Variational deep learning of equilibrium transition path ensembles.

Area of Science:

• Chemical Physics
• Computational Chemistry
• Statistical Mechanics

Background:

• Understanding rare chemical events is crucial in chemistry and materials science.
• Traditional methods struggle with efficiently calculating reaction rates for complex systems.

Purpose of the Study:

• To present a novel time-dependent variational method for learning reaction mechanisms.
• To efficiently evaluate reaction rates within a transition path ensemble.
• To gain insights into the dynamics of rare reactive events.

Main Methods:

• Utilizing a time-dependent variational approach with a neural network ansatz.
• Approximating the time-dependent commitment probability.
• Decomposing the reaction rate based on stochastic path action components.

Main Results:

• The method successfully infers reaction mechanisms and provides quantitatively accurate rate estimates.
• Accurate rates were obtained with minimal trajectory data across various systems.
• The decomposition method elucidates the contribution of reactive modes and their couplings.

Conclusions:

• The developed method offers an efficient and insightful approach to studying reactive processes.
• It provides a systematically improvable framework for rate evaluation.
• The technique offers unique insights into rare event transitions through commitment probability analysis.