Speed-up of Monte Carlo simulations by sampling of rejected states

The Markov chain Monte Carlo method is an important tool to estimate the average properties of systems with a very large number of accessible states. This technique is used extensively in fields ranging from physics to genetics and economics. The rejection of trial configurations is a central ingredient in existing Markov chain Monte Carlo simulations. I argue that the efficiency of Monte Carlo simulations can be enhanced, sometimes dramatically, by properly sampling configurations that are normally rejected. This "waste-recycling" of microstates is useful in sampling schemes in which only one of a large set of trial configurations is accepted. It differs fundamentally from schemes that extract information about the density of macrostates from virtual Monte Carlo moves. As a simple illustration, I show that the method greatly improves the calculation of the order-parameter distribution of a two-dimensional Ising model. This method should enhance the efficiency of parallel Monte Carlo simulations significantly.