Improved Estimation of Density of States for Monte Carlo Sampling via MBAR

摘要

We present a new method to calculate the density of states using the multistate Bennett acceptance ratio (MBAR) estimator. We use a combination of parallel tempering (PT) and multicanonical simulation to demonstrate the efficiency of our method in a statistical model of sampling from a two-dimensional normal mixture and also in a physical model of aggregation of lattice polymers. While MBAR has been commonly used for final estimation of thermodynamic properties, our numerical results show that the efficiency of estimation with our new approach, which uses MBAR as an intermediate step, often improves upon conventional use of MBAR. We also demonstrate that it can be beneficial in our method to use full PT samples for MBAR calculations in cases where simulation data exhibit long correlation.