The Drivers of the Decline in Supermassive Black Hole Growth at $z<2$

TL;DR

This study uses the refined SMBH accretion distributions from nine deep and wide-field surveys to pinpoint the drivers of the sharp decline in SMBH growth after $z\sim2$. By decomposing the cosmic BHAR density into $\rho_{\mathrm{BHAR}} \approx n_{\mathrm{AGN}}^{\mathrm{eff}} \times \langle\lambda_{\mathrm{Edd}}\rangle \times \langle M_{\mathrm{BH}}\rangle \times \frac{(1-\epsilon)\times1.26\times10^{38}}{\epsilon c^2}$ and using the observed evolution of $\langle\lambda_{\mathrm{Edd}}\rangle$, $\langle M_{\mathrm{BH}}\rangle$, and $n_{\mathrm{AGN}}^{\mathrm{eff}}$, the authors find that the dominant factor is a substantial decline in the typical Eddington ratio by about $1.35$ dex from $z\approx1.5-2$ to $z\approx0.2$, rather than a major shift toward lower-mass BHs. They further show that $M_\star$ mainly modulates the average outburst luminosity $\overline{L_{bol}^{AGN}}$ rather than the AGN duty cycle $f_{AGN}$, with $f_{AGN}$ evolving more strongly in massive galaxies. This work clarifies the origin of AGN downsizing and the coevolution of SMBHs with their hosts, and it outlines paths for extending the analysis with future wide-field X-ray surveys and next-generation observatories.

Abstract

It is well established that cosmic supermassive black hole (SMBH) growth peaks at $z\approx1.5-2$, followed by a strong decline of $\approx1-1.5\,\rm dex$ toward the present day, with the comoving number density of higher-luminosity active galactic nuclei (AGNs) peaking at higher redshift (referred to as "AGN downsizing"). We leverage the best current measurements of the SMBH accretion distribution, based upon data from nine well-characterized extragalactic fields with a "wedding-cake" design, to investigate and quantify the drivers of the drastic decline in cosmic SMBH growth. The decline in the typical Eddington ratio ($λ_\mathrm{Edd}$) of AGNs (decreasing by $\approx1.35\,\rm dex$ from $z\approx1.5-2$ to $z\approx0.2$) is the dominant driver for the broad decline in SMBH growth, rather than a shift of accretion activity to less-massive SMBHs. As $λ_\mathrm{Edd}$ decreases toward lower redshift, the primary contributor to the cosmic SMBH accretion density ($ρ_\mathrm{BHAR}$) has shifted from high-$λ_\mathrm{Edd}$ AGNs to low-$λ_\mathrm{Edd}$ AGNs, even though the latter always dominate the comoving AGN number density at $z<4$. We also find that the decline in SMBH growth toward lower SMBH mass in less-massive galaxies is primarily due to the decreasing outburst luminosity rather than the duty cycle.

Figures (7)

• Figure 1: Top panel: $\rho_{\mathrm{BHAR}}$ as a function of redshift. The black data points represent the total $\rho_{\mathrm{BHAR}}$ sampled by our data. The red, yellow, and blue data points represent $\rho_{\mathrm{BHAR}}$ for AGNs accreting at $\lambda_{\mathrm{Edd}}=0.01-0.1$, $\lambda_{\mathrm{Edd}}=0.1-1$, and $\lambda_{\mathrm{Edd}}=1-100$, respectively. For comparison, the cosmic SFRD MadauDickinson+2014 scaled by a factor of 5000 is shown as the dashed line to approximate the peak of $\rho_{\mathrm{BHAR}}$ at $z\approx2$. The green dotted line and the brown dashed line represent the total $\rho_{\mathrm{BHAR}}$ from Ueda+2014 and Yang+2018, respectively. Bottom panel: logarithm of the ratio of $\rho_{\mathrm{BHAR}}$ contributed by AGNs with $\lambda_{\mathrm{Edd}}>0.1$ to that contributed by AGNs with $\lambda_{\mathrm{Edd}}<0.1$. The error bars represent the 90% confidence intervals derived with our Monte Carlo method.
• Figure 2: AGN host-galaxy SMF $\phi^\mathrm{AGN}_\mathrm{M}$ at different redshifts, with colors defined in the legend. The left, middle, and right panels show the $\lambda_{\mathrm{Edd}}=0.01-0.1$, $\lambda_{\mathrm{Edd}}=0.1-1$, and $\lambda_{\mathrm{Edd}}=1-100$ bins, respectively. For comparison, the galaxy SMF at $z=0.2-0.5$ in Weaver+2023 is shown as black solid curves. The colored shaded stripes represent the $1\sigma$ uncertainty from $p(\lambda_{\mathrm{Edd}}|M_\star,z)$.
• Figure 3: Contribution to $\rho_{\mathrm{BHAR}}$ per galaxy at different redshifts assuming the median $p(\lambda_{\mathrm{Edd}}|M_\star,z)$ and $M_{\mathrm{BH}}=0.002M_\star$. The color bars represent the logarithm of $(1-\epsilon)/(\epsilon c^2)\times F(\lambda_{\mathrm{Edd}},M_\star|z)\times M_\star\lambda_{\mathrm{Edd}}\kappa(M_\star)$. The median values of $\log\lambda_{\mathrm{Edd}}$ ($\log M_{\mathrm{BH}}$) at fixed $\log M_{\mathrm{BH}}$ ($\log\lambda_{\mathrm{Edd}}$) are shown as solid (dash-dotted) curves. The $25-75\%$ quantiles of $\log\lambda_{\mathrm{Edd}}$ ($\log M_{\mathrm{BH}}$) at fixed $\log M_{\mathrm{BH}}$ ($\log\lambda_{\mathrm{Edd}}$) are shown as dotted (dashed) curves. The error bars represent the $\log\langle\lambda_{\mathrm{Edd}}\rangle$ and $\log\langle M_{\mathrm{BH}}\rangle$ that roughly span the region enclosed by the $25-75\%$ quantiles of $\log\lambda_{\mathrm{Edd}}$ and $\log M_{\mathrm{BH}}$.
• Figure 4: The evolution of $\langle\lambda_{\mathrm{Edd}}\rangle$ and $\langle M_{\mathrm{BH}}\rangle$ at different redshifts, with colors defined in the legend. Top panel: The error bars show the region bounded by the $25-75\%$ quantiles of $\log\lambda_{\mathrm{Edd}}$ and $\log\langle M_{\mathrm{BH}}\rangle$ in Figure \ref{['fig::flambM']} assuming the median $p(\lambda_{\mathrm{Edd}}|M_\star,z)$ and $M_{\mathrm{BH}}=0.002M_\star$. Bottom panel: The error bars represent the 90% confidence intervals derived with our Monte Carlo method. From $z=1.5-2.0$ to $z=0.2-0.5$, $\langle\lambda_{\mathrm{Edd}}\rangle$ and $\langle M_{\mathrm{BH}}\rangle$ both decrease; $\langle\lambda_{\mathrm{Edd}}\rangle$ decreases by $1.35\,\rm dex$, but $\langle M_{\mathrm{BH}}\rangle$ decreases only by $0.21\,\rm dex$.
• Figure 5: $n_\mathrm{AGN}$ and $n_\mathrm{AGN}^\mathrm{eff}$ as a function of redshift. The black data points represent the total $n_\mathrm{AGN}$ sampled by our data ($\lambda_{\mathrm{Edd}}>0.0013$). The red, yellow, and blue data points represent $n_\mathrm{AGN}$ for AGNs accreting at $\lambda_{\mathrm{Edd}}=0.01-0.1$, $\lambda_{\mathrm{Edd}}=0.1-1$, and $\lambda_{\mathrm{Edd}}=1-100$, respectively. The black solid line represents $n_\mathrm{AGN}^\mathrm{eff}$, and the dashed line represents $n_\mathrm{AGN}$ for AGNs within the $25-75\%$ quantile range of $\langle\lambda_{\mathrm{Edd}}\rangle$. The error bars of the data points and the grey shaded region for $n_\mathrm{AGN}^\mathrm{eff}$ represent the 90% confidence intervals derived with our Monte Carlo method. For comparison, we show the total $n_\mathrm{AGN}$ for AGNs with $\log L_\mathrm{X}>42$ from Buchner+2015 (grey hatched region representing 10--90% confidence intervals), Miyaji+2015 (green dotted line), and Peca+2023 (brown dotted-dashed line).