We have a working code which is capable of performing classical MC simulations on a variety of condensed matter systems, including molecules with several constitutents, such as the TIP5P water model. Tradiational Metropolis MC updates are implemented as well as a parallel temerping algorithm. We do not consider the code fully verified and validated at this point; more testing and debugging is required to have full confidence in the results. However, current results are reasonable, and the code is amenable to extensions and enhancements that should make it useful for future work.
We have presented preliminary results on two areas: strategies for efficient paralleization of the parallel tempering algorithm and application to liquid water.
Scheduleing in most cases can not lower simulation runtime. I is useful when there are more replics then cpus. The scheduler is complicated, but prefered when number of replicas is not too much larger then the number of availible cpus. When the number of replicas is much larger it is possible to use simpler on-demand scheduleing.
As for water simulations, we repeat that results are preliminary and should be taken as suggestive, not definitive. Nonetheless, we have reason to believe that parallel tempering may prove useful in speeding up simulations of complex liquids like water. Our results indicate, on one hand, that parallel tempering can significantly improve equilibration of a system, such as water, with many local minima that would otherwise constrain the simulation. Additionally, we hope to draw a conclusion about the ability of parallel tempering to reduce the autocorrelation time of a simulation. Present results are inconclusive.