High-redshift AGN population in radiation-hydrodynamics simulations

TL;DR

The paper develops a new suite of high-redshift, radiation-hydrodynamics simulations by embedding on-the-fly radiative transfer in the IllustrisTNG framework within a MillenniumTNG protocluster, enabling simultaneous tracking of BH growth, galaxy assembly, and IGM ionization. By varying radiative efficiency and radiation modeling, the study shows that BHs seed in overdense regions and grow more rapidly when $\epsilon_{\rm rad}$ is lowered to $0.1$, though stellar assembly typically outpaces BH growth, keeping $M_\text{BH}/M_*\lesssim3\times10^{-3}$ and BHs below the local $M_\text{BH}$-$M_*$ relation at high $z$. The authors compare to observed high-$z$ AGN and JWST results, finding general agreement for faint AGN but not for rare, luminous quasars, which motivates the quasar-boosted model that enhances AGN luminosity to study proximity effects and He ionization. The results underscore the value of on-the-fly RT in capturing feedback-driven gas dynamics and provide a framework to incorporate the missing luminous quasar population for a more complete picture of early BH growth and reionization.

Abstract

High-redshift active galactic nuclei (AGN) have long been recognized as key probes of early black hole growth and galaxy evolution. However, modeling this population remains difficult due to the wide range of luminosities and black hole masses involved, and the high computational costs of capturing the hydrodynamic response of gas and evolving radiation fields on-the-fly. In this study, we present a new suite of simulations based on the IllustrisTNG galaxy formation framework, enhanced with on-the-fly radiative transfer, to examine AGN at high redshift (z > 5) in a protocluster environment extracted from the MillenniumTNG simulation. We focus on the co-evolution of black holes and their host galaxies, as well as the radiative impact on surrounding intergalactic gas. The model predicts that black holes form in overdense regions and lie below the local black hole-stellar mass relation, with stellar mass assembly preceding significant black hole accretion. Ionizing photons are primarily produced by stars, which shape the morphology of ionized regions and drive reionization. Given the restrictive black hole growth in the original IllustrisTNG model, we reduce the radiative efficiency from 0.2 to 0.1, resulting in higher accretion rates for massive black holes, more bursty growth, and earlier AGN-driven quenching. However, the resulting AGN remain significantly fainter than observed high-redshift quasars. As such, to incorporate this missing population, we introduce a quasar boosted model, in which we artificially boost the AGN luminosity. This results in strong effects on the surrounding gas, most notably a proximity effect, and large contributions to He ionization.

Figures (13)

• Figure 1: Visual illustration of the mass assembly of black holes (top row), and galaxy stellar mass (bottom row)Background: Gas distribution, color-coded by the corresponding gas overdensity $\delta_\mathrm{g} = \rho_\mathrm{g}/\bar{\rho}_\mathrm{g} - 1$. Each panel shows a projection slice measuring 20 cMpc across and 10 cMpc in depth, centered on the most massive halo in the simulation. We show three different redshifts: $z = 9$ (left column), $z = 7.5$ (middle column) and $z = 6$ (right column). The right column also includes a 2.5 cMpc square zoom-in around the center of the slice. In the top row, black holes are depicted as circles color-coded by mass, with seed BHs shown as black triangles. In the bottom row, stellar masses of galaxies are indicated by star symbols, also color-coded by mass. The top row illustrates that the first BHs are seeded around $z \sim 9$ and tend to trace the dense, filamentary structures due to the halo-dependent seeding model (see Sec. \ref{['sect:TNG']}). By $z = 6$, the most massive BH has grown to $M_\mathrm{BH} \sim 10^{8.5} \, \mathrm{M_\odot}$ and resides in the central protocluster. The bottom row highlights that galaxies assemble their stellar mass earlier and more rapidly than BHs, with $M_\star$ typically exceeding the mass of the central BH by at least two orders of magnitude.
• Figure 2: Growth of the central SMBH (black) and its host galaxy (red) over time. Solid and dotted-dashed lines represent $\epsilon_{\rm{rad}} = 0.1$ and $\epsilon_{\rm{rad}} = 0.2$, respectively. In terms of black hole growth, at very high redshift ($z \gtrsim 7.5$, and low BH mass, $M_\mathrm{BH} \lesssim 10^7\,\mathrm{M_\odot}$), the Bondi--Hoyle accretion is highly inefficient, and BH growth is dominated by mergers. Once $M_\mathrm{BH}$ attains $M_\mathrm{BH} \gtrsim 10^7\,\mathrm{M_\odot}$, the differences between the two models ($\epsilon_\mathrm{rad} = 0.1$ and $\epsilon_\mathrm{rad} = 0.2$) grow over time, as the Bondi--Hoyle accretion rate scales as $\propto M_\mathrm{BH}^2$. The growth of the host galaxy is largely insensitive to the choice of $\epsilon_{\rm{rad}}$, apart from minor stochastic variations and the onset of quenching due to AGN kinetic feedback, which becomes effective at $z \lesssim 5.7$ in the $\epsilon_{\rm{rad}} = 0.1$ run. As a result, the co-evolutionary trajectories of the BH and its host galaxy diverge between the two models, a trend explored further in Fig. \ref{['fig:MBH-Ms']}.
• Figure 3: Effects of the radiative efficiency value on the AGN luminosities. The lines (solid: fiducial model, $\epsilon_\mathrm{rad} = 0.1$; dotted-dashed: $\epsilon_\mathrm{rad} = 0.2$) show the number density of AGN exceeding a given bolometric luminosity threshold, as indicated on the x-axis ($L_\mathrm{bol} = \epsilon_\mathrm{rad} \dot{M} c^2$, without accounting for attenuation from obscuration effects). The lines are color-coded by redshift, $z=9\rightarrow5$, in increments of $\Delta z =1$. At $z = 9$ and $z = 8$, when accretion is still highly inefficient and nearly identical in both simulations (see Fig. \ref{['fig:centralBH_z']}), the AGN in the $\epsilon_\mathrm{rad} = 0.2$ run are approximately twice as luminous as those in the $\epsilon_\mathrm{rad} = 0.1$ run -- an expected result given the direct scaling with the radiative efficiency. Once the BH masses become high enough for accretion to become efficient ($z = 7$ and $z = 6$), $\epsilon_\mathrm{rad} = 0.1$ produces more luminous AGN, due to the significantly higher accretion rates. By $z = 5$, however, AGN kinetic feedback has already been triggered in the $\epsilon_\mathrm{rad} = 0.1$ simulation, suppressing BH growth for the massive BHs and leading to reduced accretion rates and consequently, lower AGN luminosities. We overplot observational constraints from broad-line AGNMatthee2023 at $z \textcolor{black}{ \approx} 5$, and the quasar luminosity function Shen2020 at $z \approx 6$ and $z\approx 5$ (dashed lines, and extrapolations shown with dotted lines), finding good agreement with our simulations predictions.
• Figure 4: Black hole mass function (left column) and stellar mass function (right column), at three different redshifts: $z = 9$, $z = 7.5$ and $z = 6$. We show all radiation models, explained in Table \ref{['tab:prop']}, and for an easier distinction between them, we include the cumulative black hole and stellar mass functions in the bottom row. As seen in Fig. \ref{['fig:env_proj']}, the first BHs are seeded at around $z \approx 9$ (seed mass is indicated by the gray dotted line), and grow slower than their host galaxies. The radiation models show good overall convergence in both the black hole and galaxy populations, with small differences due to feedback and stochastic effects. The main difference arises in the $\epsilon_\mathrm{rad} = 0.2$ run, which shows a suppressed high-mass end of the black hole mass function at $z = 7.5$ and $z = 6$, as expected (see Sec. \ref{['sect:erad']}). The upturn at the high-mass end of the black hole mass function at $z = 6$ is driven by the dense environments surrounding the most massive BHs (see Fig. \ref{['fig:env_proj']}), which boost accretion rates, along with the strong mass dependence of the Bondi--Hoyle accretion model. However, this trend will be suppressed by $z \approx 5.5$, as AGN kinetic feedback becomes active. We compare our predictions with observational constraints from LRDs BHMFs (top-left panel, in yellow, with gray points representing data not corrected for completeness), and stellar mass functions from JWST studies (top-right panel, color-coded by redshift). Although the LRDs BHMF constraints are derived at lower redshifts than our simulations, they suggest very rapid black hole growth between $z = 6$ and $z \approx 4$, which may be difficult to reproduce with current models. We find overall good agreement with the observational constraints on the stellar mass function.
• Figure 5: The $M_{\rm{BH}}$--$M_\star$ relation for the central black hole in our simulation over $\bm{5 < z < 8}$. Lines are color-coded by redshift and represent the fiducial RT model (solid line), the RT model with $\epsilon_{\mathrm{rad}} = 0.2$ (double-dot-dashed line), and the equilibrium cooling-UVB model (dot-dashed line). The KormendyHo2013 local relation is shown as a grey dashed line with a shaded region, indicating the scatter. Observational data for $5 < z < 8.5$ from JWST, color-coded by redshift, are also overplotted (with references shown in the plot). This figure illustrates that galaxies assemble their stellar mass earlier, with the central black holes consistently lying below the local $M_\mathrm{BH}$–$M_\star$ relation, and significantly less massive than the AGN observed with JWST. The choice of radiative efficiency leads to distinct co-evolutionary tracks: while the stellar masses remain comparable, black hole growth becomes increasingly suppressed for $\epsilon_\mathrm{rad} = 0.2$ at $z \lesssim 7.5$ (see also Fig. \ref{['fig:centralBH_z']}). As seen in Fig. \ref{['fig:BHMF6']}, the overall growth of both black holes and galaxies is largely insensitive to the specific radiation model. Here, we include only two representative models (fiducial and equilibrium cooling–UVB) to highlight the relatively minor effects of feedback and stochastic variability.