DeepEMs-25: a deep-learning potential to decipher kinetic tug-of-war dictating thermal stability in energetic materials

Abstract

Atomic-scale insight into decompositions in energetic materials (EMs) is essential for harnessing energy release, which remains elusive due to both instrumental and computational limitations. Herein, we developed DeepEMs-25, a deep-learning potential trained on diverse EMs towards accurate and efficient simulations. Applying DeepEMs‑25 to an isostructural ABX₃ molecular perovskites series, with A-site organic cations, B-site alkali or ammonium cations, and X-site perchlorate anions, we probe the effect of cation size on reactivity. Arrhenius analysis of 100-ps trajectories reveals that increasing B‑site ionic radius simultaneously decreases X–A collision’s activation energy (enhancing reaction rates) and decreases X–A collision’s pre‑exponential factor (reducing collision frequency), producing opposing kinetic effects. Such “kinetic tug‑of‑war” explains why an intermediate‑sized cation yields maximal thermal stability by optimally balancing reactivity and collision dissipation. A similarly sized reactive cation promotes additional hydrogen-transfer pathways causing accelerating decomposition. Our findings link atomistic kinetics to macroscopic stability, informing next-generation EMs design.