Institute for Advanced Study/Princeton University Early Universe/Cosmology Lunch Discussion

Abstract 1: Measurements of the Hubble constant (H0) as determined via the Cepheid-supernovae distance ladder appear to provide strong evidence for physics beyond LCDM. However, in the Carnegie Chicago Hubble Program (CCHP), we used the Tip of the Red Giant Branch (TRGB) method to determine a value of H0 that is less in tension with LCDM than the Cepheid-SN value, weakening the claim for new physics. This disagreement between the TRGB and Cepheids has sparked debate over the CCHP's calibration of the TRGB zero point, the value of which is perfectly degenerate with H0. In this talk, I will present my recent findings that settle the issue through careful experimental design and state-of-the-art measurements. I will demonstrate that this contention over TRGB calibration was sourced by systematic biases (due to, e.g., metallicity, age, and dust) in literature measurements. The final result is a self-consistent TRGB calibration determined to its highest accuracy yet. The new result also confirms the existence of the Cepheid-TRGB distance discrepancy and that the likeliest explanation for it is systematic error in measurements of the most distant SN host galaxies (20-30 Mpc). The TRGB has an exciting future ahead. With data from my forthcoming Hubble Space Telescope program, the definitive Hubble Telescope calibration of the TRGB will be determined. The door will also be opened for the method to be used in the near-infrared, which promises a tenfold increase in the number of TRGB-calibrated SNe Ia—a prospect that would change the face of distance ladder H0 experiments.

Abstract 2: The era of big astronomical data grows as survey telescopes like Roman, LSST, and SPHEREx near first light. Machine learning is a crucial tool to analyze and learn from these data, but requires large and well-understood data sets for training. The Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project consists of simulations created specifically to train neural networks and other machine learning tools, with the ultimate goal of measuring the cosmology of the universe from observations. We present a new ‘hump’ for the CAMELS project, created to specifically leverage the power of galaxy clustering summary statistics to constrain cosmology while also carrying features of galaxy formation physics. The new CAMELS-SAM suite consists of 1000+ unique N-body simulations of (100 h-1 Mpc)3 and N=(640)3 particles across a very broad range of ΩM and σ8, and each with a unique iteration of the Santa Cruz semi-analytic model (SAM) for galaxy formation with varied parameters controlling stellar and AGN feedback. As a proof of concept of the possibilities with these simulations, we show how neural networks are able to robustly measure cosmology and constrain parameters of baryonic physics with the galaxy two-point correlation function, void probability function, and count-in-cells.