Efficient Screening of Organic Singlet Fission Molecules Using Graph Neural Networks.

Area of Science:

• Materials Science
• Computational Chemistry
• Renewable Energy

Background:

• Singlet fission (SF) is crucial for exceeding theoretical efficiency limits in photovoltaics.
• Discovering efficient SF materials is challenging due to limited candidates and high computational costs.
• Accelerating the identification of SF materials is vital for sustainable solar energy.

Purpose of the Study:

• To develop a high-throughput screening framework for identifying singlet fission materials.
• To leverage AI, specifically graph neural networks (GNNs), for predicting excited-state properties.
• To reduce computational expenses in the search for novel SF candidates.

Main Methods:

• Implemented a graph neural network (GNN) framework trained on the FORMED database.
• Screened over 20 million molecular structures from OE62 and QO2Mol databases.
• Employed multi-level validation, including DFT, GW approximation, and Bethe-Salpeter equation calculations.

Main Results:

• The GNN achieved high accuracy (≈0.1 eV MAE) for predicting S1, T1, and T2 excitation energies.
• Identified 180 potential SF molecules and over 1000 conformers.
• Highlighted a subset of experimentally feasible SF candidates after further assessment.

Conclusions:

• The AI-driven framework significantly accelerates the discovery of functional materials for optoelectronics.
• This approach offers an efficient strategy for identifying singlet fission candidates for sustainable solar energy.
• The study demonstrates the power of integrated computational methods and AI in materials discovery.