Erroneous Rates and False Statistical Confirmations from Infrequent Metadynamics and Other Equivalent Violations of...

Recent combinations of hyperdynamics and adaptive biasing potentials or free-energy-based bias potentials introduce an implicit assumption: the bias potential needs only be zero on the prominent free-energy barriers of interest. Herein we demonstrate that this implicit assumption is flawed. Thus, hyperdynamics in collective variables is likely to fail unless there are no hidden barriers within the initial state. Moreoever, the one-sample Kolmogorov-Smirnov test used to declare "trust" in hyperdynamics is shown to be unable to classify this failure. In fact, nonzero bias potential on hidden barriers emerges as a contradiction of the hypothesis that a Poisson distributed set of hyperdynamics escape times is a correctly distributed set of escape times. We demonstrate failure in the alanine dipeptide benchmark, and we present an apparent failure for a protein conformation change. The standards for "trustworthy" hyperdynamics in collective variables must be raised.