Data driven design of dyes with high dielectric constant for efficient optoelectronics

Abstract

The optimization of dyes with high dielectric constant is crucial to perform superior functions in communication electronics, solar cells, and battery technologies. Generally, the approaches to synthesizes compounds require significant time and effort, underlining the importance of computing tools that would help to save time. In this work, machine learning models are used to estimate dielectric constant. Random Forest is best model. yielding the least RMSE of 0.592 and highest R² of 0.676. Descriptors such as SMR_VSA10 were found from the correlation analysis to be very important. Ten thousand new dye structures were designed and dielectric constants of these dyes were predicted by the Random Forest regressor. Chemical space is visualized through clustering. The results presented here highlight how machine learning methods can be applied in the context of dye chemistry to minimize the time spent conducting experiments and to improve dye synthesis rates.