Accelerating NOvA's Feldman-Cousins procedure using high performance computing platforms

Abstract

In order to assess the compatibility between models containing physically constrained parameters and small-signal data, uncertainties often must be calculated by Monte Carlo simulation to account for non-normally distributed errors. This is the case for neutrino oscillation experiments, where neutrino-matter weak interactions are rare and beam intensity at the far site is low. The NuMI Off-axis νe Appearance (NOvA) collaboration attempts to measure the parameters governing neutrino oscillations within the PMNS oscillation model by comparing model predictions to a small data set of neutrino interactions. To account for non-normality, NOvA uses the computationally intensive Feldman-Cousins (FC) procedure, which involves fitting thousands of independent pseudoexperiments to generate empirical distribution functions that are used to calculate the significance of observations. I, along with collaborators on NOvA and Scientific Discovery through Advanced Computing: High Energy Physics Data Analytics (SciDAC-4) collaborations, have implemented the FC procedure utilizing the High Performance Computing (HPC) facilities at the National Energy Research Scientific Computing Center (NERSC). With this implementation, we have successfully processed NOvA's complete FC corrections for our recent neutrino + antineutrino appearance analysis in 36 hours: a speedup factor of 50 as compared to the methods used in previous analyses.