Implementation of the replica-exchange Wang-Landau sampling on Graphics Processing Units (GPUs)

The present study focuses on the GPU based implementation of the replica-exchange Wang-Landau sampling. We update the density of states periodically, which leads to better accuracy. To validate our algorithm, we perform simulations for the 2D Potts model with linear sizes ranging from 20 up to 200. We investigate also the performance of our GPU code compared with the single walker Wang-Landau algorithm. GPU implementation speedups of up to 90 were achieved.