This work has been published in JMMM.
The paper shows how the classical and constrained Monte Carlo algorithms can be parallelized and implemented for GPUs using domain decomposition. The method is applicable not only to simple cubic lattices, but also for realistic cubic lattices (bcc, fcc, hcp etc.) as well as any number of exchange neighbours. The work has extensively tested numerically the parallelized algorithms for calculation of temperature dependences of various magnetic parameters (magnetization, exchange stiffness, anisotropy constants, susceptibilities).