Abstract

Contributed Talk - Splinter LargeScale

Thursday, 10 September 2026, 15:00   (MW-2235)

Merging Information from Galaxy Formation with Cosmology Inference: A Lens-Equal-Source Dark Energy Survey Year 6 3 × 2-point Analysis using Physics-Informed Priors

Zhengyangguang Gong, Andrés N. Salcedo, Joao Reboucas, Fabian Hervas-Peters, Dhayaa Anbajagane, Tim Eifler
Steward Observatory, University of Arizona; Department of Physics, University of Arizona; Department of Astronomy and Astrophysics, University of Chicago; Kavli Institute for Cosmological Physics, University of Chicago

Standard 3 × 2-point cosmological analyses divide observed galaxies into separate foreground lens and background source populations, which necessitates separate characterization of each and requires independent parameterizations of their systematic uncertainties. As an alternative, we present a lens-equal-source 3 × 2-point forecast with the statistical precision and observational systematics of the Dark Energy Survey Year 6 (DES Y6) data. Because utilizing a deep, flux-limited source sample as a direct tracer of large-scale structure introduces theoretical challenges related to its complex halo occupation distribution (HOD), we supplement our likelihood inference with physics-informed priors on the linear galaxy bias. These priors are constructed by coupling the UniverseMachine empirical galaxy formation predictions with Balrog synthetic photometric injections to accurately model DES Y6 clustering and selection effects. Using the CoCoA framework, our forecasts show that the lens-equal-source paradigm with uninformative priors is highly competitive with standard 3 × 2-point analyses. Crucially, the addition of physics-informed b1 priors significantly improves the constraints on Ωm and σ8 in the ΛCDM model, and on w in the wCDM model. Additionally, this framework facilitates the self-calibration of photometric redshift uncertainties, an effect primarily driven by the galaxy-galaxy lensing signal and complemented by the galaxy clustering. By both naturally compressing the systematic parameter space and reducing the prior volume, this unified approach offers a scalable and robust framework for next-generation surveys.