Accelerating DFT Simulations for Uncertainty Quantification in Nuclear Structure

Many nuclei that participate in astrophysical processes, such as the rapid neutron capture process, are outside of experimental reach. As such, theoretical predictions with quantified uncertainties are required [1]. Density functional theory (DFT) predicts structure properties such as masses and radii, and provides inputs to decay rate calculations. Further, DFT is able to make these predictions across the nuclear landscape, although predictions such as fission yields still require model emulators [2]. Bayesian uncertainty quantification (UQ) requires hundreds-of-thousands of calculations per nucleus, making emulators even more important. I will discuss an emulation strategy based on reduced order modeling techniques that have recently been gaining popularity in the field [3]. I will show that emulator errors on the order of 0.01% in binding energies are achievable, with at least an order-of-magnitude reduction in the calculation time, using as few as 100 training points for the emulator. As such, these emulation techniques are useful for future UQ efforts in DFT.

[1] M.R. Mumpower et al., doi:10.1016/j.ppnp.2015.09.001

[2] D. Lay et al., doi:10.1103/PhysRevC.109.044305

[3] E. Bonilla et. al., doi:10.1103/PhysRevC.106.054322