A dormant overmassive black hole in the early Universe

Abstract

Recent observations have found a large number of supermassive black holes already in place in the first few hundred million years after the Big Bang, many of which seem to be overmassive relative to their host galaxy stellar mass when compared with local relation^1-9. Several different models have been proposed to explain these findings, ranging from heavy seeds to light seeds experiencing bursts of high accretion rate^10-16. Yet, current datasets are unable to differentiate between these various scenarios. Here we report the detection, from the JADES survey, of broad Hα emission in a galaxy at z = 6.68, which traces a black hole with a mass of about 4 × 10⁸M_⊙ and accreting at a rate of only 0.02 times the Eddington limit. The black hole to host galaxy stellar mass ratio is about 0.4-that is, about 1,000 times above the local relation-whereas the system is closer to the local relations in terms of dynamical mass and velocity dispersion of the host galaxy. This object is most likely an indication of a much larger population of dormant black holes around the epoch of reionization. Its properties are consistent with scenarios in which short bursts of super-Eddington accretion have resulted in black hole overgrowth and massive gas expulsion from the accretion disk; in between bursts, black holes spend most of their life in a dormant state.

Fig. 1. Prism spectrum and Hα line of GN-1001830.

a, The two-dimensional prism spectrum. b, The one-dimensional prism (black line in the bottom panel) with marked emission lines. c, The spectrum around Hα showing the presence of a broad component. The lines shown are the observed spectrum (black solid line, with grey shading indicating 1σ uncertainties) along with the best-fit line to the narrow (red dashed) and broad (green dashed) components. The [Nii] doublet is shown in blue; it is only marginally detected at 2σ. The magenta solid line shows the total fit. The grey line portion at around 4.95 μm of the spectrum shows the region that was masked because of a possible artefact or Hα emission from a lower redshift interloper. d, The fit residuals for a simple narrow Hα and [Nii] fit (black line) and the best fit, containing a broad component (purple line). The narrow-line-only fit does not account for the broad wings of the line, leaving substantial systematic residuals.

Fig. 2. Comparison of GN-1001830 with other high-z AGN and models in terms of accretion rate, black hole mass and stellar mass of the host galaxy.

a,c, Accretion rate relative to the Eddington limit, λ_Edd, versus black hole mass, log M_BH. b,d, Black hole mass versus stellar mass of the host galaxy log M_*. The green dashed lines indicate constant M_BH/M_* ratios, whereas the solid green line represents the local relation from ref. ; the shaded region shows the scatter. In all panels, GN-1001830 is indicated by a magenta circle with error bars. In a and b, comparison with other JWST-discovered AGN at high redshift is shown in blue^–, and with the QSO population at similar redshifts is shown in orange and yellow^–. The observed negative correlation between λ_Edd and M_BH is probably reflective of the dependence of Eddington luminosity on black hole mass and observational incompleteness and not a separate physical phenomenon. In c and d, comparison with the predictions (at z ≈ 7) from the semi-analytical models from refs. ^, in the scenario of Eddington-limited accretion is shown as grey points and the scenario of light or heavy seeds that can experience super-Eddington accretion as red contours. Error bars indicate 1σ uncertainties.

Fig. 3. Completeness simulation results on the Eddington ratio versus black hole mass plane.

The blue points show the previously discovered JWST sources at 6 < z < 8, as in Fig. 2. The dark green points show the simulated AGN (at z ≈ 7) in the scenario of super-Eddington bursts. GN-1001830 is indicated by a magenta circle with error bars. The colour shading indicates the completeness of the JADES spectroscopic survey in detecting black holes with a given mass and accreting at a given rate relative to Eddington. It can be readily seen that most of the low-accretion rate AGN predicted by super-Eddington bursts lie in the sub-50% completeness region and that GN-1001830 overlaps them at the edge of the high-completeness region. Error bars indicate 1σ uncertainties.

Extended Data Fig. 1. Prism spectrum of GN-1001830.

The top panel shows the 2D spectrum with the y axis representing the shutter pitch and yellower portions showing more positive flux. The bottom panel shows the extracted 1D prism spectrum with emission line locations indicated by coloured vertical lines. The (noisier) R1000 spectrum is shown in blue; the wavelength range is narrowed with respect to Fig. 1 to leave out the noisiest parts of R1000. The panel to the right shows a zoomed-in view on the blended Hγ and [O iii]λ4363 feature along with its decomposition into two Gaussian profiles. Grey shading indicates 1σ uncertainties.

Extended Data Fig. 2. Combined spectrum around Hα compared with the individual exposures.

Grey lines in the top panel show the individual exposures, with the stacked spectrum shown in black. The bottom panel shows the 2D spectrum zoomed in on the same region. It can be seen that there is an outlier in the location of the artefact at λ ≈ 4.94 μm. The 2D cutout also shows the slight spatial offset of the feature.

Extended Data Fig. 3. Grating spectrum around Hβ and [OIII] along with the best-fit model.

It can be seen that the data are well explained by single component fits to each line, indicating no significant outflows. The spike at 3.775 μm is likely a noise feature that survived sigma clipping. Grey shading indicates 1σ uncertainties.

Extended Data Fig. 4. Star-forming main sequence.

Star-formation rate versus stellar mass diagram showing the location of GN-1001830 with respect to the star-forming main sequence. The black dash-dotted line shows the star-forming main sequence fit from ref.  extrapolated to z = 6.677, with the grey shaded area representing the uncertainties. Data at 7 < z < 9 from refs. ^– and ref.  are shown with brown and red symbols. The dashed blue line indicates the limit below which it takes a galaxy more than the Hubble time at z = 6.677 to double its mass. The magenta circle shows the location of GN-1001830, which is consistent with the dashed line within 1σ. Error bars show 1σ uncertainties.

Extended Data Fig. 5. AGN-host decompositions.

The data, residual, model, and fluxes with the recovered galaxy component, for the point source and galaxy decomposition in the 8 JADES NIRCam bands. The figure shows that the galaxy+point source model has fit the data well within all bands without leaving significant residuals. The bottom row shows the modelled point source component; stacked magnitudes in each band are shown above each column. Each panel is 0.8 by 0.8 arcsec in size.

Extended Data Fig. 6. AGN-host SEDs.

The spectral energy distribution for the point source and galaxy decomposition in the 10 NIRCam bands. The figure shows that we have recovered a significant amount of the host galaxy’s flux within all bands and that the AGN SED is reddened. Horizontal bars indicate filter widths, while vertical ones show 1σ uncertainties.

Extended Data Fig. 7. Spectral energy distribution (SED) for the host galaxy fit by Prospector.

Yellow points correspond to the observed photometry. Black squares correspond to the model photometry and the model spectrum is overplotted in black. The chi distribution of the observed to model photometry is shown below. We note that we do not fit the F356W and F410M bands, shown in green, due to strong contamination from the AGN which is readily apparent in their excess flux relative to the other bands. The figure shows that Prospector has fit the observed photometry well. Horizontal bars indicate filter widths, while vertical ones show 1σ uncertainties.

Extended Data Fig. 8. Dynamical mass and velocity dispersion comparisons.

Location of GN-1001830 (magenta point) on the black hole mass versus stellar velocity dispersion (left) and versus dynamical mass of the host galaxy (right). Other high-z AGN found by JWST are shown with blue symbols. The black dash dotted line shows the local M_BH − M_bulge relation from ref. . The solid blue line shows the M_BH − σ_* relation from ref. . Shaded areas show the scatter around these relations. While not yet on the local relations, the offset of GN-1001830 is much less severe than in the BH-stellar mass diagram. Error bars show 1σ uncertainties.

Extended Data Fig. 9. Comparison between GN-1001830 and sources in the FABLE simulation.

The layout is the same as the bottom row of Fig. 2, with sources from super-Eddington simulations shown as red contours, while those from sub-Eddington ones are indicated by grey points. As in the case of the CAT models, the Eddington-limited heavy seed scenario fails to simultaneously explain the high BH-to-stellar mass ratio of GN-1001830 and the very low accretion rate. Instead, the scenario in which BHs experience super-Eddington accretion phases can match the properties of GN-1001830, although additional simulations would be need to bridge the gap between the 100 h⁻¹ Mpc box and proto-cluster zoom-in simulations. Error bars show 1σ uncertainties.

Extended Data Fig. 10. Completeness simulations.

Same as Fig. 3, except with super-Eddington models of the FABLE simulation, showing that dormant, post super-Eddington burst AGN are strongly biased against in the current surveys. Error bars show 1σ uncertainties.