Electronic Descriptors for Supervised Spectroscopic Predictions

Abstract

Spectroscopic properties of molecules hold great importance for the description of the molecular response under the effect of UV/vis electromagnetic radiation. Computationally expensive ab initio (e.g., MultiConfigurational SCF, Coupled Cluster) or TDDFT methods are commonly used by the quantum chemistry community to compute these properties. In this work, we propose a (supervised) Machine Learning approach to model the absorption spectra of organic molecules. Several supervised ML methods have been tested such as Kernel Ridge Regression (KRR), Multiperceptron Neural Networs (MLP), and Convolutional Neural Networks. [ 26328822 J. Chem. Phys. 2015 143 084111 Adv. Sci. 2019 6 1801367 ] The use of only geometrical-atomic number descriptors (e.g., Coulomb Matrix) proved to be insufficient for an accurate training. [ 26328822 J. Chem. Phys. 2015 143 084111 ] Inspired by the TDDFT theory, we propose to use a set of electronic descriptors obtained from low-cost DFT methods: orbital energy differences (Δϵ ia = ϵ a – ϵ i ), transition dipole moment between occupied and unoccupied Kohn–Sham orbitals (⟨ϕ i | r |ϕ a ⟩), and when relevant, charge-transfer character of monoexcitations ( R ia ). We demonstrate that with these electronic descriptors and the use of Neural Networks we can predict not only a density of excited states but also get a very good estimation of the absorption spectrum and charge-transfer character of the electronic excited states, reaching results close to chemical accuracy (∼2 kcal/mol or ∼0.1 eV).

Encoding Hole–Particle Information in the Multi-Channel MolOrbImage for Machine-Learned Excited-State Energies of Large Photofunctional Materials

“…For a subspace of the GDB-13 data set, the kernel ridge regression model with Coulomb matrix and MO-based Fock–Coulomb–Exchange descriptors achieves MAEs of 0.13 and 0.18 eV, respectively, for E S 1 at the ZINDO (Zerner’s intermediate neglect of differential overlap) level. As most of the studies , that report promising prediction accuracy are using small molecules of rather limited use in photofunctional devices, there is still a pressing need to deploy ML models to large molecules with practical applications.…”

“…27 Subsequent study on the same data set, using the SchNet model with 10 000 training molecules (N train ), demonstrates MAEs of 0.494 and 0.434 eV for E Sd 1 and E Sd 2 , respectively, at the PBE0/def2-TZVP level. 28 For a subspace of the GDB-13 data set, the kernel ridge As most of the studies 30,31 that report promising prediction accuracy are using small molecules of rather limited use in photofunctional devices, there is still a pressing need to deploy ML models to large molecules with practical applications. Recently, we reported a convolutional-layer-dominated NN architecture (MO-NN) that maps the ground-state quantum descriptor molecular orbital image (MolOrbImage) to highlevel ADC(2)/cc-pVTZ (the second-order algebraic diagrammatic construction) computed transition energies and intensities (Figure 1a).…”