Recovering From Structural Instability in Nuclear Power Design Optimization

Abstract

The Scientific Method—parsing a problem into isolated subproblems—is often necessarily employed in optimization efforts to reduce large and complex problems into more tractable parcels. However, adopting an ersatz action space relies on the assumption that the dependencies between each subproblem are weakly coupled. This paper first illustrates the not-insignificant distortion of objective space topology that can occur when optimizing over an ersatz action space for a lattice-scale problem. We then develop a novel method to recover from this distortion and loss of any optima therein. Sequential design and decision-making practice needs to be aware of and take steps to mitigate this potential pitfall.