Towards High-Fidelity Multi-Scale Modeling: A Study in Copper

Leveraging Bayesian inference a new uncertainty-aware equation of state model for copper has been constructed. This multiphase EOS is constrained via diamond-anvil-cell, Hugoniot, melt, speed of sound, vaporization, DFT-MD, and electrical conductivity measurements. To model the electrical response, we employ the Lee-More-Desjarlais (LMD) framework, trained on both expanded and compressed DFT-MD data alongside experimental isobaric measurements. Leveraging the DFT-MD EOS training data a machine-learned interatomic potential (MLIAP) is constructed using a robust EOS-reinforced genetic algorithm training procedure. The trained MLIAP demonstrates stability and transferability across a wide range of densities/temperatures, showing good agreement with experimental speed of sound measurements and Hugoniot data, while effectively stabilizing the experimentally confirmed high-pressure body-centered cubic (BCC) phase in shock. Such coupled EOS-atomistic approaches are critical for novel scale-bridging modeling frameworks that are aimed at resolving complex physics in more exotic warm-dense matter regimes.