Enhancing the prediction of TADF emitter properties using Δ-machine learning: A hybrid semi-empirical and deep tensor neural network approach

This study presents a machine learning (ML)-augmented framework for accurately predicting excited-state properties critical to thermally activated delayed fluorescence (TADF) emitters. By integrating the computational efficiency of semi-empirical PPP+CIS theory with a Δ-ML approach, the model overcomes the inherent limitations of PPP+CIS in predicting key properties, including singlet (S<inf>1</inf>) and triplet (T<inf>1</inf>) energies, singlet-triplet gaps (ΔE<inf>ST</inf>), and oscillator strength (f). The model demonstrated exceptional accuracy across datasets of varying sizes and diverse molecular features, notably excelling in predicting oscillator strength and ΔE<inf>ST</inf> values, including negative regions relevant to TADF molecules with inverted S<inf>1</inf>-T<inf>1</inf> gaps. This work highlights the synergy between physics-inspired models and machine learning in accelerating the design of efficient TADF emitters, providing a foundation for future studies on complex systems and advanced functional materials.