Assessing the Reliability of the Dynamics Reconstructed from Metadynamics

Sampling a molecular process characterized by an activation free energy significantly larger than k(B)T is a well-known challenge in molecular dynamics simulations. In a recent work [Tiwary and Parrinello, Phys. Rev. Lett. 2013, 111, 230602], we have demonstrated that the transition times of activated molecular transformations can be computed from well-tempered metadynamics provided that no bias is deposited in the transition state region and that the set of collective variables chosen to enhance sampling does not display hysteresis. Ensuring though that these two criteria are met may not always be simple. Here we build on the fact that the times of escape from a long-lived metastable state obey Poisson statistics. This allows us to identify quantitative measures of trustworthiness of our calculation. We test our method on a few paradigmatic examples.