Astronomy & Astrophysics manuscript no.

aa ©ESO 2024
June 27, 2024

Size matters: are we witnessing super-Eddington accretion in

high-redshift black holes from JWST?
Alessandro Lupi1,2,⋆ , Alessandro Trinca1, , Marta Volonteri3 , Massimo Dotti4,2 , and Chiara Mazzucchelli5

1
Dipartimento di Scienza e Alta Tecnologia, Università degli Studi dell’Insubria, via Valleggio 11, I-22100, Como, Italy
2
INFN, Sezione di Milano-Bicocca, Piazza della Scienza 3, I-20126 Milano, Italy
3
Institut d’Astrophysique de Paris, UMR 7095, CNRS and Sorbonne Université, 98 bis boulevard Arago, 75014 Paris, France
4
Dipartimento di Fisica “G. Occhialini”, Università degli Studi di Milano-Bicocca, Piazza della Scienza 3, I-20126 Milano, Italy
5
Instituto de Estudios Astrofísicos, Facultad de Ingeniería y Ciencias, Universidad Diego Portales, Avenida Ejercito Libertador 441,
Santiago, Chile
arXiv:2406.17847v1 [astro-ph.HE] 25 Jun 2024

Draft June 27, 2024

ABSTRACT

Observations by the James Webb Space Telescope of the Universe at z ≳ 4 have shown that massive black holes (MBHs) appear
extremely overmassive compared to the local correlation for active galactic nuclei. In some cases, these objects might even reach
half the stellar mass inferred for the galaxy. Understanding how such objects formed and grew to this masses has then become a big
challenge for theoretical models, with different ideas ranging from heavy seed to super-Eddington accretion phases. Here, we take a
different approach, and try to infer how accurate these MBH mass estimates are and whether we really need to revise our physical
models. By considering how the emerging spectrum (both the continuum and the broad lines) of an accreting MBH changes close to
and above the Eddington limit, we infer a much larger uncertainty in the MBH mass estimates relative to that of local counterparts,
up to an order of magnitude, and a potential preference for lower masses and higher accretion rates, which i) move them closer to the
local correlations, and ii) might indicate that we are witnessing for the first time a widespread phase of very rapid accretion.
Key words. accretion disks - black hole physics - galaxies: active - galaxies: high-redshift

1. Introduction compensate for the stunted growth in low-mass galaxies (see,

e.g. Anglés-Alcázar et al. 2017). Apart from the mass of these
Massive black holes (MBHs) are ubiquitously found to inhabit MBHs, another important difference with the local population
the centre of massive galaxies up to redshift z ≳ 6 (e.g. Fan et al. is that these MBHs seem to be extremely massive compared to
2006; Mortlock et al. 2011; Bañados et al. 2018; Fan et al. 2023; their galaxy hosts, lying well above the local correlations (Fa-
Maiolino et al. 2023), with masses in the range ∼ 105 − 1010 M⊙ . rina et al. 2022; Maiolino et al. 2023; Yue et al. 2024; Stone
Observationally, they are commonly identified via gas accretion et al. 2024) in some cases even weighing more than half the to-
through the conversion of gravitational energy into radiation, tal stellar mass (Juodžbalis et al. 2024). Note, however, that an
which makes them shine as Active Galactic Nuclei (AGN). They important role in the comparison is also played by the galaxy
sometimes also produce powerful collimated jets. mass employed, either the stellar mass, as in the recent JWST
MBHs are expected to gain most of their mass via radiatively results, or the dynamical mass, as in the case of ALMA ob-
efficient accretion (Soltan 1982; Marconi et al. 2004), hence they servations (Decarli et al. 2018; Izumi et al. 2021; Farina et al.
should have formed from lower-mass black hole ‘seeds’ (see, 2022). Among the theoretical efforts to explain these systems,
e.g., Inayoshi et al. 2020 and Volonteri et al. 2021 for a review). the most promising solutions consider i) a somewhat extremely
With the advent of the James Webb Space Telescope efficient heavy seed formation at high redshifts and an efficient
(JWST), we have pushed the observational limit deeply in the suppression of star formation by the accreting MBH, ii) a strong
dark ages, finding galaxies up to z ∼ 14 (Carniani et al. 2024). observational bias (Li et al. 2024), or iii) a population of pri-
Some of these galaxies also hosted MBHs, which appear to mordial MBHs (Ziparo et al. 2022; Dolgov 2024). Despite these
challenge most MBH formation mechanisms, unless one as- efforts, a clear consensus is still missing to date, partially be-
sumes an initially heavy seed (104 − 105 M⊙ ) and a continuous cause of the large uncertainties in the stellar mass and (poten-
growth at the Eddington limit. Theoretical models are further tially) in the MBH mass. This second possibilty is rarely con-
challenged by the large abundance of these (candidate) objects sidered. In fact, the MBH mass at these redshifts is commonly
(e.g. Harikane et al. 2023; Maiolino et al. 2023; Greene et al. inferred through the single-epoch method, employing the virial
2024), which implies a formation efficiency of massive seeds theorem combined with the correlations between the broad Hα,
way larger than what found in theoretical models. Several stud- Hβ, or MgII line widths and luminosities, and the emission prop-
ies have shown that this issue can be alleviated if one consid- erties of the continuum emitted by the innermost regions of the
ers the plausibility of accretion above the Eddington limit (e.g., accretion disc. These correlations have been calibrated in the lo-
Lupi et al. 2016; Pezzulli et al. 2016; Regan et al. 2019; Mas- cal Universe (z ≲ 0.3 Vestergaard & Osmer 2009; Bentz et al.
sonneau et al. 2023; Lupi et al. 2023; Shi et al. 2024), which can 2013; Reines et al. 2013; Reines & Volonteri 2015), and then
⋆ extrapolated at high redshift.
[email protected]

Article number, page 1 of 8

 A&A proofs: manuscript no. aa

Recently, King (2024) pointed out that close-to-Eddington or the latter needed to determine the broad-line emission from the
super-Eddington accreting MBHs would have (i) the emission disc properties (see, e.g. Osterbrock & Ferland 2006). For con-
from the accretion disc beamed by multiple scatterings within sistency with Kubota & Done (2019), we normalize the spectrum
the funnel created by a central thickening of the disc itself and bolometric luminosity to the value estimated from the numerical
(ii) unvirialized BLR whose dynamics is mostly dominated by integration of the slim disc solution by Sadowski (2011). This
outflows. Under such conditions, King (2024) demonstrated that normalisation allows gives us the effective radiative efficiency η
the MBH estimates inferred would be artificially biased towards for each combination of the three model parameters, that we use
high values, and argued that such an effect might be particularly in the rest of the paper to determine L/LEdd = η/ηthin Lthin /LEdd .
relevant for high-redshift AGN. Another potential source of bias With the table so created, we then built a theoretical model for
could instead result from an inaccurate estimate of the broad line the BLR emission to be compared with observations. In partic-
region (BLR) size, as suggested by recent reverberation map- ular, the observed quantities we considered are: the broad-line
ping campaigns - including the SEAMBH (Du et al. 2014) and width (either Hα or Hβ) and the luminosity (either the Hα lumi-
the SDSS-RM (Grier et al. 2017) - of multiple highly accreting nosity or the luminosity at 5100Å), according to the values re-
MBHs, Martínez-Aldama et al. (2019). ported in the corresponding observational works (Harikane et al.
In particular, these campaigns demonstrated that the time lag 2023; Maiolino et al. 2023; Übler et al. 2023; Greene et al.
of the Hβ line, which is directly associated to the size of the BLR, 2024), both with their associated uncertainties σ.
depends on the accretion rate of the MBH, and shortens for ac- Our model is defined as follows.
cretion rates above fEdd ∼ 0.3 (Wang et al. 2014, W14 hereafter).
The proposed interpretation for such effect is radiation pressure – Given a specific combination of MBH , Lthin /LEdd and aBH ,
which, for accretion rates close and above the Eddington limit, we extract L5100Å and Lion via tri-linear interpolation on our
thickens the accretion disc. Such a thicker disc is better described table.
by the slim-disc solution (Abramowicz et al. 1988) rather than by – For simplicity, we do not make any specific assumption on
a more standard radiatively efficient Shakura & Sunyaev (1973, the cloud properties in the BLR, and generically assume that
SS hereafter) disc, and results in a lower flux of ionizing pho- they are homogeneously distributed around the central MBH
tons reaching the BLR clouds compared to a radiatively efficient (Wang et al. 2014).1 Following W14, we assume that self-
AGN with identical optical spectrum. In these conditions, the shadowing is negligible within the funnel, which is defined
BLR splits in unshadowed and shadowed regions, the latter re- by an aperture
ceiving less photons and shrinking in size, which result in a net
fEdd < 8
 ◦
shorter lag. 
 90
θfun ≈  ◦ ◦
fEdd < 100 ,

Motivated by these results, in this work we explore the ef- 
118 − 33 log fEdd 8 ≥ (1)
fect of a varying BLR size, based on the aforementioned results  76◦ − 12◦ log fEdd

 fEdd ≥ 100
and on a fully physical approach, on the inferred MBH masses
in the most challenging high-redshift sources observed by JWST where fEdd ≡ ṀBH c2 /LEdd = η−1
thin Lthin /LEdd , and that the ion-
to date. In particular, we account for the possibility that the ob- ising radiation emitted within this solid angle directly im-
served luminosities might be the result of a lower mass, highly pinges on the BLR clouds. Assuming an intrinsic spectrum
accreting MBH, with the aim of assessing potential biases in the with angular distribution dF/dθ ∝ cos θ, we then determine
MBH mass estimates. the broad-line emission from clouds within the funnel solid
The manuscript is organised as follows. In Section 2 we de- angle assuming the local correlation (Greene & Ho 2005)
scribe our procedure to estimate the MBH mass, in Section 3 we
present our results, and in Section 4 we discuss potential caveats LHβ,fun xfun L5100Å
!1.133±0.005
in the analysis and draw our conclusions. −1
= (1.425 ± 0.007) , (2)
42
10 erg s 1044 erg s−1

2. Methods where xfun is the fraction of the total ionising flux within
the funnel. Outside the funnel, instead, we model self-
In order to test how relevant the evolution of the BLR size with
shadowing through Eq. (19) in W14
the Eddington ratio is in high-redshift systems, we build a theo-
retical model of the accretion disc and the BLR emissions based !−0.6
LHβ,s−s ξs−s cos θfun fEdd
on the electromagnetic spectrum of a slim disc, as defined by ≈ 0.28 , (3)
the agnslim model in xspec (Kubota & Done 2019). Of the many LHβ,fun ξfun 1 − cos θfun 50
parameters available in the model, in our work we only consid-
ered the impact of the three main ones: the MBH mass MBH , where ξs−s and ξfun are the anisotropic factors for Hβ emis-
the Eddington ratio Lthin /LEdd ≡ ηthin ṀBH c2 /LEdd , with Lthin sion from the BLR clouds in the self-shadowed region and
and ηthin being the bolometric luminosity and the radiative ef- within the funnel respectively. As these values are com-
ficiency of a SS disc, and the MBH spin aBH , leaving the oth- pletely unconstrained, but for pole-on observers (where they
ers to their default value. We sample 6250 different combina- are both equal to unity). Here, we assume for simplicity that
tions, with 25 logarithmically-spaced MBH masses between 105 they are always of the same order and remove them from
and 1010 M⊙ , 25 logarithmically-spaced Eddington ratios in the the equation. The total Hβ luminosity is finally estimated as
range 0.01 − 103 , and 10 linearly-spaced values of the MBH LHβ = LHβ,fun + LHβ,s−s .
spin between 0 and 0.998. The spectrum covers the energy range 1
This is a very simplistic assumption, as both the cloud angular dis-
0.1 eV – 100 keV, corresponding to a wavelength range 0.12Å –
tribution and their maximum distance from the source are completely
12.4 µm, in 1000 logarithmically-spaced bins. unconstrained. Previous studies hinted at a common disc-like geometry
After the spectra have been generated, for each combination for the BLR (e.g. Wills & Browne 1986; Collin-Souffrin 1987; Run-
we tabulate the luminosity at 5100Å (λLλ ), and the ionising lu- noe et al. 2013). We stress that a flatter BLR would enhance the self-
minosity Lion above E > 0.1 keV (soft-X; Kwan & Krolik 1979), shadowing effect.

Article number, page 2 of 8

 Alessandro Lupi et al.: Size matters: are we witnessing super-Eddington accretion in high-redshift black holes from JWST?

In order to determine the Hα luminosity, we assume the stan-

dard scaling from Greene & Ho (2005) 10 Yue+24
Greene+24
!1.157±0.005
LHα L5100Å,proxy Harikane+23
−1
= (5.25 ± 0.02) , (4) 9 Maiolino+23
42
10 erg s 1044 erg s−1

log(MBH,new /M )
Ubler+23
where 8
!1/(1.133±0.005)
L̃5100Å,proxy LHβ
= . 7
1044 erg s−1 (1.425 ± 0.007) × 1042 erg s−1 8

(5)
6
6
We note that, when the broad-line flux or luminosity are not
reported, as in the case of Yue et al. (2024), we directly com- 6 8
5
pare L5100Å from our model with the observed data. 5 6 7 8 9 10
– The last piece of information we need for the model is log(MBH,ref /M )
the full-width-half-maximum (FWHM) of the broad lines,
which we determine by assuming virial equilibrium in the Fig. 1. MBH estimates from the MCMC for the validation run against
BLR, which gives the MBH mass reported in the observational studies considered in this
s work. The black line corresponds to the 1:1 relation, with the grey
RBLR shaded area 0.3 dex wide. The dots correspond to the MBHs in Yue
FWHMHβ = (6) et al. (2024, blue), Greene & Ho (2005, orange), Harikane et al. (2023,
fvirialGMBH
green), Maiolino et al. (2023, red), and Übler et al. (2023, purple). In
the inset we show the results obtained for the Reines & Volonteri (2015)
for the Hβ line, where fvirial is a parameter taking into ac- data as cyan crosses.
count the unknown inclination, geometry, and kinematics of
the BLR. In this work, we consider as our ‘fiducial’ case
fvirial = 1.075 (Reines & Volonteri 2015), but also explore where Yi is the observed broad-line FWHM and the luminos-
a case in which fvirial ∝ (FWHMline,obs )−k (Mejía-Restrepo ity, Ȳi is the value predicted by our model, and si is the uncer-
et al. 2018, MR18 hereafter), with k = 1 (Hα) or k = 1.17 tainty in the observed data (assumed Gaussian). The parameters
(Hβ). In order to estimate RBLR , we employ the relations de- of our model that we aim at constraining are MBH , Lthin /LEdd ,
rived by Martínez-Aldama et al. (2019) and aBH . As priors, we assume a log-flat distribution for MBH
and Lthin /LEdd over the intervals [5, 10] and [−3, 3] respectively,
RBLR
log = α log fEdd + β, (7) and a uniform distribution for aBH between 0 and 0.998. We
RBLR,Ref ran the MCMC for 10000 steps employing 32 walkers.2 In or-
der to incorporate the uncertainties in the correlations used by
which takes into account the self-shadowing of the BLR. For our model, every time we employ one of the relations above,
the fiducial model, we set α = −0.143, β = −0.136, and we sample the slope and normalisation from a Gaussian distri-
assume RBLR,Ref as the reference Hβ BLR size estimate by bution centred on the best-fit value and with σ defined by the
Bentz et al. (2013) uncertainty of the fit.3 This choice ensures a proper coverage of
the parameter space, even with a very limited dataset given by
RBLR,Ref L5100Å
log = 1.527 ± 0.31 + 0.533+0.035
−0.033 log . only two values. In the case of a FWHM-dependent virial fac-
1lt − day 1044 erg s−1 tor, we randomly sample the virial factor for each source before
(8) starting the MCMC from a Gaussian distribution centred on the
observed broad-line FWHM with the observed uncertainty, and
In the MR18 case, we employ instead α = −0.283, β = keep it constant throughout the optimisation procedure, in line
−0.228, and fEdd = fvirial
−2
η−1
thin Lthin /LEdd . For sources where with the correlation found by Mejía-Restrepo et al. (2018).
the MBH mass is estimated from the Hα, we finally convert
FWHMHβ to the FWHMHα through the Bentz et al. (2013)
relation 3. Results
 FWHM 
Hα 3.1. Model validation
FWHMHβ = (1.07 ± 0.07) × 103 km s−1 . (9)
103 km s−1 Before running the MCMC with the fiducial model described in
In order to compare our model predictions with observations, the previous section, we decided to validate our procedure by
we employ a Markov-Chain Monte Carlo (MCMC) algorithm neglecting the effects due to the accretion disc transition to a
as implemented in the emcee package (Foreman-Mackey et al. slim-disc. In practice: i) we employed
2013). We consider here as our observational sample the sources
RBLR L5100Å
identified by Harikane et al. (2023); Maiolino et al. (2023); Yue log = 1.555 ± 0.31 + 0.542+0.035
−0.033 log (11)
et al. (2024); Übler et al. (2023), and Greene et al. (2024), in the 1lt − day 1044 erg s−1
redshift range 4 ≲ z ≲ 7. The likelihood L for the MCMC is 2
The number of steps chosen corresponds to about 100 autocorrelation
defined through time-scales, which is sufficient to guarantee a robust optimisation.
3
  When the uncertainties are asymmetric, we approximate the distri-
1 X  (Yi − Ȳi )2 bution as a Gaussian distribution with σeff the average between the two
ln L = − + i ,
2 

 2
ln(2πs )  (10)
2 i si uncertainties.

Article number, page 3 of 8

 A&A proofs: manuscript no. aa

log(MBH /M ) = 9.24+0.06
−0.08

log(L/LEdd ) = −0.16+0.14
−0.14
45 .30 .15 .00 .15
0
log(L/LEdd )
0 0
−
0
−

aBH = 0.33+0.17
−0.20
0.
−
6
0.
aBH

4
0.
2
0.
0

− .4
45

30

15
00

15

6
9.

9.

9.

9.

0.

0.

0.
9
0.

0.

0.
0.

0.
−

log(MBH /M ) log(L/LEdd ) aBH

Fig. 2. Corner plot resulting from the MCMC validation run on J1030+0524 (Yue et al. 2024) for the three physical parameters of the model MBH ,
L/LEdd (obtained by rescaling Lthin /LEdd as described in Section 2), and aBH . The blue lines correspond to the values reported in the original work.

in Eq. (6), as done in Reines & Volonteri (2015), ii) we inferred luminosity assuming the cosmology and redshift reported in the
the broad-line luminosities from the scaling relations in Bentz discovery paper, while Harikane et al. (2023) give directly the
et al. (2013, see also Eq.s 2 and 4), using our tabulated value for broad line luminosity. We will refer to the MBH masses obtained
L5100Å , and iii) we assumed a constant fvirial = 1.075 as in Reines with this procedure as ‘validation’ in the following.
& Volonteri (2015). With these assumptions, we found our best
parameters to be in line with those in the published works, as
shown in Fig. 1. The inset shows the remarkable agreement of In general, we find that the spin is very poorly constrained
our procedure with the data by Reines & Volonteri (2015). The by our MCMC, due to the limited amount of observational data
only mild discrepancy is in the data by Greene et al. (2024), we have and the moderate dependence on its value, whereas the
where the estimates show a somewhat larger scatter around the MBH mass and L/LEdd are typically well determined. As an ex-
1:1 relation. The systematic small shift of the Harikane et al. ample of the robustness of our procedure, we report in Fig. 2
(2023, above) and Maiolino et al. (2023, below) is likely related the corner plot obtained for J1030+0524 from Yue et al. (2024),
to the information provided in the respective papers. Maiolino one of the most massive sources in the sample, which is also one
et al. (2023) report the Hα flux, which is then converted into of the few validation cases in which the posterior distribution of
the MBH spin exhibits a peak rather than being almost flat. The
Article number, page 4 of 8
 Alessandro Lupi et al.: Size matters: are we witnessing super-Eddington accretion in high-redshift black holes from JWST?

blue lines in the corner plot correspond to the estimates from the present in the local AGN sample, but with variations not larger
literature, which agree well with our estimate. than 0.2 dex, about a factor of 3 smaller than the intrinsic uncer-
tainty by Reines & Volonteri (2015), and typically much smaller
than the 0.5 dex found in the high-redshift sample.
3.2. Full model
In the left panel of Fig. 3, we show the same plot of Fig. 1, but 3.3. AGN spectra
for the slim-disc model. We clearly observe that the fiducial case
is close to the 1:1 relation, but typically offset of about 0.5 dex As a final check of our procedure, we built synthetic MBH emis-
towards lower values compared to those reported in the litera- sion spectra for the analysed sources employing all of the three
ture, with correspondingly higher accretion rates, often super- models considered in this work. The best parameters to build the
Eddington. The MR18 case, because of the additional depen- spectra are defined as the average among the 10 evaluations of
dence of the virial factor on the broad-line FWHM, results in our MCMC with the maximum likelihood. For each model, we
even lower MBH masses. The Ṁ/ ṀEdd ratio is shown in the right extracted the continuum spectrum from our tables and added on
panel, where ṀEdd ≡ 10LEdd /c2 , assuming the fiducial and the top the emission of the broad line (but for the sources in Yue et al.
MR18 cases of our slim-disc model (red dots and purple squares 2024, where we employed the luminosity at 5100Å). In order to
respectively) and the validation run (black crosses). Despite the consistently compare with observed spectra, we also accounted
differences in the two slim-disc models, we find that the distri- for dust extinction following the attenuation law by Calzetti et al.
bution of Eddington ratios is quite similar, with the least massive (2000), assuming RV = 4.05 for the source by Harikane et al.
MBHs more often preferring higher accretion rates. The MR18 (2023), and the Small Magellanic Cloud value RV = 2.74 for
case, consistently with the lower MBH masses reported in the the sources by Maiolino et al. (2023) and Greene et al. (2024),
left panel, almost always prefer super-Eddington rates, with val- to be consistent with the assumptions in the different studies. For
ues between 10 and a 100 times Eddington. Interestingly, the the sources observed by the EIGER program (Yue et al. 2024) we
most massive MBHs from Yue et al. (2024) also prefer super- did not include any attenuation. The results are reported in Fig. 5
Eddington accretion rates, well above the Eddington limit, which for 4 selected sources: CEERS_02782 (Harikane et al. 2023),
might hint at an ineffective self-regulation of their growth via JADES_000954 (Maiolino et al. 2023), J0100+2802 (Yue et al.
feedback processes. Another interesting aspect is that, even in 2024), and UNCOVER_13821 (Greene et al. 2024). We clearly
the validation run, some MBHs seem to lie above the Eddington see that our models can always recover the spectral properties
limit, especially those with the lowest masses, suggesting that of the sources, both the continuum region and the broad Hα line
the estimate of their properties according to the local correla- intensity and width, independently of the assumptions. The only
tions might be biased towards higher MBH masses. Finally, note peculiar case is J0100+2802, where the complexity of the broad
that discriminating such high accretion rates from more typical Hβ line profile, not symmetric and with potential hints of off-
cases is not easy, as the luminosity of these objects would never set components, together with the missing modelling of the Iron
exceed, even in the most extreme cases, a few times the Edding- emission in our model, does not allow us to recover the exact
ton luminosity (5-10). spectrum. Nonetheless, we find that our model very well repro-
As already discussed above, also with slim-disc model, the duces the power-law continuum, but for a mildly higher normal-
MBH and the Eddington ratio in our analysis are typically con- isation, simply due to the use in our MCMC of the total contin-
strained within one order of magnitude, whereas the MBH spin uum luminosity reported in Yue et al. (2024) instead of the con-
is almost always uniformly distributed, which suggests that our tribution of the power-law component only.4 This confirms i) the
model can accommodate the observed data almost independently robustness of our procedure, and ii) that the dependence of the
of the spin. At super-Eddington rates this is expected, as the ef- BLR emission on the accretion disc structure and the Eddington
fective radiative efficiency does not depend on the spin (Madau ratio is somewhat degenerate, resulting in potentially significant
et al. 2014). Below the Eddington limit, instead, this suggests differences in the MBH mass estimate if not properly taken into
that the information available is not sufficient to actually disen- account.
tangle the spin from the other two parameters.
Finally, we can assess how the correlations between MBHs
and their hosts would change considering the results of our full 4. Discussion and conclusions
model. The results are shown in Fig. 4, for the fiducial case (left
In this work, we have built a semi-empirical model of the BLR
panel) and the MR18 one (right panel). We can observe that our
emission of MBHs in different accretion regimes. By combin-
estimates are closer to the local relations, and this effect is more
ing theoretical models of the emission of thin and slim accre-
relevant for the MR18 case. Interestingly, this decrease does not
tion discs (Kubota & Done 2019) with observed scaling relations
completely realign the MBHs with the local correlations, but
at low-redshift which naturally account for different accretion
suggests that the current estimates, especially for the lowest mass
regimes, we have built a versatile model that can be applied to
MBHs observed, could have a much larger uncertainties than re-
high-redshift sources as those recently observed by JWST.
ported in the literature, and their overmassiveness relative to the
We have incorporated our model in a MCMC tool that we
host should be considered in the light of what we found in this
used to re-analyse some recent candidate MBHs from JWST
work, besides observational biases.
observations. Our results showed that, in many cases, a super-
Even though not reported here, as an estimate of the stel-
Eddington accreting MBH is preferred with respect to the stan-
lar mass is not available, we performed our analysis also on the
dard SS accretion disc, which translates in MBH masses of up
sources by Matthee et al. (2024), finding similar variations in the
to an order of magnitude lower. This is in contrast with local
MBH mass to those just discussed. In order to check whether
the inclusion of a slim disc emission produced a rigid shift of the 4
As a check, we re-ran our MCMC on J0100+2802 with a 5% lower
MBH masses also for the local sources, we reanalysed the Reines luminosity at 5100Å (consistent with the expected power-law contri-
& Volonteri (2015) sample using our fiducial model. We found bution), and found that with almost identical MBH mass estimates the
that, on average, a decrease in the inferred MBH mass was also agreement with the power-law fit was remarkable, as expected.

Article number, page 5 of 8

 A&A proofs: manuscript no. aa

10 fiducial
MR18 2
9
log(MBH,slim /M )

log(Ṁ /ṀEdd )
8
0

7 −1

−2 validation
6
fiducial
MR18
−3
5
5 6 7 8 9 10 6 7 8 9 10
log(MBH,Ref /M ) log(MBH /M )

Fig. 3. Left panel: same as Fig. 1, but for our full model, with the estimates of the entire sample shown as red dots (fiducial) and purple squares
(MR18). The black dashed line is to guide the eye and corresponds to a 0.5 dex offset relative to the 1:1 relation. Right panel: Eddington ratio
distribution for our fiducial model (red dots), the MR18 case (purple squares), and the validation run (black crosses) as a function of the estimated
MBH mass. The thick grey dashed line corresponds to the Eddington limit.

1010 Reines+15 1010 Reines+15

Reines+15 (literature) Reines+15 (literature)
literature literature
109 fiducial 109 MR18
log(MBH /M )

log(MBH /M )

108 108

107 107

106 106

105 7 105 7
10 108 109 1010 1011 1012 10 108 109 1010 1011 1012
log(Mstar /M ) log(Mstar /M )

Fig. 4. MBH mass–stellar mass relation for the source in our sample. We show the local AGN from Reines & Volonteri (2015) as blue stars and
orange diamonds, with the underlying shaded area correspond to the 1-σ and 2-σ uncertainties around the best fits to the local samples (grey and
cyan for inactive and active galaxies respectively). The original data from the literature is shown as green circles, whereas our new estimates are
reported as red dots (left panel) and purple crosses (right panel) for the two virial factors considered. For completeness, we also show as magenta
dotted lines constant mass ratios of 0.01 and 0.1.

sources as those by Reines & Volonteri (2015), where more than nosities of some of these sources compared to a standard SS disc,
95 per cent of the AGN are sub-Eddington and our fiducial model without being for this reason more physically plausible. More-
almost perfectly recovers the masses reported in the literature. over, any difference in the structure of the BLR (different ge-
We also note that the missing detection in X-rays of many of ometry of the clouds, different density, etc.), as well as different
these sources might be compatible with a slim accretion disc, inclinations, might in principle produce similar effects without
but we leave this aspect to future investigations. requiring a highly super-Eddington accretion rate. All these un-
certainties enter the virial factor, whose definition can produce
Despite the extreme relevance of potentially detecting and variations in the MBH mass estimate up to one order of magni-
identifying highly super-Eddington sources, the sustainability of tude, as we have shown here, especially in high-redshift systems
this accretion phase over long time-scales is unclear (see, e.g. for which only a limited amount of information is available.
Regan et al. 2019; Massonneau et al. 2023; Lupi et al. 2023).
In particular, there is a potential degeneracy between the MBH As for our model, King (2024) pointed out that high-redshift
mass and the Eddington ratio, and we cannot completely exclude MBH mass estimates could be biased toward too high values.
a biased preference for super-Eddington accretion in low-mass Differently from King (2024), in our analysis we did not consider
systems. In fact, because of the radiation trapping in the inner- any radiation beaming nor the possibility that the BLR might
most regions of the accretion disc, which suppresses the increase be mainly dominated by unvirialized outflows. Considering the
in ionising and bolometric luminosity, a slim disc model has more likely super-Eddington nature of many observed sources,
more freedom to match the combination of FWHM and lumi- and the fact that in these conditions radiation beaming as well
Article number, page 6 of 8
 Alessandro Lupi et al.: Size matters: are we witnessing super-Eddington accretion in high-redshift black holes from JWST?

CEERS 02782 (z = 5.24): log(MBH /M ) = 7.62

Fλ (10−20 ergs−1 cm−2 Å−1 )
validation (7.8; −0.77; 0.46)
2 Hα
fiducial (7.7; −0.67; 0.52)
MR18 (7.4; 0.32; 0.64)

1
Hβ

0
2.0 2.5 3.0 3.5 4.0 4.5 4.05 4.10 4.15
λ (µm)
JADES 000954 (z = 6.76): log(MBH /M ) = 7.9
Fλ (10−20 ergs−1 cm−2 Å−1 )

validation (7.6; −0.32; 0.46) Hα

15
fiducial (7.4; 0.20; 0.42)
MR18 (7.0; 1.15; 0.41)
10

Hβ
5

0
3.0 3.5 4.0 4.5 5.0 5.05 5.10 5.15
λ (µm)
J0100+2802 (z = 6.33): log(MBH /M ) = 10.06
Fλ (10−18 ergs−1 cm−2 Å−1 )

15 Hβ [O III]

10

validation (10; −0.37; 0.42)

5
fiducial (9.1; 1.22; 0.26)
MR18 (8.7; 2.35; 0.21)
0
3.3 3.4 3.5 3.6 3.7 3.8 3.9
λ (µm)
UNCOVER 13821 (z = 6.34): log(MBH /M ) = 8.1
validation (8.0; −1.09; 0.40) Hα
20
fiducial (7.7; −0.42; 0.49)
MR18 (7.2; 1.02; 0.59)
fλ

10
Hβ

0
1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 4.8 4.9
λ (µm)
Fig. 5. Reconstructed spectra for 4 selected sources in our sample: CEERS_02782, JADES_000954, J0100+2802, and UNCOVER_13821. The
observed spectra (obtained from the public data release of the different programs, but for the UNCOVER source, which has been extracted from
the published paper) are shown as black solid lines (with the right panel showing a zoom on the Hα line), the blue dashed, orange dash-dotted, and
green dotted lines refer to our validation, fiducial, and MR18 models respectively. The cyan vertical line in J0100+2802 corresponds to λ = 5100Å
redshifted to the observer frame, which we used to constrain the models. The grey line corresponds to the power-law continuum component from
the fit by Yue et al. (2024). All but the UNCOVER source report absolute fluxes, whereas in the UNCOVER case the flux is normalised to the
luminosity at 2500Å, as done in Greene et al. (2024). The numbers reported in the legend correspond to the parameters employed for each model
log(MMBH /M⊙ ), log(L/LEdd ), and aBH , whereas the mass estimates above each panel correspond to those in the corresponding observational papers.

Article number, page 7 of 8

 A&A proofs: manuscript no. aa

as nuclear ouflows becomes more significant, we expect the un-

certainties in the mass estimate to become even larger. Unfor-
tunately, the limited data available does not allow us to confirm
whether a bias in the mass is real or not, and whether such a
bias might realign MBHs with the local correlation. However, it
provides some insights on the impact of detailed accretion disc
physics on the MBH mass estimates. In the future, we will incor-
porate additional information from the observed spectra that will
help us to better constrain the actual mass through our physically
motivated model.
Acknowledgements. AL and AT acknowledge support from PRIN MUR
"2022935STW". AL thanks the organisers of the "Massive black holes in the
first billion years" conference Micheal Tremmel and John Regan, as well as Ri-
carda Beckmann, Amy Reines, and Alberto Sesana for useful discussions that
inspired this work.

References
Abramowicz, M. A., Czerny, B., Lasota, J. P., & Szuszkiewicz, E. 1988, ApJ,
332, 646
Anglés-Alcázar, D., Faucher-Giguère, C.-A., Quataert, E., et al. 2017, MNRAS,
472, L109
Bañados, E., Venemans, B. P., Mazzucchelli, C., et al. 2018, Nature, 553, 473
Bentz, M. C., Denney, K. D., Grier, C. J., et al. 2013, ApJ, 767, 149
Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682
Carniani, S., Hainline, K., D’Eugenio, F., et al. 2024, arXiv e-prints,
arXiv:2405.18485
Collin-Souffrin, S. 1987, A&A, 179, 60
Decarli, R., Walter, F., Venemans, B. P., et al. 2018, ApJ, 854, 97
Dolgov, A. D. 2024, arXiv e-prints, arXiv:2401.06882
Du, P., Hu, C., Lu, K.-X., et al. 2014, ApJ, 782, 45
Fan, X., Bañados, E., & Simcoe, R. A. 2023, ARA&A, 61, 373
Fan, X., Strauss, M. A., Richards, G. T., et al. 2006, AJ, 131, 1203
Farina, E. P., Schindler, J.-T., Walter, F., et al. 2022, ApJ, 941, 106
Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125,
306
Greene, J. E. & Ho, L. C. 2005, ApJ, 630, 122
Greene, J. E., Labbe, I., Goulding, A. D., et al. 2024, ApJ, 964, 39
Grier, C. J., Trump, J. R., Shen, Y., et al. 2017, ApJ, 851, 21
Harikane, Y., Zhang, Y., Nakajima, K., et al. 2023, ApJ, 959, 39
Inayoshi, K., Visbal, E., & Haiman, Z. 2020, ARA&A, 58, 27
Izumi, T., Matsuoka, Y., Fujimoto, S., et al. 2021, ApJ, 914, 36
Juodžbalis, I., Maiolino, R., Baker, W. M., et al. 2024, arXiv e-prints,
arXiv:2403.03872
King, A. 2024, MNRAS, 531, 550
Kubota, A. & Done, C. 2019, MNRAS, 489, 524
Kwan, J. & Krolik, J. H. 1979, ApJ, 233, L91
Li, J., Silverman, J. D., Shen, Y., et al. 2024, arXiv e-prints, arXiv:2403.00074
Lupi, A., Haardt, F., Dotti, M., et al. 2016, MNRAS, 456, 2993
Lupi, A., Quadri, G., Volonteri, M., Colpi, M., & Regan, J. A. 2023, arXiv e-
prints, arXiv:2312.08422
Madau, P., Haardt, F., & Dotti, M. 2014, ApJ, 784, L38
Maiolino, R., Scholtz, J., Curtis-Lake, E., et al. 2023, arXiv e-prints,
arXiv:2308.01230
Marconi, A., Risaliti, G., Gilli, R., et al. 2004, MNRAS, 351, 169
Martínez-Aldama, M. L., Czerny, B., Kawka, D., et al. 2019, ApJ, 883, 170
Massonneau, W., Volonteri, M., Dubois, Y., & Beckmann, R. S. 2023, A&A,
670, A180
Matthee, J., Naidu, R. P., Brammer, G., et al. 2024, ApJ, 963, 129
Mejía-Restrepo, J. E., Lira, P., Netzer, H., Trakhtenbrot, B., & Capellupo, D. M.
2018, Nature Astronomy, 2, 63
Mortlock, D. J., Warren, S. J., Venemans, B. P., et al. 2011, Nature, 474, 616
Osterbrock, D. E. & Ferland, G. J. 2006, Astrophysics of gaseous nebulae and
active galactic nuclei
Pezzulli, E., Valiante, R., & Schneider, R. 2016, MNRAS, 458, 3047
Regan, J. A., Downes, T. P., Volonteri, M., et al. 2019, MNRAS, 486, 3892
Reines, A. E., Greene, J. E., & Geha, M. 2013, ApJ, 775, 116
Reines, A. E. & Volonteri, M. 2015, ApJ, 813, 82
Runnoe, J. C., Shang, Z., & Brotherton, M. S. 2013, MNRAS, 435, 3251
Sadowski, A. 2011, arXiv e-prints, arXiv:1108.0396
Shakura, N. I. & Sunyaev, R. A. 1973, A&A, 24, 337
Shi, Y., Kremer, K., & Hopkins, P. F. 2024, arXiv e-prints, arXiv:2405.12164
Soltan, A. 1982, MNRAS, 200, 115
Stone, M. A., Lyu, J., Rieke, G. H., Alberts, S., & Hainline, K. N. 2024, ApJ,
964, 90
Übler, H., Maiolino, R., Curtis-Lake, E., et al. 2023, A&A, 677, A145
Vestergaard, M. & Osmer, P. S. 2009, ApJ, 699, 800
Volonteri, M., Habouzit, M., & Colpi, M. 2021, Nature Reviews Physics, 3, 732
Wang, J.-M., Qiu, J., Du, P., & Ho, L. C. 2014, ApJ, 797, 65
Wills, B. J. & Browne, I. W. A. 1986, ApJ, 302, 56
Yue, M., Eilers, A.-C., Simcoe, R. A., et al. 2024, ApJ, 966, 176
Ziparo, F., Gallerani, S., Ferrara, A., & Vito, F. 2022, MNRAS, 517, 1086

Article number, page 8 of 8