Speaker
Description
Rare transitions between long-lived states underlie many important processes in condensed matter, from nucleation to protein folding, yet their simulation remains challenging due to the vast separation of timescales involved. In this talk, I will discuss how machine learning can enhance rare event simulations. A central theme is the committor function, the ideal reaction coordinate, which can be iteratively learned from transition path sampling data and used to guide the generation of new reactive trajectories. Symbolic regression then distills interpretable analytical expressions that reveal the relevant collective variables driving the transition. Identifying the few collective variables that matter is essentially a problem of physically informed dimensionality reduction. Applied to ice nucleation, this approach shows that nucleus size alone provides a non-Markovian description of the process, while structural descriptors such as crystalline order and tetrahedrality yield an improved reaction coordinate. I will also discuss how conditioned normalizing flows can generate independent shooting points for transition path sampling, removing correlations between sampled paths and enabling parallelization of the sampling process.