Generative organic electronic molecular design informed by quantum chemistry.

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning

Background:

• Generative models accelerate molecular discovery but often rely on low-cost evaluations unsuitable for complex materials.
• Functional molecular materials require accurate property prediction, posing challenges for traditional generative approaches.

Purpose of the Study:

• To develop a computational framework linking reinforcement learning with quantum chemistry for designing functional molecular materials.
• To discover molecules with specific excited state energy levels for singlet fission and triplet-triplet annihilation applications.

Main Methods:

• Integration of the REINVENT reinforcement learning framework with excited state quantum chemistry calculations.
• Implementation of a two-step curriculum strategy for molecule discovery and focused optimization.
• Evaluation of generated molecules based on targeted excited state properties and synthesizability.

Main Results:

• The framework successfully identified molecules with desired excited state energy levels.
• Improved Pareto fronts for targeted properties versus synthesizability were achieved.
• Several established and novel design principles for singlet fission and triplet-triplet annihilation materials were uncovered.

Conclusions:

• The developed framework offers a powerful approach for the *de novo* design of functional molecular materials.
• Combining generative models with accurate quantum chemistry enables efficient exploration of complex chemical spaces.
• This method facilitates the discovery of materials with tailored photophysical properties and provides insights into molecular design strategies.