Cross-correlation of Luminous Red Galaxies with ML-selected AGN in HSC-SSP III: HOD Parameters for Type I and Type II Quasars

TL;DR

We analyze the angular cross-correlation of 1.5 million LRGs with ~28k ML-selected AGN in three HSC-SSP fields to infer halo occupation parameters for Type I and Type II quasars at z ≈ 0.7–1.0. By applying a 3-parameter HOD to the LRG autocorrelation and then fitting LRG–AGN cross-correlations, we find Type I AGN reside in halos about 3× more massive than Type II, while Type I AGN have a markedly smaller satellite fraction. The one-halo term slopes differ between spectral types, with Type I showing a shallower intra-halo clustering than Type II, suggesting environmental differences beyond a strict unified model. These results imply the quasar spectral class encodes information about the host halo environment, prompting future work linking accretion physics to halo growth using semi-empirical models and upcoming spectroscopic surveys.

Abstract

Understanding the dark matter (DM) halo environment in which galaxies that host active galactic nuclei (AGN) reside is a window into the nature of supermassive black hole (SMBH) accretion. We apply halo occupation distribution (HOD) modeling tools to interpret the angular cross-correlation functions between $1.5\times10^6$ luminous red galaxies (LRGs) and our $\sim28,500$ Hyper Suprime-Cam + Wide-field Infrared Survey Explorer-selected (and $L_{6 μm}$-limited) AGN to infer the halo properties of distinct quasar samples at physical scales $s>0.1\,{\rm Mpc}$, for $z\in0.7-1.0$. We find that Type I (unobscured) and Type II (obscured) AGN cluster differently, both on small and large physical scales. The derived HODs imply that Type I AGN reside, on average, in substantially ($\sim3\times$) more massive halos ($M_h \sim 10^{13.4} M_\odot$) than Type II AGN ($M_h \sim 10^{12.9} M_\odot$) at $>5σ$ significance. While Type II AGN show one-halo correlations similar to that of galaxies of their average halo mass, the Type I AGN intra-halo clustering signal is significantly shallower. We interpret this observation with HOD methods and find Type I AGN are significantly less likely ($f_{sat}\sim0.05^{+1}_{-0.05}\%$) to be found in satellite galaxies than Type II AGN. We find reddened + obscured AGN to have typical satellite fractions for their inferred average halo mass ($\sim10^{13} M_\odot$), with $f_{sat} \sim 20^{+10}_{-5}\%$. Taken together, these results pose a significant challenge to the strict unified AGN morphological model, and instead suggest that a quasar's spectral class is strongly correlated with its host galaxy's dark matter halo environment. These intriguing results have provided a more complex picture of the SMBH -- DM halo connection, and motivate future analyses of the intrinsic galaxy and accretion properties of AGN.

Figures (12)

• Figure 1: Pedagogical example of the effect of changing the values of the 3-parameter HOD model that we project to its angular correlation function ($\omega(\theta)$) form, as outlined in § \ref{['sec:HOD_method']}. The baseline values are $\log M_{\rm min} = 13 \log M_\odot$, $\log M_{1} = 14 \log M_\odot$, $\alpha = 1.5$. The physical scales are converted assuming $z=0.8$. The axes are shared between panels, the middle panel shows the relative contributions of the one- and two-halo terms, while the dotted-dashed line is the minimum scale we will fit data to with this model. We iteratively change one parameter by $\pm1.0$ in each panel, as indicated by the text in the upper right of each panel. The colored shading relates the shift in the parameter value to the model $\omega(\theta)$ it produces. $\log M_{\rm min}$ and $\log M_{1}$ have relatively straightforward (and opposite) effects on the amplitudes of the models, while $\alpha$ affects the one-halo term as a $\theta$-dependent scaling.
• Figure 2: Top: The measured HSC LRG projected angular autocorrelation for our three HSC fields. The $1\sigma$ uncertainties are drawn from the square root of the diagonal of the jackknife covariance matrix for each sample. The open symbols represent the per-bin inverse variance weighted mean and error across the fields. The dashed line represents the joint-field best-fit 3-parameter HOD model. The gray dash dotted line represents the minimum scale for which we fit the data, $s > 0.1\,{\rm Mpc}$. Bottom: Residuals from each field and their inverse variance-weighted mean, highlighting the poor fit on large scales.
• Figure 3: MCMC-derived posteriors for the 3-parameter HOD model fit to the LRG autocorrelations in our $z\in0.7-1.0$ bin (see definitions in § \ref{['sec:HOD_method']}). Contours are shown for the 1, 2, and $3\sigma$ 2-D confidence levels ($39.4\%,\, 86.5\%,\,98.9\%$). We maximize the joint likelihood by summing the likelihood of each subfield while requiring that a single HOD model fit all the data. We recover HOD parameters for our magnitude-limited ($r<24$) LRG sample, finding that our derived parameters are $b_g = 2.07 \pm 0.01$, $\log \langle M_h\rangle= 13.48\pm0.01\, \log M_\odot$, and $f_{sat} = 11.1 \pm 0.5\%$ .
• Figure 4: Top: The inverse-variance weighted mean of the measured LRG $\times$ full HSC+WISE AGN sample projected angular autocorrelation. We co-add the three HSC fields' measured cross-correlations and illustrate them as black triangles. The $1\sigma$ uncertainties are the inverse variance weighted error. The black dashed line represents the joint-field best-fit 3-parameter HOD model, while the dotted turquoise line is the same for the 5-parameter HOD model. The gray dash dotted line represents the minimum scale for which we fit the data, $s > 0.1\,{\rm Mpc}$. Bottom: Residual plots for the 3- and 5-parameter best-fit HOD models (colors as in the above panel).
• Figure 5: Posteriors for the 3-parameter HOD model fit to the cross-correlation of LRGs and the complete AGN sample in our $z\in0.7-1.0$ bin. Contours are shown for the 1, 2, and $3\sigma$ 2-D confidence levels ($39.4\%,\, 86.5\%,\,98.9\%$). After fixing the cross terms for the LRGs from our fits to their autocorrelation, we maximize the joint likelihood by summing the likelihood of each subfield while requiring a single AGN HOD model fit all the data. Our derived parameters are $b_g = 1.7 \pm 0.1$, $\log \langle M_h\rangle= 13.17\pm0.03\, \log M_\odot$, and $f_{sat} = 13^{+3}_{-2} \%$.