Abstract
Accurate and efficient prediction of the optical properties of two-dimensional (2D) materials is crucial for photonic applications but remains a challenging task due to discrepancies between theoretical and experimental approaches. Here, we present a physics-guided machine learning (ML) framework for accelerated screening of 2D materials. It combines first-principles density functional theory (DFT) calculations and graph neural network models backed by experimental spectroscopic validation and Cauchy-model integration. Within this framework, we collected a database of more than 1000 monolayers of transition metal dichalcogenides (TMDs) along with their optical properties. We also proposed a universal method for defining a physically meaningful thickness of two-dimensional structures, which enabled a correction of the optical properties obtained from PBE-based density functional theory. Using the collected database, we developed a Cauchy-model-based machine learning model to calculate refractive indices in the near-infrared (near-IR) region (755-1064 nm). The developed approach reflects correlations between the atomic structures of monolayers and their optical properties, which are confirmed by extensive testing against independent 2D materials databases. In this way, our ML-driven strategy offers a powerful tool for the rapid screening of novel monolayer materials with tailored optical functionalities, significantly accelerating the discovery and design of next-generation photonic materials. As an application, we further demonstrate how high-refractive-index candidates such as Bi(2)Te(2)Se enable enhanced field confinement and long crosstalk lengths in monolayer waveguides, highlighting their promise for integrated photonics.