Cosmic Density Field Reconstruction with a Sparsity Prior Using Images of Distant Galaxies

Abstract

Future astronomical observations are expected to deliver multi- peta byte images of stars, galaxies, and supernovae. Processing the sheer volume of imaging and spectroscopic data is technically challenging, and producing scientific outputs from the big data will remain a key task in the next decade. We develop novel methods based on modern machine learning and deep learning to analyze data from Subaru Hyper Suprime-Cam. In this contribution, we focus on reconstruction of cosmic density field. We use the observation of gravitational lensing effect that causes slight deformation of shapes of galaxies. The collective effect can be used to reconstruct the large-scale density distribution. Our novel technique assuming a sparsity prior allows to reconstruct the density field in full three dimensions. Statistical analysis of cosmic structure enables accurate determination of a few fundamental quantities called cosmological parameters that describe the contents and the evolution of the Universe.