We therefore selected a subset of 10 pertinent databases from the GMTKN30 database to test the B3LYP-DCP approach for their ability to predict properties based on these interactions. We performed calculations using the following sub-databses: BHPERI (barrier heights for pericyclic reactions), 56 DARC (Diels-Alder reaction energies), 57 BSR36 (bond separation energies of saturated hydrocarbons), 58 ISO34 (isomerization energies of small and medium sized organic molecules), 59 ISOL22 (isomerization energies of large organic molecules), 60 PCONF (relative energies of phenylalanyl-glycyl-gycine tripeptide conformers), 61 ACONF (relative energies of alkane conformers), 62 SCONF (relative energies of sugar conformers), 42,63 IDISP (intramolecular dispersion interactions), 42,43 and ADIM6 (interactions of n-alkane dimers). 49 A summary of the performance of B3LYP/6-31+G(2d,2p) without DCPs, with the DCPs of reference 19, and those developed in this work are given in Table 6 in the form of mean absolute errors relative to the reference data in each of the sub-databases. For comparison, we also present the results of calculations using B3LYP-D3/6-31+G(2d,2p). The explicit data for each of the relevant entries are provided in the ESM. The data presented in Table 6 and the ESM do not include data for species that contain elements other than H, C, N, and O atoms. Table 6 . Mean absolute errors (MAE, kcal/mol) of the B3LYP/6-31+G(2d,2p) approach on the sub-databases of the GMTKN30 database, without and with various dispersion-correction schemes. The lowest MAE for each database is indicated in bold. Sub-database name B3LYP B3LYP-DCP (ref. 19) B3LYP-DCP (this work) B3LYP-D3 BHPERI 3.44 3.36 2.48 1.67 DARC 10.71 13.02 4.44 3.54 BSR36 9.77 3.29 1.34 2.41 ISO34 2.02 3.16 1.60 1.49 ISOL22 7.34 5.46 2.39 4.58 PCONF 3.71 0.64 0.46 0.57 ACONF 1.02 0.17 0.06 0.07 SCONF 0.93 1.07 0.77 0.36 IDISP 15.64 2.99 1.45 2.53 ADIM6 4.81 0.14 0.09 0.34
19 The data in Table 6 shows that unadorned B3LYP/6-31+G(2d,2p) predicts fairly large MAEs for the 10 sub-databases. Particularly large MAEs are found for the ISOL22 and IDSIP databases. This is not surprising in light of the fact that the reference systems in these sub- databases are largely dominated by dispersion interactions, which B3LYP cannot model properly. The Diels-Alder sub-database barrier heights, DARC, is also poorly treated by unadorned B3LYP, suggesting that short-range interactions that occur at the transition states are strongly influenced by non-covalent interactions. The inclusion of dispersion corrections via the DCPs of reference 19 offers improvement to most, but not all, of the sub-databases. The MAEs for ISO34, DARC and SCONF increase with the use of the C-DCPs of reference 19 (as does that for SCONF), while that for BHPERI is only very slightly reduced. Large reductions in MAEs are seen for the remaining databases. These observations broaden the scope of systems that are poorly modeled with the C-DCPs of reference 19 and support our conclusion that they adversely affect the predicted energies associated with C-C bond making/breaking chemistry.
You've reached the end of your free preview.
Want to read all 44 pages?
- Fall '19
- dr. ahmed