The relationships between experimental and theoretical 13C NMR chemical shifts of a pristine fullerene C60, monoadducts from [2 + n] cycloaddition (n = 1–3), and one [2 + 1] bis‐adduct are systematically analyzed for the first time by using diverse quantum‐chemical levels of theory. These levels involved B3LYP, B3PW91, B97‐2, mPW1PW91, PBE1PBE, and X3LYP hybrid functionals combined with 3‐21G, 6‐31G, 6‐31G(d), 6‐31G(d,p), 6‐31G(d,2p), LanL2DZ, and SDDAll basis sets. X3LYP/6‐31G approach is determined to have the lowest deviations from the 13C NMR experimental data compared to the other methods for all the fullerene compounds (mean absolute error value is 0.856 ppm and root mean squared error value is 1.197 ppm). The highest deviations are characteristic for α (sp2 C2/C5/C8/C10) and β (sp2 C6/C7/C11/C12) carbon atoms relative to a functionalization site and for those (sp3 C1/C9) directly attached with a side fragment in the [2 + n] monoadducts (n = 1–3). A probable reason of such deviation is that the approaches do not take into account a contribution of paramagnetic ring currents to 13C NMR chemical shifts. The results will be useful in design of novel fullerene derivatives and in performing unambiguous 13C NMR chemical shift assignments with modern quantum chemistry calculations.

Real‐valued models based on deep artificial neural networks were proposed to predict 13C NMR chemical shifts of fullerene C60 core carbon atoms for computer‐aided structure elucidation of complex fullerene C60 mono‐adducts. We showed that parametric rectified linear units could be successfully used as activation functions in hidden layers of artificial neural networks for decision of complex physical‐chemical tasks. A total of 400 artificial neural networks were trained and tested in order to reveal the best‐fitted models. The best prediction accuracy of real‐valued models was achieved with MAEP = 1.83 ppm/RMSEP = 2.60 ppm using artificial neural network model which has 110 and 120 hidden units, respectively, with parametric rectified linear unit as activation function.

No abstract