Exploring the enantioselective synthesis mechanism of ammonium cations in solution using deep learning potential

Asymmetric synthesis is fundamental to modern organic chemistry. Predicting the stereoselectivity of catalytic reactions in solution remains a challenging problem, as it requires a comprehensive understanding of the underlying mechanistic, energetic, and kinetic factors. To address this challenge, we propose an active learning workflow that integrates iterative cycles of <i>ab initio</i> molecular dynamics (AIMD) and deep learning potential molecular dynamics (DLPMD) simulations. To demonstrate its utility, this workflow is applied to investigate the enantioselective synthesis of quaternary ammonium cations catalyzed by 1,1'-bi-2-naphthol scaffolds (BINOL), with a focus on simulations of large molecular systems over extended timescales. The results of the simulations successfully reproduce the experimentally observed chirality of the major product molecules, as confirmed by ¹HNMR spectroscopy. Furthermore, these simulations provide detailed insight into the reaction pathways and reveal that the chirality of the primary product is thermodynamically controlled under experimental conditions. Consequently, this workflow offers a promising strategy for exploring complex reaction mechanisms and enhancing the predictive accuracy of asymmetric synthesis in complex solutions using deep learning techniques.