JADES reveals a large population of low mass black holes at high redshift

TL;DR

This work uses stacking of ~600 JWST/NIRSpec spectra from JADES to uncover a population of low-mass black holes at $3<z<7$, traced by faint broad H$\alpha$ emission from the BLR in galaxies not detected as AGN individually. By combining continuum subtraction, emission-line fitting, and SN/VMS and outflow tests, the authors infer BH masses around $\log(M_{BH}/M_\odot)\approx 6.2$–$6.6$ with $L/L_{Edd}\sim 0.02$–$0.1$, and construct a BHMF that rises toward low masses and aligns with models invoking short, super-Eddington accretion and heavy seeds. The results place many high-redshift BHs near the local $M_{BH}-M_*$ relation, while also indicating an overmassive tail and a substantial dormant BH population, highlighting selection biases in previous JWST AGN detections. Overall, the study provides critical empirical constraints on BH seeding and early growth channels and motivates future high-resolution, lensing-assisted, or dynamical mass measurements to extend the census to even smaller BHs. These findings have significant implications for our understanding of SMBH formation and galaxy co-evolution in the early universe.

Abstract

JWST has revealed a large population of active galactic nuclei (AGN) in the distant universe, which are challenging our understanding of early massive black hole seeding and growth. We expand the exploration of this population to lower luminosities by stacking $\sim 600$ NIRSpec grating spectra from the JWST Advanced Deep Extragalactic Survey (JADES) at $3<z<7$, in bins of redshift, [OIII]5007 luminosity and equivalent width, UV luminosity and stellar mass. In various stacks, we detect a broad component of H$α$ without a counterpart in [OIII], implying that it is not due to outflows but is tracing the Broad Line Region (BLR) of a large population of low-luminosity AGN not detected in individual spectra. We also consider the possible contribution from Supernovae (SNe) and Very Massive Stars and conclude that while this is very unlikely, we cannot exclude some potential contribution by SNe to some of the stacks. The detection, in some stacks, of high [OIII]4363/H$γ$, typical of AGN, further confirms that such stacks reveal a large population of AGN. We infer that the stacks probe black holes with masses of a few times $10^6~M_\odot$ accreting at rates $L/L_{Edd}\sim 0.02-0.1$, i.e. a low mass and dormant parameter space poorly explored by previous studies on individual targets. We identify populations of black holes that fall within the scatter of the local $M_{BH}-M_{*}$ scaling relation, indicating that there is a population of high-z BHs that are not overmassive relative to their host galaxies and which have been mostly missed in previous JWST observations. Yet, on average, the stacks are still overmassive relative the local relation, with some of them 1-2 dex above it. We infer that the BH mass function (BHMF) at $3<z<5$ rises steeply at low masses. The BHMF is consistent with models in which BHs evolve through short bursts of super-Eddington accretion.

Figures (25)

• Figure 1: Stacked spectra of our sample of medium resolution JADES galaxies that have not been previously identified as AGN at $3<z<5$ (top) and $5<z<7$ (bottom). These are mean stacks with no weighting or normalisation. The $3<z<5$ stack contains 424 sources and the $5<z<7$ stack contains 152 sources.
• Figure 2: Zoom of the $3<z<5$ stack (left) and $5<z<7$ stack (right) around the H$\alpha$. The observed spectrum is given by the blue histogram and shaded region is the error. The pink line shows the fit with a narrow component only. The orange line is the fit of the [N II] doublet with velocity and width tied to the H$\alpha$ line. The dashed line shows the total fit. The bottom panel shows the residuals of the fit where the dotted line indicates the $\pm 1 \sigma$ levels. The zoom in clearly shows the broad residuals that the narrow-only model fails to reproduce around H$\alpha$ and the excess flux in these regions is also clear from the significant residuals in the bottom panel on both the blue and red sides of the central H$\alpha$ wavelength.
• Figure 3: Zoom around the [O][iii] and H$\alpha$ lines of the $3<z<5$ total stack. In contrast to Fig.\ref{['fig:halpha']}, the H$\alpha$ fit (top-right) includes a broad component (violet line), with the residuals shown in the bottom panel. On the left the [O][iii] doublet is fit with narrow and broad components tied to the H$\alpha$ line, with the residuals shown in the bottom panel (the dotted horizontal lines show again the $\pm 1\sigma$ deviation). The H$\alpha$ is well fit, as also indicated by the $\Delta BIC = 221$ in favour of the broad component. On the contrary, the [OIII] has clear systemic residuals indicating that the broad component of H$\alpha$ is inadequate to reproduce the [OIII]. The $\Delta BIC$ for [OIII] strongly favours a freely fitted [O][iii] broad component (with FWHM = 672), indicating that any broad components in [O][iii] do not have the same origin as the H$\alpha$ broad component.
• Figure 4: Same as Fig.\ref{['fig:halpha_broad1']}, but for the total stack in the $5<z<7$ redshift bin. The broad component is clearly a good fit to the H$\alpha$ line. However, in this case, the same broad component added to [O][iii] is also a good fit and has a similar value of the $BIC$ of a freely fitted [O][iii] broad component. This could indicate that in this case the broad $H\alpha$ could be due to outflows rather than an AGN BLR.
• Figure 5: Same as Fig.\ref{['fig:halpha_broad1']}, but for the stack in the $3<z<5$ redshift range and in the highest [OIII] luminosity bin. In this case the middle panel show the residuals without the broad component, while the bottom panel shows the residuals with the inclusion of the broad component (in the case of [OIII] tied to the H$\alpha$ line). Note that the scale on the residuals for the [O][iii] line and the H$\alpha$ lines are different, but we plot the $\pm 1\sigma$ levels as a dotted line for clarity. The FWHM of the H$\alpha$ broad component, which provides a good fit, is $\gtrsim1000$$\rm km\,s^{-1}$, supporting the BLR hypothesis. In the case of [OIII] the same broad component leaves strong residuals.