Fig. 9: Performance of the artificial intelligence–quantum mechanical method 1 (AIQM1) for excited states.
From: Artificial intelligence-enhanced quantum chemical method with broad applicability
a Mean absolute error (MAD) in vertical excitation energies for Thiel’s benchmark set66 (AIQM1 in red, time-dependent density functional theory, TDDFT, in blue). b MAD in bond lengths for the ExGeom benchmark set26, 66 (AIQM1 vs approximate coupled cluster singles-and-doubles method (CC2) in red, AIQM1 vs. TDDFT in blue, TDDFT vs. CC2 in black). c Lengths of C–O bonds in formaldehyde and acetaldehyde as compared to experiment (experiment in black, AIQM1 in red, TDDFT in blue, CC2 in green; one value for TDDFT is missing due to the failed geometry optimization at this level66). Values for TDDFT (B3LYP/TZVP), CC2, TDDFT vs. CC2, and experiment are taken from ref. 66.
