Analysis of influencing factors and the most probable transition pathway in the narrow escape problem for molecular systems based on deep learning method

This study employs physics-informed neural networks (PINNs) to investigate the narrow escape problem in irregular domains, aiming to understand how key parameters influence molecular escape behavior and to analyze the most probable transition pathway of molecules. We focus on two critical metrics: mean exit time and escape probability, characterizing escape behavior in stochastic systems. Using PINNs, we effectively address the domain's complexities and examine the effects of parameters such as diffusion coefficient, angular velocity, annular area, and absorption domain size on mean exit time and escape probability. Moreover, by computing the most probable transition pathway, we further uncovered the underlying mechanisms that govern molecular motion in complex environments. An interesting observation was that increasing the diffusion coefficient expanded the high-probability escape region but decreased the overall escape probability. The results provide valuable insights for optimizing escape efficiency in practical applications and highlight the potential of PINNs for studying complex diffusion problems.