2024A&A...692A.161B

Context. The assembly and co-evolution of super-massive black holes (SMBHs) and their host galaxy stellar population is one of the key open questions in modern galaxy evolution. Observationally constraining this question is challenging. Important parameters of galaxies, such as the stellar mass (M_⋆) and star formation rate (SFR), are inferred by modeling the spectral energy distribution (SED), with templates constructed on the basis of various assumptions on stellar evolution. In the case of galaxies triggering SMBH activity, the active galactic nucleus (AGN) contaminates the light of the host galaxy at all wavelengths, hampering inferences of host galaxy parameters. Underestimating the AGN contribution due to incomplete AGN templates results in a systematic overestimation of the stellar mass, biasing our understanding of AGN and galaxy co-evolution. This challenge has gained further attention with the advent of sensitive wide-area surveys with millions of newly detected luminous AGN, including those by eROSITA, Euclid, and LSST.
Aims. We aim to robustly estimate the accuracy, bias, scatter, and uncertainty of AGN host galaxy parameters, including stellar masses, and improve these measurements relative to previously used techniques.
Methods. This work makes two important contributions. Firstly, we present a new SED fitting code, GRAHSP, with an AGN model composed of a flexible power-law continuum with empirically determined broad and narrow lines and a FeII forest component, a flexible infrared torus that can reproduce the diverse dust temperature distributions, and appropriate attenuation on the galaxy and AGN light components. We verify that this model reproduces published X-ray to infrared SEDs of AGN to better than 20% accuracy. A fully Bayesian fit includes uncertainties in the model and the data, making the inference highly robust. The model is constrained with a fast nested sampling inference procedure supporting the many free model parameters. Secondly, we created a benchmark photometric data set where optically selected pure quasars are paired with non-AGN pure galaxies at the same redshift. Their photometry flux is summed into a hybrid (Chimera) object but with known galaxy and AGN properties. Based on this data-driven benchmark, true and retrieved stellar masses, SFR, and AGN luminosities can be compared, allowing for the evaluation and quantification of biases and uncertainties inherent in any given SED fitting methodology.
Results. The Chimera benchmark, which we release with this paper, shows that previous codes systematically overestimate M_⋆ and SFR by 0.5 dex with a wide scatter of 0.7 dex at AGN luminosities above 10⁴⁴ erg s^–1. In 20% of cases, the estimated error bars lie completely outside a 1 dex-wide band centreed around the true value, which we consider an outlier. In contrast, GRAHSP shows no measurable bias on M_⋆ and SFR, with an outlier fraction of only about 5%. GRAHSP also estimates more realistic uncertainties.
Conclusions. Unbiased characterization of galaxies hosting AGN enables characterization of the environmental conditions conducive to black hole growth, whether star formation is suppressed at high black hole activity, and identifying the mechanisms that prevent overluminous AGN relative to the host galaxy mass. It can also shed light on the long-standing questions of whether AGN obscuration is primarily an orientation effect or related to phases in galaxy evolution.