Efficacy of Computational Models of Dense Plasmas

Computational models must balance physics fidelity with computational cost. Because many important applications cannot be modeled with the highest-fidelity models, it is important to assess boundaries in parameter space for which lower-fidelity models still provide useful information other-wise unobtainable. Here, we perform a metastudy in which data from a wide range of computational models used in the high energy-density physics community is examined to reveal physics regimes in which they confer little advantage over simpler models. Model fidelity is measured by comparing high-fidelity predictions with new predictions from two very simple pair potential models. Error metrics are defined, and patterns in the data are sought. This data-driven approach reveals the surprising result that simpler models become applicable not because of higher temperature and/or lower density, but rather based on relative ionization level $\langle Z\rangle/Z$. Moreover, we find that the simpler models tend to fail abruptly as the role of atomic and molecular physics plays an increasing role, suggesting a fairly narrow “transition” between residual chemistry and disordered plasma behavior.