Cosmology-dependent covariance in galaxy cluster number counts: consequences for parameter inference

Abstract

Galaxy clusters provide constraints on cosmology through their abundance as a function of mass and redshift. Parameter inference from cluster counts requires modelling the covariance entering the likelihood, including contributions from Poisson shot noise and super-sample covariance (SSC) induced by long-wavelength density fluctuations. Since evaluating the full covariance during parameter inference can be computationally expensive, particularly for SSC terms, many analyses compute it at a fiducial cosmology and keep it fixed. In this work, we investigate the impact of covariance misspecification on the estimation of $\Omega_c$, $\sigma_8$, and $w$. We perform a systematic analysis in which the covariance is either varied consistently with the sampled cosmology or fixed at displaced cosmological models, including intermediate strategies where only selected components, such as SSC, are held fixed. Our analysis incorporates observational effects relevant for LSST-like optical surveys, including mass-proxy scatter and photometric redshift uncertainties. We find that the estimators of $\Omega_c$, $\sigma_8$, and $w$ remain unbiased even when the covariance is evaluated at an incorrect cosmology. However, fixing the covariance can significantly over- or underestimate confidence regions. The magnitude and sign of this effect are driven primarily by amplitude-related parameters such as $S_8$. For LSST-like surveys, an inconsistent covariance specification can artificially modify the apparent $S_8$ tension inferred from cluster counts. We further show that a single covariance update evaluated at the recovered best-fit cosmology is sufficient to restore the correct uncertainty normalization. These results indicate that fixed-covariance approximations may be adequate for some single-probe analyses, but a fully cosmology-dependent treatment is required for consistent multi-probe studies. ABRIDGED