Extrapolative Analysis of Fast-Switching Free Energy Estimates in a Molecular System

Abstract

We perform an extrapolative analysis of "fast-growth" free-energy-difference (DF) estimates of a computer-modeled, fully-solvated ethane<->methanol transformation. The results suggest that extrapolation can greatly reduce the systematic error in DF estimated from a small number of very fast switches. Our extrapolation procedure uses block-averages of finite-data estimates, and appears to be particularly useful for broad, non-Gaussian distributions of data which produce substantial systematic errors with insufficient data. In every tested case, the extrapolative results were better than direct estimates.