Numerical Optimization of Computationally Expensive Functions at Argonne National Laboratory

The increase in computational resources in the past decades has resulted in computationally expensive numerical simulations appearing in nearly all scientific domains. Often times, domain scientists wish to identify inputs to the simulation that produce some form of optimal behavior. In this talk, we focus on algorithmic approaches for optimizing such computationally expensive functions, and highlight motivating problems from Argonne National Laboratory. No foreknowledge of numerical optimization is assumed; pictures will be used throughout the presentation.