Data Driven Discovery of Singlet Fission Materials

<p dir="ltr">Singlet fission (SF) is a spin-allowed photophysical process in which one high-energy singlet exciton splits into two lower-energy triplet excitons, offering a promising route to surpass the Shockley–Queisser limit in photovoltaic efficiency. Among the various forms, intermolecular singlet fission occurring in crystalline solids is particularly desirable for device integration. However, the development of SF-enabled solar cells is currently limited by the scarcity of known molecular crystals that exhibit efficient SF in the solid state. Computational screening presents a scalable and cost-effective approach to accelerate SF material discovery. </p><p dir="ltr">In this thesis, we develop a multi-fidelity computational workflow, the Materials Machine Learning (MatML) workflow, for the discovery of SF candidates by integrating quantum mechanical simulations with machine learning (ML) and database mining. High-accuracy excited-state properties are computed using the many?body perturbation theory approach combining the ���� approximation and the Bethe–Salpeter equation (����+BSE), which we have used to build a first-of-its-kind dataset (PAH101) of polycyclic aromatic hydrocarbon (PAH) crystals. These data serve as a foundation for identifying correlations between low-cost density functional theory (DFT) quantities and high-level excited-state properties. To bridge cost and accuracy, we train ML models using the SISSO algorithm to predict excited state optoelectronic descriptors such as the singlet and triplet excitation energies, the singlet–triplet gap, and the exciton binding energy. The PAH101 dataset and SISSO models can be valuable in analyzing other interstate transition phenomena other than SF, including Thermally Activated Delayed Fluorescence (TADF) and Triplet-Triplet Annihilation (TTA) and beyond.</p>