The dataset of covalent bond lengths resulting from the first-principle calculations

The non-empirical dataset of the covalent bond lengths is created via the automated procedure combining geometries of 26050 small closed-shell molecules optimized by one of three DFT methods (B3LYP/6-31G*, PBEh-3c, B97-3c) with the covalent bond graphs recovered from the MP2 first-order density matrices by recently introduced property-oriented orbital localization procedure (CLPO). The dataset (https://dx.doi.org/10.5281/zenodo.2631844) contains the distances between H, B, C, N, O, F, Si, P, S, Cl, Ge, As, Se, Br atoms linked by 213404 single, 38732 double and 5031 triple bonds. The statistical distributions of lengths of the covalent bonds were analyzed for different pairs of chemical elements and the influence of electron delocalization on the bond lengths was studied. The average lengths of the bonds not affected by electron delocalization, obtained from the proposed dataset, have been found to be in good agreement with the average bond lengths derived from the available experimental data in the gas-phase.