Simulation of Spin Models in Multicanonical Ensemble with Collective Updates

Abstract

We propose a Monte Carlo method which performs a random walk in energy space using cluster-like collective updates. By imposing that bond probabilities depend continuously on the microcanonical temperature, we obtain dynamic exponents close to their ideal random walk values. The method proves remarkably powerful when applied to models governed by long-range interactions, where it straightforwardly combines with the efficient Luijten-Bloete cluster algorithm to yield a dramatic reduction in the computation load.