The committor function is a central object in transition path theory: for a stochastic system moving between two metastable states, it gives the probability that the system first reaches one state rather than the other. Researchers use it to analyze rare transitions in physical, chemical, and biological systems.

The new preprint introduces an iterative neural variational solver for this function in the context of under-damped Langevin dynamics. The approach combines neural network representations with a variational formulation, and the iteration is designed to refine the solution. The abstract does not provide implementation details or numerical results.

This is a single-source digest, so there are no independent results to compare. The contribution is methodological: it extends neural variational techniques to a dynamical setting where the committor function is typically hard to compute. Readers interested in the specifics will need to consult the full paper.