Parallel Monte Carlo simulations using a residence weight algorithm -: art. no. 016701

A parallel Monte Carlo (MC) algorithm, the residence weight algorithm, is proposed. This algorithm is an extension to continuous ensembles of the residence time algorithm. The proposed algorithm consists of generating several trial MC moves in parallel, selecting one of them using appropriate relative probabilities and correcting the non-Boltzmannian sampling procedure by means of appropriate configuration weights. The corresponding parallel Boltzmannian scheme is based on a configurational-bias Monte Carlo scheme and will also be considered. The efficiency of both parallel algorithms has been compared in a case study: the slow relaxation dynamics in a model silicate. For a small number of trial moves generated in parallel, we observe that the residence weight algorithm performs less efficiently than its corresponding configurational-bias scheme by a factor of about 2. However, when the number of parallel trial moves increases and becomes larger than ten, we observe that the residence weight algorithm and the corresponding configurational-bias scheme perform with equivalent efficiencies.