Impact of MCMC convergence behavior on MMC parallelization

Multicanonical Monte Carlo (MMC) is a technique to accelerate simulations by using adaptive importance sampling (IS). Because the adaption algorithm is system independent, MMC is a practical, handy tool that can be used in many situations and in many fields of research. Combination of MMC with supercomputer infrastructures can support increasingly complex systems. A supercomputer works with parallelized algorithms. Efficient MMC relies on Markov Chain Monte Carlo (MCMC) as a key enabler in generating samples from biased distributions. While Monte Carlo simulations are embarrassingly parallelizable, MCMC is inherently serial in nature and a priori difficult to parallelize. In this article, we will examine three diverse systems to explore how MMC can benefit from parallelization. We will uncover some hints to parameterize the parallel algorithm to compromise between speed and accuracy.