Abstract:
H and C NMR shieldings of 7 representative small hydrocarbons are calculated using GIAO with up to 27 Pople and Dunning basis sets on the levels HF, MP2, and B3LYP for experimental re geometries. Basis set effects are described and related to basis set properties.
By means of regressions of the results with experimental gas phase shieldings as well as with zero-point vibrationally corrected values, methods and basis sets best suited for exact calculation of shieldings are chosen using average deviations, standard errors (precision) and slopes (accuracy). By sorting and graphical plotting of the indicators, groups related to basis set properties show up.
The results differ depending on the considered indicator. Recommended for calculation of shieldings is the correction of the zero-point vibration, furthermore the use of MP2 with large Dunning basis sets. For highest accuracy, TZ basis sets are adequate. Inferior to MP2, but mostly superior to HF is B3LYP, at which already 6 311G* delivers good results, but cc-pVnZ with n>2 becomes too inaccurate. For highest quality, H and C have to be calculated separately with different basis sets.
For H, smallest deviations are delivered by B3LYP, followed by MP2 and HF. The most precise method is MP2 prior to B3LYP and HF. MP2 is superior in accuracy, too. With vibrational correction, the deviations are approximately halved and MP2 outruns B3LYP, while slopes still improve in the sequence HF < B3LYP < MP2.
For C, deviations are smallest for HF and much larger for MP2, which however catches up clearly on vibrational correction. Besides, HF yields the most accurate slopes. MP2 shows highest precision and all indicators follow the same trends.
The derivation of relative chemical shifts from shieldings causes a significant error cancellation. HF/6 311G* becomes feasible. Using methane as reference instead of TMS, even smaller deviations are obtained.
Optimized geometries of constant quality as basis are superior to experimental geometries from different sources. For unknown reasons, regressions with solution NMR data also deliver better correlations.
Shieldings calculated with Dunning basis sets are extrapolated to the basis set limit by fitting with exponential functions. The extrapolation is advantageous for H, for C less advisable, for methane referenced shifts not reasonable.
The geometry dependence of calculated chemical shifts is investigated on 18 molecules from methane to TMS by variation of the C-H and C-C distances.
The dependence of the H shifts from C-H distances is relatively distinct for saturated molecules (35 ppm/A for methane), and weaker for molecules with multiple bonds (13 ppm/A for benzene). It shows slightly concave, almost linear curves. Dependent of the C-H distances, the C shifts show almost linear, slightly convex curves with slopes between 6 (ethyne) and 204 ppm/A (ethane).
The variation of C-C single bonds hardly influences H shifts; double bonds cause stronger effects, which often can be fitted with r^4. In both cases, the curves are convex. Triple bonds result in linear dependencies with slopes around 6 ppm/A. On C-C variation, the C shifts also show convex curves, whose slopes increase with coordination number, while the curvatures decrease. The shifts behave nearly proportional to r^2n with the bond order n.
The two dimensional variation of C-H and C-C bonds proofs that both variations hardly interact with each other.
Already the small differences of various experimental distances result in significantly different shifts, however, no type of distance generally delivers better shifts than re.
The interrelation of the geometry dependence with molecular vibrations is investigated by correlating the zero point amplitudes of the total symmetric C-H and C-C valence vibrations with shift changes along these amplitudes. C-H vibrations do not reveal patterns in the H shift changes. For C intervals, classes of substances can be distinguished; with increasing C-C bond order the shift change decreases. The H shift changes caused by C-C vibrations are small. Again, the saturated molecules differ from the double bonded ones. The C shift intervals are pronounced stronger, but do not show any interpretable trend.