Flat-histogram algorithms: optimal parameters and extended application

Abstract

We provide analysis of the convergence properties and applicability extensions of flat-histogram algorithms, with a particular focus on the Wang-Landau algorithms (exemplified by converging stochastic approximation Monte Carlo (SAMC)) and multicanonical (MUCA) algorithms. Our investigation reveals that the optimal decay rate of the modification factor in SAMC algorithms is influenced by the number of energy bins rather than the width of the energy range. Despite the frequent naming of these algorithms based on the histogram flatness, our findings indicate that flatness demonstrates a limited correlation with estimation accuracy. We explore the implications of integrating the importance sampling technique with flat-histogram algorithms, demonstrating that this combination yields comparable or better accuracy in density of states estimations, almost independent of specific algorithmic parameters within certain bounds. Furthermore, our research extends the possibilities of the flat-histogram and importance sampling combination for investigating a range of underlying system parameters simultaneously within a single simulation. These system parameters could both originate from the potential, e.g., various relative contributions of different energy terms or characteristic interaction range, and characterize the accessible configurations, e.g., through the size of the simulation box.