Item

Abstract

This article looks at Skilling’s nested sampling from a physical perspective and interprets it as a microcanonical demon algorithm. Using key quantities of statistical physics we investigate the performance of nested sampling on complex systems such as Ising, Potts and protein models. We show that releasing multiple demons helps to smooth the truncated prior and eases sampling from it because the demons keep the particle off the constraint boundary. For continuous systems it is straightforward to extend this approach and formulate a phase space version of nested sampling that benefits from correlated explorations guided by Hamiltonian dynamics.