Sobolev Sampling of Free Energy Landscapes

Abstract

A family of fast sampling methods is introduced here for molecular simulations of systems having rugged free energy landscapes. The methods represent a generalization of a strategy consisting of adjusting a model for the free energy as a function of one- or more collective variables as a simulation proceeds. Such a model is gradually built as a system evolves through phase space from both the frequency of visits to distinct states and generalized force estimates corresponding to such states. A common feature of the methods is that the underlying functional models and their gradients are easily expressed in terms of the same parameters, thereby providing faster and more effective fitting of the model from simulation data than other available sampling techniques. They also eliminate the need to train simultaneously separate neural networks, while retaining the advantage of generating smooth and continuous functional estimates that enable biasing outside the support grid. Implementation of the methods is relatively simple and, more importantly, they are found to provide gains of up to several orders of magnitude in computational efficiency over existing approaches.