Often realistic models in chemical compound space may be multi-scale, and require the use of high-performance computing clusters for their evaluation. Predictive calculations, such as a rigorous sensitivity analysis of materials design parameters, amount to a global optimization over design parameter space -- potentially requiring an already prohibitively large simulation to be performed hundreds, if not thousands, of times. Massively parallel, distributed, and asynchronously coupled computational methods must be developed to help minimize the computational time required to produce accurate predictions on the structure and properties of all but the simplest of materials. The need to prepare, schedule, and monitor thousands of model evaluations, and interactively explore and analyze data is a challenging problem that requires a sophisticated software framework capable of distributing and managing computations on large-scale heterogeneous resources. I will present an optimization framework, and also a framework for heterogeneous computing, that when utilized together, can make computationally intractable sensitivity and optimization problems much more tractable. The optimization framework provides global search algorithms that have been extended to parallel, where evaluations of the model can be distributed to appropriate large-scale resources, while the optimizer centrally manages their interactions and navigates the objective function. Additionally, new algorithms have been developed that launch multiple optimizers in parallel, thus allowing highly efficient local search algorithms to provide fast global optimization. In this way, parallelism in optimization also can allow us to not only find global minima, but to simultaneously find all local minima and transition points – thus providing a much more efficient means of mapping out a potential energy surface.
Copyright. All Rights Reserved.