Accelerated discovery of new organic photovoltaic dyes for OPVs using light absorbance as the primary screening criterion via machine learning and DFT

Abstract

This research provides an analysis of the light absorption properties of organic dyes in different organic solvents. By employing state-of-the-art machine learning (ML) techniques, including multi-output Gaussian process regression and ensemble methods like XGBoost and Random Forest regressors, we successfully predicted solvent-specific absorbance characteristics. XGBoost demonstrated outstanding predictive efficiency, and interpretation via SHapley Additive exPlanations (SHAP) analysis identified the topological polar surface area as the most critical molecular descriptor. For the de novo design of novel dyes, we developed a Transformer-Assisted Orientation (TAO) approach, generating three iterative rounds of new structures. The photovoltaic potential of these newly designed dyes was validated through density functional theory (DFT) and time-dependent DFT (TD-DFT) calculations. Geometry optimizations and electronic property calculations were performed at the ωB97XD/LanL2DZ level, while electronic spectra were simulated using the CAM-B3LYP/6-31G+(d,p) method with a polarizable continuum model (PCM) for acetonitrile. This integrated ML/DFT pipeline yielded dyes with remarkable predicted photovoltaic parameters, including a peak open-circuit voltage (V_oc) of 0.96 V, a light harvesting efficiency (LHE) of 95%, a fill factor (FF) of 0.87, and a short-circuit current density (J_sc) of 28.75 mA cm⁻². This study establishes a robust, data-driven framework for the rapid discovery and design of high-performance organic photovoltaic materials.