Cosmological inference with halo clustering reconstructed from the redshift-space galaxy distribution

TL;DR

This paper demonstrates that halo reconstruction from galaxy distributions using the cylinder grouping method allows for unbiased, more precise cosmological parameter inference from redshift-space clustering data, even on smaller scales.

Contribution

It shows that reconstruction-induced systematics can be modeled within EFT, improving robustness and precision in cosmological constraints from galaxy clustering.

Findings

Reconstruction yields unbiased estimates of $f\sigma_8$ and other parameters.

A simple multipole-dependent rescaling captures dominant reconstruction effects.

Uncertainty in $f\sigma_8$ is reduced by over 20% when extending the fit to smaller scales.

Abstract

Accurate modeling of small-scale redshift-space clustering is crucial for full shape RSD analyses, where satellite galaxies contribute to 1-halo terms and Finger-of-God distortions. We investigate halo reconstruction based on the cylinder grouping (CG) method of Okumura et al. (2017), which selects an effective halo center tracer from the observed galaxy distribution, and how it impacts cosmological parameter inference. Using DESI-like luminous red galaxy mock catalogs from the AbacusSummit simulations at $z=1.1$, we perform effective field theory (EFT)-based full-shape modeling of the power spectrum of the reconstructed-halo sample. We show that the dominant reconstruction-induced systematics can be described and incorporated within the standard EFT framework. In particular, a simple multipole-dependent rescaling inferred directly from the data on large scales captures the dominant effect, while residual small-scale changes are absorbed by the standard counterterm and stochastic sector, without introducing additional reconstruction-specific parameters. The reconstructed-halo sample yields unbiased constraints on cosmological parameters, including the growth rate $f\sigma_8$ and Alcock-Paczynski parameters. Compared to the galaxy sample, it enables both improved robustness and increased statistical precision: the inferred $f\sigma_8$ remains stable when extending the fit beyond $k_{\max}\simeq 0.2\,h\,{\rm Mpc}^{-1}$, with its uncertainty reduced by more than $20\%$.