Uncertainty estimation for molecular dynamics and sampling

Machine-learning models have emerged as a very effective strategy to sidestep time-consuming electronic-structure calculations, enabling accurate simulations of greater size, time scale, and complexity. Given the interpolative nature these models, reliability predictions depends on position in phase space, it is crucial obtain an estimate error that derives from finite number reference structures included during model training. When using machine-learning potential sample finite-temperature ensemble, uncertainty individual configurations translates into thermodynamic averages leads loss accuracy when simulation enters previously unexplored region. Here, we discuss how quantification can be used, together with baseline energy model, or more robust but less interatomic potential, resilient support active-learning strategies. Furthermore, introduce on-the-fly reweighing scheme makes possible extracted long trajectories. We present examples covering different types structural properties systems diverse water liquid gallium.