Predicting experimentally measured optical gap of conjugated polymers with DFT and machine learning

Bin Liu, Yunrui Yan, Mingjie Liu*

University of florida

Conjugated polymers (CPs), characterized by their alternating σ and π bonds, have garnered significant attention for their diverse structures and adjustable electronic properties. However, achieving optimal electronic properties, particularly the optical gap (E_opt^exp ) of CPs with high efficiency and accuracy remains challenging due to the expansive chemical space. This study introduces a rational model that combines density functional theory (DFT) calculations and data-driven machine learning (ML) approach to predict the experimentally measured E_opt^exp of CPs. We demonstrated that by removing alkyl side chains and extending the conjugated backbone, the obtained oligomer structure can effectively capture the electronic properties of a CP, significantly improving the correlation between the DFT-calculated HOMO-LUMO gap (E_gap^DFT) and E_opt^exp (R2=0.51). In contrast, E_gap^DFT calculated from the monomer units of CPs showed a much lower correlation (R2=0.15). To further improve the prediction accuracy of E_opt^exp, we trained various conventional ML models using E_gap^DFT from oligomers and multiple molecular representations obtained from monomer units. The HBGR model emerged as the most effective, achieving an R² of 0.755 and an RMSE of 0.1 eV in predicting E_opt^exp. Remarkably, when applied to a new, unseen CP dataset, our model exhibited superior transferability compared to baseline models that relied solely on molecular representations. For the first time, we demonstrated a novel and effective strategy for predicting the experimentally measured fundamental properties of CPs, paving the way for the accelerated design and development of high-performance CPs in photoelectronic applications.