Fast and accurate emulators for nuclear scattering

Applications of reduced basis method emulators are increasing in nuclear physics. They enable fast and accurate sampling of parametric high-fidelity calculations, which is key for applying Bayesian statistical methods for calibration, sensitivity analyses, uncertainty propagation, and more. These emulators are trained using a selection of high-fidelity solutions, generically called snapshots, with parameters (e.g., low-energy couplings of a nuclear potential) chosen such that a subspace is spanned that accurately represents the relevant part of the full solution space. But how should one choose these training snapshots? In this talk, I will present an active learning approach that iteratively places training snapshots in the parameter space where the estimated error is the largest, thereby greedily minimizing the emulator’s global error. It enables emulator predictions with quantified errors and is broadly applicable to large-scale linear systems. I will discuss applications of this approach to nuclear two-body scattering and other recent developments that set the stage for Bayesian parameter estimation of chiral two- and three-body forces using fast and accurate scattering emulators.