Modeling Galaxy Surveys with Hybrid SBI

TL;DR

This paper introduces HySBI, a hybrid simulation-based inference framework combining perturbation theory and small-scale simulations, to efficiently analyze galaxy surveys and improve cosmological parameter constraints.

Contribution

It develops the HySBI framework for galaxy clustering, enabling extraction of more information from large datasets without extensive simulations.

Findings

HySBI yields 20% tighter constraints on Ω_m.

HySBI yields 60% tighter constraints on σ_8.

Joint modeling improves parameter estimation accuracy.

Abstract

Simulation-based inference (SBI) has emerged as a powerful tool for extracting cosmological information from galaxy surveys deep into the non-linear regime. Despite its great promise, its application is limited by the computational cost of running simulations that can describe the increasingly-large cosmological datasets. Recent work proposed a hybrid SBI framework (HySBI), which combines SBI on small-scales with perturbation theory (PT) on large-scales, allowing information to be extracted from high-resolution observations without large-volume simulations. In this work, we lay out the HySBI framework for galaxy clustering, a key step towards its application to next-generation datasets. We study the choice of priors on the parameters for modeling galaxies in PT analysis and in simulation-based analyses, as well as investigate their cosmology dependence. By jointly modeling large- and small-scale statistics and their associated nuisance parameters, we show that HySBI can obtain 20\% and 60\% tighter constraints on $\Omega_m$ and $\sigma_8$, respectively, compared to traditional PT analyses, thus demonstrating the efficacy of this approach to maximally extract information from upcoming spectroscopic datasets.