DeepLSS: Breaking Parameter Degeneracies in Large-Scale Structure with Deep-Learning Analysis of Combined Probes

• Published in: Physical review. X Vol. 12; no. 3; p. 031029
• Main Authors: Kacprzak, Tomasz, Fluri, Janis
• Format: Journal Article
• Language: English
• Published: College Park American Physical Society 01.08.2022
• Subjects: Alignment; Artificial intelligence; Astronomical models; Astrophysics; Big Bang theory; Clustering; Cosmology; Deep learning; Figure of merit; Galactic evolution; Galaxies; Gravitational lenses; Large scale structure of the universe; Legal issues; Machine learning; Parameters; Red shift; Simulation; Sky surveys (astronomy); Stars & galaxies; Universe
• ISSN: 2160-3308; 2160-3308
• DOI: 10.1103/PhysRevX.12.031029

In classical cosmological analysis of large-scale structure surveys with two-point functions, the parameter measurement precision is limited by several key degeneracies within the cosmology and astrophysics sectors. For cosmic shear, clustering amplitudeσ8and matter densityΩmroughly follow theS8=σ8(Ωm/0.3)0.5relation. In turn,S8is highly correlated with the intrinsic galaxy alignment amplitudeAIA. For galaxy clustering, the biasbgis degenerate with bothσ8andΩm, as well as the stochasticityrg. Moreover, the redshift evolution of intrinsic alignment (IA) and bias can cause further parameter confusion. A tomographic two-point probe combination can partially lift these degeneracies. In this work we demonstrate that a deep-learning analysis of combined probes of weak gravitational lensing and galaxy clustering, which we call DeepLSS, can effectively break these degeneracies and yield significantly more precise constraints onσ8,Ωm,AIA,bg,rg, and IA redshift evolution parameterηIA. In a simulated forecast for a stage-III survey, we find that the most significant gains are in the IA sector: the precision ofAIAis increased by approximately 8 times and is almost perfectly decorrelated fromS8. Galaxy biasbgis improved by 1.5 times, stochasticityrgby 3 times, and the redshift evolutionηIAandηbby 1.6 times. Breaking these degeneracies leads to a significant gain in constraining power forσ8andΩm, with the figure of merit improved by 15 times. We give an intuitive explanation for the origin of this information gain using sensitivity maps. These results indicate that the fully numerical, map-based forward-modeling approach to cosmological inference with machine learning may play an important role in upcoming large-scale structure surveys. We discuss perspectives and challenges in its practical deployment for a full survey analysis.