Machine learning photodynamics uncover blocked non-radiative mechanisms in aggregation-induced emission

Aggregation-induced emission (AIE) is a photophysical phenomenon in which weakly luminescent organic chromophores become strongly luminescent in aggregate. The reduced non-radiative decay in aggregates is often cited as the explanation of the AIE. However, the mechanism of competing non-radiative decay pathways is not resolved due to the lack of excited-state structural information in the time-resolved experiments and prohibitively expensive quantum mechanical calculations for photodynamics simulations. We investigated the excited-state dynamics of classic AIE molecules in aggregate, hexaphenylsilole (HPS), tetraphenylsilole (TPS), and cyclooctatetrathiophene (COTh) with a multiscale machine learning accelerated photodynamics approach, integrating neural networks, semiempirical methods, and molecular mechanics. Our simulations predict 263-, 5-, and 12-fold fluorescence enhancement of HPS, TPS, and COTh in good agreement with the experiments (255, 3, and 12). We identified a shared non-radiative decay mechanism involving πCC torsions in these molecules. These torsions are blocked in aggregate due to intermolecular hindrance between substituents, promoting AIE.