Can force fields developed for carbon nanomaterials describe the isomerization energies of fullerenes?

We evaluate the performance of carbon force fields for 1811 C60 PW6B95-D3/Def2-QZVP isomerization energies. Several force fields (most notably the machine-learning GAP-20 potential) exhibit a high statistical correlation with the DFT isomerization energies. Therefore, linear scaling of the isomerization energies can significantly improve the accuracy. The best scaled force fields attain mean-absolute deviations of 8.5 (GAP-20), 12.3 (LCBOP-I and REBO-II), and 13.3 (ABOP) kcal mol−1, which translate to mean-absolute relative deviations of 4.7% (GAP-20), 6.5% (LCBOP-I), 6.6% (REBO-II) and 7.1% (ABOP). Therefore, these force fields offer a computationally economical way for exploring the relative energies of fullerenes.