Group Contribution Method Supervised Neural Network for Precise Design of Organic Nonlinear Optical Materials.

Area of Science:

• Materials Science
• Computational Chemistry
• Organic Chemistry

Background:

• Designing organic small-molecule nonlinear optical (NLO) materials requires rational approaches.
• Predicting molecular optical properties is crucial for material design.
• Existing methods may require extensive data or lack efficiency.

Purpose of the Study:

• To develop a theory-guided machine learning framework for rational design of D-π-A type organic small-molecule NLO materials.
• To establish a method for accurately and efficiently predicting molecular optical properties.
• To enable efficient structural search for novel NLO materials.

Main Methods:

• Developed a Lewis-mode group contribution method (LGC).
• Integrated LGC with multistage Bayesian neural networks (msBNN) and evolutionary algorithms (EA) into an interactive framework (LGC-msBNN-EA).
• Utilized EA specifically designed for LGC to facilitate structural search.

Main Results:

• The LGC-msBNN-EA framework accurately predicts various optical properties of molecules.
• The framework requires only a small dataset for effective training.
• Efficient structural search for new materials is achievable using the EA component.
• The framework demonstrates satisfying performance in predicting NLO material properties.

Conclusions:

• The developed LGC-msBNN-EA framework offers a rational and efficient approach to designing organic small-molecule NLO materials.
• Combining chemical principles with data-driven tools enhances predictive accuracy and design efficiency.
• This framework is expected to be valuable for structure design in related scientific fields.