of molecules with BPW91/LANL2DZ revealed no benefit in supplementing this basis set by the addition of diffuse functions, nor of the counterpoise correction. There appear to be specific incompatibilities with some method/basis set combinations, and even the continuing availability of 3-21G for these metals is questionable. Because of its accuracy and reliability, the combination BPW91/LANL2DZ is recommended for these noble-metal systems (and to extensions such as the cluster-model approach to adsorption of a molecule on a metal surface).