A comparison of density functional methods for the estimation of proton chemical shifts with chemical accuracy

Fifteen procedures based on hybrid density functional theory were used to c alculate magnetic properties for the carbon-bound hydrogen nuclei of SO sma ll to modest-sized organic molecules. The predicted isotropic shieldings de rived from the various methods were compared with each other and also with solution experimental data. The computational methods investigated included the IGAIM and GIAO procedures, the 6-311++G(d,p), 6-311++G(2df,p), and 6-3 11++G(3df,2p) basis sets, the B3LYP, B3P86, and B3PW91 hybrid density funct ionals, and molecular geometries optimized using both MP2 and B3LYP methods . Although agreement with experiment consistently improved as the basis set was enlarged, the improvement upon going from 6-311++G(2df,p) to 6-311++G( 3df,2p) was extremely small, and even the difference between 6-311++G(d,p) and 6-311++G(2df,p) was of a. modest size. The CIAO and IGAIM procedures yi elded very similar results in conjunction with the largest basis set, but C IAO suffered considerably less degradation than did IGAIM as the basis set size was decreased. The three functionals B3LYP, B3P86, and B3PW91 performe d in al extremely similar fashion, although B3LYP proved marginally superio r to the others. The method of geometry optimization also was found to make little difference. Of the computational methods investigated, the GIAO/B3L YP/6-311++G(d,p)//B3LYP/6-31+G(d) procedure probably represents the best co mpromise between accuracy and expense and yielded proton chemical shifts ha ving a root-mean-square error of 0.15 ppm in comparison with solution exper imental values after empirical linear scaling. The more expensive GIAO/B3LY P/6-311++G(2df,p)//B3LYP/6-31+G(d) method provided only a slightly lower ro ot-mean-square error of 0.14 ppm.