A 1% accurate method to include baryonic effects in galaxy-galaxy lensing models

The clustering of galaxies and galaxy-galaxy lensing are two of the main observational probes in Stage-IV large-scale structure surveys, such as Euclid and LSST. Unfortunately, the complicated relationship between galaxies and matter greatly limits the exploitation of this data. Sophisticated theoretical galaxy bias models – such as the hybrid Lagrangian bias expansion – allow describing galaxy clustering down to scales as small as k = 0.7 h Mpc−1. However, the galaxy-matter cross-power spectra are already affected by baryons on these scales, directly impacting the modelling of galaxy-galaxy lensing. In this work, we propose a way to extend state-of-the-art models of the galaxy-matter cross-power spectrum Pgm(k) (currently only accounting for dark matter) by including a baryonic correction term inferred from the matter component (the suppression Smm(k) = Pmm, full physics/Pmm, gravity only), so that Pgm, full \, physics (k) = √Sₘₘ Pgm, gravity \, only. We use the FLAMINGO hydrodynamical simulations to measure the effect of baryons on the galaxy-matter cross-power spectrum and to assess the performance of our model. Specifically, we perform a Bayesian analysis of synthetic data, implementing a model based on BACCO’s hybrid Lagrangian bias expansion (for the nonlinear galaxy bias) and Baryon Correction Model (for the baryon suppression of the matter power spectrum). Ignoring the effect of baryons on the galaxy-matter cross-power spectrum leads to a biased inference of the galaxy bias parameters, while ignoring baryons in both the galaxy-matter and matter-matter power spectra leads to a biased inference of both the galaxy bias and cosmological parameters. In contrast, our method is 1% accurate compared to all physics variations in FLAMINGO and on all scales described by hybrid perturbative models (k < 0.7 h Mpc−1). Moreover, our model leads to inferred bias and cosmological parameters compatible within 1σ with their reference values. We anticipate that our method will be a promising candidate for analysing forthcoming Stage-IV survey data.