Euclid preparation

Abstract

Euclid’s photometric galaxy cluster survey has the potential to be a very competitive cosmological probe. The main cosmological probe with observations of clusters is their number count, within which the halo mass function (HMF) is a key theoretical quantity. We present a new calibration of the analytic HMF, at the level of accuracy and precision required for the uncertainty in this quantity to be subdominant with respect to other sources of uncertainty in recovering cosmological parameters from Euclid cluster counts. Our model is calibrated against a suite of N-body simulations using a Bayesian approach taking into account systematic errors arising from numerical effects in the simulation. First, we test the convergence of HMF predictions from different N-body codes, by using initial conditions generated with different orders of Lagrangian Perturbation theory, and adopting different simulation box sizes and mass resolution. Then, we quantify the effect of using different halo finder algorithms, and how the resulting differences propagate to the cosmological constraints. In order to trace the violation of universality in the HMF, we also analyse simulations based on initial conditions characterised by scale-free power spectra with different spectral indexes, assuming both Einstein–de Sitter and standard ΛCDM expansion histories. Based on these results, we construct a fitting function for the HMF that we demonstrate to be sub-percent accurate in reproducing results from 9 different variants of the ΛCDM model including massive neutrinos cosmologies. The calibration systematic uncertainty is largely sub-dominant with respect to the expected precision of future mass–observation relations; with the only notable exception of the effect due to the halo finder, that could lead to biased cosmological inference.

Robust Field-level Likelihood-free Inference with Galaxies

“…The quest to extract the maximum information from galaxy redshift surveys has motivated the development of many different approaches (Efron 1982;Feldman et al 1994;Taylor et al 2013;Abramo et al 2016;Heavens et al 2017;Hahn et al 2020 Euclid Collaboration: Castro et al 2023;Pontoppidan et al 2022) is driving forward this field of research. While we do not have a final answer to this question, ML techniques are appearing as a promising tool to tackle this problem (Ravanbakhsh et al 2017;Hassan et al 2020;Mangena et al 2020;Ntampaka et al 2020;Villaescusa-Navarro et al 2021a;Cole et al 2022;Perez et al 2022).…”