NMR chemical shifts in MNDO approximation: Parameters and results for H, C, N, and O

The evaluation of the NMR chemical shift tensor has been implemented at the semiempirical MNDO level (modified neglect of diatomic overlap) for an spd basis using gauge-including atomic orbitals and analytic derivative theory . AU relevant contributions to the chemical shifts of nuclei and points in space (nuclear independent chemical shifts,NICS), including three-center te rms, can be computed. The execution time is dominated by the evaluation of the three-center terms, which typically require more than 90% of the total effort. Calculations with standard MNDO parameters overestimate the variati on of the paramagnetic contributions, leading to an unsatisfactory descript ion of the trends in chemical shifts. Agreement with experiment is improved by reoptimizing the electronic structure parameters to reproduce the exper imental chemical shifts for 97 small molecules and ions. One-center. energi es, orbital exponents, and resonance beta parameters of H, C, N, and O have been adjusted, for a total of 16 parameters. The reference set covers chem ical shift ranges of 31, 346, 933, and 1650 parts per million (ppm) for H, C, N, and O, respectively. The final RMS deviations from experiment are 0.6 4, 13.6, 39.6, and. 80.6 ppm for II, C, N, and O, respectively, down from 1 .93, 36.8, 120.0, and 119.0 ppm for the original MNDO parameters. This corr esponds to less than 5% of the total chemical shift range for each element. A significant fraction of the total error is due to small molecules with u nusual bonding. On a larger set of solution NMR data for 384 common organic molecules, the RMS errors are reduced to 11.9, 38.8, and 61.7 ppm for C, N , and O, respectively. Three-center terms typically contribute a few ppm to the total chemical shift for all fowl elements studied, but they are essen tial for a qualitatively correct description of H-1 chemical shifts and NIC S. These terms are less important for C, N, and O, and can usually be omitt ed for these elements without significantly degrading the results. The over all quality of the results obtained with the NMR-specific parameterization is competitive with ab initio Hartree-Fock calculations. The relatively low computational cost permits studies of NMR chemical shifts and their trends in large molecules. Possible directions for future semiempirical NMR devel opments are outlined based on an analysis of some shortcomings of the prese nt approach. (C) 1999 John Wiley & Sons, Inc.