Entropy estimation from simulations of non-diffusive systems

Abstract

Simulations generating a statistical ensemble of configurations of a molecular system provide information to derive entropy. For cases where the spatial distribution is bounded, the entropy can be derived from the average of the logarithm of the spatial probability density, but will be subject to systematic errors due to statistical fluctuations. A method is given to eliminate these errors by using different interval sizes and fitting the results to a simple theoretical expression. The method is feasible when the multidimensional case can be broken up into a sum of low dimensional contributions. The validity of the latter approximation and ways to correct for correlations are discussed.