Mathematical Modeling and Analysis
Mathematical Modeling and Analysis
Predictability of complex phenomena in physical systems with uncertain (under- determined by data) parameters is of central importance for many LANL programs. In practical applications, system parameters are often sampled at selected locations and their values elsewhere on the numerical grid are inferred through interpolation techniques, such as Kriging. This results in parameter distributions that are often much smoother than is realistic. We aim to replace the currently used interpolation techniques with an approach that uses risk-based optimization to populate the parameter space. Unlike traditional approaches that estimate parameter distributions without regard for critical behavior of a system, our parameter estimation approach can yield parameter distributions that correspond to system failure.
