Description

Abstract

Wang-Landau sampling method to an annealed version and present a new annealed Wang-Landau Monte Carlo (AWLMC)-based tracking method . When the Wang Landau sampling method obtains diverse samples in a whole state space, it needs a  vast number of samples. To enhance the efficiency of the sampling process, our method sequentially reduces the state space into a smaller one, which contains the target states compactly. We utilize the marginal likelihood and DOS information to reduce the state space. We extend the WLMC-based tracking method into a new N-Fold Wang-Landau (NFWL)-based one to maintain good performance when the dimension of the state space increases. The performance of the WLMC-based tracking method and its annealed version depends on the accuracy of the DOS estimate. However, given a fixed number of samples, the accuracy substantially decreases as the dimension of the state space increases.To preserve good performance in the high-dimensional state space, we adopt the N-Fold way algorithm  which can estimate the DOS with a very small number of samples. Moreover, the NFWL-based tracking method is a rejection-free algorithm. It always accepts the proposed states using its efficient proposal density. This property also enables the method to need a smaller number of samples.