MNRAS 000, 1–15 (2015) Preprint 21 May 2023 Compiled using MNRAS LATEX style file v3.0
Population statistics of intermediate-mass black holes in
dwarf galaxies using the NewHorizon simulation
Beckmann, R.S.1?, Dubois, Y.2, Volonteri, M.2, Dong-Pa´ez, C. A.2,
Trebitsch, M.3, Devriendt, J.4, Kaviraj, S.5, Kimm, T.6, Peirani, S.7
1Institute of Astronomy and Kavli Institute for Cosmology, University of Cambridge, Madingley Road, Cambridge, CB3 0HA, UK
2Institut d’Astrophysique de Paris, CNRS, Sorbonne Universite´, UMR7095, 98bis bd Arago, 75014 Paris, France
3Kapteyn Astronomical Institute, University of Groningen, P.O. Box 800, 9700 AV Groningen, The Netherlands
4 University of Oxford, Astrophysics, Denys Wilkinson Building, Keble Road, Oxford OX1 3RH, UK
5 Centre for Astrophysics Research, Department of Physics, Astronomy and Mathematics, University of Hertfordshire, Hatfield, AL10 9AB, UK
6 Department of Astronomy, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul 03722, Republic of Korea
7Universite´ Coˆte d’Azur, Observatoire de la Coˆte d’Azur, CNRS, Laboratoire Lagrange, Bd de l’Observatoire,CS 34229,
06304 Nice Cedex 4, France
Accepted XXX. Received YYY; in original form ZZZ
ABSTRACT
While it is well established that supermassive black holes (SMBHs) co-evolve with their
host galaxy, it is currently less clear how lower-mass black holes, so-called intermediate-
mass black holes (IMBHs), evolve within their dwarf galaxy hosts. In this paper, we
present results on the evolution of a large sample of IMBHs from the NewHorizon zoom
volume, which has a radius of 10 comoving Mpc. We show that occupation fractions of
IMBHs in dwarf galaxies are at least 50 percent for galaxies with stellar masses down
to 106 M, but BH growth is very limited in dwarf galaxies. In NewHorizon, IMBHs
growth is somewhat more efficient at high redshift z = 3 but in general, IMBHs do
not grow significantly until their host galaxy leaves the dwarf regime. As a result,
NewHorizon under-predicts observed AGN luminosity function and AGN fractions.
We show that the difficulties of IMBHs to remain attached to the centres of their host
galaxies plays an important role in limiting their mass growth, and that this dynamic
evolution away from galactic centres becomes stronger at lower redshift.
Key words: black hole physics – galaxies: dwarf – Methods: numerical
1 INTRODUCTION
Supermassive black holes (SMBHs) with masses of MBH >
107M or above, are well known to tightly correlate with
properties of their host galaxy, such as the stellar bulge
mass and the stellar velocity dispersion (see Kormendy & Ho
2013, for a review.). While these correlations are firmly es-
tablished observationally for SMBHs, little data is available
for black holes (BHs) in the intermediate BH mass range
104 < MBH < 106M, so-called intermediate-mass black
holes (IMBHs, in the following the acronym ‘BHs’ refers to
black holes of all masses). If extrapolations of the correla-
tions into the intermediate-mass regime hold, such IMBHs
could play the same role in dwarf galaxies that SMBHs play
in massive galaxies.
IMBHs, and their potential role in shaping dwarf galax-
ies, is less well understood from both a simulation and an
? E-mail: ricarda.beckmann@ast.cam.ac.uk
observation point of view. Observationally, the inherently
lower luminosity of IMBHs due to their low BH masses make
them difficult to detect and to distinguish from star forma-
tion (see Greene et al. 2019, for a review). From a theoretical
point of view, the high resolution required to resolve the in-
ternal structure of simulated dwarf galaxies means they are
often unresolved in large scale galaxy evolution simulations,
which can therefore not be used to study the coevolution of
IMBHs and their host galaxies (Haidar et al. 2022). Instead,
specific simulations targeting the dwarf galaxy regime are
required to study such galaxies and their central BHs.
Over the last decade, great progress has been made to
expand the sample of observed IMBHs, as has been sum-
marised in the recent reviews by Greene et al. (2019) and
Mezcua (2017). There are two main methods for detecting
IMBHs: kinematical studies detect BHs via the impact of
their gravitational potential on host galaxy stars. This tech-
nique makes it possible to detect quiescent BHs and to di-
rectly measure the BH mass, but is limited to nearby galax-
© 2015 The Authors
2 Beckmann, R.S.
ies. Alternatively, IMBHs can be detected when accreting as
active galactic nuclei (AGN) in a variety of wavelengths: in
the optical (Molina et al. 2021; Mezcua et al. 2018; Mezcua
& Sa´nchez 2020; Manzano-King & Canalizo 2020), the X-
ray (Chilingarian et al. 2018; Latimer et al. 2021; Toptun
et al. 2022), the radio (Davis et al. 2022; Yang et al. 2020;
Reines et al. 2019) or the infrared (Lupi et al. 2020) to high-
light a few recent studies. Other recent detection methods
being explored are gamma-ray bursts potentially lensed by
an IMBH (Paynter et al. 2021), short-term variability (Shin
et al. 2022). In the future, gravitational wave observations
will provide another opportunity to detect IMBHs (Sesana
et al. 2005; Ricarte & Natarajan 2018; Bellovary et al. 2018;
Amaro-Seoane 2018; Katz et al. 2020; Valiante et al. 2021;
De Cun et al. 2023; Gair et al. 2011). The sample of IMBHs
in dwarf galaxies is therefore very much still in the process of
being assembled, and it is expected to expand significantly
in coming years.
Observational surveys to date have shown that AGN
fractions in dwarf galaxies are the typically observed to be
in the range of 0.1 − 5 percent (Latimer et al. 2021; Mezcua
& Sa´nchez 2020; Wylezalek et al. 2018; Reines et al. 2013;
Pardo et al. 2016; Aird et al. 2018; Birchall et al. 2020) but
may be as high as 30 percent (Kaviraj et al. 2019; Dickey
et al. 2019; Davis et al. 2022) and depend strongly on the
AGN selection method (Mezcua & Sa´nchez 2020; Lupi et al.
2020; Greene et al. 2019) and the observational proxy for
BH mass (Gallo & Sesana 2019) chosen. Firm IMBH mass
detections are only available for a small sample of objects,
and even fewer also have kinematics measurements of the
host galaxy. For those that have both, evidence is mounting
that the MBH − σ relation shows no evidence of a break in
slope in the dwarf galaxy regime (Greene et al. 2019; Bal-
dassare et al. 2020; Nguyen et al. 2019). This lack of break
has been argued to be a sign that IMBHs in dwarf galax-
ies regulate the evolution of their host galaxy through feed-
back in the same way that massive galaxies are regulated
by SMBHs (King & Nealon 2021). Further evidence for the
theory of IMBH-regulated dwarf galaxies comes from studies
of gas kinematics in dwarf that show significantly more dis-
turbed morphologies for those with AGN than those without
(Manzano-King & Canalizo 2020).
In simulations, our current picture of the co-evolution
between BH and host galaxy is more mixed: some show a
break in the correlations around the transition from dwarf
galaxy to massive galaxy (Sharma et al. 2019; Koudmani
et al. 2021), while others merely find an increase in scat-
ter and no break (Ricarte et al. 2019; Barai & de Gouveia
Dal Pino 2019; Sharma et al. 2022). As many groups find
that the growth of BHs in low-mass galaxies is regulated by
supernova (SN) feedback (Dubois et al. 2015; Habouzit et al.
2017; Bower et al. 2016; Angle´s-Alca´zar et al. 2017; Trebitsch
et al. 2018), whether BHs are over-massive or under-massive
in comparison to the correlation depends strongly on how
both supernova feedback and BH accretion and feedback are
modelled (Koudmani et al. 2022). With current SN feedback
models, BH growth in dwarf galaxies is restricted mostly to
high redshift (Barai & de Gouveia Dal Pino 2019; Koudmani
et al. 2021) but recent evidence from simulations suggests
AGN feedback in dwarf galaxies might continue to impact
both star formation and galactic outflows with with strong
SN feedback (Koudmani et al. 2019; Nelson et al. 2019) and
that a wider variety of feedback models can lead to AGN
feedback playing an even more important role in the evolu-
tion of dwarf galaxies (Koudmani et al. 2022). Analytic mod-
els also argue that the fraction of active BHs in dwarf galax-
ies could be higher than currenlty observed in X-ray samples
(Pacucci et al. 2021). On the other hand, widespread growth
of IMBHs in dwarf galaxies would lead to an overproduction
of faint AGN in tension with the observed AGN luminosity
function (Habouzit et al. 2017; Tillman et al. 2022).
Both observations (Mezcua & Sa´nchez 2020) and simu-
lations (Bellovary et al. 2018, 2021) frequently show IMBHs
in galaxies that are not located at the center of the galaxy.
The discussion around the reason for this phenomenon is
ongoing, but is likely linked to the evolution history of the
host galaxy. Boldrini et al. (2020) use isolated halos to show
that DM subhalos falling onto dwarf galaxies can displace
the central IMBH by 100 pc or more, while Bellovary et al.
(2021) show that the off-centre location in their sample of
dwarf galaxies in a cosmological environment is due to merg-
ers. While there generally is a large ‘hidden’ population of
BHs in dwarf galaxies which are not sufficient active to be
observable (Volonteri & Natarajan 2009), the percentage of
hidden BHs that are off-centre is particularly high (Sharma
et al. 2022).
In this paper, we present the sample of IMBHs in
dwarf galaxies, which are here defined to have a mass of
Mdwarf = 3×109 M, in the NewHorizon simulation1 (Dubois
et al. 2021). NewHorizon is a cosmological zoom simulation
of an average density volume of the Universe that has suffi-
ciently high resolution to resolve galaxies down to a stellar
mass of 106M. We use this sample to study correlations and
population statistics of IMBHs and their dwarf galaxy hosts
to expand our understanding of the coevolution of massive
galaxies and their BHs into the dwarf galaxy regime. The
paper is structured as follows: the simulation setup is briefly
recapped in Section 2. BH mass evolution is discussed in Sec-
tion 3.1, BH occupation ratios are shown in Section 3.2 and
mass functions are discussed in Section 3.3. The detectabil-
ity of BHs is discussed in Sections 3.4 and 3.5, while the
distribution of IMBHs within their host galaxies is analysed
in Section 3.6. Section 4 summarises the paper.
2 SIMULATION
NewHorizon is a high-resolution resimulation of an aver-
age spherical sub-volume with a radius of 10 comoving
Mpc of the Horizon-AGN simulation (Dubois et al. 2014b).
NewHorizon has been presented in detail in (Dubois et al.
2021). NewHorizon was run until z = 0.25.
The simulation assumes a ΛCDM cosmology consistent
with WMAP-7 data (Komatsu et al. 2011) with a dark
energy density ΩΛ = 0.728, baryon density Ωb = 0.045,
total matter density Ωm = 0.272, a Hubble constant of
H0 = 70.4 km s−1 mpc−1, and an amplitude of the mat-
ter power spectrum and power-law index of the primordial
power spectrum of σ8 = 0.81 and ns = 0.967 respectively. A
high-resolution region of radius of 10 comoving Mpc with a
1 https://new.horizon-simulation.org/
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 3
DM mass resolution of 1.2 × 106M is embedded within the
142 a side comoving Mpc box of Horizon-AGN.
All simulations within the Horizon suite were performed
using ramses (Teyssier 2002), using a second-order un-
split Godunov scheme for solving the Euler equations, and
an HLLC Riemann solver with a MinMod Total Variation
Diminishing scheme to reconstruct interpolated variables.
In NewHorizon refinement proceeds according to a quasi-
Lagrangian scheme up to a maximum resolution of 34pc at
z=0, where a cell is refined if its mass exceeds 8 times the ini-
tial mass resolution. The minimum cell size is kept approx-
imately constant throughout by adding an extra level of re-
finement at expansion factor aexp = 0.1, 0.2, 0.4 and 0.8. This
is supplemented with a super-Lagrangian refinement crite-
rion that enforces refinement of cells whose size is smaller
than one Jeans length wherever the gas number density is
larger than 5 H cm−3.
The gas follows an equation of state for an ideal
monoatomic gas with an adiabatic index of γad = 5/3. Gas
cooling is modelled using cooling curves from Sutherland &
Dopita (1993) down to 104 K, assuming equilibrium chem-
istry. Heating from a uniform UV background takes place
after redshift zreion = 10 following Haardt & Madau (1996).
Star formation occurs in cells whose hydrogen gas num-
ber density exceeds n0 = 10 H cm−3 following a thermo-
turbulent sub-grid algorithm in combination with a Schmidt
law (Kimm et al. 2017; Trebitsch et al. 2017, 2021). The stel-
lar mass resolution is 1.3 × 104 M), and stars are assumed
to have a Chabrier (Chabrier 2005) initial mass function
with cutoffs at 0.1 and 150 M. Stellar feedback is modelled
following Kimm et al. (2015), which separately tracks the
momentum and energy conserving phase of the explosion,
which reproduces the stellar-mass-to-halo-mass relation in
dwarfs (Dubois et al. 2021).
New BHs are formed in any cell in which the gas and
stellar density exceeds the threshold for star formation,
which has a stellar velocity dispersion of more than 20 kms−1
and that is located more than 50 kpc from any existing BH.
Each BH is formed with a mass of MBH,0 = 104 M and an
initial spin of a = 0. We note that this is slightly below the
stellar mass resolution of 1.3×104 M which might have im-
portant consequences for how well the dynamics of BH near
their seed mass is resolved in NewHorizon. To avoid spu-
rious motions of BHs due to finite force resolution effects,
we include an explicit drag force of the gas onto the BH,
following Ostriker (1999). Two BH particles are allowed to
merge into a single BH particle when they get closer than
4∆x (∼ 150 pc) and when the relative velocity of the pair
is smaller than the escape velocity of the binary. A detailed
analysis of BH mergers in NewHorizon is presented in Volon-
teri et al. (2020).
BHs grow through un-boosted Bondi-Hoyle-Lyttleton
accretion
ÛMBHL = 4pi(GMBH)
2ρ
(c2s + v2rel)3/2
(1)
where ρ, cs and vrel are the local mass-weighted, kernel-
weighted gas density, sound speed and relative velocity be-
tween gas and BH. The accretion rate is capped at the BH’s
Eddington rate ÛMEdd, which is calculated using the spin-
dependent radiative efficiency εr:
εr = fatt (1 − Eisco) = fatt
(
1 −
√
1 − 2/(3risco)
)
(2)
where risco = Risco/Rg is the radius of the innermost sta-
ble circular orbit (ISCO) in reduced units and Rg is half
the Schwarzschild radius of the BH. Risco depends on spin
a. For the radio mode, the radiative efficiency used in the
effective growth of the BH is attenuated by a factor fatt =
min(χ/χtrans, 1) following Benson & Babul (2009), where χ is
the Eddington ratio.
During accretion, a fraction of the accreted energy is
returned to the gas in one of two AGN feedback modes: 1. A
quasar mode at Eddington ratios χ > 0.01, in which energy
is injected isotropically around the BH as thermal energy
using a fixed efficiency of 15% 2. A jet mode at Eddington
ratios χ < 0.01, in which energy is injected as kinetic energy
in bipolar jets aligned with the BH’s spin axis using a spin-
dependent efficiency that is higher for high spin values (see
Dubois et al. 2021, for details).
BH spin is followed on-the-fly in the simulation, taking
into account the angular momentum of accreted gas, BH-
BH mergers and BH spindown during jet mode feedback.
The BH spin model is presented in detail in Dubois et al.
(2014a). When the BH is accreting with an Eddington frac-
tion χ > 0.01, BHs are either spun up or down depending
on whether the angular momentum of the accreted gas feeds
an aligned or misaligned sub-grid disc (King et al. 2005).
For accretion at lower luminosity, the BH-driven jets are
assumed to be powered by energy extraction from BH rota-
tion (Blandford & Znajek 1977), and as a consequence the
BH spin magnitude can only decrease. During mergers, the
spin of the remnant is calculated according to the spin of
the initial BHs, and the angular momentum of the binary,
according to Rezzolla et al. (2008).
All physics included in NewHorizon are described in
further detail in (Dubois et al. 2021). The spin evolution
of black holes in NewHorizon is analysed in a companion
publication (Beckmann et al. 2022).
2.1 Black hole and galaxy catalogue
The sample of BHs discussed in this paper includes all BHs
contained within NewHorizon that are associated with a
host galaxy from the galaxy catalogue, which in turn has
to be associated with a host halo from the halo catalogue.
The DM halo catalogue consists of all un-contaminated
DM halos identified by the structure-finding algorithm HOP
Eisenstein & Hut (1998). A halo is considered uncontami-
nated if all DM particles contained within it originate from
the zoom region, i.e. are at the maximum DM mass resolu-
tion.
Galaxies are identified using HOP applied to the star
particles of the simulation. The galaxy consists of all galax-
ies flagged as level 1 (i.e. main or central structure for a
given DM halo), or that have a stellar mass above 108M if
they are satellites. Galaxies are only included in the galaxy
catalogue if they are located within the central 0.1 virial
radii of a halo contained in the halo catalogue. The cen-
tres of galaxies and halos are identified using an iterative
shrinking-sphere approach (Power et al. 2003).
To identify BHs with their host galaxy, we cycle over all
MNRAS 000, 1–15 (2015)
4 Beckmann, R.S.
100 kpc
Figure 1. Stellar density projection of the central region of NewHorizon at z = 2. With a radius of 10 comoving Mpc for the total
zoom region (equivalent to 3.3 Mpc at z = 2), only a small fraction of the simulated volume and sample of galaxies is shown here.
Main BHs are overplotted in white, while secondary BHs are shown in grey.
galaxies in the catalogue from most to least massive, and for
each galaxy identify the most massive BH to be contained
within two effective radii of the galaxy’s centre. Galaxy ef-
fective radii are defined to be the geometric mean of the
half-mass radius of the projected stellar densities along each
of the Cartesian axis. This BH is flagged as the galaxy’s main
BH and removed from the sample of unallocated galaxies.
We then repeat the loop over all galaxies from most to least
massive, identifying all as-of-yet unallocated BHs contained
within 2 effective radii of the galaxy as secondary BHs of
that galaxy. A galaxy can contain multiple BHs, but a BH
can only be associated with a single galaxy. The full sample
of BHs discussed in this paper contains all BHs associated
with host galaxies. BHs contained in contaminated galax-
ies and halos, and “wandering” BHs that are far from the
luminous part of any galaxy are excluded from the sample.
The distribution of main BH in and around galaxies for the
central region of the box at z = 2 can be seen in Fig. 1.
3 RESULTS
3.1 Black hole - galaxy correlations
As can be seen in the top panel of Fig. 2, BHs in dwarf galax-
ies (Mstar < Mdwarf = 3 × 109 M) in NewHorizon grow little
beyond their seed mass of 104 M by z = 0.25. It is only
once their host galaxy leaves the dwarf regime that main
BHs grow efficiently onto the observed Mstar − MBH corre-
lation. Secondary BHs continue to struggle to grow beyond
105 M. As a result, the MBH − Mstar relation shows a clear
break in slope around Mdwarf .
In the bottom panel of Fig. 2, we study in more detail
how efficient BHs in dwarf galaxies grow over time by plot-
ting MBH,acc, the accreted mass. MBH,acc is the mass of each
BH gained through gas accretion alone. It is found by tak-
ing the fiducial BH mass, MBH and subtracting both the seed
mass MBH,0 = 104 M as well as any mass gained through
BH-BH mergers. As can be seen by the slope α of the lin-
ear fit to MBH,acc vs Mstar (solid lines), BHs in more massive
dwarf galaxies do grow somewhat more than in low-mass
dwarfs, but the slope of this trend is much shallower than
for main BHs in massive galaxies. It is, however, very similar
to that for secondary BHs in galaxies of all stellar masses.
We caution that while the BH seed mass MBH,0 is no longer
included in MBH,acc, the distribution of MBH,acc still depend
on MBH,0 as the BH accretion rate depends explicitly on the
instantaneous BH mass at any given point in time (see Eq.1).
If we had decided to set MBH,0 to a higher value initially, our
BH would have grown faster than a BH with lower MBH,0 in
the same environment.
This change in slope is different to tentative observa-
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 5
10
4
10
5
10
6
10
7
10
8
10
9
10
10
M
BH
 [M
]
BM19
newH stacked sample
newH z = 0.25, main
RV15 (AGN)
RV15 (dyn)
Greene20
Greene20, upper limits
10
6
10
7
10
8
10
9
10
10
10
11
Mstar [M ]
10
1
10
2
10
3
10
4
10
5
10
6
10
7
10
8
M
BH
,,
ac
c 
[M
]
main BH, dwarf galaxy
= 0.15
secondary BH, all galaxies
= 0.12
main BH, massive galaxy
= 1.79
Figure 2. Correlations between mass and host galaxy properties
at z = 0.25: [Top] Galaxy stellar mass versus BH mass MBH for
all main BHs. The grey distribution shows that stacked sample
at all redshifts. [Bottom] Galaxy stellar mass versus accereted
BH mass MBH,acc for all BHs. Shown on both plots for compari-
son are observational data from Reines & Volonteri (2015) (RV15,
green triangles), Baron & Me´nard (2019) (BM19, brown contours)
and Greene et al. (2019) (Greene20, blue markers and limits).The
same observations are shown on both panels. α denotes the slope
of the fits for each population of BHs. Errorbars for RV15 are
omitted for clarity. Galaxies left of the dotted black line are con-
sidered dwarf galaxies.
tional conclusions, which show no evidence so far for a break
in slope in the MBH − Mstar relation during the transition
from dwarf to massive galaxy (see Greene et al. 2019, for a
review), and observational data points plotted for compari-
son in Fig. 2. The lack of change in slope could be due to an
observational bias, as over-massive BHs are easier to observe
for a given galaxy stellar mass than under-massive ones. No
such bias exists in simulations, which could potentially skew
the comparison.
10
4
10
5
10
6
10
7
10
8
10
9
10
10
M
BH
 [M
]
newH z = 0.25
KH13, Eq 7
Greene2019
Greene20, all, limits
KH13
Greene2019, upper limits
Greene2019, upper limits
Baldassare2020
10
1
10
2
 [km/s]
10
8
10
9
10
10
10
11
M
st
ar
 [M
]
Cappellari2013
Oh2020
Barat2019
NewH, z = 0.25
Figure 3. [Top] BH mass versus stellar velocity dispersion σ,
for all galaxies with a stellar mass above 108M at z = 0.25.
Observations shown are from Greene et al. (2019)(Greene2019,
blue markers and shaded region), Baldassare et al. (2020) (Bal-
dassare2020, green crosses) and Kormendy & Ho (2013) (KH13,
purple markers and shaded region). [Bottom] Faber-Jackson rela-
tion for redshift z = 0.15. Shown for comparison are observational
fits from from Cappellari et al. (2013); Barat et al. (2019) and
Oh et al. (2020).
Alternatively BH growth in galaxies in
NewHorizon might be artificially suppressed due to
an AGN feedback efficiency that could have been set too
high, as the BH feedback efficiency controls the normalisa-
tion of the correlations between BH and their host galaxy
properties Dubois et al. (2012). Evidence for this hypothesis
comes from both the MBH − Mstar relation (top panel, Fig.
2) and the MBH − σ relation (top panel, Fig 3) where BHs
in NewHorizon fall consistently below observational values
and fits. The bottom panel of Fig 3 confirms that this is
not due to the fact that stellar velocity dispersions σ are
systematically over-estimated in NewHorizon, as galaxies
MNRAS 000, 1–15 (2015)
6 Beckmann, R.S.
at this redshift fall within observational constraints. While
BHs seem to fall on the observed MBH-σ relation at the
low-mass end, we caution that the mass of any BH that has
not grown by at least an order of magnitude is likely to still
be dominated by the numerical seed mass.
Another possible explanation for the lack of BH growth
in NewHorizon is due to the position of BHs in galaxies.
It is suspicious that the slope of the MBH,acc − Mstar re-
lation is very similar for main BH in dwarf galaxies and
secondary BH in massive galaxies, which could suggest
that BH in dwarf galaxies fail to grow as they are insuf-
ficiently close to the centre of their host galaxy. This effect
in NewHorizon might be enhanced due to the similarity
of the BH seed mass, MBH,0 = 104 M, and the stellar mass
resolution of 1.3×104 M, which leads to artificial scattering
of BH trajectories by interactions with the stellar particles
although this two-body effect is somewhat mitigated by the
multi-grid gravity solver. We explore the spatial distribution
of BH, and its impact on BH growth, further in Sec. 3.6.
Whether the break in the MBH − Mstar relation is real
remains an open question. With its lower seed mass of
104 M, an order of magnitude lower than previous sim-
ulations of dwarf galaxies (see e.g. Koudmani et al. 2021;
Sharma et al. 2022, who used a seed mass of 105 M)
and two orders of magnitude lower than the typical seed
masses of 106 M of large-scale cosmological simulations (see
Haidar et al. 2022, for a comparison of different simulations),
NewHorizon probes the co-evolution between IMBHs and
dwarf galaxies for a larger range of galaxy masses. Using a
seed mass of 106 M, Koudmani et al. (2021) find tentative
evidence for a flattening of the MBH − Mstar relation, while
Sharma et al. (2022) do not, but instead discuss that this
could be an artifact of their high seed mass, which artificially
boosts accretion onto overmassive BHs in galaxies.
In the rest of the paper, we will explore what drives
the (lack of) mass growth of BHs in dwarf galaxies,
quantify whether the observable population of IMBHs in
NewHorizon reproduces observed AGN in dwarf galaxies,
and how the population of observable IMBHs in dwarf galax-
ies compares to the full sample.
3.2 Occupation fractions
One key question to understand the population of IMBHs in
dwarf galaxies is how many dwarf galaxies contain a main
BH. There are several ways to measure occupation fractions,
as illustrated by the different panels in Fig. 4 (as a function
of galaxy mass). The top panel shows the fraction of galaxies
(halos) that contain at least one BH, the second panel takes
BH multiplicity into account and shows the average number
of BH per galaxy (halo), while the third panel denotes the
fraction of galaxies that contain two or more BHs. In this
section, we discuss occupation fractions of galaxies and halos
which measure the presence of BHs in a given galaxy or halo,
regardless of their luminosity. For the fraction of active BHs
(AGN fractions), see Section 3.5.
As already seen in the MBH − Mstar relation in Fig. 2,
BHs are formed in galaxies with stellar masses as low as
Mstar = 106 M. As can be seen in Fig. 4 occupation frac-
tions for galaxies and halos fall with decreasing galaxy (halo)
mass but remain significant across the dwarf galaxies, with
10
0
3 × 10
1
4 × 10
1
6 × 10
1
f o
cc
fit
data
10
0
10
1
<n
BH
>
z = 0.25
z = 1
z = 2
z = 3
10
6
10
7
10
8
10
9
10
10
10
11
Mstar [ M ]
10
2
10
1
10
0
f m
ul
tip
le
Figure 4. Fraction of galaxies that contain at least one BH [top],
average number of BHs per galaxy [middle] and fraction of galax-
ies with multiple BHs [bottom] as a function of galaxy mass Mstar.
A white background denotes dwarf galaxies. Solid lines on the top
panel denote a fit of Eq. 3, with free parameters at each redshift
listed in Table 1. Dashed lines connect data points. Error bars
show Poisson errors.
even the lowest mass galaxies having a minimum occupation
fraction of focc > 0.37. There is little redshift evolution for
all BHs across the mass range, with 57% of dwarf galax-
ies containing at least one BH at z = 3, in comparison to
54 % at z = 0.25. Multiple occupation of galaxies is fairly
common for massive dwarfs (with more than 26 percent of
dwarf galaxies with Mstar > 108 M hosting more than one
BH at z = 0.25), but becomes increasingly uncommon at
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 7
z α M′ 
z = 0.25 0.43 8.9 39.3
z = 1 0.48 8.67 26.43
z = 2 0.52 8.78 29.49
z = 3 0.46 8.55 33.75
Table 1. Fitting parameters for focc from Eq. 3. Fits are shown
in the top panel of Fig. 4 as solid lines.
lower galaxy stellar mass. The lack of growth of IMBHs in
dwarf galaxies is therefore not due to low number statistics.
NewHorizon contains 376 BHs distributed across 308 dwarf
galaxies, out of a total of 552 dwarf galaxies at z = 0.25.
In comparison to Sharma et al. (2022), who use the Ro-
mulus25 simulation to study the population of BHs in dwarf
galaxies, we also report a sharp decline in the occupation
fraction with galaxy stellar mass at low redshift. At higher
redshift, however, occupation fractions in NewHorizon are
considerably flatter that in Romulus25. This could be due
to the lower stellar mass resolution in NewHorizon (1.3 ×
104 M) in comparison to Romulus25 (2.1×105 M), and the
corresponding lower seed mass (104 M in NewHorizon in
comparison to 106 M for Romulus25), which allows the co-
evolution of a given galaxy and its BH to be followed from
earlier on in its evolution. As the intrinsic occupation frac-
tion cannot be observed, we defer a comparison to observa-
tions to the fraction of active BHs in Section 3.5.
For ease of comparison to other datasets, the fraction
of occupied galaxies has been fit with a function of the form
f (M) = 1 − α
1 +
(
log (M)
M′
) (3)
where M is the galaxy stellar mass Mstar, and the free pa-
rameters of the fit α,  and M ′ are listed in Table 1 for
each redshift and both galaxy samples. Similar fits for halo
occupation fractions are found in Appendix A.
3.3 Black hole mass functions
One consequence of the efficient seeding and low BH growth
in dwarf galaxies is that there is a large number of BHs at or
near the seed mass. As a result, the BH mass function shown
in Fig. 5 steepens strongly at low BH masses at all redshifts,
in comparison to predictions based on observed galaxy stel-
lar mass function (grey and black lines from Gallo & Sesana
2019 and Greene et al. 2019 respectively). Even when re-
stricting our BH sample to more readily observable objects,
by taking only main BHs of galaxies with stellar masses
above 108M into account (solid line), our BH mass func-
tions remain very high for BHs below 105M. This is due to
a combination of an efficient seeding algorithm, which cre-
ates high occupation fractions of BHs in low mass galaxies,
and lack of sustained growth of the BH.
Conversely, NewHorizon under-predicts the number of
more massive BHs, i.e. falls below observational limits for
MBH > 105 M, suggesting that despite ample seeding,
BHs struggle to grow beyond their seed masses in the en-
vironment probed here. We note that due to our compara-
tively small simulation volume combined with the fact that
10
4
10
5
10
6
10
7
10
8
MBH [M ]
10
4
10
3
10
2
10
1
dn
/d
lo
g(
M
BH
) [
M
pc
3  
de
x
1 ]
Gallo&Sesana19, Eq4
Greene2020, linear
Greene2020, nuclear cluster
z = 0.25
z = 1.0
z = 2.0
z = 3.0
all galaxies
galaxies with M * > 108 M  only
Figure 5. Evolution of the mass function of main BHs with
redshift. Shown in comparison are fits based on observations
from Gallo & Sesana (2019) (black dotted) and Greene et al.
(2019)(grey shaded). The two models for Greene et al. (2019)
are derived using a linear occupation fraction (light grey) and a
nuclear cluster occupation fraction (dark grey) respectively. BH
mass functions are annotated with Poisson error bars. Solid lines
include main BHs of galaxies with a stellar mass above 108M,
while dotted lines include all main BHs.
NewHorizon probes an average volume rather than an over-
dense one, number statistics for massive galaxies and their
high mass BH are poor, and errorbars therefore compara-
tively large. We observe no clear evolution of the BH mass
function with redshift for the mass bins probed at a given
redshift, with the only evolution coming from BHs growing
to a more massive regime over time.
3.4 Black holes in dwarf galaxies as AGN
As can be seen in Fig. 6, the ability of BHs to accrete ef-
fectively in dwarf galaxies, and the associated luminosity
of AGN in dwarf galaxies, evolves strongly with redshift. In
this section we differentiate between the instantaneous bolo-
metric luminosity of a BH, LBH,bol, measured at the target
redshift, a mean bolometric luminosity of a BH, LBH,bol,mean,
which is calculated by taking the mean of log(LBH,bol) for all
values within ∆t ± 100 Myr around the target redshift, and
a peak bolometric luminosity, LBH,bol,peak which is defined
to be the maximum luminosity of a BH within ∆t ± 100 Myr
around the target redshift. We used the BH spin-based radia-
tive efficiencies to compute the BH luminosities, as detailed
in Dubois et al. (2021). As shown in the companion paper
on BH spin for NewHorizon BHs (Beckmann et al. 2022),
spin-based radiative efficiencies do modulate the bolometric
luminosity for individual BHs by up to a factor of ∼ 3, but
have on average little impact on the distribution of luminosi-
ties for the whole population. This increased feedback effi-
ciency, and consequent reduced mass growth of BHs, could
explain why all main BHs in massive galaxies are found at
the lower end of observations.
At high redshift, there is a wide range of accretion ef-
ficiencies across the sample of BHs. The most efficiently
MNRAS 000, 1–15 (2015)
8 Beckmann, R.S.
10
36
10
38
10
40
10
42
L B
H,
bo
l [
er
g 
 s
1 ]
z = 3 z = 2
10
4
10
5
MBH [M ]
10
36
10
38
10
40
10
42
L B
H,
bo
l [
er
g 
 s
1 ]
z = 1
10
4
10
5
MBH [M ]
z = 0.25
mean luminosity
peak luminosity
0
1
2
3
4
5
d c
en
tre
 [k
pc
]
Figure 6. BH bolometric luminosity versus BH mass for all main BHs in dwarf galaxies, i.e., with stellar mass below Mdwarf = 3×109 M.
Points are coloured by the distance between the BH and the centre of the host galaxy. Mean luminosities over ±∆t = 100 Myr are shown
for all redshifts (crosses), with BHs with mean Lbol < 1035 erg s−1 plotted at 1035 erg s−1 as upper limits. For z ≤ 1, peak luminosities over
the same ∆t are also shown (dots). Constant Eddington fractions fEdd are shown as black lines.
accreting BH growth with time-average Eddington ratios
fEdd,mean = L¯BH,bol,mean/LEdd > 0.1 but such BH are rare
as the majority of the sample grow so inefficiently that their
mass growth is negligible: 76.6 percent have a fEdd ≤ 3×10−3,
i.e. a mass growth timescale greater than the Hubble time
and a luminosity far too faint to be detectable in AGN sur-
veys. However, even among low accretors, brief accretion
spikes with significantly higher luminosities are common:
23.6 percent of BHs at z = 3 have peak luminosities of
fedd,peak > 0.1, compared to only 1.3 percent of the sample
whose fedd,mean is that high. If such accretion spikes leave a
longer-lasting signal, they might enhance the observability
of high-redshift BH.
With decreasing redshift, BH activity decreases
markedly. At z = 0.25, no BH in a dwarf galaxy has a mean
Eddington fraction larger than fEdd = 10−4 which trans-
lates an effective minimum Salpeter time of 450 Gyr: by
z = 0.25, growth for BHs in dwarf galaxies has effectively
stopped in NewHorizon. It is well known that BH growth
slows down over time (Dubois et al. 2012; Volonteri et al.
2016; Habouzit et al. 2022). However, there is growing ev-
idence that BH growth in simulated dwarf galaxies slows
down even faster than the slowdown in stellar mass growth,
as similar trends to those in NewHorizon are also found
in other simulations of IMBHs in dwarf galaxies, (Barai &
de Gouveia Dal Pino 2019; Koudmani et al. 2021; Sharma
et al. 2022), who all concluded that AGN activity in dwarf
galaxies decreases strongly with redshift. A question that
remains to be addressed is if, statistically, BH growth in
simulated dwarf galaxies slows down faster than for more
massive galaxies.
As can be seen by the colour-coding of datapoints in Fig.
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 9
6, the location of BHs plays a role in their decreased activity
at low redshift: efficiently growing BHs are on average very
close to the centre of their host galaxies, while the accretion
efficiency of those further out drops markedly. We discuss
the impact of BH location within their host galaxy further
in Sec. 3.6.
Despite their low average growth rates, BHs continue
to see brief bursts of activity even at low redshift, as can be
seen by the peak luminosities plotted in the lower two panels
of Fig. 6, which can be more than an order of magnitude
higher than the mean luminosity. As a result, the bolometric
luminosity functions in Fig. 7 do not drop to zero even at low
redshift. At each redshift in Fig. 7, luminosities plotted are
derived from a stacked sample of LBH,bol within ∆t = ±100
Myr of the target redshift, to account for the variation in
AGN luminosity. Results do not depend sensitively on the
choice of ∆t. Luminosity functions are shown for both the
sample of all galaxies in NewHorizon, and for dwarf galaxies
only. Due to the mass evolution of galaxies over time, the two
samples are indistinguishable at high redshift (z ≥ 2) except
at the very bright end, but the decrease in AGN activity for
BHs in dwarf galaxies can be clearly seen at z ≤ 1.
At high redshift, the luminosity function shows a clear
peak around 1042 erg s−1, which is approximately equal to
the Eddington luminosity for our seed BHs. The peak in
the luminosity function at this luminosity shows that many
BHs at z ≥ 2 are accreting as efficiently as permitted by our
accretion algorithm (see Sec. 2 for details) One intriguing
possibility is that BHs in NewHorizon are undermassive at
low redshift because accretion is capped at the Eddington
limit at high redshift. Even brief super-Eddington episodes
could give BHs an early mass boost (Regan et al. 2018) that
could influence their evolution later, although Massonneau
et al. (2023) have shown that super-Eddington accretion ac-
tually reduced the overall growth of BHs in massive compact
galaxies.
By z < 1, the peak in the luminosity function around
1041−1042 ergs−1 has disappeared. At this redshift, the lumi-
nosity function at Lbol < 1041 erg s−1 is still almost entirely
due to BHs in dwarf galaxies, while BHs in massive galaxies
dominate at higher luminosity. By z = 0.25, BHs in mas-
sive galaxies dominate as far down the luminosity function
as 1037 erg s−1, with only a small contribution from BHs in
dwarf galaxies in this luminosity range due to brief peaks of
AGN activity (see Fig. 6). At all redshifts, the bright end
(Lbol > 1041 erg s−1) of the luminosity function is generally
lower than the values by Shen et al. (2020) but in good
agreement with observations by Hopkins et al. (2007), with
the exception of z = 0.25 where NewHorizon under-predicts
the luminosity function for L < 1043 erg s−1. This most likely
simply reflects the fact that there are no overmassive BHs
in NewHorizon and many galaxies host under-massive BHs
(see Fig. 2). We note that given that NewHorizon is a zoom
simulation, luminosity functions should be treated with cau-
tion as we do not have a statistically significant sample of
massive galaxies, and their SMBH. As the low-luminosity
end of the luminosity function could be populated by both
highly accreting IMBHs and inefficiently accreting SMBHs,
the absence of SMBHs in the sample will be felt across the
whole luminosity function.
3.5 AGN fractions
To assess the detectability of our AGN, we also compute the
X-ray luminosity of the host galaxy from both X-ray bina-
ries (XRBs) and hot gas. For XRBs, we compute the total
X-ray luminosity LXRB by adding the contributions from soft
(0.5 − 2 keV) and hard (2 − 10 keV) X-rays according to the
redshift dependent relation from Lehmer et al. (2016). For
X-ray emission from hot gas Lgas in the host galaxy, we com-
pute the emission in the soft X-ray band using the relation
from Mineo et al. (2012), and extrapolate to the hard X-ray
band assuming a photon index of Λ = 3 following Mezcua
et al. (2018). The total X-ray luminosity of the host galaxy is
LXray,gal = LXRB+Lgas. A system is considered observable for
a given luminosity cut Lcut if LXray,BH + LXray,gal > Lcut, and
detectable as an AGN if also LXray,BH > 2LXray,gal following
Birchall et al. (2020). The resulting distribution of LXray,gal
versus LXray,BH can be seen in Fig. 8 for different redshifts.
For the BH X-ray luminosity, we calculate the 0.5 −
10 keV luminosity using the bolometric correction from Shen
et al. (2020). For each galaxy, the BH X-ray luminosity
LXray,BH is the sum of the instantaneous X-ray luminosities
of all BHs associated with that galaxy, and is, hence, the
sum of the galaxy’s main BH as well as its secondary BHs.
LXray,BH is therefore the BH luminosity that would be ob-
served using a telescope with insufficient angular resolution
to separate individual BHs in the galaxy. Typically, LXray,BH
is dominated by the main BH. This is partially because not
every galaxy even has a secondary BH (see the discussion
on BH multiplicity in Section 3.2) and because main BHs
are much more active: main BH are 2 (z ≥ 1) to 4 (z = 0.25)
times more likely to appear on Fig. 8, i.e. to have an instan-
taneous X-ray luminosity in excess of the solar luminosity
1034 erg s−1 than secondary BHs. As a result, only 5 percent
(z = 0.3) to 11 percent (z = 0.25) of galaxies have secondary
BHs that contribute more than 20 percent to LXray,BH.
The number of BHs that meet the Birchall et al. (2020)
criterion decreases strongly with redshift: while there are
plenty of BHs identifiable as AGN at z ≥ 2, there are no BHs
in dwarf galaxies at z = 0.25 that would be recognisable as
AGN using the Birchall et al. (2020) requirement (dashed
line), no matter the luminosity cut. This would remain true
in even the most optimistic case if we considered the peak
BH luminosity within a ∆t ± 100 Myr window around the
target redshift (not shown on the plot).
As a result, the fraction of AGN in dwarf galaxies drops
strongly with decreasing redshift, as can be seen in Fig. 9,
from fAGN ∼ 17 percent at z = 3 to fAGN = 0 percent at z =
0.25 using the instantaneous X-ray BH luminosity LBH,xray
and a luminosity cut of Lcut > 1039 erg s−1. This is in line with
previous simulation results by Barai & de Gouveia Dal Pino
(2019) and Koudmani et al. (2021), who both showed that
AGN activity in dwarf galaxies is significantly higher at high
redshift. Like in NewHorizon, there are no remaining active
BHs in dwarf galaxies in the FABLE simulations at z ≤ 0.25
(Koudmani et al. 2021). This disagrees with observations
from Birchall et al. (2020), who report an AGN fraction of
4 % at z < 0.25. We note that the observed AGN fractions
in dwarf galaxies in literature vary strongly depending on
how an AGN is defined and how the X-ray luminosity of
the host galaxy is accounted for. As a result, some X-ray
surveys find values as high as 30 percent (Zhang et al. 2009;
MNRAS 000, 1–15 (2015)
10 Beckmann, R.S.
10
5
10
4
10
3
10
2
dn
/d
lo
g(
L b
ol
) [
M
pc
3  
de
x
1 ]
S20, z = 6
H07, z = 6
z = 6
S20, z = 4
H07, z = 4
z = 4
10
5
10
4
10
3
10
2
dn
/d
lo
g(
L b
ol
) [
M
pc
3  
de
x
1 ]
S20, z = 3
H07, z = 3
z = 3
S20, z = 2
H07, z = 2
z = 2
10
36
10
38
10
40
10
42
10
44
10
46
Lbol [erg s 1]
10
5
10
4
10
3
10
2
dn
/d
lo
g(
L b
ol
) [
M
pc
3  
de
x
1 ]
S20, z = 0.8
H07, z = 1
z = 1
10
36
10
38
10
40
10
42
10
44
10
46
Lbol [erg s 1]
S20, z = 0.2
H07, z = 0.1
z = 0.25
all
Mstar < 3 × 109M
Figure 7. Bolometric BH luminosity functions with Poisson error bars for the full sample of galaxies (solid) and dwarf galaxies (dashed).
At z > 3, all galaxies are dwarf galaxies so only one line is shown for clarity. Luminosities shown are derived from a stacked sample within
±100 Myr of the target redshift. Shaded regions show fits to observational luminosity functions by Hopkins et al. (2007) (H07) and Shen
et al. (2020) (S20) for comparison.
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 11
10
37
10
38
10
39
10
40
10
41
10
42
LXray, gal [erg s 1]
10
34
10
36
10
38
10
40
10
42
10
44
L X
ra
y,
BH
 [e
rg
 s
1 ]
AGN criterion
dwarf galaxy
massive galaxy
10
37
10
38
10
39
10
40
10
41
10
42
LXray, gal [erg s 1]
z = 0.25
z = 1
z = 2
z = 3
Figure 8. Distribution of X-ray luminosity of BHs versus that of their host galaxies using the mean BH luminosity (left z = 3&2, right
z = 1&0.25). Data-points above the dashed line are detectable as AGN, as they have LXray,BH > 2LXray,gal, if the combined luminosity of
BH and galaxy exceeds the luminosity cut, i.e. if also LXray,BH + LXray,gal > Lcut for a given Lcut.
0.5 1.0 1.5 2.0 2.5 3.0
z
0
5
10
15
20
25
30
35
f A
GN
 [%
]
Lcut = 1039 erg s 1
Lcut = 1037 erg s 1
peak, Lcut = 1039 erg s 1
Birchall2020
Figure 9. Fraction of observable AGN as a function of red-
shift for all dwarf galaxies with a total X-ray luminosity above
1039 erg s−1 (blue) and 1037 erg s−1 (grey). Solid lines use the instan-
taneous Lxray,BH, while dotted lines use the peak Lxray,BH within
∆t ± 100 Myr. Coloured markers show observational data mea-
sured at z < 0.25 by Birchall et al. (2020), for luminosity cuts of
> 1039 erg s−1.
She et al. 2017). Here we restrict the comparison to Birchall
et al. (2020) as we have designed our AGN observability
criteria to reflect theirs.
AGN fractions depend strongly on the luminosity cut,
as can be seen by comparing the 1039 ergs−1 to the even
more optimistic 1037 ergs−1, which shows lower fAGN at all
redshifts probed here. This is due to the fact that with in-
creasing luminosity cut, the number of AGN in the sample
increases but the number of galaxies increases even faster.
A direct comparison to observations of this redshift evolu-
tion is difficult as observing AGN with luminosity cuts as
low as 1039 erg s−1, let alone the lower cut of 1037 erg s−1,
is currently only possible at very low redshift. Instead, ob-
servations of the redshift evolution of fAGN such as those
presented in Mezcua et al. (2018) use luminosity cuts of the
order of 1040.5 erg s−1, which no BH in dwarf galaxies achieve
in NewHorizon, even at z = 3. At such higher luminosity
cuts, observations show a predominantly flat evolution of
the AGN fraction with decreasing redshift at z < 1 (Mezcua
et al. 2018; Birchall et al. 2020), which suggests that cur-
rent simulations over-quench AGN. This hypothesis is fur-
ther supported by the fact that there are no bright AGN at
all in NewHorizon. This conclusion assumes that AGN are
observed at their instantaneous luminosity at z = 0.25. If
we make the generous assumption that AGN activity leaves
an observable signal that persists for longer than the AGN
outburst itself, and that we can therefore observe AGN at
their peak activity within a ∆t = 100 Myr window (similar
to the analysis in Fig. 7) then two potential AGN do ap-
pear in the sample at z = 0.25. This raises the observable
AGN fraction to fAGN = 1.5%. Despite this being a generous
assumption, the resulting AGN fraction still lies below the
observed fraction of 4 % by Birchall et al. (2020). We also
caution that this assumes that the observable AGN signal
does not fade at all for up 100 Myr after the outburst has
ended, and therefore represents an upper limit.
Overall, we conclude that mass growth for BH with seed
MNRAS 000, 1–15 (2015)
12 Beckmann, R.S.
Figure 10. Distribution of the distance between BHs and the
centre of their host galaxies, dBH for main (solid) and secondary
(dashed) BHs. The in-set pan shows a zoom in on the range
dBH < 5 kpc and the median distance of main (secondary) BHs as
triangles (crosses) at each redshift.
masses of 104 M in NewHorizon is stifled below observed
limits, particularly at low redshift. The time window for ef-
ficient BH growth is too short for BHs to compensate for
the lack of earlier growth at the seed mass. This could sug-
gests that BHs seeds in dwarf galaxies are formed as mas-
sive seed BHs (> 104 M), or that current SN prescrip-
tions are too strong to allow for the observed BH growth.
Other potential numerical effects, besides SN prescriptions,
that might lead to the observed over-quenching, is an over-
estimated AGN feedback efficiency. However, it is not as
simple as simply decreasing the SN feedback strength, as
doing so will over-predict the stellar-to-halo mass relation,
which for NewHorizon is already at the upper end of em-
pirical constraints. Instead, it might be more a question of
where SN energy is deposited, how turbulence is injected
in the interstellar medium and how SN and possibly AGN
feedback can continue to regulate the galaxy mass content
without over-quenching IMBHs in dwarf galaxies (see Koud-
mani et al. 2022, for a detailed investigation of the impact of
SN feedback strength on IMBH growth in dwarf galaxies).
Alternatively, the lack of BH growth could be a sign of other
relevant missing physics, such as the suppression of cooling
from cosmic rays which have been shown to suppress the
star formation rate (SFR), and resulting SN injection rate,
by a factor of 2-3 (Dashyan & Dubois 2020).
3.6 Distribution of black holes in dwarf galaxies
To grow efficiently, BHs must be able to tap into an abun-
dant local supply of cold gas. The lack of growth for BHs in
NewHorizon could therefore be due to one of two reasons:
either dwarf galaxies in NewHorizon are cold gas poor, or
BHs are not located where the cold gas is. As can be seen by
Figure 11. Log-normalised distribution φ of the distance be-
tween main BHs and the centre of their host galaxies, dBH as a
function of the host galaxy mass Mstar (left) and BH mass MBH
(right) for a stacked sample of all BHs at z = 3, 2, 1 and 0.25.
the colour-coding of datapoints in Fig. 6, the location of BHs
certainly plays a role in their decreased activity: efficiently
growing BHs are on average very close to the centre of their
host galaxies, while the ability to accrete of those further out
drops markedly. The distribution of BHs from the centre of
their host galaxy evolves with redshift, as can be seen qual-
itatively in Fig. 6 and more quantitatively in Fig. 10. The
sample of BHs plotted in Fig. 10 separately analyses the dis-
tribution of main and secondary BHs, unlike Fig. 8 which
only shows main BHs. At all redshifts, main BHs dominate
the sample, with 86% of BHs at z = 3 classified as ‘main’
which decreases to 62% at z = 0.25.
In general, main BHs are closer to the centre of their
host galaxy than secondary BHs, as can be seen in Fig. 10.
We note that main BHs being closer to the centre of their
host galaxy is not by design as the ‘main’ BH of a galaxy
is not selected to be the one located closest to the centre
of the galaxy. Instead, it is defined to be the most massive
BH that meets the criterion of being identified with a given
galaxy (see Sec. 2.1 for more details).
Both main BHs and secondary BHs become less cen-
trally located over time. The median distance of BHs to the
centre of their galaxy increases from 0.68 kpc at z = 3 to
2.96 kpc at z = 0.25 for main BHs, and from 1.48 kpc at z = 3
to 4.55 kpc at z = 0.25 for secondary BHs. This increase in
separation between galaxy centre and BH for both main and
secondary BH is partially due to the increased size of galax-
ies at low redshift but not exclusively. Accounting for the in-
crease in average galaxy size, here measured by the galaxies’
effective radius reff with decreasing redshift, the median dis-
tance still increases from 1.0reff at z = 3 to 1.3reff at z = 0.25
for main BHs. We therefore conclude that main BHs strug-
gle to remain attached to the centre of their host galaxy for
long periods of time, and both main and secondary BH be-
come less central over time. As will be discussed in detail
in a companion publication entirely focused on the dynam-
ics of BHs in NewHorizon Beckmann et al. (2023), BHs
predominantly become displaced from their galactic center
during galaxy mergers. Dynamical timescales for such low
mass BHs are extremely long, BHs struggle to settle back
into galactic centers following such disturbances.
As can be seen in Fig. 11 it is predominantly low-mass
BHs in low-mass galaxies which struggle to remain attached
to the centre. While BHs close to the centre of their host
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 13
galaxy exist at all BH and galaxy masses, it is only above a
threshold BH mass (MBH & 105 M) and a threshold host
galaxy mass (Mstar & 1010 M) that main BH can be reliably
found in the centre of the galaxy.
From previous work, BHs in galaxies that are not lo-
cated in the centre of galaxies is not unexpected. Bellovary
et al. (2021) found that BHs in dwarf galaxies are fre-
quently displaced from the centre of their host galaxy fol-
lowing galaxy mergers, and Sharma et al. (2022) report
that a significant fraction of their BHs in dwarf galaxies
are off-centre. Observationally, there is also evidence for the
fact that BHs in dwarf galaxies wander. Observationally,
Reines et al. (2019) reported that the majority of their radio-
selected AGN in dwarf galaxies are off-centre with respect
to the host galaxy. Part of this is due to the fact that dwarf
galaxies frequently show disturbed morphologies, so there is
no clear ‘centre’ for BHs to be located in. However, the dis-
placement of BHs in dwarf galaxies goes beyond this effect,
as the fraction of disturbed morphologies in dwarf galaxies
falls with decreasing redshift (in NewHorizon, from 20 per-
cent at z = 3 to 5 percent at z = 1, see Martin et al. 2020),
while the mean distance of main and secondary BHs to the
centre of the galaxy increases with decreasing redshift.
By z = 0.25, only 27 percent of main BHs (and 30 per-
cent of all BHs) remain within 1 kpc of the centre, which
is lower than in previous studies of BHs in dwarf galaxies
such as Sharma et al. (2022). This is through a mix of BHs
being displaced by galaxy mergers (Bellovary et al. 2021)
but most likely also a consequence of NewHorizon lower
BH seed mass, as dynamical timescales for BHs to settle
back to galactic centres is directly proportional to the mass
of the BH (Pfister et al. 2019), and lower-mass BHs strug-
gle more than higher mass BHs to sink back to the centre
(Ma et al. 2021; Bellovary et al. 2021) (see also Fig. 11). In
NewHorizon, BH might additionally be prone to wandering
as their seed mass of MBH,0 = 104 M is close to the stel-
lar mass resolution of 1.3 × 104 M of NewHorizon, which
induces stochastic effects in their orbits. Additionally, we
do not analytically model the unresolved dynamical friction
from star and DM particles, which has been shown to play a
strong role in allowing BHs to sink to the centre of galaxies
(Pfister et al. 2019; Chen et al. 2021), and the dynamical
friction from gas becomes less efficient at increased resolu-
tion due to instabilities in the wake (Beckmann et al. 2018)
and the turbulent nature of the gas (Lescaudron et al. 2022).
3.7 Discussion
One of the clear take-aways of the results presented here is
how strongly observational signatures of IMBHs at low red-
shift are influenced by the BH dynamics, which has signifi-
cant consequences on both the MBH−Mstar and observability
of IMBHs at low redshift. As can be seen in Fig. 7, the radial
distance of BHs from the centre of their host galaxy is anti-
correlated with the accretion efficiency, and therefore regu-
lates the long-term mass evolution: In NewHorizon , BHs
struggle to grow until the gain masses of ∼ 5 × 105 M, at
which point BHs sink efficiently to the centre of galaxies (see
Fig. 11) and grow onto the observed correlations (see Fig. 2).
This causes a clear break in the MBH − Mstar relation, which
is not reported in other works on IMBHs in dwarf galaxies
that either force the BH to remain attached to the centres
of galaxies (Barai & de Gouveia Dal Pino 2019), or do not
study the low-mass regime by using a high seed masses (e.g.
Sharma et al. 2022, who use a seed mass of 106 M). Both
works note that their lack of dynamics and high seed mass
respectively mean their results are likely an upper limit on
IMBH growth in dwarf galaxies. By the same argument, the
wandering of BHs in NewHorizon means our results likely
present a lower limit on the growth of IMBHs in dwarf galax-
ies. In this context it is interesting to note that Koudmani
et al. (2019), who use a lower seed mass of 105 M as well as
a repositioning scheme also report some evidence for flatten-
ing of the MBH − Mstar relation. Like NewHorizon, they also
under-predict observed X-ray AGN fractions at low redshift
(see Fig. 9), despite the repositioning scheme that forces
their IMBHs into galactic centres. Overall this shows that
IMBHs in dwarf galaxies are extremely sensitive probes of
BH and galaxy coevolution, and that the dynamics of the
BHs lie at the heart of when and how IMBHs and dwarf
galaxies coevolve.
This points to a big open problem in the field: if low-
mass BHs struggle to remain attached to the centres of
galaxies, and not being attached to the centres of galax-
ies means BHs struggle to grow, how can we explain the
observed, active IMBHs in dwarf galaxies? Either there is
something that is fundamentally missing in our understand-
ing (or numerical modelling) of BH dynamics, or all seed
BHs must be sufficiently massive to avoid such dynamical
difficulties. We will further explore the dynamics of both
main and secondary BHs in NewHorizon galaxies, and its
impact on long-term IMBH mass growth, in detail in an
upcoming companion paper Beckmann et al. (2023).
4 CONCLUSIONS
In this paper we studied the evolution of BHs in dwarf galax-
ies in the NewHorizon large-volume zoom simulation. We
found that
(i) BHs do not start growing efficiently until their host
galaxy leaves the dwarf galaxy regime (i.e. when Mstar >
3 × 109 M. As a result, most BHs in dwarf galaxies remain
near their seed mass for long periods of time (here MBH,0 =
104 M. There is a weak trend for BHs in more massive
dwarf galaxies to grow more through accretion than in low-
mass dwarfs.
(ii) Occupation fractions of BHs in dwarf galaxies remain
high and show little evolution with redshift. The fraction of
galaxies hosting multiple BHs increases strongly with host
galaxy stellar mass and reaches unity as galaxies become
massive.
(iii) BHs grow much more actively at high redshift (z ≥
2), where there are a significant number of objects that have
time-averaged high Eddingtion ratios. At low redshift (≤ 1,
average Eddington ratios fall very low but brief bursts of
higher activity remain common. These bursts are too infre-
quent to contribute significant mass growth to the BH, but
do make BHs intermittently observable.
(iv) When looking at the X-ray luminosity of BHs and
their host galaxies, most BHs are insufficiently luminous to
outshine their host even at high redshift. This means that
MNRAS 000, 1–15 (2015)
14 Beckmann, R.S.
even at high redshift (z = 3), the fraction of dwarf galaxies
that host an AGN that can be detected above an optimistic
luminosity threshold of 1039 ergs−1 is only ∼ 17 percent.
(v) At lower redshift, the fraction of AGN for a given
X-ray luminosity cut decreases, with no observable AGN
remaining at z ∼ 0.25.
(vi) Due to their low seed mass, BHs in
NewHorizon struggle to remain attached to the cen-
tres of their host galaxies, with the average distance
between BHs and galaxy centre increasing from 0.68 kpc at
z = 3 to 2.96 kpc at z = 0.25. BHs in massive galaxies sink
efficiently to the centre.
Overall, the evolution of the BH population in
NewHorizon shows that the lower seed mass exacerbates
many of the processes that limit BH growth. Previous sim-
ulation work with higher seed mass should consequently be
seen as an upper limit to how much BHs in dwarf galaxies
can grow given our current model of BH dynamics, BH ac-
cretion and SN feedback. As such, dwarf galaxies remain a
promising laboratory to constrain stellar feedback and BH
physics.
DATA AVAILABILITY
All data used in this paper is available upon request to the
first author.
ACKNOWLEDGEMENTS
This work was granted access to the HPC resources of
CINES under the allocations c2016047637, A0020407637
and A0070402192 by Genci, KSC-2017-G2-0003 by KISTI,
and as a “Grand Challenge” project granted by GENCI
on the AMD Rome extension of the Joliot Curie super-
computer at TGCC. This research is part of the Spin(e)
ANR-13-BS05-0005 (http://cosmicorigin.org), Segal ANR-
19-CE31-0017 (http://secular-evolution.org) and Horizon-
UK projects. This work has made use of the Infinity clus-
ter on which the simulation was post-processed, hosted by
the Institut d’Astrophysique de Paris. RSB would like to
thank Newnham College, Cambridge, for financial support.
TK was supported by the National Research Foundation of
Korea (NRF) grant funded by the Korea government (No.
2020R1C1C1007079 and No. 2022R1A6A1A03053472). We
warmly thank S. Rouberol for running it smoothly. The large
data transfer was supported by KREONET which is man-
aged and operated by KISTI. RSB gratefully acknowledges
funding from Newnham College, Cambridge. SK acknowl-
edges support from the STFC [ST/S00615X/1] and a Senior
Research Fellowship from Worcester College, Oxford.
REFERENCES
Aird J., Coil A. L., Georgakakis A., 2018, MNRAS, 474, 1225
Amaro-Seoane P., 2018, Phys. Rev. D, 98, 063018
Angle´s-Alca´zar D., Faucher-Gigue`re C.-A., Quataert E., Hopkins
P. F., Feldmann R., Torrey P., Wetzel A., Keresˇ D., 2017,
MNRAS, 472, L109
Baldassare V. F., Dickey C., Geha M., Reines A. E., 2020, ApJ,
898, L3
Barai P., de Gouveia Dal Pino E. M., 2019, MNRAS, 487, 5549
Barat D., et al., 2019, MNRAS, 487, 2924
Baron D., Me´nard B., 2019, MNRAS, 487, 3404
Beckmann R. S., Slyz A., Devriendt J., 2018, MNRAS, 478, 995
Beckmann R. S., Dubois Y., Volonteri M., 2022, in prep
Beckmann R. S., Volonteri M., Dubois Y., 2023, in prep
Bellovary J., Cleary C., Munshi F., Tremmel M., Christensen C.,
Brooks A., Quinn T., 2018, MNRAS, 482, 2913
Bellovary J. M., et al., 2021, MNRAS, 505, 5129
Benson A. J., Babul A., 2009, MNRAS, 397, 1302
Birchall K. L., Watson M. G., Aird J., 2020, MNRAS, 492, 2268
Blandford R. D., Znajek R. L., 1977, MNRAS, 179, 433
Boldrini P., Mohayaee R., Silk J., 2020, MNRAS, 495, L12
Bower R., Schaye J., Frenk C. S., Theuns T., Schaller M., Crain
R. A., McAlpine S., 2016, MNRAS, 465, 32
Cappellari M., et al., 2013, MNRAS, 432, 1862
Chabrier G., 2005, The Initial Mass Function: From Salpeter 1955
to 2005. Vol. 327, Springer Netherlands, doi:10.1007/978-
1-4020-3407-7 5, http://link.springer.com/10.1007/
978-1-4020-3407-7_5
Chen N., Ni Y., Tremmel M., Matteo T. D., Bird S., DeGraf C.,
Feng Y., 2021, MNRAS, 510, 531
Chilingarian I. V., Katkov I. Y., Zolotukhin I. Y., Grishin K. A.,
Beletsky Y., Boutsia K., Osip D. J., 2018, ApJ, 863, 1
Dashyan G., Dubois Y., 2020, A&A, 638, A123
Davis F., et al., 2022, MNRAS, 511, 4109
De Cun V. I., Bellovary J. M., Katz M. L., 2023, MNRAS, 520,
3916
Dickey C. M., Geha M., Wetzel A., El-Badry K., 2019, ApJ, 884,
180
Dubois Y., Devriendt J., Slyz A., Teyssier R., 2012, MNRAS, 420,
2662
Dubois Y., Volonteri M., Silk J., Devriendt J., Slyz A., 2014a,
MNRAS, 440, 2333
Dubois Y., et al., 2014b, MNRAS, 444, 1453
Dubois Y., Volonteri M., Silk J., Devriendt J., Slyz A., Teyssier
R., 2015, MNRAS, 452, 1502
Dubois Y., et al., 2021, A&A, 651, A109
Eisenstein D. J., Hut P., 1998, ApJ, 498, 137
Gair J. R., Mandel I., Miller M. C., Volonteri M., 2011, Gen.
Relativ. Gravit., 43, 485
Gallo E., Sesana A., 2019, ApJ, 883, L18
Greene J. E., Strader J., Ho L. C., 2019, ARA&A, 58, 257
Haardt F., Madau P., 1996, ApJ, 461, 20
Habouzit M. M., Volonteri M., Dubois Y., 2017, MNRAS, 468,
3935
Habouzit M., et al., 2022, MNRAS, 509, 3015
Haidar H., et al., 2022, MNRAS, 514, 4912
Hopkins P. F., Richards G. T., Hernquist L., 2007, ApJ, 654, 731
Katz M. L., Kelley L. Z., Dosopoulou F., Berry S., Blecha L.,
Larson S. L., 2020, MNRAS, 491, 2301
Kaviraj S., Martin G., Silk J., 2019, MNRAS, 489, L12
Kimm T., Cen R., Devriendt J., Dubois Y., Slyz A., 2015, MN-
RAS, 451, 2900
Kimm T., Katz H., Haehnelt M., Rosdahl J., Devriendt J., Slyz
A., 2017, MNRAS, 466, stx052
King A., Nealon R., 2021, MNRAS, 502, L1
King A. R., Lubow S. H., Ogilvie G. I., Pringle J. E., 2005, MN-
RAS, 363, 49
Komatsu E., et al., 2011, ApJSupplement Series, 192, 18
Kormendy J., Ho L. C., 2013, ARA&A, 51, 511
Koudmani S., Sijacki D., Bourne M. A., Smith M. C., 2019, MN-
RAS, 484, 2047
Koudmani S., Henden N. A., Sijacki D., 2021, MNRAS, 503, 3568
Koudmani S., Sijacki D., Smith M. C., 2022, MNRAS, 516, 2112
Latimer C. J., Reines A. E., Bogdan A., Kraft R., 2021, ApJ, 922,
L40
Lehmer B. D., et al., 2016, ApJ, 825, 7
MNRAS 000, 1–15 (2015)
IMBHs in dwarf galaxies 15
Lescaudron S., Dubois Y., Beckmann R. S., Volonteri M., 2022,
Accepted in A&A, p. arXiv:2209.13548
Lupi A., Sbarrato T., Carniani S., 2020, MNRAS, 492, 2528
Ma L., Hopkins P. F., Ma X., Angle´s-Alca´zar D., Faucher-Gigue`re
C.-A., Kelley L. Z., 2021, MNRAS, 508, 1973
Manzano-King C. M., Canalizo G., 2020, MNRAS, 498, 4562
Martin G., et al., 2020, MNRAS, 500, 4937
Massonneau W., Volonteri M., Dubois Y., Beckmann R. S., 2023,
A&A, 670, A180
Mezcua M., 2017, Int. J. Mod. Phys. D, 26, 1730021
Mezcua M., Sa´nchez H. D., 2020, ApJ, 898, L30
Mezcua M., Civano F., Marchesi S., Suh H., Fabbiano G., Volon-
teri M., 2018, MNRAS, 478, 2576
Mineo S., Gilfanov M., Sunyaev R., 2012, MNRAS, 426, 1870
Molina M., Reines A. E., Latimer C. J., Baldassare V., Salehirad
S., 2021, ApJ, 922, 155
Nelson D., et al., 2019, MNRAS, 490, 3234
Nguyen D. D., et al., 2019, ApJ, 872, 104
Oh S., et al., 2020, MNRAS, 495, 4638
Ostriker E. C., 1999, ApJ, 513, 252
Pacucci F., Mezcua M., Regan J. A., 2021, ApJ, 920, 134
Pardo K., et al., 2016, ApJ, 831, 203
Paynter J., Webster R., Thrane E., 2021, Nat. Astron, 5, 560
Pfister H., Volonteri M., Dubois Y., Dotti M., Colpi M., 2019,
MNRAS, 486, 101
Power C., Navarro J. F., Jenkins A., Frenk C. S., White S. D. M.,
Springel V., Stadel J., Quinn T., 2003, MNRAS, 338, 14
Regan J. A., Downes T. P., Volonteri M., Beckmann R., Lupi A.,
Trebitsch M., Dubois Y., 2018, MNRAS, 486
Reines A. E., Volonteri M., 2015, ApJ, 813, 82
Reines A. E., Greene J. E., Geha M., 2013, ApJ, 775, 116
Reines A., Condon J., Darling J., Greene J., 2019, ApJ, 888, 36
Rezzolla L., Barausse E., Dorband E. N., Pollney D., Reisswig C.,
Seiler J., Husa S., 2008, Phys. Rev. D, 78, 044002
Ricarte A., Natarajan P., 2018, MNRAS, 481, 3278
Ricarte A., Tremmel M., Natarajan P., Quinn T., 2019, MNRAS,
489, 802
Sesana A., Haardt F., Madau P., Volonteri M., 2005, ApJ, 623,
23
Sharma R. S., Brooks A. M., Somerville R. S., Tremmel M.,
Bellovary J., Wright A. C., Quinn T. R., 2019, ApJ, 897, 103
Sharma R. S., Brooks A. M., Tremmel M., Bellovary J., Ricarte
A., Quinn T. R., 2022, ApJ, 936, 82
She R., Ho L. C., Feng H., 2017, ApJ, 835, 223
Shen X., Hopkins P. F., Faucher-Gigue`re C.-A., Alexander D. M.,
Richards G. T., Ross N. P., Hickox R. C., 2020, MNRAS, 495,
3252
Shin L., Woo J.-H., Son D., Cho H., Kim T., Gallo E., Kang W.,
2022, AJ, 163, 73
Sutherland R. S., Dopita M. A., 1993, ApJSupplement Series, 88,
253
Teyssier R., 2002, doi:10.1051/0004-6361:20011817, 385, 337
Tillman M. T., Wellons S., Faucher-Gigue`re C.-A., Kelley L. Z.,
Angle´s-Alca´zar D., 2022, MNRAS, 511, 5756
Toptun V., Chilingarian I., Grishin K., Katkov I., Zolotukhin
I., Goradzhanov V., Demianenko M., Kuzmun I., 2022.
Crossref, pp 304–306, doi:10.51194/VAK2021.2022.1.1.117,
http://www.inasan.ru/wp-content/uploads/2022/02/VAK_
2021_.pdf#page=305
Trebitsch M., Blaizot J., Rosdahl J., Devriendt J., Slyz A., 2017,
MNRAS, 470, 224
Trebitsch M., Volonteri M., Dubois Y., Madau P., 2018, MNRAS,
478, 5607
Trebitsch M., et al., 2021, A&A, 653, A154
Valiante R., et al., 2021, MNRAS, 500, 4095
Volonteri M., Natarajan P., 2009, MNRAS, 400, 1911
Volonteri M., Dubois Y., Pichon C., Devriendt J., 2016, MNRAS,
460, 2979
z α M′ 
z = 0.25 0.93 9.65 9.98
z = 1 0.61 10.36 28.45
z = 2 0.67 10.31 21.25
z = 3 0.48 10.5 36.2
Table A1. Fitting parameters for focc from Eq. 3 as a function
of halo mass Mhalo. Fits are shown in the top panel of Fig. A1 as
solid lines.
Volonteri M., et al., 2020, MNRAS, 498, 2219
Wylezalek D., Zakamska N. L., Greene J. E., Riffel R. A., Drory
N., Andrews B. H., Merloni A., Thomas D., 2018, MNRAS,
474, 1499
Yang J., Gurvits L. I., Paragi Z., Frey S., Conway J. E., Liu X.,
Cui L., 2020, MNRAS, 495, L71
Zhang W. M., Soria R., Zhang S. N., Swartz D. A., Liu J., 2009,
ApJ, 699, 281
APPENDIX A: HALO OCCUPATION
FRACTIONS
Fig. A1 shows the halo occupation fraction and the average
number of BHs per halo. The analysis shown here is equiva-
lent to that for galaxies in Section 3.2, but as a function of
DM halo mass rather than galaxy stellar mass. As before,
occupation fractions are fit with a function of the form of
Eq. 3 where M is now the DM halo mass Mhalo, with fitting
parameters shown in Tab. A1.
This paper has been typeset from a TEX/LATEX file prepared by
the author.
MNRAS 000, 1–15 (2015)
16 Beckmann, R.S.
10
0
f o
cc
fit
data
10
8
10
9
10
10
10
11
10
12
10
13
Mhalo [ M ]
10
0
10
1
10
2
<n
BH
>
z = 0.25
z = 1
z = 2
z = 3
Figure A1. Fraction of halos that contain at least one BH [top]
and average number of BHs per halo[bottom] as a function of halo
mass Mhalo. Solid lines on the top panel denote a fit of Eq. 3, with
free parameters at each redshift listed in Table A1. Dashed lines
connect data points. Error bars show Poisson errors.
MNRAS 000, 1–15 (2015)