Fig. 2: The design of the artificial intelligence–quantum mechanical method 1 (AIQM1).
From: Artificial intelligence-enhanced quantum chemical method with broad applicability
a Flowchart of training the neural network (NN) part of the AIQM1, AIQM1@DFT*, and AIQM1@DFT methods. b Schematic representation of the components of the AIQM1, AIQM1@DFT*, and AIQM1@DFT methods (yellow). DFT denotes density functional theory data at ωB97X/def2-TZVPP, CC— approximation to coupled cluster with single, double, and perturbative triple excitations with complete basis set extrapolation scheme (CCSD(T)*/CBS), E energies, F forces, ODM2*—semiempirical quantum mechanical (SQM) orthogonalization- and dispersion-corrected method 2 (light brown), D4—the fourth generation of dispersion (subscript disp) corrections (green), NN – neural networks of ANI-type (blue; corrections to learn also blue). Molecular data sets are in gray.
