Title:Rethinking Metadynamics

Abstract:Arguably the mother of all enhanced sampling methods is umbrella sampling. In this approach, the exploration of the configurational space is expedited by adding to the potential energy a bias that is a function of few selected degrees of freedom. Soon after its introduction, in the so called adaptive umbrella sampling, it was suggested to build such bias adaptively. This strategy has been followed by many other methods. Among the many, metadynamics has recently found considerable fortune. In metadynamics the bias is constructed directly while in adaptive umbrella sampling the probability distribution is first reconstructed, and from this the bias is derived. Here we take this second point of view and first reconstruct the probability distribution of the collective variables using a kernel density estimation in a metadynamics-like framework. This has considerable advantages and leads to performances that are better than either adaptive umbrella sampling or metadynamics, while keeping the number of free parameters to a bare minimum. The new method comes in two variants. In the first, one aims at a very accurate estimation of the free energy, while in the second the main focus is on a rapid exploration of the free energy landscape. We demonstrate the performance of these two methods in a number of representative examples.