High-throughput computation and machine learning of refractive index of polymers

Abstract

Refractive index (RI) of polymers plays a crucial role in the design of optoelectronic devices, including displays and image sensors. We have developed a framework for (1) high-throughput computation of RI values for computationally synthesized amorphous polymer structures based on a generalized polarizable reactive force-field (ReaxPQ+) model, which is orders-of-magnitude faster than quantum-mechanical methods; (2) prediction of composition–structure–RI relationships based on a machine-learning model based on graph attention neural network; and (3) computation of frequency-dependent RI combining ReaxPQ+ and Lorentz-oscillator models. The framework has been tested on a computational database of amorphous polymers.