• Nuclear, Particle, Astroparticle and Cosmology (NUPAC) Seminars

September 29, 2026 2:00 PM
PAIS 3205

Host:

Tim Dolch

Presenter:

Ryan Curry (LANL)

Computational approaches to solving the many-body Schrödinger equation have reached the level where they can reliably be used to compute observables to within sub-percent accuracy. At the same time, modern descriptions of nuclear forces, such as those derived from chiral EFT, carry inherent uncertainties that must be accounted for in many-body calculations. The natural approach for such a task is the use of Bayesian statistics, which allow for the propagation of uncertainties through to the final posteriors. However, most many-body methods do not lend themselves easily to Bayesian inference, as the thousands of likelihood evaluations required quickly exhaust available computational resources. One possible solution to this bottleneck is the use of machine learning surrogate models, or emulators, which can accurately reproduce the results of high-fidelity many-body calculations, for a small fraction of the computational cost. Today, I will discuss recent work developing a framework that uses emulators for few- and many-body calculations to perform Bayesian inference of the low-energy couplings from chiral EFT and then, importantly, to propagate the theoretical and experimental uncertainties from the level of the interaction to the final many-body predictions.