Chemical Kinetics take place on timescales several orders of magnitude larger than atomic motions. This can make direct estimation of rates prohibitively expensive. We introduce a new-data driven algorithm, Dynamical Galerkin Approximation, that allows chemical rates to be extracted using only short quantities. This work justifies and generalizes MSM estimates of dynamical statistics such as mean-first-passage-times and committors. This work also requires new approaches for dealing with loss of Markovianity, leading us to introduce Delay Embedding into this new context. Time permitting, we will also discuss recent work on error analysis and lag-time selection for VAC.
Back to Machine Learning for Physics and the Physics of Learning