The End of Easy Phenomenology for CMB Experiments: A Case Study in the Dark Sector

TL;DR

This paper demonstrates that ignoring non-linear corrections in CMB lensing data leads to biases in cosmological parameters and limits the constraining power for dark sector models, emphasizing the need for accurate modeling.

Contribution

It provides a case study on the impact of non-linear corrections in CMB lensing for dark sector models, highlighting biases and the importance of proper modeling tools.

Findings

Ignoring non-linear corrections causes 0.2-0.6σ biases in key cosmological parameters.

Including lensing data can improve constraints on dark matter interactions by up to 50%.

Common non-linear correction codes can be unstable for extended dark sector models.

Abstract

The precision of Cosmic Microwave Background (CMB) experiments, specifically its lensing reconstruction, has reached the limit where non-linear corrections cannot be ignored. Neglecting these corrections results in biased constraints on cosmological parameters. In this work, we use lensing data from Planck and the South Pole Telescope third generation camera (SPT-3G) taken in 2018 to highlight the impact of these biases in two ways. First, we estimate the shifts due to ignoring non-linear corrections in $\Lambda$CDM. We find 0.2-0.6$\sigma$ shifts in the Dark Matter (DM) fraction, the Hubble constant, and the amplitude of matter fluctuations at $8 h^{-1}$ Mpc. Second, we estimate the loss in constraining power for not including data sensitive to non-linear corrections. As a case study, we consider two interacting DM models, for which such corrections are not readily available in existing CMB Boltzmann codes. The first one is DM interacting with baryons, while the second is DM interacting with Dark Radiation (DR). For the former case, when we add primary CMB data from SPT-3G 2018 observations, we find that constraints on model parameters improve by 10-20% compared to previous studies. However, we forecast a further 50% improvement on these constraints if one could include current or upcoming SPT-3G lensing data. For the case of DM interacting with DR, no meaningful constraints on the model parameters are found without including information from CMB lensing. We also highlight that the codes used to calculate non-linear corrections in $\Lambda$CDM, specifically HaloFit and HMCode, provide unstable results when improperly used for these extended models. These outcomes constitute a reason for caution if using CMB lensing data when constraining such models, as well as a motivation for estimating their non-linear corrections.