Monte Carlo Sampling with Layered Auxiliary Potentials

In this talk, we describe two new Monte Carlo schemes, Cascading Auxiliary Potential Sampling (CAPS) and Rosenbluth-CAPS, that accelerate the computation of equilibrium system properties. The fundamental premise of both approaches is to generate a Markov chain of configurations for a ``true’’ potential energy surface (PES) using random walks on auxiliary potentials and an appropriate acceptance criterion. The power of the methods derive from the choice of auxiliary potentials, which may reduce the cost of the calculation and/or enable more efficient exploration of phase space; when the auxiliary potentials are equivalent to the ``true’’ PES, the methodologies reduce to standard MC sampling. We demonstrate the application of these schemes and discuss their performance in a series of both toy and molecular systems. It is shown that CAPS and RCAPS are competitive with other sampling methods, such as parallel tempering and metadynamics, but CAPS and RCAPS afford a degree of flexibility that make them uniquely suited to some applications for which other enhanced sampling methods are less obvious. Importantly, because CAPS and RCAPS simply generate configurations according to the Boltzmann distribution, they can be further complemented by existing enhanced sampling strategies.