My research on transition path sampling addressed the problem of simulating the trajectories of molecular systems during reactions. Development of a new level of understanding of the mathematics of such systems and their simulations. Applied the mathematics to devise a new machine-learning powered algorithm that overcomes previous limitations.
One of the most useful metrics of an algorithm’s performance is the relative entropy between its result and the desired answer, which measure the information difference between them. My co-author and I developed the first method for fully computing the relative entropy when estimating the paths of transition processes. We demonstrated that our result can be used to train models that sample from the transition paths, a significant contribution to the field.