Singling out modified gravity parameters and data sets reveals a dichotomy between Planck and lensing

An important route to testing general relativity (GR) at cosmological scales is usually done by constraining modified gravity (MG) parameters added to the Einstein perturbed equations. Most studies have analysed so far constraints on pairs of MG parameters, but here, we explore constraints on one parameter at a time while fixing the other at its GR value. This allows us to analyse various models while benefiting from a stronger constraining power from the data. We also explore which specific data sets are in tension with GR. We find that models with (mu = 1, eta) and (mu, eta = 1) exhibit a 3.9 sigma and 3.8 sigma departure from GR when using Planck18 + Supernovae type Ia (SNe) + Baryon acoustic oscillation (BAO), while (mu, eta) shows a tension of 3.4 sigma. We find no tension with GR for models with the MG parameter Sigma fixed to its GR value. Using a Bayesian model selection analysis, we find that some one-parameter MG models are moderately favoured over Lambda CDM when using all data set combinations except Planck cosmic microwave background lensing and dark energy survey data. Namely, Planck18 shows a moderate tension with GR that only increases when adding any combination of redshift space distortion, SNe, or BAO. However, adding lensing diminishes or removes these tensions, which can be attributed to the ability of lensing in constraining the MG parameter Sigma. The two overall groups of data sets are found to have a dichotomy when performing consistency tests with GR, which may be due to systematic effects, lack of constraining power, or modelling. These findings warrant further investigation using more precise data from ongoing and future surveys.

Automated summary

Testing general relativity at cosmological scales typically involves analyzing constraints on modified gravity parameters. This study introduces a new method for constraining one parameter at a time and identifies tensions with GR in different data set combinations.