Publications

A. Kannan and S. M. Wild, "Benefits of Deeper Analysis in Simulation-Based Groundwater Optimization Problems," Preprint ANL/MCS-P2043-0212, February 2012. [pdf]

A common dilemma in many decision, design, and modeling problems involving water resources is determining the level of simulation output to be passed to an optimization solver. A pure black-box approach allows for the easiest interface between the solver and the underlying simulator(s) but often requires a large number of simulation evaluations. On the other hand, full knowledge of the underlying computation enables the use of more specialized solvers but is labor intensive, increases a solver overhead, and can be intractable or unrealistic. We explore trade offs occurring between these two extremes on a groundwater problem based on the Lockwood Solvent Groundwater Plume Site. We make explicit use of knowledge about the problem nonsmoothness and additional outputs from the underlying simulator. We propose an augmented Lagrangian framework to solve the resulting problem under three different levels of information. Our results show both the benet of working with richer output from the underlying simulator and the trade offs between computational complexity and accuracy with this increase in information.