Machine Learning Aided Design of Polymer with Targeted Band Gap Based on DFT Computation

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Polymer band gap is crucial for determining electrical conductivity.
• Predicting band gaps computationally is essential for materials design.
• Density Functional Theory (DFT) and molecular descriptors are key inputs.

Purpose of the Study:

• To develop a machine learning model for predicting polymer band gaps.
• To identify key molecular features influencing polymer band gaps.
• To facilitate rapid design of new polymers with targeted band gaps.

Main Methods:

• Utilized support vector regression (SVR) for prediction.
• Generated molecular descriptors using Dragon software.
• Employed maximum relevance minimum redundancy for feature selection.
• Trained and validated the model using DFT-computed band gap data.

Main Results:

• An optimal SVR model was achieved using 16 key features.
• High prediction accuracy was demonstrated with R² values of 0.824 (cross-validation) and 0.925 (independent test).
• Feature analysis identified critical descriptors influencing band gaps.

Conclusions:

• The developed SVR model accurately predicts polymer band gaps.
• The model enables efficient screening of polymers for specific band gaps.
• This approach accelerates the discovery and design of novel polymers.