Determination of the chemical potential using energy-biased sampling

摘要

An energy-biased method to evaluate ensemble averages requiring test-particle insertion is presented.The method is based on biasing the sampling within the subdomains of the test-particle configurational space with energies smaller than a given value freely assigned.These energy wells are located via unbiased random insertion over the whole configurational space and are sampled using the so-called Hit-and-Run algorithm,which uniformly samples compact regions of any shape immersed in a space of arbitrary dimensions.Because the bias is defined in terms of the energy landscape it can be exactly corrected to obtain the unbiased distribution.The test-particle energy distribution is then combined with the Bennett relation for the evaluation of the chemical potential.We apply this protocol to a system with relatively small probability of low-energy test-particle insertion,liquid argon at high density and low temperature,and show that the energy-biased Bennett method is around five times more efficient than the standard Bennett method.A similar performance gain is observed in the reconstruction of the energy distribution.