Development of a Ta Machine Learned Interatomic Potential: towards molecular dynamics simulations of Z

Machine Learned Interatomic Potentials (MLIPs) have introduced unprecedented levels of accuracy into molecular dynamics (MD) simulations. However, ensuring that accuracy is consistent across wide ranges of phase space (temperature, pressure) is still a major challenge. To address this challenge, we apply in-house optimization methods to the development of a Ta MLIP. Previous Tri-lab Ta strength studies including those at Sandia National Laboratories' Z Pulsed Power Facility, referred to simply as 'Z', provide a wealth of experimental data for comparison with MLIP properties and performance. We will discuss our strategies for training set development and hyperparameter optimization toward physical MLIP behavior, high accuracy, and agreement with experiment. Molecular Dynamics simulations will demonstrate how the potential responds to temperature and pressure changes towards describing behavior on Z.