Draft version June 23, 2014

Preprint typeset using LATEX style emulateapj v. 5/2/11

THE CASE FOR SUPER-CRITICAL ACCRETION ON MASSIVE BLACK HOLES AT HIGH REDSHIFT
Marta Volonteri1,2 & Joseph Silk1,3,4
Draft version June 23, 2014

ABSTRACT
Short-lived intermittent phases of super-critical (super-Eddington) growth, coupled with star for-
mation via feedback, may account for early growth of massive black holes (MBH) and coevolution
arXiv:1401.3513v1 [astro-ph.GA] 15 Jan 2014

with their host spheroids. We estimate the possible growth rates and duty cycles of these episodes,
assuming slim disk solutions, where the emerging luminosity depends logarithmically on the accretion
rate. As long as radiation is trapped in the disk, the effective radiative efficiency is low and several
short episodes ( 104 years) are possible, especially if energy flows through collimated funnels and
feedback is limited. When the host galaxy is able to sustain inflow rates at 1100 M /yr, replenishing
and circulation lead to a sequence of short episodes that may increase the growth rates of MBHs, with
respect to an Eddington-limited case, by several orders of magnitude in 107 years. Our model for e
MBH growth predicts that the ratio of MBH accretion rate to star formation rate is 102 or higher,
naturally leading to a ratio of MBH to stellar mass higher than the canonical value of 103 at
early epochs, in agreement with current observations.
Subject headings: accretion, accretion disks galaxies: high-redshift black hole physics

1. INTRODUCTION In this paper we use the following notation: L =

The presence of even a few billion solar mass MBHs M c2 fEdd LEdd ; tEdd M c2 /LEdd = 0.45 Gyr;
at high redshift (e.g., De Rosa et al. 2013, and refer- M Edd LEdd /c2 = M/tEdd (note the absence of in
ences therein) challenges conventional Eddington-limited = M /M Edd, where m
this definition); m is a dimension-
growth models for MBHs. Assuming Eddington- less ratio of accretion rates, not an accretion rate itself.
limited accretion, a black hole mass grows as
M = M0 exp [(1 /)(t/tEdd )], where tEdd = 2. SUPERCRITICAL SLIM DISKS
T c/(4 G mp ) = 0.45 Gyr, and is the radiative ef- In general, if material flows through the inner edge of
ficiency, normally related to black hole spin in radia-
tively efficient accretion disks, ranging from = 0.057 the disk at a rate M , a fraction is radiated away. It
to = 0.32 for spin parameters ranging from 0 to 0.998. then follows that:
Given the age of the Universe at z = 6 7 and the 1 M M
M BH = fEdd = (1 )m , (1)
estimated MBH masses, > 109 M , constant Eddington- tEdd tEdd
limited accretion is implied if 103 < M0 < 105 M and a
duty cycle 50% for M0 > 106 M (e.g., Johnson et al. and fEdd = m. In our terminology, super-critical or
2012, and references therein). super-Eddington accretion rates are defined in terms of
From a mathematical point of view, one could imagine the accretion rate, m, rather than luminosity (in prin-
resolving the problem by picking the lowest radiative ef- ciple, given a sufficiently low efficiency a super-critical
ficiency, = 0.057. From the astrophysical point of view, MBH may be emitting at sub-Eddington luminosity).
however, the real issue consists in guaranteeing that the Given an initial mass M0 and a final mass, M , the
host galaxy can continuously provide gas at the exact average values of fEdd , are related as follows:
Eddington rate for the MBH, despite feedback effects 1 t tH
(Dubois et al. 2013), and ensure that the MBH is always fEdd = m(1
) , (2)
tEdd tEdd
able to accept this gas.
We envisage that the inflow rate of gas may at times be where t
tH and tH is the Hubble time. If we assume we
much larger than the MBH Eddington rate, and explore estimate M from observations, there are four unknowns,
the consequences for MBH growth (see also Begelman M0 , m,
and (f and m are related through : f = m),

2012b). We develop arguments in favor of short-lived in- or:
termittent phases of super-critical growth, coupled with M (t) = (1 )M tH + M0 . (3)
star formation via feedback, in order to account for
early growth of MBH and coevolution with their host Let us assume that a MBH accreting super-critically
spheroids. (m
> 1) develops a slim accretion disk (but see
Coughlin & Begelman 2013), where, for instance:
1 Institut dAstrophysique, UMR 7095 CNRS, Universit e   
Pierre et Marie Curie, 98bis Blvd Arago, 75014 Paris, France L m
2 Department of Astronomy, University of Michigan, Ann Ar- 2 1 + ln , (4)
bor, MI, USA LEdd 50
3 Department of Physics and Astronomy, The Johns Hopkins
University Homewood Campus, Baltimore, MD 21218, USA for m > 50, and L/LEdd = m/25 otherwise. We
4 Beecroft Institute for Cosmology and Particle Astrophysics,
adopted here the expression in Mineshige et al. (2000),
University of Oxford, Keble Road, Oxford OX1 3RH but the logarithmic dependence is a common feature in
2 Volonteri & Silk
Figure 1. Eddington ratio, radiative efficiency from Eq. 5, and
inflow
MBH mass as a function of time assuming fixed inflow rates M
of 1 M /yr (dashed) or 100 M /yr (solid) and an initial MBH
mass of 105 M . The dotted green curve marks Eddington-limited
growth. Within 107 yr the MBH growth may be boosted by 24
orders of magnitude with respect to the Eddington-limited case.
super-critical disk models, e.g. Paczynski (1982) and
Wang & Netzer (2003), and the exact value does not
have a strong influence on the arguments presented in
this paper. From fEdd = m it follows that the effec-
tive radiative efficiency is:
 1  
1 m m tacc /tEdd 6 22 (/c) 2 105 2
= ln , (5)
25 50 50
and = 1/25 for m < 50. Therefore, while the accre-
tion rate can be highly super-Eddingon, the emergent
luminosity is only mildly super-Eddingon, because of the
logarithmic dependence and low radiative efficiency (see
also McKinney et al. 2013).
If we ignore feedback (see section 7), and assume that
the accretion rate is M inflow 100 M /yr, while M Edd =
0.02M7 M /yr, normalizing the MBH mass to 107 M ,
we can use Eq. 3 to estimate 102 , and a growth
time 107 yr. If we self-consistently evolve m, M and
at fixed M inflow we find the results shown in Fig. 1,
compatible with the previous simple estimate.
3. RADIATION TRAPPING
Trapping of radiation occurs when the photon diffusion
time, i.e., the time for photons to escape the disk, exceeds
the timescale for accretion. When photons are trapped,
they end-up being advected inward with the gas, rather
than diffuse out of the disk surface. Plausibly, as long
as radiation is trapped in the disk, the emergent lumi-
nosity does not exceed much the Eddington limit. The
radius at which radiation is trapped can be defined as
the locus where the infall speed of the gas equals the
diffusion speed of the radiation (Begelman 1978). For
m > 1: Rtr = mR s = 2 106 m M7 pc, where Rs is the
Schwarzschild radius of the MBH (this expression may be
corrected by a factor H/R, of order unity for discs puffed
up by the trapped radiation, Ohsuga et al. 2002). We
can estimate a time-scale for a single accretion episode
by requiring that the whole disks radiation is trapped,
i.e. the centrifugal barrier that defines where the ac-
cretion disk forms, RD , is inside the trapping radius:
RD 6 Rtr (see Volonteri & Rees 2005). Such condition
is shown to produce stable radiation trapping in Ohsuga
et al. (2005, their computational domain is smaller than
the trapping radius characteristic of their accretion rates
and disk geometries). By requiring that Rtr > RD we
take a very conservative approach, and a likely lower
limit to the duration of a super-critical episode, as for in-
stance McKinney et al. (2013) find stable super-critical
accretion in disks with Rtr < RD .
Let us assume that material is captured near the sphere
of influence of the MBH, Rg = G M/ 2 = 4.5 M7 100 2
1
pc, where 100 /100 km s . The material, having
angular momentum, does not fall directly into the hole,
and it forms an accretion disk. By conserving specific
angular momentum one finds that the outer radius of
the accretion disk scales with Rg , RD = 2 Rg , where
6 1 indicates the fraction of angular momentum that
is retained (it can also signify that only material with
angular momentum 1/ times smaller than average is
captured). For all radiation to be trapped, RD 6 Rtr .
This condition translates into:
2  c 2 M 1
6 1. (6)
2 tEdd M
We can now simplify this expression by approximating
M = M tacc . Rearranging the inequality:
2
2
0.1 100 , (7)
For = 0.1 and = 100 km s1 , tacc 9000 yr. With
M inflow 100 M /yr, at the end of the episode MBH
106 M , and RD = Rtr 6 0.01 pc, a value roughly com-
patible with the half-light radii of accretion disks ob-
tained from microlensing, 1016 cm (Blackburne et al.
2011). With the conservative choice = 0.1 a MBH
would grow > 106 M in < 104 yr. A smaller would im-
ply a more compact accretion disk, i.e., a smaller RD , and
super-critical accretion would proceed for longer, grow-
ing the MBH to larger masses.
4. QUENCHING
In section 3 we have assumed that and are con-
stant for a given system. In general, the disk grows
by accretion within an evolving galaxy, and the disks
growth will eventually quench when its radius exceeds
the capture radius Rg , where the gravitational poten-
tial is no longer dominated by the MBH. Low angular
momentum gas may still be accreted from larger radii
but we ignore this for the moment as it is likely to be
inefficient for any plausible angular momentum distri-
bution. Equating disk radius with photon trapping ra-
dius: RD /Rs = (M /M Edd) and Rg /Rs = (c/)2 . Now
2
M = L/c and MEdd = M/tEdd. How inefficient is the
radiative efficiency within the disk radius? We obtain an
 Supercritical accretion
expression for the radiative efficiency,
   
Rg L 2
= .
RD LEdd c inflow,100 ,
For the Polish doughnut disk model L 5LEdd
(Paczynski 1982), and therefore = 5(/c)2 5 105 .
The disk size grows as mass is accreted, and the grow-
ing disk radius (which by definition coincides with the
trapping radius) can be found by equating the mass ac-
cretion rate withq the mass flux across the MBH capture
radius: RD = tacc M / , where is the opacity and
c/v is the optical depth. Most of this material will
not reach the MBH on a dynamical time-scale because
of the gas angular momentum. However it plausibly
provides the reservoir for MBH growth. We infer that
M = c2 tacc / , whence m = ctacc /Rs . This translates
into a radiative efficiency
L Rs c
= .
LEdd RD v
The optically thick disk may become unstable on
a dynamical time, v/RD due to various instabilities
(convective, MRI, etc., e.g., Balbus & Hawley 1991;
Stone & Norman 1994; Blandford & Begelman 2004).
We then obtain an estimate of how super-critical the ac-
cretion may be, namely
RD c 1 3/2
m
= RD M 3/2 (10)
Rs v
at fixed optical depth, with v the keplerian velocity.
Since the disk dynamical time-scale varies as
3/2
RD M 1/2 , the growth boost per episode of transient ac-
cretion is of order that achievable at the Eddington rate
for steady accretion and may be most important for seed
MBH growth. One can have in principle many episodes
of short-lived but highly super-critical accretion, up to
1000 times the Eddington rate.
From the definition of trapping radius, and c/v,
we can calculate the length of the accretion episode, if
once again we assume that v is the Keplerian velocity:
1/2
c c RD c M Edd tacc 105 M M 100 tacc,3 , (15)
= 1/2
= 1/2 Rs1/2 = m
m 1/2 .
v (G M ) (G M )1/2
(11)
From Eq. 4 and the quenching condition:
q based on circulation and replenishment arguments.
RD = tacc M / = Rg = GM 2 , (12)
implying that
GM c 1/2
tacc = = 105 yrM7 100
4
m
5e3 , (13)
4 4 m
1/2
suggesting timescales of 102 104 yr.
After quenching, there is a delay until the accretion
flow re-establishes itself. In order for super-critical ac-
cretion to be relevant, a MBH mass should grow by 2
orders of magnitude. In any one episode, a MBH would
grow by 104 106 M (assuming the above time-scale
and an accretion rate of 100 M /yr). A relatively small
number of accretion episodes is needed.
3
One relevant time-scale requires accumulation of
enough mass in the nucleus to feed the MBH. or
(8) inflow = 105 yrM7 M 1
trepl = M/M
(14)
at the very minimum (here the inflow rate is in units of
100 M/yr). If only low angular momentum gas is able
to form the accretion disk, then only a fraction of the
infalling gas will be available, lengthening the replenish-
ment time-scale.
5. CIRCULATION AND DUTY CYCLES
Jet-driven feedback or more generally any ultrafast
outflow from the Active Galactic Nucleus (AGN) drives
a bow shock into the inhomogeneous multiphase accret-
ing gas (Wagner et al. 2012, 2013). The resulting shocks
generate entropy gradients which in turn generate vortic-
ity. A simple derivation is due to application of Croccos
theorem (Antonuccio-Delogu & Silk 2010) and verified in
3-d numerical simulations (Cielo et al. 2013). Suppose
the typical scale, associated with density inhomogeneities
(9)
in the flow is L 0.01L. Turbulent diffusion to larger
scales results in circulation that feeds the MBH along
the minor axis of the thick disk. The diffusion coefficient
is L v/3. The time to regenerate the feeding of the
2
MBH is 3 (L/L ) (L/v) (RD /L) . Plausible numbers
suggest a crossing time-scale L/v 1023 yr, and a
regeneration time 1045 yr.
We have argued that each accretion episode lasts
1035 yr. Eq. 7 assumes that the disk is initially smaller
than the trapping radius, and it grows until it becomes
larger than rtr . Eq. 13 is derived assuming that the disk
radius coincides with the trapping radius, and quenching
occurs when the disk reaches the capture radius. Un-
der both conditions the disk is expected to be stable,
although the accretion rate onto the MBH may differ
from the inflow rate, as mass is lost in outflows/winds
or circulation in the disk (Blandford & Begelman 1999;
Ohsuga et al. 2005; Begelman 2012a, note that classi-
cal slim disk models have spatially constant m). Both
conditions, in any case, lead to similar behavior and
timescales.
Per accretion episode, the MBH grows in mass by:
M = M tacc = m
where the accretion rate is in units of 100 M/yr and
time in units of 103 yr. Each accretion episode is followed
by a flow regeneration period of order treg 1045 yr
Since treg > tacc , the maximum number of super-
critical accretion episodes will be: N = tH /treg
104 treg,5 yr for treg in units of 105 yr. If at each episode
the MBH grows in mass by M 105 M , then the to-
tal growth is N M 109 M , and the duty cycle is
= N tacc /tH = 0.01 tacc,3/treg,5.
As the MBH grows in mass, the importance of
super-critical accretion episodes, which bring in similar
amounts of mass, decreases (see also sections 3 and 4
on how disk properties and growth may limit the impor-
tance of this growth channel to MBH < 107 M ). While
for low mass MBHs ( 1024 M ), the mass doubling
time decreases from roughly a Salpeter time in the case
of Eddington-limited accretion ( 107 yr) to tacc 1034
4 Volonteri & Silk
Figure 2. MBH accretion rate vs galaxy star formation rate for
heterogeneous samples. AGN and quasars are shown as triangles,
galaxies as circles. Small triangles represent AGN in COSMOS
(Bongiorno et al. 2012) and small circles are mid-infrared or sub-
mm selected galaxies (Rafferty et al. 2011; Symeonidis et al. 2011).
Large filled circles are results of X-ray stacking on main sequence
galaxies (Mullaney et al. 2012). Large empty triangles are z = 4.8
AGN (Netzer et al. 2013, the largest triangle marks the stack of
Herschel-undetected sources), and large filled triangles are z 6
AGN (Willott et al. 2013, and references therein).
yr, for a large MBH ( 1078 M ), the mass doubling
time is of order the Salpeter time.
6. MBH ACCRETION AND STAR FORMATION
Our model involves intense but brief phases of super-
critical gas accretion that feed the MBH but are also
likely to feed intense bursts of star formation due to
the high gas inflow rates. If Ly-alpha cooling dominates
(Spaans & Silk 2006), accretion onto the central MBH is
likely to be dominant over fragmentation.
Silk (2013) develop a simple feedback model that re-
lates the star formation rate (SFR) to the MBH accretion
rate. The model couples the two rates by incorporating
outflows by jets or winds that induce pressure-enhanced
star formation. We modify this model for the case of
super-critical MBH accretion, and discuss the implica-
tions in the following. In the notation adopted in the
present paper, the triggered star formation rate averaged
over the disk half-light radius is estimated as
M = fp M , (16)
where is the radiative efficiency, = mSN vc/2ESN
is the SNe boost factor, fp is the mechanical ad-
vantage factor of the wind or jet-driven bow shock
(Wagner et al. 2013). We define the supernova energy
ESN = 1051 E51 ergs, the mass in stars formed per type
II supernova mSN = 150m150 M , and the velocity at
which SN-driven blast waves enter the cooling phase
v = 400v400 km/s. Plausible estimates for the parame-
ters are 0.2, = 180m150 v400 /E51 , fp 30.
For the radiatively-efficient Eddington-limited case,
Silk (2013) infer that M 103 M , thereby maintain-
ing the canonical ratio of MBH mass to stellar mass ob-
served at low z. Studies of large AGN-selected samples
show little or no correlation between star formation rate
and AGN accretion rate. However plotting M versus M
for objects over a wide range of host galaxy or MBH
mass, Eddington ratio and z is likely to hide any under-
lying correlation, as indeed seems to be the case (e.g.,
Rosario et al. 2012, 2013, and references therein). That
statistically the ratio M /M may be approximately con-
stant over a wide range in z was recognized in an early
study (Silverman et al. 2008), and in a recent analysis of
SFR-selected and X-ray stacked galaxies at z 0.5 2
(Mullaney et al. 2012; Chen et al. 2013). To illustrate
this point, especially at high z where the samples are
dominated (and biased) by high host galaxy mass and
Eddington ratios, we show in Fig. 2 some data taken
from the literature. This figure is meant only to illus-
trate the observational range that can now be covered
(see the discussion in Netzer et al. 2013, for details and
additional references). We note the following, however.
If we disregard the large scatter in the unstacked low z
data for the reasons cited above, the ratio of M to M
at z 5 6 (the sources whose MBH growth we en-
deavour to describe with the present paper) may be an
order of magnitude higher than at z 1 2. This is con-
sistent with measurements of the trend in the ratio of
MBH mass to stellar mass with z, suggested to increase
by an order of magnitude by z > 4 (e.g., Wang et al. 2010;
Targett et al. 2012).
For super-critical growth that is extremely radiatively
inefficient, we infer that 1, and M 103 M .
MBHs that grew at least in part through radiatively-
inefficient super-critical accretion would therefore grow
at a faster rate than the stellar mass of their host galax-
ies. Our model therefore naturally accommodates the
suggestion that high redshift MBHs account for a higher
fraction of the stellar mass of the host. Most of the
z 6 AGN have fEdd > 0.3, the limit above which the
thin-disk approximation breaks down (Laor & Netzer
1989). Indeed, we note that by applying self-consistent
accretion-disk models to a sample of Sloan Digital Sky
Survey AGN, Netzer & Trakhtenbrot (2013) infer that
many AGN may have slim disks, rather than thin disks.
7. CONCLUSIONS
The possibility of widespread super-critical accretion
has often been advocated to describe either local (e.g.,
Collin et al. 2002; Wang & Netzer 2003) or high-redshift
sources (Volonteri & Rees 2005; Wyithe & Loeb 2012; Li
2012), or through recent revisions of the Soltan argument
(Novak 2013).
Recent simulations (Ohsuga & Mineshige 2011;
Sadowski et al. 2013; McKinney et al. 2013) suggest
that MBHs accreting at super-critical rates are most
likely to be characterized by strongly collimated outflows
or jets. This has several consequences. Firstly, detection
of sources accreting at highly super-critical rates would
be suppressed by a factor corresponding to the square
of the jet opening angle. For a funnel opening angle
of 0.1 rad (Sadowski et al. 2013) only one in 30
sources would be detected, and radio emission may be
 Supercritical accretion
suppressed by interaction with the cosmic microwave
background (Ghisellini et al. 2013). Secondly, jetted
sources likely cause limited feedback in the vicinity of
the MBH (Wang et al. 2006) or in the host galaxy, thus
allowing longer-lived super-critical accretion episodes.
For gas-rich hosts, relevant at early epochs, the AGN
jet (or wind) driven by the MBH can be strongly
coupled to the star formation rate, presumably be-
cause of bow shock pressure-enhanced star formation
(Gaibler et al. 2012; Wagner et al. 2013). Moreover, re-
cent data (Matsuoka et al. 2013) suggests that the most
luminous phase of the AGN correlates with star forma-
tion and most likely precedes any phase of negative feed-
back.
Finally, we note that in the case of super-critical accre-
tion, the radiative efficiency is not determined by MBH
spin, thus the expected spin-up caused by prolonged ac-
cretion phases would not hinder MBH growth through a
high radiative efficiency.
MV and JS thank M. Begelman for useful comments.
MV thanks A. Bongiorno for kindly providing the galaxy
and AGN properties of her sample. MV acknowledges
funding support from NASA, through Award Number
ATP NNX10AC84G; from SAO, through Award Num-
ber TM1-12007X, from NSF, through Award Number
AST 1107675, and from a Marie Curie Career Integra-
tion Grant (PCIG10-GA-2011-303609). The research of
JS has been supported at IAP by the ERC project 267117
(DARK) hosted by Universite Pierre et Marie Curie -
Paris 6 and at JHU by NSF grant OIA-1124403.
REFERENCES
Antonuccio-Delogu, V., & Silk, J. 2010, MNRAS, 405, 1303
Balbus, S. A., & Hawley, J. F. 1991, ApJ, 376, 214
Begelman, M. C. 1978, MNRAS, 184, 53
Begelman, M. C. 2012a, MNRAS, 420, 2912
Begelman, M. C. 2012b, ApJ, 749, L3
Blackburne, J. A., Pooley, D., Rappaport, S., & Schechter, P. L.
2011, ApJ, 729, 34
Blandford, R. D., & Begelman, M. C. 1999, MNRAS, 303, L1
Blandford, R. D., & Begelman, M. C. 2004, MNRAS, 349, 68
Bongiorno, A. et al. 2012, MNRAS, 427, 3103
Chen, C.-T. J., Hickox, R. C., Alberts, S., et al. 2013, ApJ, 773, 3
Cielo, S., Antonuccio-Delogu, V., Macci`o, A. V., Romeo, A. D., &
Silk, J. 2013, arXiv:1311.5562
Collin, S. , Boisson, C., Mouchet, M., Dumont, A.-M., Coup e, S.,
Porquet, D., & Rokaki, E. 2002, A&A, 388, 771
Coughlin, E. R., & Begelman, M. C. 2013, arXiv:1312.5314
5
De Rosa, G. , Venemans, B. P., Decarli, R., Gennaro, M., Simcoe,
R. A., Dietrich, M., Peterson, B. M., Walter, F., Frank, S.,
McMahon, R. G., Hewett, P. C., Mortlock, D. J., & Simpson,
C. 2013, ArXiv e-prints
Dubois, Y. , Pichon, C., Devriendt, J., Silk, J., Haehnelt, M.,
Kimm, T., & Slyz, A. 2013, MNRAS, 428, 2885
Gaibler, V., Khochfar, S., Krause, M., & Silk, J. 2012, MNRAS,
425, 438
Ghisellini, G. Celotti, A., Tavecchio, F., Haardt, F., & Sbarrato,
T. 2013, arXiv:1311.7147
Johnson, J. L. , Whalen, D. J., Fryer, C. L., & Li, H. 2012, ApJ,
750, 66
Laor, A., & Netzer, H. 1989, MNRAS, 238, 897
Li, L.-X. 2012, MNRAS, 424, 1461
Matsuoka, Y., Strauss, M. A., Price, T. N., III, & DiDonato,
M. S. 2013, arXiv:1312.2417
McKinney, J. C., Tchekhovskoy, A., Sadowski, A., & Narayan, R.
2013, arXiv:1312.6127
Mineshige, S. Kawaguchi, T., Takeuchi, M., & Hayashida, K.
2000, PASJ, 52, 499
Mullaney, J. R. , Daddi, E., B ethermin, M., Elbaz, D., Juneau,
S., Pannella, M., Sargent, M. T., Alexander, D. M., & Hickox,
R. C. 2012, ApJ, 753, L30
Netzer, H., & Trakhtenbrot, B. 2013, MNRAS, 2900
Netzer, H. , Mor, R., Trakhtenbrot, B., Shemmer, O., & Lira, P.
2013, ArXiv e-prints, arXiv:1308.0012
Novak, G. S. 2013, ArXiv e-prints, arXiv:1310.3833
Ohsuga, K., Mineshige, S., Mori, M., & Umemura, M. 2002, ApJ,
574, 315
Ohsuga, K., Mori, M., Nakamoto, T., & Mineshige, S. 2005, ApJ,
628, 368
Ohsuga, K., & Mineshige, S. 2011, ApJ, 736, 2
Paczynski, B. 1982, Mitteilungen der Astronomischen
Gesellschaft Hamburg, 57, 27
Rafferty, D. A. , Brandt, W. N., Alexander, D. M., Xue, Y. Q.,
Bauer, F. E., Lehmer, B. D., Luo, B., & Papovich, C. 2011,
ApJ, 742, 3
Rosario, D. J., Santini, P., Lutz, D., et al. 2012, A&A, 545, A45
Rosario, D. J., Trakhtenbrot, B., Lutz, D., et al. 2013, A&A, 560,
A72
Sadowski, A., Narayan, R., McKinney, J. C., & Tchekhovskoy, A.
2013, arXiv:1311.5900
Silk, J. 2013, ApJ, 772, 112
Silverman, J. D., Green, P. J., Barkhouse, W. A., et al. 2008,
ApJ, 679, 118
Spaans, M., & Silk, J. 2006, ApJ, 652, 902
Stone, J. M., & Norman, M. L. 1994, ApJ, 433, 746
Symeonidis, M. et al. 2011, MNRAS, 417, 2239
Targett, T. A., Dunlop, J. S., & McLure, R. J. 2012, MNRAS,
420, 3621
Volonteri, M. & Rees, M. J. 2005, ApJ, 633, 624
Wagner, A. Y., Bicknell, G. V., & Umemura, M. 2012, ApJ, 757,
136
Wagner, A., Umemura, M., & Bicknell, G. 2013, ApJ, 763, L18
Wang, J.-M., Chen, Y.-M., & Hu, C. 2006, ApJ, 637, L85
Wang, J.-M. & Netzer, H. 2003, A&A, 398, 927
Wang, R. , Carilli, C. L., Neri, R., Riechers, D. A., Wagg, J.,
Walter, F., Bertoldi, F., Menten, K. M., Omont, A., Cox, P., &
Fan, X. 2010, ApJ, 714, 699
Willott, C. J., Omont, A., & Bergeron, J. 2013, ApJ, 770, 13
Wyithe, J. S. B. & Loeb, A. 2012, MNRAS, 425, 2892