Application of Optimal Experimental Design to High-Energy-Density Physics

In many HED experiments quantities of interest are not measured directly. To address this, it’s common to compare measurable data with simulation or analytic model output and identify where the models are consistent with experimental results. These methods often require the use of an inverse problem. These approximations introduce uncertainty and make error propagation difficult. [1] Recently, Bayesian methods have been used to improve data analysis and experimental design (often with the goal of optimizing model output). While these methods have improved experimental outputs, e.g. yield, unknowns and uncertain physics remain.

Here, we use Optimal Experimental Design (OED) to maximize information gain on unknown physics. OED quantifies expected information gain, i.e. expected uncertainty from prior to posterior, of the unknown model parameters provided by an experiment and maximizes it across a design space. We apply OED to three analytical models whose physics is representative of integral systems present in HED. A shock compression study, an EOS study, and an RMI study. We define experimental designs and unknown model parameters. The estimated expected utility is calculated for each model and maximized across the design space. These optimal designs provide tighter posteriors on uncertain parameters which improves approximations on quantities of interest thus resulting in better constrained models.



[1] P. F. Knapp and W. E. Lewis, Review of Scientific Instruments 94, 061103 (2023).