Modeling the Nonradiative Decay of Cyclooctatetrathiophene in Solution and Crystal Phases Using Machine Learning

Modeling excited-state dynamics of molecular aggregates via multiscale calculations remains challenging due to the resource-intensive excited-state electronic structure calculations. This study presents a direct combination of machine learning (ML)-accelerated excited-state calculations and semiempirical-level molecular mechanics methods, specifically GFN-FF, within the ONIOM scheme. The ML-photodynamics simulations reveal the critical role of neighboring π_CC and σ_CC bond torsion in controlling the nonradiative decay of the well-known aggregation-induced emission (AIE) molecule cyclooctatetrathiophene (COTh) in THF solution and COTh crystal. The predicted fluorescence quantum yield enhancement factors, ranging from 3- to 22-fold in the solution to the crystal, are in good agreement with the experimental results. The trajectory analysis revealed increasing restrictions on the π_CC bond torsion of COTh from the THF solution to the COTh crystal, which blocked the nonradiative decay pathways. Our approach provides a quantitative understanding of the AIE mechanisms of COTh and is expected to be applied in the rational design of AIE materials in the future.