Understanding the Key Factors for Photoinduced Radical Generation in Crystalline Triphenylamines Using Experiment and Machine Learning

In this study, we combine experiments, calculated properties, and machine learning (ML) to design new triphenylamine-based (TPA) molecules that have a high photoinduced radical (PIR) generation in crystals. A data set of 34 crystal structures was extracted from the Cambridge Crystallographic Data Centre. Eighteen structures with experimentally reported PIR values from 0 to 0.85% were used to build an ML model trained using Random Forest that achieves an average leave-one-out test set error of 0.173% PIR. The ML model was used to screen the remaining 16 compounds, of which 4 were selected and subsequently compared with the experimentally measured PIR%. The predicted PIR% demonstrated good agreement with the measured values of TPA bis-urea macrocycle host–guest complexes and non-macrocyclic compounds of TPAs. Examining a broad set of molecular architectures/scaffolds allows for investigating the structural and electronic properties that lead to high PIR generation. We found very different trends for macrocycles, linear TPAs, and mono TPAs, where mono TPAs consistently have the lowest PIR generation. Macrocycles tend to have the highest PIR generation, especially for systems with benzene and fluorobenzene guests. Although linear analogs overall perform worse than macrocycles, they display clear trends with increasing excited-state dipole moment, oscillator strength, and electron–hole covariance, while decreasing ionization potential and interatomic distance are generally correlated with higher PIRs. What is consistently observed is that higher PIRs are seen for brominated analogs. Our study, therefore, provides guidelines for future design strategies of TPAs for PIR generation.