This is a fork of the package SpinMC.jl. It adds a recursive tuning of the parallel tempering temperature distribution that converges to a set of temperatures where exchange acceptance rates between replicas are approximately the same (as described in "Markov Chain Monte Carlo Simulations and Their Statistical Analysis", Section 6.3).
Usage of the package remains completly the same (refer to the original package), to tune parallel tempering temperature set add the following kwargs to the MonteCarlo constructor:
recursionIterations: number of iterations the recursive tuning algorithm will perform, ideally, the point where the temperature set has converged (default: 0).recursionRate: Rate at which the temperatures will be tuned, in units of sweeps (default: 1000).
Experimentation with these parameters is required for every specific problem, as these kind of algorithms can easily become unstable.