Machine Learning Guided Discovery of Non-Linear Optical Materials

Abstract

Nonlinear optical(NLO) materials are crucial in achieving desired frequencies in solid-state lasers. So far, new NLO materials have been discovered using high-throughput calculations or chemical intuition. This study demonstrates the effectiveness of utilizing a high refractive index as a proxy for a high second harmonic generation(SHG) coefficient. It also emphasizes the importance of hardness in screening balanced NLO materials. Two machine learning models are developed to predict refractive indices and Vickers hardness. By applying these models to the OQMD database, potential NLO candidates are identified based on non-centrosymmetricity, refractive index, hardness value, and bandgap properties. These findings are validated using density functional theory(DFT) calculations. Notably, this approach successfully identifies several already established NLO materials, reinforcing the validity of the methodology.