From high-dimensional committors to reactive insights.

Area of Science:

• Physical Chemistry
• Computational Chemistry
• Statistical Mechanics

Background:

• Transition path theory (TPT) is a key framework for analyzing rare dynamical transitions between metastable states.
• High-dimensional systems pose challenges for TPT due to the curse of dimensionality.
• Advanced function approximators like neural networks show potential for computing TPT's core object, the committor function.

Purpose of the Study:

• To develop a method for extracting mechanistic insights from high-dimensional committor functions.
• To illustrate a straightforward yet powerful approach for tracking individual dynamical coordinate evolution during reactive events.

Main Methods:

• Utilizing high-dimensional function approximators (e.g., neural networks) to compute the committor function.
• Developing a strategy based on marginalizing the reactive ensemble.
• Analyzing the evolution of individual dynamical coordinate distributions during transitions.

Main Results:

• A novel method for tracking individual dynamical coordinate evolution during reactive events is presented.
• The approach effectively captures the distribution's evolution, not just its mean.
• Demonstrates the utility of high-dimensional committor functions for mechanistic analysis.

Conclusions:

• The proposed method provides a powerful way to gain mechanistic understanding from complex, high-dimensional systems.
• Marginalizing the reactive ensemble offers a natural way to analyze coordinate dynamics.
• This work extends the applicability of TPT in high-dimensional regimes.