Twenty-First-Century Statistical and Computational Challenges in Astrophysics

Abstract

Modern astronomy has been rapidly increasing our ability to see deeper into the Universe, acquiring enormous samples of cosmic populations. Gaining astrophysical insights from these data sets requires a wide range of sophisticated statistical and machine learning methods. Long-standing problems in cosmology include characterization of galaxy clustering and estimation of galaxy distances from photometric colors. Bayesian inference, central to linking astronomical data to nonlinear astrophysical models, addresses problems in solar physics, properties of star clusters, and exoplanet systems. Likelihood-free methods are growing in importance. Detection of faint signals in complicated noise is needed to find periodic behaviors in stars and detect explosive gravitational wave events. Open issues concern treatment of heteroscedastic measurement errors and understanding probability distributions characterizing astrophysical systems. The field of astrostatistics needsincreased collaboration with statisticians in the design and analysis stages of research projects, and joint development of new statistical methodologies. This collaboration will yield more astrophysical insights into astronomical populations and the cosmos itself.