Quiescent galaxies at $z \gtrsim 2.5$: observations vs. models

The presence of massive quiescent galaxies at high redshifts is still a challenge for most models of galaxy formation. The aim of this work is to compare the observed number density and properties of these galaxies with the predictions of state-of-the-art models. The sample of massive quiescent galaxies has been selected from the COSMOS2015 photometric catalogue with $z_{\rm phot}\geq 2.5$, $\log (M_*/M_\odot)\geq 10.5$ and $\log(\mathrm{sSFR\,[yr^{-1}]})\le -11$. The photometric SEDs of the selected galaxies have been thoroughly analyzed based on different stellar population synthesis models. The final sample includes only those galaxies qualified as quiescent in all SED fitting runs. The observed properties have been compared to theoretical models: the number density of quiescent galaxies with $10.5 \leq \log(M_*/M_\odot) < 10.8$ is reproduced by some models, although there is a large scatter in their predictions. Instead, very massive $\log(M_{*}/M_{\odot}) \geq 10.8$ are underpredicted by most of the current models of galaxy formation: some of them, built on the CARNage simulation, are consistent with data up to $z \sim 4$, while at higher redshifts the volume of the considered simulation is too small to find such rare objects. Simulated galaxies which match the observed properties in the $\mathrm{sSFR}-M_*$ plane at $z\sim 3$ have been analyzed by reconstructing their evolutionary paths: their merger trees suggest that AGN feedback could be the key process allowing for a rapid quenching of the star formation at $z\gtrsim 4$ and that its treatment should be improved in models.