Automated, Parallel Optimization Algorithms for Stochastic Functions

The optimization algorithms for stochastic functions are desired specifically for real-world and simulation applications where results are obtained from sampling, and contain experimental error or random noise. We have developed a series of stochastic optimization algorithms based on the well-known classical down hill simplex algorithm. Our parallel implementation of these optimization algorithms, using a framework called MW, is based on a master-worker architecture where each worker runs a massively parallel program. This parallel implementation allows the sampling to proceed independently on many processors as demonstrated by scaling up to more than 100 vertices and 300 cores. This framework is highly suitable for clusters with an ever increasing number of cores per node. The new algorithms have been successfully applied to the reparameterization of a model for liquid water, achieving thermodynamic and structural results for liquid water that are better than a standard model used in molecular simulations, with the the advantage of a fully automated parameterization process