Development and Bayesian Uncertainty Quantification of Coarse-Grained Models of Metals Based on Embedded Atom Method Potentials

Coarse-grained (CG) molecular dynamics (MD) simulations have emerged as a powerful and cost-effective approach for modeling materials by simplifying atomic structures into CG beads. However, accurately parametrizing interatomic potential models (force fields, FFs) that can reliably reproduce material properties and quantifying the uncertainties associated with both the model parameters and their predictions remains a major challenge. In this study, we developed coarse-grained embedded atom method (CG EAM) potentials to model interatomic interactions in face-centered cubic (FCC) metals, including palladium (Pd), gold (Au), silver (Ag), copper (Cu), and platinum (Pt). The CG EAM potentials combine the physical interpretability of a traditional EAM with the computational efficiency of coarse-graining. We first employed a Particle Swarm Optimization (PSO) framework integrated with CG MD simulations to explore a 14-dimensional parameter space and identify CG EAM parameters that reproduce key physical, mechanical, and thermodynamic properties, such as cohesive energy, lattice constants, and elastic moduli. These parameters were subsequently refined using a Bayesian uncertainty quantification (BUQ) approach, which allowed the systematic assessment of uncertainties in both the FF parameters and the predicted properties. For all five metals, this framework yielded robust parameter ranges within which the predicted properties generally remained within their 95% confidence intervals. Overall, this integrated parameter optimization and BUQ approach provides an effective strategy for developing accurate and reliable interatomic potentials while offering a generalizable framework for designing both hard and soft materials with targeted properties.