The accretion of quasars at the epoch of reionisation: $JWST$ catches the primeval monsters slowly feasting

TL;DR

This study uses JWST/NIRSpec spectroscopy of eight luminous quasars at $z\geq 5.9$ to self-consistently constrain black-hole masses $M_{BH}$, accretion-disc luminosities $L_{AD}$, and Eddington ratios $λ_{Edd}$ via accretion-disc (AD) modelling. By combining broad-line spectral fitting, single-epoch mass calibrations, bolometric corrections, andGR-corrected AD models (including KERRBB and SLIMBH) with model averaging and Bayesian cross-checks, the authors demonstrate that AD-based measurements yield smaller systematic uncertainties ($\sim$0.2 dex for $M_{BH}$ and $\sim$0.1 dex for $L_{AD}$) than traditional SE methods. The resulting $λ_{Edd}$ distributions are predominantly sub-Eddington (mean around $\log(λ_{Edd})\sim -0.86$) with a tiny super-Eddington fraction (≈0.2%), challenging the view that bright high‑$z$ QSOs routinely accrete near or above the Eddington limit. The work confirms JWST’s capability to test AD models at $z\gtrsim 4$ and suggests that bright blue QSOs at the epoch of reionisation largely reside in sub-Eddington accretion states, with implications for early SMBH growth and feedback processes.

Abstract

Quasars (QSOs) emit an enormous amount of light as a result of the accretion of gas onto supermassive black holes (SMBHs). Thanks to their luminosity, the most distant known QSOs allow us to trace the growth of SMBHs deep into the epoch of reionisation. In this work, we employed $JWST$/NIRSpec observations of eight luminous (log$(L_{3000\,A^{\circ}}/(erg \, s^{-1}))>$45.7) QSOs at $z\geq$5.9 to constrain their accretion properties, namely black hole mass, accretion disc (AD) luminosity, and Eddington ratio ($M_{BH}$, $L_{AD}$, $λ_{Edd}$), by fitting the rest-frame UV and optical emission with different AD models. This method provided self-consistent measurements of both $M_{BH}$ and $L_{AD}$. The uncertainties on $M_{BH}$ and $L_{AD}$, obtained within the AD-modelling framework ($σ^{AD}_{M_{BH}}\sim$0.2 dex; $σ^{AD}_{L_{AD}}\sim$0.1 dex), are significantly smaller than the systematic uncertainties associated with single-epoch $M_{BH}$ ($\sim$0.4 dex) and $L_{AD}$ derived via bolometric corrections ($\sim$0.2 dex). Based on these results, in our sample we found an average Eddington ratio of $\langle \log(λ_{Edd}) \rangle=-0.9$, with a dispersion of $\sim$0.2 dex. Assuming that our high-z QSOs are representative of optically-selected bright blue QSOs, we derive a fraction of systems accreting above the Eddington limit of $\sim$0.2%. In conclusion, this work i) demonstrates the suitability of $JWST$ to test AD models on high-redshift ($z\gtrsim$4) QSOs, thanks to the large NIRSpec spectral coverage; ii) shows that AD modelling can yield robust $M_{\rm BH}$ and $L_{\rm AD}$ measurements, with smaller uncertainties than the typical calibrations; and iii) provides compelling evidence for sub-Eddington accretion in bright high-$z$ QSOs, challenging the widespread paradigm of near- or super-Eddington accretion occurring in these sources.

Figures (11)

• Figure 1: QSOs presented in this work (red stars) and other QSO surveys at different redshifts, namely XQz5 lai2024xqz5, ZQ100 lopez2016, the Gemini/GNIRS sample of shen2019gemini, the $z$$\sim$ 6.3--7.6 sample of yang2021probing, and HYPERION zappacosta2023hyperluminous. Quasars at $z$$\lesssim$ 2.6 from the SDSS DR16Q (wu2022catalog) are shown as black contours. Some high-$z$ QSOs are shared among multiple surveys.
• Figure 2: Spectral fits of @series Mg ii Mg ii Mg ii, H$\beta$, and H$\alpha$ lines of J1425$+$3254. Fluxes are shown in the rest frame. All the components are colour-coded according to the legend. The remaining spectrum after the best-fit model subtraction is shown as a black line. The dotted lines show the expected wavelengths for the most prominent emission lines. For the @series Mg ii Mg ii Mg ii line here we only show the best fit obtained assuming a Lorentzian profile.
• Figure 3: Left: Accretion-disc modelling of J1425$+$3254. The continuum points employed in the fit are marked as cyan dots. The shaded bands represent the 16$\rm^{th}$--84$\rm^{th}$ percentiles of the distribution of 10,000 randomly extracted best-fit models. The dashed lines mark the broad emission lines (@series Mg ii Mg ii Mg ii, H$\beta$, H$\alpha$) used to compute $M_{\rm BH}$ via SE relations. The shaded area highlights the region bluewards of Ly$\alpha$, affected by IGM absorption and thus excluded from the fit. Right: Joint posterior distribution of $M_{\rm BH}$ and $L_{\rm AD}$.
• Figure 4: Top left, top right, bottom left: comparison between the $M_{\rm BH}$ values estimated from SE calibrations and AD fitting for each QSO. Different colours and symbols represent different prescriptions as listed in Table \ref{['tbl:MBH_calibrations']}. Only the fiducial TN12 and N19 calibrations are shown with full coloured symbols. In the case of the DB20 and DB25 calibrations, we employed their recipes for both FWHM and $\sigma$ of the line. For the @series Mg ii Mg ii Mg ii calibrations, we used the broad Gaussian (gau) as well as the Lorentzian (lor) line widths. The AD estimate is marked by a magenta star across all panels. Measurement uncertainties are typically smaller than the point size. The errorbar in each panel highlights the 0.4 dex systematic uncertainty (see Section \ref{['sec:SE_MBH']}). Points are slightly shifted horizontally for visualisation purposes. Bottom right: comparison between the $L_{\rm bol}$ values estimated from bolometric corrections and AD fitting for each QSO. The systematic uncertainty of 0.2 dex is shown as an errorbar on the left. The worst agreement is found for J2239+0207, for which the discrepancy with respect to the N19 expectation is 0.25 dex.
• Figure 5: Distribution of $\lambda_{\rm Edd}$ in our sample. The dashed histograms show the best-fit values distributions, while the solid ones are smoothed for the systematic and statistical uncertainties. Blue and red lines represent the $\lambda_{\rm Edd}$ distributions derived from calibrated SE relations and AD modelling, respectively. In magenta we present the total $\lambda_{\rm Edd}$ distribution for the joint capellupo2015active, cheng2019modelling, and campitiello2020estimating samples at $z$$\lesssim$ 1.5. The cyan histogram highlights the effect of adopting the richards2006spectral bolometric correction. We also report the super-Eddington (sup-Edd) fractions of our sample according to different recipes using the same colour-code as the distributions.