Efficient simulated tempering with approximated weights: Applications to first-order phase transitions

Simulated tempering (ST) has attracted a great deal of attention in the last years, due to its capability to allow systems with complex dynamics to escape from regions separated by large entropic barriers. However its performance is strongly dependent on basic ingredients, such as the choice of the set of temperatures and their associated weights. Since the weight evaluations are not trivial tasks, an alternative approximated approach was proposed by Park and Pande (2007) to circumvent this difficulty. Here we present a detailed study about this procedure by comparing its performance with exact (free-energy) weights and other methods, its dependence on the total replica number R and on the temperature set. The ideas above are analyzed in four distinct lattice models presenting strong first-order phase transitions, hence constituting ideal examples in which the performance of algorithm is fundamental. In all cases, our results reveal that approximated weights work properly in the regime of larger R's. On the other hand, for sufficiently small R its performance is reduced and the systems do not cross properly the free-energy barriers. Finally, for estimating reliable temperature sets, we consider a simple protocol proposed by Valentim et al. (2014).