Machine-learning assisted design principle search for singlet fission: an example study of cibalackrot
Fabian Weber, Hirotoshi Mori
Abstract
Abstract This work uses quantum chemistry calculations and machine learning to explore design rules for singlet fission in a chemical space of four million indigoid derivatives. We identify ~400,000 derivatives of 2,2′-diethenyl cibalackrot, which theoretically fulfil the energy conditions for exoergic singlet fission above the silicon band gap energy. Probing this database with a random forest classifier, we observe that small substituents with positive mesomeric effects and weak negative inductive effects reinforce the desired energetic conditions when placed at specific positions. Finally, a subset of molecules that reflects the random forest classifier’s rules are investigated for their quantum chemical properties to translate the desirable structural motifs into wavefunction-based design rules. Here, direct correlations between the energetic condition for singlet fission, the biradical character and the charge and triplet spin density in prominent molecular regions are identified, providing insights that may serve as a guide for singlet fission core structure development.