Δ-machine learning of triplet excitation energies in organic chromophores

Abstract

Here, we report a Δ-machine learning approach for predicting triplet excitation energies (T₁) of diverse organic chromophores by combining high quality reference data with quantum chemical calculations. A directed message passing neural network corrects TDDFT, ΔSCF and xTB/sTDA prediction to achieve near-chemical accuracy while substantially reducing computational cost. Notably, Δ-ML corrected xTB/sTDA predictions approach TDDFT-level accuracy, enabling rapid high-throughput screening of T₁ energies.