Title (Oral): Data-Driven Discovery of Nonlinear Optical Properties in Two-Dimensional Materials

We search for van der Waals (vdW) materials that exhibit nonlinear optical (NLO) phenomena using data-driven techniques that integrate density functional theory (DFT) and machine learning (ML). Our goal is to accelerate the discovery of stable vdW materials exhibiting significant second-order susceptibility, a key property for NLO applications.

Our study utilizes the Computational 2D Materials Database (C2DB), from which we extract second-harmonic generation (SHG) spectra to build a dataset for training and evaluating ML models. We employ a random forest classifier, to predict whether a given material exhibits NLO properties, followed by a regression model, to estimate the second-order susceptibility. We find that the choice of material descriptors, derived from atomic properties, plays a critical role in model accuracy. In addition, ML creates avenues to gain physical insight into the behavior of vdW materials with nonlinear optical properties. Our work paves the way for the rapid identification of promising candidates for photonic and optoelectronic applications.