Robust Pulsar Timing Inference with Non-Gaussian Distributions

• Activity Number: 102 - SAMSI-ASTRO: New Innovations and Challenges in Astrostatistics
• Type: Invited
• Date/Time: Monday, July 30, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Statistical and Applied Mathematical Sciences Institute
• Author(s): Hyungsuk Tak* and Justin A. Ellis and Sujit Ghosh
• Companies: SAMSI and West Virginia University and North Carolina State Univ.
• Keywords: Outlier; Markov chain Monte Carlo; Gaussian process; Astronomy; Gravitational waves

Abstract

Recent pulsar timing data releases by the NANOgrav collaboration team (NANOgrav, 2015) have shown that we may be approaching the limits of Gaussian models, considering tons of outlying observations that Gaussian heteroskedastic measurement errors cannot explain. For example, 11,097 observations out of 16,219 (about 70\% of the data) for PSR J1455-3330 are removed due to their low signal-to-noise ratios, which enabled the team to fit their Gaussian error model, i.e., a parametric model built on a Gaussian measurement error assumption. These discarded data, however, are clear evidence of non-Gaussian data generation process of pulsar timing array, requiring non-Gaussian error models that can extract information from the whole data without removing any of them. We use a mixture of Gaussian and Student's t measurement errors to analyze pulsar timing data, empirically showing that this mixture error enables robust and accurate inference by leveraging the best of robust Student's t error and efficient Gaussian error.