MNRAS 000, 1–21 (2015) Preprint 4 July 2023 Compiled using MNRAS LATEX style file v3.

The incidence of AGN in galaxies with different stellar population ages

Q. Ni,1,2★ J. Aird,1 A. Merloni,2 K. L. Birchall,3 J. Buchner,2 M. Salvato,2 G. Yang4,5
1 Institute
for Astronomy, University of Edinburgh, Royal Observatory, Edinburgh EH9 3HJ, UK
2 Max-Planck-Institut für extraterrestrische Physik (MPE), Gießenbachstraße 1, D-85748 Garching bei München, Germany
3 School of Physics & Astronomy, University of Leicester, University Road, Leicester LE1 7RH, UK
4 Kapteyn Astronomical Institute, University of Groningen, P.O. Box 800, 9700 AV Groningen, The Netherlands
5 SRON Netherlands Institute for Space Research, Postbus 800, 9700 AV Groningen, The Netherlands
arXiv:2307.00051v1 [astro-ph.GA] 30 Jun 2023

Accepted XXX. Received YYY; in original form ZZZ

ABSTRACT
It has been argued that recycled gas from stellar mass loss in galaxies might serve as an important fuelling source for black holes
(BHs) in their centers. Utilizing spectroscopic samples of galaxies from the Sloan Digital Sky Survey (SDSS) at 𝑧 = 0–0.35 and
the Large Early Galaxy Astrophysics Census (LEGA-C) survey at 𝑧 = 0.6–1 that have X-ray coverage from XMM-Newton or
Chandra, we test this stellar mass loss fuelling scenario by investigating how AGN activity and BH growth vary with the break
strength at 4000 Å, Dn 4000 (which is closely related to the age of stellar populations), as younger galaxies are considered to
have higher stellar mass loss rates. We found that when controlling for host-galaxy properties, the fraction of log 𝐿 X /𝑀★ > 32
(which roughly corresponds to Eddington ratios ≳ 1%) AGN and sample-averaged black hole accretion rate (BHAR) decrease
with Dn 4000 among Dn 4000 ≲ 1.9 galaxies, suggesting a higher level of AGN activity among younger galaxies, which supports
the stellar mass loss fuelling scenario. For the oldest and most massive galaxies at 𝑧 = 0–0.35, this decreasing trend is not present
anymore. We found that, among these most massive galaxies at low redshift, the fraction of low specific-accretion-rate (31 <
log 𝐿 X /𝑀★ < 32) AGNs increases with Dn 4000, which may be associated with additional fuelling from hot halo gas and/or
enhanced accretion capability.
Key words: galaxies: active – galaxies: evolution – galaxies: nuclei – X-rays: galaxies

1 INTRODUCTION & Ho 2013). However, the noticeable fraction of AGNs in quiescent

galaxies (given their low level of star formation activity and thus
In the past decades, the understanding of how galaxies evolve over
inferred low level of global cold gas content) indicates that additional
cosmic history has progressed rapidly as a result of accumulating
mechanisms may fuel the growth of central BHs after quenching
data from various imaging and spectroscopic surveys (e.g. Kauff-
(e.g. Kocevski et al. 2017; Wang et al. 2017; Aird et al. 2019, 2022).
mann et al. 2003; Faber et al. 2007; Madau & Dickinson 2014; van
Among galaxies where the cold gas supply is not sufficient, stellar
der Wel et al. 2014; Barro et al. 2017; Wu et al. 2018). Galaxies
mass loss may provide an important, additional source of fuel for
appear to follow a range of evolutionary pathways, whereby their
accretion onto the BH (e.g. Hopkins & Hernquist 2006; Ciotti &
stellar populations and gas properties change over time, although the
Ostriker 2007; Kauffmann & Heckman 2009). In this scenario, as
physical processes that drive these transformations are still unclear.
galaxies with younger stellar populations have stronger stellar winds
While supermassive black holes (BHs) only occupy a small space
and higher mass loss rates that provide more fuel for the central BHs,
in the galaxy centers, they are thought to play an important role in
BH growth is expected to vary among galaxies with different ages of
galaxy evolution. As the mass of BHs is found to be tightly correlated
stellar populations. Through characterizing the AGN activity across
with the mass of their host bulges (e.g. Magorrian et al. 1998; Mar-
the lifecycle of galaxies, we can investigate the role of stellar mass
coni & Hunt 2003; McConnell & Ma 2013; Kormendy & Ho 2013),
loss in fuelling BHs.
BHs appear to coevolve with their host galaxies. While we know that
central supermassive BHs grow primarily by accreting gas and can As spectra can provide direct information about the age of stellar
be seen as Active Galactic Nuclei (AGNs), the exact feeding mech- populations through features such as the break strength at 4000 Å,
anism of BHs remains unclear. It is important to investigate what Dn 4000, large samples of galaxies and AGNs with spectroscopic
drives the growth of BHs, which will ultimately reveal the physical coverage are needed to investigate the potential fuelling through stel-
mechanisms behind the potential coevolution scenario. lar mass loss. In the local universe, the Sloan Digital Sky Survey
It has been found that BH accretion rate tracks the star formation (SDSS) provides a large legacy sample of galaxies (e.g. Strauss et al.
rate over cosmic history, suggesting cold gas supply as a common 2002), and the X-ray information provided by serendipitous cata-
fuel for both the galaxy and the BH (e.g. Aird et al. 2010; Kormendy logs from accumulating X-ray observations over the entire sky (e.g.
Evans et al. 2010; Webb et al. 2020) provides opportunities to effec-
tively identify AGNs among this large galaxy sample (e.g. Brandt &
★ E-mail: [email protected] Alexander 2015; Brandt & Yang 2021). More recently, deep spectro-

© 2015 The Authors

2
scopic surveys (e.g. van der Wel et al. 2016; McLure et al. 2018) have We use the RapidXMM database (Ruiz et al. 2022) to select galax-
started to help build representative samples of galaxies at intermedi- ies in MPA-JHU catalog within XMM-Newton coverage. RapidXMM
ate reshifts. When deep X-ray coverage is available, these surveys can provides XMM-Newton aperture photometry as well as upper limits
also be effectively utilized to study the incidence of AGNs in galaxies within HEALPix cells of size ≈ 3 arcsec. As the 4XMM-DR11
with different stellar population ages (e.g. Mountrichas et al. 2022; catalog (Webb et al. 2020) is utilized to match galaxies with their
Georgantopoulos et al. 2023). XMM-Newton counterparts for X-ray AGN identification (see Sec-
In this paper, we use samples of galaxies from SDSS and the Large tion 2.1.2), we limit our sample to SDSS galaxies with XMM-Newton
Early Galaxy Census (LEGA-C; e.g. van der Wel et al. 2016) survey point observations made before 2020 December 17.
to investigate how AGN activity and BH growth vary with Dn 4000
across the galaxy lifecycle at 𝑧 = 0−0.35 and 𝑧 = 0.6−1. We also test
whether AGN activity and BH growth vary with Dn 4000 when other 2.1.2 Matching selected galaxies with XMM-Newton counterparts
galaxy properties are controlled, thus revealing whether the age of
The 4XMM-DR11 catalog used in this study to provide X-ray coun-
stellar populations has a fundamental influence on AGN activity/BH
terparts to SDSS galaxies contains sources drawn from a total of
growth.
12210 XMM-Newton observations made between 2000 February 1
The paper is structured as follows. In Section 2, we describe the
and 2020 December 17, including ∼ 600,000 unique X-ray sources
sample construction process. In Section 3, we detail the analysis
over ≈ 1239 deg2 .
results and discuss what they imply in Section 4. The conclusions
As SDSS galaxies in the MPA-JHU catalog are a subset of SDSS
are presented in Section 5. Throughout this paper, stellar masses (𝑀★)
objects, we first match X-ray sources in the 4XMM DR11 catalog
are given in units of 𝑀⊙ ; star formation rates (SFRs) and sample-
with SDSS DR8 (Adelman-McCarthy & et al. 2011) and catWISE
averaged black hole accretion rates (BHAR) are given in units of
(Marocco et al. 2021) catalogs, and use the obtained optical/NIR
𝑀⊙ yr −1 . 𝐿 𝑋 represents X-ray luminosity at rest-frame 2–10 keV in
counterparts positions to further match with MPA-JHU galaxies.
units of erg s −1 . Quoted uncertainties are at the 1𝜎 (68%) confidence
When we perform the source matching, we only consider 4XMM
level. A cosmology with 𝐻0 = 70 km s −1 Mpc −1 , Ω 𝑀 = 0.3, and
sources that are within 30 arcmin of MPA-JHU sources, thus restrict-
ΩΛ = 0.7 is assumed.
ing to 4XMM sources that lie within the footprint of the MPA-JHU
catalog. The input SDSS and WISE catalogs only include objects
within 1 arcmin of 4XMM sources, to save computation time. Fol-
2 DATA & SAMPLE SELECTION lowing the method in Ni et al. (2021a), we use NWAY (Salvato et al.
2018) to perform the source matching, with priors obtained from
Two samples of X-ray AGNs with spectroscopic coverage are utilized sources reported in the Chandra Source Catalog (CSC) 2.0 (Evans
in this study: one sample includes SDSS galaxies with XMM-Newton et al. 2010) that are matched to the 4XMM sources within the 95%
coverage (see Section 2.1); one sample includes LEGA-C galaxies uncertainties (as Chandra detections have better positional accu-
with Chandra coverage in the COSMOS field (see Section 2.2). In racy than XMM-Newton detections). Utilizing the SDSS and WISE
Section 2.4, we discussed how the AGN fraction and BHAR are sources that are matched to these Chandra sources within 5′′ , we
estimated. obtain priors in the WISE W1+W2 vs. W1−W2 color space and in
the 𝑖-band magnitude space. With these priors, we match 4XMM
sources to their optical/NIR counterparts. Matched primary counter-
2.1 Constructing a sample of SDSS galaxies with XMM-Newton
parts with 𝑝 any > 0.1 and 𝑝 i > 0.1 are adopted as reliable matches.
coverage
We then match MPA-JHU galaxies to these obtained optical/NIR
2.1.1 Selecting galaxies in MPA-JHU catalog with XMM-Newton positions of 4XMM sources with a 1′′ matching radius. To assess the
coverage matching accuracy, we use the 4XMM sources that have Chandra
counterparts, and compare the level of agreement between matched
The SDSS galaxy/AGN sample utilized in the study is built upon the MPA-JHU sources when either 4XMM or CSC position is used. The
MPA-JHU value-added catalog.1 The MPA-JHU catalog provides matched MPA-JHU counterparts of 4XMM sources have a ≈96%
galaxy property measurements for SDSS DR8 spectra classified by agreement with those of CSC sources.
the pipeline as a galaxy (SPECTROTYPE = GALAXY), including
Dn 4000, 𝑀★, and SFR used in this study. We only consider objects
observed as the prime targets in the SDSS Legacy Survey Main 2.2 LEGA-C galaxies in the COSMOS field
Galaxy Sample in this study.2
Dn 4000 in the MPA-JHU catalog is measured according to the To extend the redshift range probed in this study, we also include
Balogh et al. (1999) definition, as the ratio of the flux in the 4000– galaxies/AGNs in the LEGA-C Data Release 3 (van der Wel et al.
4100 Å continuum to that in the 3850–3950 Å continuum. 𝑀★ values 2021) which have Chandra X-ray coverage from the COSMOS-
are measured following model grids based on SDSS photometry, as Legacy survey (Civano et al. 2016).3 We only utilize sources in
described in Kauffmann et al. (2003). SFR values are measured the LEGA-C catalog with PRIMARY = 1, FLAG_MORPH = 0, and
by combining emission line measurements from Brinchmann et al. FLAG_SPEC < 2; Dn 4000 values are available for 2129 of these
(2004) with aperture corrections as described in Salim et al. (2007). sources, measured according to the Balogh et al. (1999) definition
𝑀★ and SFR values in the MPA-JHU catalog are largely consistent as well. We also examine the spectra visually and remove spectra
with those derived from Galex-SDSS-WISE SED fitting (e.g. Salim with broad emission lines from the sample. We cross-match these
et al. 2016), and have been widely adopted. sources to the photometric sample of galaxies in the COSMOS field
in Ni et al. (2021b), which provides 𝑀★ and SFR measurements for

1 http://www.mpa-garching.mpg.de/SDSS/DR7/
2 Galaxies in the SDSS Legacy Survey are observed with 3′′ -diameter fibers. 3 Galaxies in the LEGA-C survey are observed with 1′′ × 8′′ slits.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 3

Table 1. Summary of sample properties. (1) Name of the sample. (2) Redshift Dn 4000 measurements is small for AGNs in our sample, and will not
range of the sample. (3) 𝑀★ range of the sample. (4) Number of galaxies in materially affect our results.
the sample. (5) Number of log 𝐿X /𝑀★ > 32 AGNs.

Sample Redshift Mass Number of Number of

Name Range Range Galaxies AGNs
(1) (2) (3) (4) (5)

4XMM 0–0.35 log𝑀★ > 9 22576 89 2.4 Obtaining AGN fraction and sample-averaged black-hole
COSMOS 0.6–1.0 log𝑀★ > 10 1496 38 accretion rate
The AGN fraction is defined in terms of 𝐿 X /𝑀★ (as advocated by
Bongiorno et al. 2016; Aird et al. 2018, 2019, 2022; Birchall et al.
sources within both the COSMOS and UltraVISTA regions utiliz- 2022), which measures the rate of black hole growth relative to the
ing photometric data in 38 bands (including 24 broad bands) from stellar mass of the host galaxy (i.e. the “specific black hole accretion
NUV to FIR (Laigle et al. 2016) and the SED-fitting code CIGALE rate”) and thus accounts for the overall stellar-mass-selection bias
(e.g. Boquien et al. 2019; Yang et al. 2020). X-ray counterparts from whereby weakly accreting AGN in more massive galaxies have a
Chandra are also matched for this photometric sample of galaxies in higher 𝐿 X and are thus easier to detect (see Aird et al. 2012). AGN
Ni et al. (2021b). The detailed process to obtain 𝑀★ and SFR mea- fractions throughout this work, unless otherwise stated, refer to ob-
surements can be seen in section 2.1 of Ni et al. (2021b). An AGN jects with log 𝐿 X /𝑀★ ⩾ 32, which roughly corresponds to Eddington
component is included in the SED fitting, in addition to the galaxy ratios of ≳ 1% following the conversion factors from Equation 2 of
component. As discussed in Ni et al. (2021b), adding the X-ray infor- Aird et al. (2018). For the 4XMM sample, we adopt the X-ray fluxes
mation or not during the SED fitting does not significantly affect the of detected XMM-Newton sources from the 4XMM DR11 catalog
Bayesian 𝑀★ and SFR measurements. Uncertainties of the 𝑀★ and (Webb et al. 2020). We convert the X-ray fluxes to 𝐿 X assuming a
SFR values obtained are also discussed in section 2.1 and appendix power-law model with Galactic absorption and Γ = 1.7 following
A of Ni et al. (2021b). We note that 𝑀★ measurements are generally the preference order of 4.5–12 keV band, and 0.2–12 keV band, thus
robust when comparing results from different SED fitting methods, minimizing the effects of X-ray obscuration. About 90% of the X-ray
with a scatter of ≈ 0.1 dex; the scatter of SFR measurements can be sources in our 4XMM sample have 4.5–12 keV band flux measure-
up to ≈ 0.4 dex. This finally provides a sample of 1792 galaxies in ments available. For the COSMOS sample, we adopt 𝐿 X calculated
the COSMOS field. from Ni et al. (2021b), which is converted from X-ray fluxes fol-
lowing the 2–7 keV, 0.5–7 keV, and 0.5–2 keV order, also assuming
the Γ = 1.7 power-law model. For X-ray sources in the COSMOS
sample, ≈ 70% of them have 2–7 keV band flux measurements avail-
2.3 Sample properties
able, and 0.5–7 keV/0.5–2 keV band flux measurements are used
In Figure 1, we present the 𝑀★ vs. 𝑧 distributions for the selected for ≈ 27%/3% of them. In the left panels of Figure 4, we present
SDSS galaxies and LEGA-C galaxies. For SDSS galaxies with the 𝐿 X /𝑀★ distribution of X-ray detected sources which have 𝐿 X
XMM-Newton coverage, we plot the 90% 𝑀★ completeness curve of values greater than the contributions from X-ray binaries (XRBs) in
galaxies in the SDSS main galaxy sample. The limiting 𝑀★ is derived our samples. The XRB luminosity (𝐿 X,XRB ) is estimated through
following section 3.2 of Ilbert et al. (2013) given the 𝑟-band Petrosian a redshift-dependent function of 𝑀★ and SFR (model 269, Fragos
magnitude limit of 17.77 of the SDSS main galaxy sample (Strauss et al. 2013), which is derived utilizing observations in Lehmer et al.
et al. 2002). In our study, we only utilize SDSS galaxies above this (2016).
mass-completeness curve, and we refer to this galaxy sample as the When deriving the AGN fraction, we correct 𝐿 X to account for
4XMM sample throughout the remainder of this work. The LEGA-C the modest systematic effect from obscuration with correction factors
primary targets are drawn from a 𝐾 𝑠 -band selected parent sample; detailed below. Utilizing X-ray sources in Chandra Deep Field-South
these targets are representative of the parent sample when taking into (Luo et al. 2017) that have similar X-ray flux level as Chandra COS-
account the selection correction factor SCOR provided in the LEGA-C MOS sources but with more counts, Yang et al. (2018) compared the
catalog (see Appendix A of van der Wel et al. 2021 for details). This intrinsic 𝐿 X from spectral modeling with 𝐿 X calculated following
selection correction factor is utilized in our analyses to weight each the scheme mentioned above, and found that the overall underesti-
galaxy, although we find that excluding this factor does not impact mation of X-ray emission due to obscuration is ≈ 20%. We apply
our results materially (see Section 3.1 for details). The 90% 𝑀★ com- this correction factor throughout this work for all the 𝐿 X values of
pleteness curve for the parent sample (which has a redshift-dependent X-ray detected sources in the COSMOS sample when calculating
𝐾 𝑠 -band limiting magnitude of 20.7−7.5×log((1+𝑧)/1.8)) is shown the AGN fraction. The obscuration correction factor is obtained in a
on the plot. In our study, we only utilize LEGA-C galaxies above this similar manner for the 4XMM sample, utilizing the XMMFITCAT-Z
mass-completeness curve of the parent sample, and we refer to this spectral fit catalag (Ruiz et al. 2021) that provides intrinsic 𝐿 X mea-
galaxy sample as the COSMOS sample throughout the remainder of surements for 3XMM-DR6 sources. We match X-ray sources in our
this work. We carry out our studies with mass-complete samples to sample with sources in the XMMFITCAT-Z catalog that have spec-𝑧
avoid any potential bias associated with an incomplete characteriza- available. With this matched sample, we derive the obscuration cor-
tion of galaxy populations. In Table 1, we present the properties of rection factor needed by obtaining the average value of the intrinsic
the 4XMM sample and the COSMOS sample. In Figures 2 and 3, 𝐿 X reported divided by 𝐿 X calculated in this work; 𝐿 X -dependent
we present how galaxies/AGNs in our samples distribute on the SFR weights are applied to the matched sample to recover the 𝐿 X distribu-
vs. 𝑀★ plane, SFR vs. Dn 4000 plane, and Dn 4000 vs. 𝑀★ plane; tion of the X-ray AGNs in the whole 4XMM sample. We found that
we also present the Dn 4000 distribution of galaxies/AGNs. In Ap- the overall underestimation of X-ray emission due to obscuration is
pendix A, we show that the contamination from AGN emission to ≈ 10%. We apply this correction factor throughout this work for all

MNRAS 000, 1–21 (2015)

4

Figure 1. 𝑀★ as a function of 𝑧 for SDSS main sample galaxies in the MPA-JHU catalog with XMM-Newton coverage (blue dots) and galaxies included the
LEGA-C survey (green dots). X-ray AGNs with log 𝐿X /𝑀★ > 32 are marked as orange stars. The dashed curve indicates the 90% 𝑀★ completeness limit as a
function of redshift.

the 𝐿 X values of X-ray detected sources in the 4XMM sample when sample with replacement) 1000 times. For each bootstrapped sample,
calculating the AGN fraction.4 the AGN fraction is calculated, and the 16th and 84th percentiles of
When deriving the AGN fraction, we have also taken into account the obtained AGN fraction distribution give the estimation of the 1𝜎
the varying sensitivity of X-ray observations that provide X-ray cov- uncertainty. When no AGN is detected in a sample, we report the 1𝜎
erage to our samples. For the 4XMM sample, we derive the sensitivity confidence upper limits derived following Cameron (2011).
upper limit of the relevant RapidXMM HealPix from the background We also estimate the long-term average BH growth from BHAR of
level reported at the position of each MPA-JHU galaxy. Following a given sample of galaxies sharing similar properties, following the
Equations 3 and 4 of Chen et al. (2018), we derive the minimum num- method described in Ni et al. (2019, 2021b) that includes contribu-
ber of counts required for a source to be detected in the 0.2–12 keV tions from both X-ray detected sources and X-ray undetected sources.
band given the background level, and derive the corresponding flux We apply the obscuration correction factor mentioned earlier (1.1 for
sensitivity with the corresponding energy conversion factor (ECF) the 4XMM sample and 1.2 for the COSMOS sample) when taking
which are derived assuming a power-law spectrum with Γ = 1.7.5 into account the X-ray emission from X-ray detected sources.6 The
Since the 4XMM catalog uses DET_ML = 6 in the 0.2–12 keV band X-ray emission of a group of X-ray undetected sources is taken into
as the source detection criterion, we set the probability of the detected account via X-ray stacking techniques in the 0.2–12 keV band for the
source being a random Poisson fluctuation due to the background as 4XMM sample and the 0.5–7 keV band for the COSMOS sample.
2.5 × 10 −3 when utilizing the equations. For the COSMOS sample, With the source counts rate and background counts rate reported in
we derive the sensitivity map following the method in Aird et al. RapidXMM, the net 0.2–12 keV count rate at each galaxy position
(2017). in the 4XMM sample can be obtained, which is then converted to
Combining with the redshift information, the lower limit of the 𝐿 X the 0.2–12 keV flux with the corresponding energy conversion factor
of a source in order to be detected, 𝐿 X,limit , can be obtained for every derived assuming a power-law spectrum with Γ = 1.7. For the COS-
galaxy in both the 4XMM sample and the COSMOS sample. We MOS sample, the stacking process is described in Ni et al. (2021b),
note that a power-law model with Galactic absorption and Γ = 1.7 is which gives the 0.5–7 keV net count rate/flux at each galaxy position.
assumed through the whole conversion process. In the right panels We derive the average X-ray luminosity 𝐿 X,stack from the average
of Figure 4, we present the 𝐿 X,limit /𝑀★ distribution of galaxies in flux and the average redshift of the stacked sample.
the 4XMM and COSMOS samples. We only derive AGN fraction With 𝐿 X for individual X-ray detected sources and 𝐿 X,stack for X-
utilizing galaxies with log 𝐿 X,limit /𝑀★ ⩽ 32, i.e. these where we ray undetected sources, we can obtain sample-averaged AGN bolo-
have the sensitivities to detect an AGN with log 𝐿 X /𝑀★ > 32, if it metric luminosity following Equation 3 of Ni et al. (2021b) assuming
exists in the given galaxy: the 𝐿 X -dependent bolometric correction from Hopkins et al. (2007):
𝑁det,log 𝐿X /𝑀⋆ >32
𝑓AGN,log 𝐿X /𝑀⋆ >32 = . (1) 𝑁 
𝑁galaxy,log 𝐿X,limit /𝑀⋆ ⩽32 Ídet
(𝐿 X − 𝐿 X,XRB )𝑘 bol + (𝐿 X,stack − 𝐿 X,XRB )𝑁non 𝑘 bol
The uncertainty of the AGN fraction is obtained via bootstrapping the 𝑛=0
𝐿 bol =
sample (i.e. randomly drawing the same number of objects from the 𝑁det + 𝑁non
(2)
4 We note that the change in AGN fraction associated with applying the We also subtract the contributions from X-ray binaries (XRBs) from
obscuration correction factor is generally much smaller than the statistical
uncertainty of AGN fraction. Thus, X-ray absorption should not bias our
results materially. 6 We note that the change in BHAR associated with applying the obscuration
5 https://www.cosmos.esa.int/web/xmm-newton/ correction factor is generally smaller than the statistical uncertainty of BHAR.
epic-upper-limits Thus, X-ray absorption should not bias our results materially.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 5

4XMM sample
2 2
log SFR

log SFR
0 0

2 2
9 10 11 12 1.0 1.5 2.0
log M Dn4000

2.0 2

Density
Dn4000

1.5
1
1.0
9 10 11 12 0 1.0 1.5 2.0
log M Dn4000
Figure 2. Galaxies in the 4XMM sample in the SFR vs. 𝑀★ plane (upper-left), SFR vs. Dn 4000 plane (upper-right), and Dn 4000 vs. 𝑀★ plane (bottom-left).
The contours encircle 68 per cent, 80 per cent, 90 per cent, and 95 per cent of galaxies. Log 𝐿X /𝑀★ > 32 AGNs are represented by the orange stars. In the
bottom-right panel, the Dn 4000 distribution of galaxies is represented by the blue histogram; the Dn 4000 distribution of log 𝐿X /𝑀★ > 32 AGNs is represented
by the orange histogram.

𝐿 X and 𝐿 X,stack before applying the bolometric correction. The con- Dn 4000 (see Figure 5c), which shows an overall similar trend as
tributions from XRBs are generally small compared to the overall BHAR.
X-ray luminosity. Then, sample-averaged AGN bolometric luminos- Similarly, we bin objects in the COSMOS sample into 4 bins
ity can be converted to BHAR adopting a constant radiative efficiency according to their Dn 4000 values, with equal number of objects per
of 0.1 following Equation 4 in Ni et al. (2021b). The uncertainty of bin. For the COSMOS sample, when we calculate AGN fraction
BHAR is obtained via bootstrapping the sample 1000 times. For each as well as BHAR for a given subsample in this study, we weight
bootstrapped sample, BHAR is calculated, and the 16th and 84th per- the contribution from each object by the SCOR parameter in the
centiles of the obtained BHAR distribution give the estimation of the LEGA-C catalog, which accounts for the selection effects from the
1𝜎 uncertainty associated with BHAR of the sample. parent sample; we also verified that the analysis results do not vary
materially when we do not weight each object by SCOR. Figure 5d
shows that, at the Dn 4000 range we probe (Dn 4000 ≈ 1.25–1.9),
3 ANALYSIS RESULTS AGN fraction decreases with Dn 4000 in general. In terms of BHAR
as well as BHAR/𝑀★ (see Figure 5e and Figure 5f), similarly, a
3.1 AGN fraction and BHAR as a function of Dn 4000 decreasing trend in general is observed with increasing Dn 4000. We
For objects in the 4XMM sample, we bin them into 6 bins with equal note that, at Dn 4000 ≲ 1.5, this decreasing trend of AGN fraction,
number of objects per bin according to their Dn 4000 values. Figure 5a BHAR, and BHAR/𝑀★ is not very significant – the large error bars
shows that the AGN fraction presents a clear decreasing trend with as a result of the limited sample size prohibits us from drawing any
Dn 4000 at Dn 4000 ≲ 1.85, and slightly increases at Dn 4000 ≳ significant conclusion.
1.85. We note that this result will not be materially affected by We further investigate AGN fraction and BHAR in different 𝑀★
the 𝐿 X /𝑀★ threshold we adopt (see Appendix B for details). We ranges for both the 4XMM sample and the COSMOS sample; the
further plot BHAR as a function of Dn 4000 for the 4XMM sample results can be seen in Figure 6. We can see that for different 𝑀★
(see Figure 5b). We can see that BHAR also decreases as Dn 4000 ranges in the 4XMM sample, AGN fraction, BHAR, and BHAR/𝑀★
increases from ∼ 1.25 to ∼ 1.85. At Dn 4000 ≳ 1.85, the trend also decrease with Dn 4000 at Dn 4000 ≲ 1.9. The increasing trend
appears to reverse, and BHAR significantly increases with increasing of BHAR at larger Dn 4000 is more prominent among objects with
Dn 4000. To account for differences in the average 𝑀★ for galaxies 11 < log 𝑀★ ⩽ 12. For different 𝑀★ ranges in the COSMOS sample,
across our Dn 4000 bins, we also plot BHAR/𝑀★ as a function of we also observe decreasing trends of AGN fraction, BHAR, and

MNRAS 000, 1–21 (2015)

6

COSMOS sample

log SFR 2 2

log SFR
0 0

2 2
10.0 10.5 11.0 11.5 1.0 1.5 2.0
log M Dn4000

2.0 3

Density
Dn4000

1.5 2
1
1.0
10.0 10.5 11.0 11.5 0 1.0 1.5 2.0
log M Dn4000
Figure 3. Galaxies in the COSMOS sample in the SFR vs. 𝑀★ plane (upper-left), SFR vs. Dn 4000 plane (upper-right), and Dn 4000 vs. 𝑀★ plane (bottom-left).
The contours encircle 68 per cent, 80 per cent, 90 per cent, and 95 per cent of galaxies. Log 𝐿X /𝑀★ > 32 AGNs are represented by the orange stars. In the
bottom-right panel, the Dn 4000 distribution of galaxies is represented by the green histogram; the Dn 4000 distribution of log 𝐿X /𝑀★ > 32 AGNs is represented
by the orange histogram.

BHAR/𝑀★ with Dn 4000, though we note that for Dn 4000 ≲ 1.5 right. We then select the nearest neighbour of objects in the reference
objects with 10 < log 𝑀★ ⩽ 11, the decreasing trend is not very bin among objects in its left/right bin in the 𝑀★, SFR, and 𝑧 space
significant, which might be caused by the limited sample size. utilizing the NearestNeighbors algorithm in the scikit-learn
python package, to constitute two comparison samples with similar
𝑀★, SFR, and 𝑧 properties, but one with smaller Dn 4000 and one
3.2 AGN fraction and BHAR as a function of Dn 4000 when with larger Dn 4000. In Figure 7, we show AGN fractions of all these
controlling for other host-galaxy parameters subsamples, with each set of subsamples sharing similar 𝑀★, SFR,
and 𝑧 values represented by different colors and symbols. Compar-
We note that, while in Section 3.1 we characterized the incidence of
ing within each subsample set reveals how AGN fraction varies with
X-ray AGNs among galaxies with different stellar ages (as Dn 4000
Dn 4000 when controlling for other host-galaxy parameters, and a
is closely associated with the age of the stellar populations), we can
significant decreasing trend is observed within each subsample set
still not quantify how the difference in Dn 4000 (which indicates the
at Dn 4000 ≲ 1.9.8 We also demonstrate that our results hold when
difference in the mean stellar population age) directly affects AGN
defining the AGN fraction by an 𝐿 X limit of 1042 erg s −1 in the Ap-
activity and BH growth, as host-galaxy properties such as 𝑀★ and
pendix C. We show BHAR and BHAR/𝑀★ of all these subsamples
SFR vary across different Dn 4000 bins, which are known to closely
as well in Figure 7, with each set of subsamples sharing similar 𝑀★,
related with AGN activity and BH growth.
SFR, and 𝑧 values represented by different colors and symbols. Com-
We thus would like to study AGN fraction and BHAR as a func-
paring within each subsample set shows that BHAR and BHAR/𝑀★
tion of Dn 4000 when controlling for other host-galaxy parameters.
also decrease with Dn 4000 at Dn 4000 ≲ 1.9 when controlling for
To achieve this, for each bin (except the first and last bins) of galaxies
in the 4XMM sample in Figure 5, we sort this subsample with their
Dn 4000 values, and keep the central 1/3 of the objects to form a
reference bin.7 The first 1/3 of the objects are merged with the bin on are necessary to ensure sufficient sources with similar 𝑀★ , SFR and 𝑧 values
the left, and the last 1/3 of the objects are merged with the bin on the in the adjacent bins that are used to create the comparison samples.
8 We note that while AGN fraction varies with D 4000, it also varies when
n
other host-galaxy properties change, so that at a fixed Dn 4000, the AGN
7 While choosing this relatively small subsample size reduces the statistical fraction from different subsamples differs due to the differences in 𝑀★ and
power compared to the larger bin size used in Section 3.1, these narrower bins SFR.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 7

4XMM AGN 4XMM AGN/galaxies

20000
250

Number of sources
Number of sources

200 15000
150
10000
100
5000
50
0 029 30 31 32 33 34
30 32 34
log(LX/M ) log(LX, limit/M )
COSMOS AGN COSMOS AGN/galaxies
1200
20 1000
Number of sources
Number of sources

15 800
10 600
400
5 200
0 30 029 30 31 32 33 34
31 32 33 34
log(LX/M ) log(LX, limit/M )
Figure 4. Left panels: The observed distribution of log 𝐿X /𝑀★ for X-ray detected sources in the 4XMM (top left) and COSMOS (bottom left) samples. Right
panels: The cumulative distribution of the detection limit, log 𝐿X,limit /𝑀★ above which an AGN could be detected for all galaxies in the 4XMM (top right)
and COSMOS (bottom right) samples. The vertical lines represent the 𝐿X /𝑀★ limit adopted in our work when calculating AGN fraction. While a significant
number of AGNs detected have log 𝐿X /𝑀★ < 32, we note that a large fraction of galaxies in both samples have log 𝐿X,limit /𝑀★ ⩽ 32 to enable an unbiased
characterization of AGN fraction.

𝑀★, SFR, and 𝑧.9 At Dn 4000 ≳ 1.9, BHAR and BHAR/𝑀★ increase at Dn 4000 ≳ 1.85, the AGN fraction does not increase significantly,
with Dn 4000 when controlling for 𝑀★, SFR, and 𝑧. which might be caused by our definition for AGNs, as we only look at
Similarly, we study how AGN fraction and BH growth vary with log 𝐿 X /𝑀★ > 32 AGNs in Section 3.1 and do not take low-accretion-
Dn 4000 within subsample sets when controlling for other host-galaxy rate AGNs into account when calculating the AGN fraction. This
parameters for the COSMOS sample. The results are shown in Fig- inspired us to examine the incidence of low-accretion-rate AGN as a
ure 8. We also observe a trend of decreasing AGN fraction, BHAR function of Dn 4000. As stated in Section 2.4, we only consider an X-
and BHAR/𝑀★ with Dn 4000, similar to that seen at lower redshifts ray detected source as AGN when 𝐿 X is greater than the contribution
in our 4XMM sample, although at Dn 4000 ≲ 1.5 this trend has a from XRBs. As can be seen in Figure 4, we have a considerable num-
relatively low significance level. ber of AGNs with log 𝐿 X /𝑀★ ⩽ 32 detected in the 4XMM sample,
We also show that host morphological properties do not affect the and there are also a large number of galaxies in the 4XMM sample
results in this subsection materially in Appendix D. with log 𝐿 X,limit /𝑀★ ⩽ 32. We thus use these objects to study how
the fraction of AGNs with 31 < log 𝐿 X /𝑀★ < 32 varies as a function
of Dn 4000, and the result is shown in Figure 9. We can see that the
3.3 The incidence of low-accretion-rate AGN as a function of fraction increases at Dn 4000 ≳ 1.85 obviously.
Dn 4000 compared with log 𝐿 X /𝑀★ > 32 AGN
In Section 3.1, we find that for objects in the 4XMM sample, We further probe if this increase of low-accretion-rate AGN frac-
while BHAR and BHAR/𝑀★ significantly increase with Dn 4000 tion with Dn 4000 is linked with 𝑀★, by studying the low-accretion-
rate AGN fraction as a function of Dn 4000 in different 𝑀★ bins. As
9 While X-ray emission from normal star-forming galaxies has potential we can see in Figure 10, for galaxies/AGN with log 𝑀★ < 11.5, AGNs
dependence on stellar ages (e.g. Gilbertson et al. 2022), the XRB contribution tend to live among younger galaxies. In contrast, for galaxies/AGNs
is little among the objects we investigated (∼ 5–15%), so should not bias our with log 𝑀★ > 11.5, the fraction of AGNs with 31 < log 𝐿 X /𝑀★ <
results. 32 increases with Dn 4000. When we perform the same analyses for

MNRAS 000, 1–21 (2015)

8

4XMM sample 7 COSMOS sample

1.6 logM :
logSFR:
10.5
0.3
10.8
-0.3
10.8 10.9 10.9 11.1
-1.0 -1.2 -1.2 -1.2
logM :
logSFR:
10.9
1.3
11.0
0.5
11.1
-0.6
11.2
-1.3
1.4 6
1.2 5
AGN fraction (%)

AGN fraction (%)

1.0 4
0.8 3
0.6
2
0.4
0.2 1
0.0 3227 3321 3295 3271 3245 3165 0 279 290 276 279
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
(a) (d)

1.0 logM :
logSFR:
10.5
0.3
10.8
-0.3
10.8 10.9 10.9 11.1
-1.0 -1.2 -1.2 -1.2 1.0 logM :
logSFR:
10.9
1.3
11.0
0.5
11.1
-0.6
11.2
-1.3

1.5 1.5
log BHAR (M /yr)

2.0 log BHAR (M /yr) 2.0

2.5 2.5
3.0 3.0
3.5 3.5
4.0 4.0
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
(b) (e)

3.0
4.0
3.5
log BHAR/M (Gyr 1)

log BHAR/M (Gyr 1)

4.5 4.0
4.5
5.0 5.0
5.5 5.5
6.0
6.0 6.5
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
(c) (f)
Figure 5. (a): AGN fraction as a function of Dn 4000 among galaxies in the 4XMM sample. The horizontal position of each data point represents the median
Dn 4000 of the sources in each bin, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars
represent the 1𝜎 confidence interval of AGN fraction from bootstrapping. We also list the median log 𝑀★ and log SFR of each bin on the top of the plot. The
numbers in the bottom of the plot represent the number of galaxies in each bin used to derive the AGN fraction (with log 𝐿X,limit /𝑀★ ⩽ 32). (b): BHAR as
a function of Dn 4000 among galaxies in the 4XMM sample. The horizontal position of each data point represents the median Dn 4000 of the sources in each
bin, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars represent the 1𝜎 confidence
interval of BHAR from bootstrapping. We also list the median log 𝑀★ and log SFR of each bin on the top of the plot. (c): BHAR/𝑀★ a function of Dn 4000
among galaxies in the 4XMM sample. The horizontal position of each data point represents the median Dn 4000 of the sources in each bin, with 𝑥-axis error
bars demonstrating the 16th and 84th percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars represent the 1𝜎 confidence interval of BHAR/𝑀★
from bootstrapping. (d): Similar to panel (a), but for AGN fraction as a function of Dn 4000 among galaxies in the COSMOS sample. (e): Similar to panel (b),
but for BHAR as a function of Dn 4000 among galaxies in the COSMOS sample. (f): Similar to panel (c), but for BHAR/𝑀★ as a function of Dn 4000 among
galaxies in the COSMOS sample.
MNRAS 000, 1–21 (2015)
 AGNs in the lifecycle of galaxies 9

3.0 4XMM sample COSMOS sample

10 < log M 11 10 < log M 11
11 < log M 12 11 < log M 12
2.5 8
2.0
AGN fraction (%)

AGN fraction (%)

6
1.5
4
1.0
0.5 2
0.0 0
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
(a) (d)

1.0 1.0
1.5 1.5
2.0
log BHAR (M /yr)

2.0 log BHAR (M /yr)

2.5 2.5
3.0 3.0
3.5 3.5
4.0 4.0
4.5 4.5
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
(b) (e)

3 3
log BHAR/M (Gyr 1)

log BHAR/M (Gyr 1)

4 4

5 5

6 6

1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
(c) (f)
Figure 6. (a): AGN fraction as a function of Dn 4000 among galaxies in two 𝑀★ bins (represented by different colors and symbols) in the 4XMM sample.
The horizontal position of each data point represents the median Dn 4000 of the sources in each bin, with 𝑥-axis error bars demonstrating the 16th and 84th
percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars represent the 1𝜎 confidence interval of AGN fraction from bootstrapping. (b): BHAR as
a function of Dn 4000 among galaxies in the 4XMM sample. The horizontal position of each data point represents the median Dn 4000 of the sources in each
bin, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars represent the 1𝜎 confidence
interval of BHAR from bootstrapping. (c): BHAR/𝑀★ as a function of Dn 4000 among galaxies in the 4XMM sample. The horizontal position of each data
point represents the median Dn 4000 of the sources in each bin, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values in each
bin. The 𝑦-axis error bars represent the 1𝜎 confidence interval of BHAR/𝑀★ from bootstrapping. (d): Similar to panel (a), but for AGN fraction as a function
of Dn 4000 among galaxies in two 𝑀★ bins in the COSMOS sample. (e): Similar to panel (b), but for BHAR as a function of Dn 4000 among galaxies in two 𝑀★
bins in the COSMOS sample. (f): Similar to panel (c), but for BHAR/𝑀★ as a function of Dn 4000 among galaxies in two 𝑀★ bins in the COSMOS sample.

MNRAS 000, 1–21 (2015)

10

4XMM sample COSMOS sample

logM : 12 logM :
2.0 logSFR: logSFR:
10
1.5
AGN fraction (%)

AGN fraction (%)

8
1.0 6
4
0.5
2
0.0 0
1.4 1.6 1.8 2.0 1.3 1.4 1.5 1.6 1.7 1.8
Dn4000 Dn4000
logM : 0.5 logM :
2.0 logSFR: logSFR:
1.0
2.5
log BHAR (M /yr)

log BHAR (M /yr)

1.5
3.0 2.0
3.5 2.5
3.0
4.0
3.5
1.4 1.6 1.8 2.0 1.3 1.4 1.5 1.6 1.7 1.8
Dn4000 Dn4000
4.50 2.5 logM :
logM : logSFR:
logSFR:
4.75 3.0
log BHAR/M (Gyr 1)

log BHAR/M (Gyr 1)

5.00 3.5
5.25
4.0
5.50
5.75 4.5
6.00 5.0
6.25 5.5
6.50 1.4 1.6 1.8 2.0 1.3 1.4 1.5 1.6 1.7 1.8
Dn4000 Dn4000
Figure 7. Top: AGN fraction as a function of Dn 4000 among galaxies in the Figure 8. Top: Similar to the top panel of Figure 7, but for AGN fraction as
4XMM sample when controlling for 𝑀★ , SFR, and 𝑧. Different symbols and a function of Dn 4000 when controlling for 𝑀★ , SFR, and 𝑧 among galaxies
colors represent a set of subsamples with similar 𝑀★ , SFR, and 𝑧 values (as in the COSMOS sample. Middle: Similar to the middle panel of Figure 7,
listed on top of the panel with the same color). The horizontal position of but for BHAR as a function of Dn 4000 when controlling for 𝑀★ , SFR, and 𝑧
each data point represents the median Dn 4000 of the sources in each sample, among galaxies in the COSMOS sample. Bottom: Similar to the middle panel
with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the of Figure 7, but for BHAR/𝑀★ as a function of Dn 4000 when controlling for
Dn 4000 values. The 𝑦-axis error bars represent the 1𝜎 confidence interval 𝑀★ , SFR, and 𝑧 among galaxies in the COSMOS sample.
of AGN fraction from bootstrapping. Middle: Similar to the top panel, but for
BHAR as a function of Dn 4000 among galaxies in the 4XMM sample when
controlling for 𝑀★ , SFR, and 𝑧. Bottom: Similar to the top panel, but for
BHAR/𝑀★ as a function of Dn 4000 among galaxies in the 4XMM sample
when controlling for 𝑀★ , SFR, and 𝑧.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 11
6 0.30
31 < log LX/M < 32 AGN fraction (%) logM : 10.5
logSFR: 0.3
10.8
-0.3
10.8 10.9 10.9 11.1
-1.0 -1.2 -1.2 -1.2
logM :
logSFR:
5 0.25

R90 MIR AGN fraction (%)

4 0.20
3 0.15
2 0.10
1 0.05
0 1.2 1.4 1.6 1.8 2.0 0.00 1.4 1.6 1.8 2.0
Dn4000 Dn4000
Figure 9. 31 < log 𝐿X /𝑀★ < 32 AGN fraction as a function of Dn 4000 Figure 11. MIR AGN fraction as a function of Dn 4000 among SDSS galaxies
among galaxies in the 4XMM sample. The horizontal position of each data when controlling for 𝑀★ , SFR, and 𝑧. Different colors represent a set of
point represents the median Dn 4000 of the sources in each bin, with 𝑥-axis subsamples with similar 𝑀★ , SFR, and 𝑧 values (as listed on top of the panel
error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values with the same color). The horizontal position of each data point represents
in each bin. The 𝑦-axis error bars represent the 1𝜎 confidence interval of the median Dn 4000 of the sources in each sample, with 𝑥-axis error bars
AGN fraction from bootstrapping. We also list the median log 𝑀★ and log demonstrating the 16th and 84th percentiles of the Dn 4000 values. The 𝑦-
SFR value of each bin on the top of the plot. axis error bars represent the 1𝜎 confidence interval of AGN fraction from
bootstrapping.

16
31 < log LX/M < 32 AGN fraction (%)

10 < logM < 11

14 11 < logM < 11.5 analyses above. Extending this more robust approach to MIR- and
11.5 < logM < 12 radio-selected samples is deferred to a future work.
12
10 3.4.1 Comparing with the incidence of AGNs selected at MIR
8 We obtain a MIR-selected AGN sample among MPA-JHU galaxies
6 in the main SDSS galaxy sample from the R90 catalog of Assef et al.
(2018), which consists of AGN candidates with 90% reliability. We
4 perform the same analyses as those in Section 3.2 to check how the
2 fraction of MIR-selected AGNs varies with Dn 4000 when control-
ling for host-galaxy properties, and whether the observed trend is
0 1.2 1.4 1.6 1.8 2.0 consistent with what we observed in the X-ray. The analysis results
Dn4000 are shown in Figure 11. We observe a similar trend as these in Sec-
tion 3.2: when controlling for host-galaxy properties, the fraction
Figure 10. 31 < log 𝐿X /𝑀★ < 32 AGN fraction as a function of Dn 4000 of MIR-selected AGNs decreases with Dn 4000 at Dn 4000 ≲ 1.9.
among galaxies with different 𝑀★ ranges in the 4XMM sample; different sym- As MIR-selected AGNs are generally biased against AGNs in most
bols and colors represent subsamples of galaxies with different 𝑀★ ranges. massive galaxies, the MIR-selected AGN fraction at high Dn 4000 in
The horizontal position of each data point represents the median Dn 4000 of Figure 11 is too small to observe any trend.
the sources in each bin, with 𝑥-axis error bars demonstrating the 16th and
84th percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars
represent the 1𝜎 confidence interval of AGN fraction from bootstrapping. 3.4.2 Comparing with the incidence of AGNs selected at radio
wavelength
To study how the fraction of radio-selected AGNs varies with
log 𝐿 X /𝑀★ > 32 AGNs, we found that the fractions of log 𝐿 X /𝑀★ > Dn 4000, we utilize the radio AGN sample from Best & Heckman
32 AGNs at different 𝑀★ ranges all drop with Dn 4000. (2012), constructed by combining SDSS data with the NRAO (Na-
tional Radio Astronomy Observatory) VLA (Very Large Array) Sky
Survey (NVSS) and the Faint Images of the Radio Sky at Twenty cen-
3.4 Comparing with the incidence of AGNs selected at different
timetres (FIRST) survey. The classification of radio AGNs into high-
wavelength bands
excitation radio galaxies (HERGs) and low-excitation radio galaxies
We note that our results in previous subsections are based on X- (LERGs) is available for this sample. This radio AGN sample mainly
ray-selected AGNs. While X-ray selection is known to be able to consists of LERGs (that have small Eddington ratios; < 1%), with a
provide the most complete and unbiased sample of AGNs, we would small fraction of HERGs reported. We match these radio AGNs to
like to test how different AGN selection methods could potentially MPA-JHU galaxies in the main SDSS galaxy sample, and perform
affect our results. In this subsection, we select AGNs based purely on the same analyses as those in Section 3.2 to check how the fraction
detection in a waveband in a given catalog, in contrast to the careful of radio-selected AGNs in general, HERGs, and LERGs, vary with
measurement of AGN fraction to specific black hole accretion rate Dn 4000 when controlling for host-galaxy properties. The analysis
limits (corrected for any incompleteness) adopted in our X-ray-based results are shown in Figure 12.

MNRAS 000, 1–21 (2015)

12
Unlike X-ray-selected AGNs and MIR-selected AGNs, radio- roughly traces the stellar mass loss per solar mass predicted by the
selected AGNs in this sample (dominated by the LERG population) FSPS model divided by a factor of ∼ 5000 among almost all galaxies
are more likely to be found among old galaxies with large Dn 4000 in the 4XMM sample (except for the Dn 4000 ≈ 2 bin), indicating that
values. This has been known for decades, as the radio AGN fraction stellar mass loss might be the fuelling source for these galaxies: not
is strongly linked with 𝑀★ (e.g. Best et al. 2005). As can be seen in only for the quiescent galaxies, but also for the star-forming galaxies.
the right panel of Figure 12, HERG fraction always decreases with For the whole 4XMM sample, the log BHAR/SFR value is ≈ −2.5,
Dn 4000 when controlling for host-galaxy properties at Dn 4000 ≲ and the scaling factor of 5000 is consistent with this value assuming
1.9, similar to what we found for the X-ray-selected AGN fraction. that stellar mass loss also serves as the major fuel for star formation
In the middle panel of Figure 12, we can see that at Dn 4000 ≲ 1.7, and the fraction of gas that turns into stars is ∼ 0.03 − 0.4 (e.g. Ciotti
the LERG fraction also decreases with Dn 4000 when controlling & Ostriker 2007). As the galaxies in our 4XMM sample and COS-
for 𝑀★, SFR, and 𝑧. In contrast, at high Dn 4000 values (Dn 4000 ≳ MOS sample have similar median/mean 𝑀★ and velocity dispersion,
1.7), the LERG fraction increases with Dn 4000 when controlling for we assume that the fraction of stellar mass loss that can be captured
other parameters, similar to what we found among low-accretion-rate by the central BH is similar for these two samples. If we compare the
X-ray-selected AGNs among massive (log 𝑀★ ≳ 11.5) galaxies. BHAR per solar mass in the COSMOS sample with the stellar mass
loss rate scaled by a factor of 1/5000, we could see that BHAR/𝑀★
only traces stellar mass loss rate well among the oldest galaxies in the
COSMOS sample. Among most galaxies in the COSMOS sample,
4 DISCUSSIONS
it is likely that the cold gas in the galaxy with origins other than the
In Section 3.1, we characterize how the log 𝐿 X /𝑀★ > 32 AGN stellar mass loss is capable of serving as the major fuel for both the
fraction, BHAR, and BHAR/𝑀★ vary with Dn 4000 at two different BH and the star formation process.
redshift ranges; in Section 3.2, we found that when controlling for The decreasing trend of X-ray AGN fraction and BHAR with
host-galaxy properties (𝑀★, SFR, and 𝑧), the fraction of log 𝐿 X /𝑀★ Dn 4000 when controlling for other host-galaxy properties among
> 32 AGNs and BHAR decrease with Dn 4000 among galaxies with most galaxies in the 4XMM sample and COSMOS sample observed
Dn 4000 ≲ 1.9. We discuss the potential reason for these findings in in Section 3.2 further supports this scenario. A similar trend is also
Section 4.1. In Section 3.3, we found that among the most massive observed in the case of MIR-selected AGN fraction and HERG frac-
galaxies at low redshift, the fraction of 31 < log 𝐿 X /𝑀★ < 32 AGNs tion at low redshift (see Figures 11 and 12). As stellar population
increases with Dn 4000, and we discuss the potential reason for this synthesis models predict that stellar mass loss rates decline as a
as well as the increase of BHAR/𝑀★ among the oldest galaxies at function of mean stellar age (e.g. see the background dots in Fig-
low redshift in Section 4.2. ure 13), we could see that the difference in X-ray AGN fraction (or
BHAR) when controlling for 𝑀★, SFR, and 𝑧 is smaller at larger
Dn 4000. From the bottom panel of Figure 7 (4XMM sample), we
4.1 Stellar mass loss as a potential fuel for BH growth observe a ∼ 1×10 −5 /5×10 −6 /2×10 −6 Gyr −1 drop in BHAR/𝑀★
associated with a ∼ 3×10 −2 /8×10 −3 /4×10 −3 Gyr −1 drop in stel-
We showed in Section 3.1 that galaxies in both the 4XMM sample
lar mass loss rate (inferred from the difference in Dn 4000) in the
and COSMOS sample display a decrease of AGN fraction, BHAR,
first/second/third Dn 4000 subsample set. In the fourth Dn 4000 sub-
or BHAR/𝑀★ with Dn 4000 at Dn 4000 ≲ 1.9, suggesting a higher
sample set, we observe an increase in BHAR when Dn 4000 in-
level of AGN activity and BH growth among younger galaxies. It is
creases. From the bottom panel of Figure 8 (COSMOS sample), we
plausible that among younger galaxies, higher amounts of fuels are
observe a ∼ 5×10 −4 /3×10 −5 Gyr −1 drop in BHAR/𝑀★ associated
available for the central BHs. Kauffmann & Heckman (2009) argue
with a ∼ 9×10 −2 /2×10 −2 Gyr −1 drop in stellar mass loss rate in the
that stellar mass loss may serve as an important source of fuel for
first/second Dn 4000 subsample set. Generally, higher levels of BH
BHs when the cold gas supply is not plentiful. As younger galaxies
growth are associated with more stellar mass loss. At the same time,
have higher stellar mass loss rates, a higher level of AGN activity and
the relation between the difference in BHAR/𝑀★ associated with
BH growth among younger galaxies is expected if recycled gas from
the difference in stellar mass loss rate is not exactly linear, i.e. the
stellar mass loss serves as an important fuelling source, consistent
predicted change in the stellar mass loss rate based on the change in
with our finding. We further plot BHAR/𝑀★ among star-forming
Dn 4000 does not result in a consistent change in BHAR/𝑀★ across
galaxies and quiescent galaxies in both the 4XMM sample and the
all the subsamples used in Figures 7 and 8. While stellar mass loss
COSMOS sample as a function of Dn 4000 separately in Figure 13,
could be one potential fuelling mechanism, it is not always the dom-
and compare with stellar mass loss rate expected at a given Dn 4000.10
inant fuelling mechanism. Also, the fraction of stellar mass loss that
We use python-FSPS (Conroy et al. 2009; Conroy & Gunn 2010)
could be accreted by the central BH is likely to vary among galax-
to predict how stellar mass loss rates vary as a function of Dn 4000.
ies, depending on other galaxy properties such as their total stellar
We adopt the universal initial mass function as parametrized by
masses (which is clearly indicated in Figure 6: more massive galaxies
Chabrier (2003). Star-forming histories are generated using a range
tend to accrete stellar mass loss more efficiently, which is expected
of formation times, with exponential decline time-scales or delayed-
as they have larger potential wells and larger black hole masses),
exponential time-scales ranging from 0.1 Gyr to 3 Gyr. In the back-
morphologies, and star formation histories.
ground of Figure 13, we show the predicted stellar mass loss rate (in
In Figure 14, we plot the stellar mass loss per stellar mass unit
units of the fraction of the stellar mass returned per Gyr) scaled by
per Gyr as a function of Dn 4000, as well as scaled BHAR/𝑀★ of
a factor of 1/5000 as a function of Dn 4000 for galaxies with differ-
galaxies in subsamples with different 𝑀★ ranges (the scaling factor
ent star-forming histories. We could see that BHAR per solar mass
is chosen so that the lowest BHAR/𝑀★ align with stellar mass loss
rate) for both the 4XMM sample and COSMOS sample.11 For the
10 The separation between star-forming galaxies and quiescent galaxies is
performed following the criterion in Section 2.4 of Ni et al. (2021b), which
utilizes the star formation main sequence. 11 We note that it is also plausible that for the bins we used for scaling

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 13
4.0 3.5
logM : logM : logM :
3.5 logSFR: 3.0 logSFR: 0.20 logSFR:
Radio AGN fraction (%)

3.0 2.5

HERG fraction (%)

LERG fraction (%)
2.5 0.15
2.0
2.0
1.5 0.10
1.5
1.0 1.0
0.05
0.5 0.5
0.0 1.4 1.5 1.6 1.7 1.8 1.9 2.0 0.0 1.4 1.5 1.6 1.7 1.8 1.9 2.0 0.00 1.4 1.5 1.6 1.7 1.8 1.9 2.0
Dn4000 Dn4000 Dn4000

Figure 12. Left panel: Radio AGN fraction as a function of Dn 4000 among SDSS galaxies when controlling for 𝑀★ , SFR, and 𝑧. Different colors represent a
set of subsamples with similar 𝑀★ , SFR, and 𝑧 values (as listed on top of the panel with the same color). The horizontal position of each data point represents
the median Dn 4000 of the sources in each sample, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values. The 𝑦-axis error
bars represent the 1𝜎 confidence interval of AGN fraction from bootstrapping. Middle panel: Similar to the left panel, but for LERGs. Right panel: Similar to
the left panel, but for HERGs.

4XMM sample COSMOS sample

2.5 2.5
Quiescent Quiescent
3.0 Star-forming 3.0 Star-forming
3.5 3.5
log BHAR/M (Gyr 1)

log BHAR/M (Gyr 1)

4.0 4.0
4.5 4.5
5.0
5.0
5.5
5.5
6.0
6.0 6.5
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
Figure 13. Left panel: BHAR/𝑀★ as a function of Dn 4000 among star-forming/quiescent galaxies in the 4XMM sample, represented by the blue/red symbols.
The horizontal position of each data point represents the median Dn 4000 of the sources in each bin, with 𝑥-axis error bars demonstrating the 16th and 84th
percentiles of the Dn 4000 values in each bin. The 𝑦-axis error bars represent the 1𝜎 confidence interval of BHAR/𝑀★ from bootstrapping. In the background,
stellar mass loss rate per stellar mass unit per Gyr scaled by a factor of 1/5000 is shown as grey dots. Right panel: Similar to the left panel, but for BHAR/𝑀★
as a function of Dn 4000 among star-forming/quiescent galaxies in the COSMOS sample.

4XMM sample, BHAR/𝑀★ does not track stellar mass loss well of the accumulated hot recycled gas is not a process that happens
among log 𝑀★ > 11 galaxies. This might be due to the fact that in a less than 1 Gyr timescale, it could explain the high BHAR/𝑀★
the accretion efficiency from recycled gas is closely related to other among the relatively young massive galaxies in the 4XMM sample.
factors in addition to 𝑀★, or there are other gas sources for the most If this is the case, it is also plausible that recycled gas contributes
massive galaxies at low redshift. It is also plausible that the onset significantly to the fuel of BHs among star-forming galaxies in the
of the cooling of recycled gas takes several Gyrs to happen among COSMOS sample (see the right panel of Figure 14).
these most massive galaxies due to, e.g., AGN feedback (more mas- We note that it is also plausible that metallicity plays a role here,
sive galaxies tend to host more luminous AGNs that drive stronger as the higher the Dn 4000, the higher the metallicity (e.g. Gallazzi
outflows and exhibit more powerful jets), so that BHAR/𝑀★ does et al. 2005), and it has been argued that BH growth might be more
not track instantaneous stellar mass loss rate at the scale of ≲ 100 efficient in the low-metallicity regime (e.g. Toyouchi et al. 2019).
Myrs, but stellar mass loss accumulated in the past several Gyrs. In Disentangling the effects of age and metallicity would need a large
the background of Figure 14, we also plot the average stellar mass and complete high-signal-to-noise spectroscopic sample from future
loss rate of stellar populations with different star formation histories surveys (e.g. DESI Collaboration et al. 2016; de Jong et al. 2019).
over the past 1 Gyr, 2 Gyrs, and 3 Gyrs, represented by the yellow, Since Dn 4000 is primarily utilized as an age-sensitive parameter
orange, and red dots. We can see that if the cooling and accretion (e.g. Kauffmann et al. 2003; Gallazzi et al. 2005; Kauffmann &
Heckman 2009; Wu et al. 2018) and we do not observe similarly
significant trends when utilizing more metallicity-sensitive parame-
calibration with Dn 4000 ≈ 1.9, the BHAR/𝑀★ is below the prediction from ters (e.g. [MgFe]’, [Mg2 Fe]; see Gallazzi et al. 2005 and references
stellar mass loss rate due to AGN feedback from jets. therein), we interpret the link between AGN activity/BH growth with

MNRAS 000, 1–21 (2015)

14
Dn 4000 mainly a result of the variation of AGN activity/BH growth among these galaxies (e.g. Gaspari et al. 2015; McDonald et al.
among galaxies with different stellar population ages. 2021). Using hydrodynamic simulations, Gaspari et al. (2015) sug-
We also caution that we do not observe a solid decreasing AGN gest that among massive galaxies, chaotic cold accretion of con-
fraction or BHAR trend with Dn 4000 at Dn 4000 ≲ 1.5 for the COS- densed hot gas is less efficient when a rotating disk is present. Thus,
MOS sample. The lack of a clear trend might be due to the limited BHs in dispersion-dominated systems might accrete a larger fraction
sample size; it might be due to gas with origins other than stellar of gas supply than in rotation-dominated systems. As the fraction
mass loss serving as the dominating fuel which “washes out” the role of elliptical galaxies increases with Dn 4000, this might also explain
of stellar mass loss; it might also be attributed to a scenario where the increasing number of low-accretion-rate AGNs as well as the
the recycled gas from stellar mass loss can not easily cool down or increasing BHAR with Dn 4000 among old, massive galaxies in the
reach the BH in the galaxy center that does not significantly affect 4XMM sample.
our lower redshift (4XMM) sample. One plausible reason for this
scenario could be supernova (SN) feedback. SN can create a rarefied
and hot environment, and SN winds can expel recycled gas from
the galaxy. The core-collapse SN rate should directly trace the SFR
(and type Ia SN rate is small compared to the core-collapse SN; e.g.
Dekel et al. 2019). The average SFR among star-forming galaxies in 5 SUMMARY AND CONCLUSIONS
the COSMOS sample is much higher than that among the 4XMM
sample (consistent with the fact that galaxies in the COSMOS sample Utilizing spectroscopic samples of galaxies with X-ray data coverage,
have higher redshift), which is associated with stronger SN feedback. we studied the incidence of AGNs among galaxies with different
We also note that for 10 < log 𝑀★ ⩽ 11 galaxies in the 4XMM sam- mean stellar population ages in this work. The main points from this
ple, the decline of BHAR/𝑀★ with Dn 4000 among relatively young paper are the following:
galaxies is not as steep as that among 11 < log 𝑀★ ⩽ 12 galaxies,
which might also be attributed to SN winds that are more effective at (i) We built two samples of galaxies/AGNs with both spectro-
expulsion when the central surface mass density is low (e.g. Hopkins scopic coverage and X-ray coverage. One sample (4XMM sample)
et al. 2022). includes SDSS galaxies with XMM-Newton coverage at 𝑧 = 0–0.35;
our other sample (COSMOS sample) includes LEGA-C galaxies with
Chandra coverage at 𝑧 = 0.6–1.0. Dn 4000 measurements from spec-
4.2 Additional fuelling mechanism and/or enhanced accretion tra are adopted as a tracer of the mean stellar population age of the
capability among old, massive galaxies in the local universe galaxy. X-ray observations are utilized to estimate AGN fraction and
We have shown in Figure 13 that BHAR among the oldest galax- BHAR for samples of galaxies (see Section 2).
ies in the 4XMM sample also does not track the stellar mass loss (ii) In Section 3.1, we characterized how the AGN fraction, BHAR
rate, and appears to be higher than expected from the stellar mass as well as BHAR/𝑀★ vary with Dn 4000 among galaxies in the
loss fuelling. Among these most massive systems at Dn 4000 ∼ 2, 4XMM sample and COSMOS sample. In Section 4.1, we show that
additional fuelling mechanisms (such as the fuelling from hot gas BHAR/𝑀★ as a function of Dn 4000 roughly traces the scaled stellar
in the halo) may take place, leading to the high BHAR/𝑀★.12 The mass loss rate predicted by Dn 4000 among galaxies in the 4XMM
additional fuelling mechanism may also be particularly effective in sample (except for the oldest galaxies) as well as old/quiescent galax-
triggering low-accretion-rate AGNs among old, massive galaxies at ies in the COSMOS sample, indicating stellar mass loss as a poten-
low redshift, as can be seen in Figure 10. Hot halo gas fuelling has tially important (and possibly dominant) fuelling source.
long been introduced as a source for powering radio AGNs, partic- (iii) In Section 3.2, we found that when controlling for host-galaxy
ularly LERGs, which tend to have old and massive hosts (e.g. Best properties (𝑀★, SFR, and 𝑧), the fraction of log 𝐿 X /𝑀★ > 32 AGNs
& Heckman 2012). In Figure 12, we can see that the fraction of and BHAR decrease with Dn 4000 among galaxies in the 4XMM
LERG increases with Dn 4000. Also, the LERG population among sample (except for the oldest/most massive galaxies) and COSMOS
old galaxies tends to increase with Dn 4000 when controlling for 𝑀★, sample, suggesting higher numbers of AGNs and higher levels of BH
SFR, 𝑧 (in contrast, the HERG population and the LERG population growth among younger galaxies. We also observed similar trends in
among young galaxies tend to decrease with Dn 4000 when control- terms of the MIR-selected AGN fraction and the HERG fraction
ling for 𝑀★, SFR, 𝑧). It is plausible that the fuelling from the hot halo among SDSS galaxies in Section 3.4. These results further support
gas could explain both the increasing fraction of low-accretion-rate the scenario of stellar mass loss as a potential fuelling source for
X-ray AGN and the increasing fraction of LERG among old galaxies. AGN (see Section 4.1).
With this additional fuelling mechanism that is more effective among (iv) In Section 3.1, we observed a slight increase of BHAR and
old galaxies, we also do not observe any sign of a decreasing trend BHAR/𝑀★ among the oldest galaxies in the local universe; in Sec-
of BH growth associated with Dn 4000 among the oldest galaxies in tion 3.3, we found that among the most massive galaxies in the local
the 4XMM sample in Figure 7. universe, the fraction of low specific-accretion-rate AGNs (31 < log
It is also plausible that the relatively high BHAR among old galax- 𝐿 X /𝑀★ < 32) increases significantly with Dn 4000. The LERG frac-
ies at low redshift is linked with enhanced capability of BH accretion tion in the local universe also increases with Dn 4000 among the old,
massive galaxies (see Section 3.4). Additional fuelling from the hot
12
halo gas and potentially enhanced accretion capability among old,
We note that hot gas among giant ellipticals also shines in the X-ray (e.g. massive galaxies may explain these trends (see Section 4.2).
Boroson et al. 2011), but the contribution from the diffuse hot gas alone can
not account for the excess amount of BHAR observed in the largest Dn 4000
bin in the 4XMM sample. For ellipticals with log 𝑀★ ∼ 11, we expect diffuse Our work shows that stellar mass loss may be an important fuelling
hot emission with log 𝐿X ∼ 40. Converting it to BHAR, it is at the level of source to trigger AGN activity and thus drive ongoing BH growth,
log BHAR ∼ −5 and log BHAR/𝑀★ ∼ −7 Gyr −1 , far below the values we not only among old quiescent galaxies, but also among young star-
observed. forming galaxies at low redshift.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 15

4XMM sample COSMOS sample

1.0 10 < log M 11 1.0 10 < log M 11
11 < log M 12 11 < log M 12
0.5 0.5
0.0 0.0
log M /M (Gyr 1)

log M /M (Gyr 1)
0.5 0.5
1.0 1.0
1.5 1.5
2.0 2.0
2.5 2.5
1.2 1.4 1.6 1.8 2.0 1.2 1.4 1.6 1.8 2.0
Dn4000 Dn4000
Figure 14. Left panel: Scaled BHAR/𝑀★ as a function of Dn 4000 among galaxies in 4XMM sample in different 𝑀★ ranges (the scaling factor is selected to
align the Dn 4000 bin with the lowest BHAR/𝑀★ value with stellar mass loss rate), plotted against stellar mass loss rate per stellar mass unit per Gyr (grey
dots), as well as the average stellar mass loss rate over the past 1 Gyr (yellow dots), 2 Gyrs (orange dots), and 3 Gyrs (red dots). The horizontal position of each
data point represents the median Dn 4000 of the sources in each bin, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles of the Dn 4000 values in
each bin. The 𝑦-axis error bars represent the 1𝜎 confidence interval of BHAR/𝑀★ from bootstrapping. Right panel: Similar to the left panel, but for scaled
BHAR/𝑀★ as a function of Dn 4000 among galaxies in the COSMOS sample in different 𝑀★ ranges.

ACKNOWLEDGEMENTS Boquien M., Burgarella D., Roehlly Y., Buat V., Ciesla L., Corre D., Inoue
A. K., Salas H., 2019, A&A, 622, A103
We thank the anonymous referee for constructive feedback. We thank Boroson B., Kim D.-W., Fabbiano G., 2011, ApJ, 729, 12
Johan Comparat for the helpful discussion. QN and JA acknowl- Brandt W. N., Alexander D. M., 2015, A&ARv, 23, 1
edge support from a UKRI Future Leaders Fellowship (grant code: Brandt W. N., Yang G., 2021, arXiv e-prints, p. arXiv:2111.01156
MR/T020989/1). KLB acknowledges funding from a Horizon 2020 Brinchmann J., Charlot S., White S. D. M., Tremonti C., Kauffmann G.,
grant (XMM2Athena). For the purpose of open access, the authors Heckman T., Brinkmann J., 2004, MNRAS, 351, 1151
have applied a Creative Commons Attribution (CC BY) licence to any Cameron E., 2011, Publ. Astron. Soc. Australia, 28, 128
Author Accepted Manuscript version arising from this submission. Chabrier G., 2003, PASP, 115, 763
Chen C. T. J., et al., 2018, MNRAS, 478, 2132
Ciotti L., Ostriker J. P., 2007, ApJ, 665, 1038
Civano F., et al., 2016, ApJ, 819, 62
DATA AVAILABILITY Conroy C., Gunn J. E., 2010, ApJ, 712, 833
Conroy C., Gunn J. E., White M., 2009, ApJ, 699, 486
The data underlying this article were accessed from the XMM-Newton DESI Collaboration et al., 2016, arXiv e-prints, p. arXiv:1611.00036
Science Archive, SDSS data releases, Chandra data archive, and ESO Dekel A., Sarkar K. C., Jiang F., Bournaud F., Krumholz M. R., Ceverino D.,
data portal. The derived data generated in this research will be shared Primack J. R., 2019, MNRAS, 488, 4753
upon reasonable request to the corresponding author. Evans I. N., et al., 2010, ApJS, 189, 37
Faber S. M., et al., 2007, ApJ, 665, 265
Fragos T., Lehmer B. D., Naoz S., Zezas A., Basu-Zych A., 2013, ApJ, 776,
L31
REFERENCES
Gallazzi A., Charlot S., Brinchmann J., White S. D. M., Tremonti C. A., 2005,
Adelman-McCarthy J. K., et al. 2011, VizieR Online Data Catalog, p. II/306 MNRAS, 362, 41
Aird J., et al., 2010, MNRAS, 401, 2531 Gaspari M., Brighenti F., Temi P., 2015, A&A, 579, A62
Aird J., et al., 2012, ApJ, 746, 90 Georgantopoulos I., Pouliasis E., Mountrichas G., Van der Wel A., Marchesi
Aird J., Coil A. L., Georgakakis A., 2017, MNRAS, 465, 3390 S., Lanzuisi G., 2023, arXiv e-prints, p. arXiv:2302.00530
Aird J., Coil A. L., Georgakakis A., 2018, MNRAS, 474, 1225 Gilbertson W., et al., 2022, ApJ, 926, 28
Aird J., Coil A. L., Georgakakis A., 2019, MNRAS, 484, 4360 Hopkins P. F., Hernquist L., 2006, ApJS, 166, 1
Aird J., Coil A. L., Kocevski D. D., 2022, arXiv e-prints, p. arXiv:2201.11756 Hopkins P. F., Richards G. T., Hernquist L., 2007, ApJ, 654, 731
Assef R. J., Stern D., Noirot G., Jun H. D., Cutri R. M., Eisenhardt P. R. M., Hopkins P. F., Wellons S., Anglés-Alcázar D., Faucher-Giguère C.-A., Grudić
2018, ApJS, 234, 23 M. Y., 2022, MNRAS, 510, 630
Balogh M. L., Morris S. L., Yee H. K. C., Carlberg R. G., Ellingson E., 1999, Ilbert O., et al., 2013, A&A, 556, A55
ApJ, 527, 54 Kauffmann G., Heckman T. M., 2009, MNRAS, 397, 135
Barro G., et al., 2017, ApJ, 840, 47 Kauffmann G., et al., 2003, MNRAS, 341, 33
Best P. N., Heckman T. M., 2012, MNRAS, 421, 1569 Kocevski D. D., et al., 2017, ApJ, 846, 112
Best P. N., Kauffmann G., Heckman T. M., Brinchmann J., Charlot S., Ivezić Kormendy J., Ho L. C., 2013, ARA&A, 51, 511
Ž., White S. D. M., 2005, MNRAS, 362, 25 Laigle C., et al., 2016, ApJS, 224, 24
Birchall K. L., Watson M. G., Aird J., Starling R. L. C., 2022, MNRAS, 510, Land K., et al., 2008, MNRAS, 388, 1686
4556 Lehmer B. D., et al., 2016, ApJ, 825, 7
Bongiorno A., et al., 2016, A&A, 588, A78 Lintott C. J., et al., 2008, MNRAS, 389, 1179

MNRAS 000, 1–21 (2015)

16
Luo B., et al., 2017, ApJS, 228, 2 by the composite galaxy spectrum from bins with larger Dn 4000
Madau P., Dickinson M., 2014, ARA&A, 52, 415 values and the Vanden Berk et al. (2001) quasar spectrum, but can
Magorrian J., et al., 1998, AJ, 115, 2285 mimic the Dn 4000 value of this panel. We can see that including
Marconi A., Hunt L. K., 2003, ApJ, 589, L21 the quasar emission will lead to a small bump around the H𝛽 line
Marocco F., et al., 2021, ApJS, 253, 8 region, which is not obvious among AGN composite spectra in our
McConnell N. J., Ma C.-P., 2013, ApJ, 764, 184
sample. We also perform the same procedures for the AGN and
McDonald M., McNamara B. R., Calzadilla M. S., Chen C.-T., Gaspari M.,
Hickox R. C., Kara E., Korchagin I., 2021, ApJ, 908, 85
galaxy composite spectra in the COSMOS sample, and the results
McLure R. J., et al., 2018, MNRAS, 479, 25 are presented in Figure A5. Including the quasar emission will make
Mountrichas G., et al., 2022, A&A, 667, A145 the bump around the H𝛾 line region noticeable in the spectra, which is
Ni Q., Yang G., Brandt W. N., Alexander D. M., Chen C. T. J., Luo B., Vito not obvious among our AGN composite spectra. The lack of apparent
F., Xue Y. Q., 2019, MNRAS, 490, 1135 differences in the broad emission-line regions of our AGN composite
Ni Q., et al., 2021a, ApJS, 256, 21 spectra compared to galaxy composite spectra indicates that severe
Ni Q., et al., 2021b, MNRAS, 500, 4989 contamination is unlikely.
Ruiz A., Georgantopoulos I., Corral A., 2021, A&A, 645, A74 To quantify the bias in Dn 4000 measurements of AGNs related to
Ruiz A., Georgakakis A., Gerakakis S., Saxton R., Kretschmar P., Akylas A.,
“hidden” AGN emission, we utilize the ratio of the integrated flux
Georgantopoulos I., 2022, MNRAS, 511, 4265
in the shaded regions presented in Figures A3, A4, or A5 between
Salim S., et al., 2007, ApJS, 173, 267
Salim S., et al., 2016, ApJS, 227, 2 the composite AGN spectrum and galaxy spectrum; these regions
Salvato M., et al., 2018, MNRAS, 473, 4937 characterize the broad H𝛽 wings for SDSS spectra and the broad H𝛾
Strauss M. A., et al., 2002, AJ, 124, 1810 wings for LEGA-C spectra. We create galaxy composite spectra in
Toyouchi D., Hosokawa T., Sugimura K., Nakatani R., Kuiper R., 2019, the Dn 4000 grid with a step of 0.01, and mix the galaxy spectra with
MNRAS, 483, 2031 quasar template to make the Dn 4000 value equal to that of the AGN
Vanden Berk D. E., et al., 2001, AJ, 122, 549 composite spectrum in the given Dn 4000 bin. For a given Dn 4000 bin,
Wang T., et al., 2017, A&A, 601, A63 when the ratio of the integrated flux in the shaded regions between
Webb N. A., et al., 2020, A&A, 641, A136 the synthetic quasar plus galaxy spectrum and the composite galaxy
Wu P.-F., et al., 2018, ApJ, 868, 37
spectrum is close to that between the composite AGN spectrum
Yang G., Brandt W. N., Darvish B., Chen C. T. J., Vito F., Alexander D. M.,
and the composite galaxy spectrum, we think Dn 4000 of the galaxy
Bauer F. E., Trump J. R., 2018, MNRAS, 480, 1022
Yang G., et al., 2020, MNRAS, 491, 740 component in the synthetic spectrum represents the true Dn 4000 of
de Jong R. S., et al., 2019, The Messenger, 175, 3 the AGN composite spectrum, so that the bias can be estimated. The
van der Wel A., et al., 2014, ApJ, 788, 28 uncertainty of the bias could be obtained by bootstrapping AGNs
van der Wel A., et al., 2016, ApJS, 223, 29 in different Dn 4000 bins, creating different composite AGN spectra,
van der Wel A., et al., 2021, ApJS, 256, 44 and repeating the above procedures. The results are presented in
Figure A6 for the 4XMM sample and Figure A7 for the COSMOS
sample. We can see that the bias is generally small (≲ 0.1–0.2),
and “calibrating” the measured Dn 4000 values of AGNs with bias
APPENDIX A: ASSESSING THE RELIABILITY OF Dn 4000
will not change the general Dn 4000 trend. We have also verified that
MEASUREMENTS FOR AGNS
our results do not change qualitatively when utilizing “calibrated”
As we limit our sample to objects with galaxy-like spectra in this Dn 4000 values (i.e., the measured values plus the bias estimated
study (i.e. quasar-like sources with prominent broad emission lines from the above method) for AGNs.
are excluded), X-ray AGNs in our sample are type 2 AGNs with ob-
scured disk emission. Thus, the AGN power-law continuum should
have little contribution to the optical spectra. However, it is possible
for a galaxy to look younger in the spectrum when an AGN compo-
nent (even if the contribution is small) is present, as quasars have a
bluer continuum compared to galaxies. In this appendix, we assess
whether the AGN disk emission affects the reliability of Dn 4000
measurements for AGNs in our sample.
In Figure A1, we present the composite spectra of galaxies/AGNs APPENDIX B: THE PROBABILITY DENSITY
in the 4XMM sample in different Dn 4000 bins (Dn 4000 = 1.1–2.0 DISTRIBUTION OF A GALAXY HOSTING AN AGN AS A
with a step of 0.1 and a bin size of 0.1), with the composite quasar FUNCTION OF 𝐿 X /𝑀★ IN THE 4XMM SAMPLE
spectrum from Vanden Berk et al. (2001) shown as well. In Figure A2,
we present the composite spectra of galaxies/AGNs in the COSMOS We note that, in Section 3.1, the AGN fraction is defined as a single
sample in different Dn 4000 bins (Dn 4000 = 1.3–1.7 with a step of fraction with 𝐿 X /𝑀★ greater than a given value, which we take to be
0.1 and a bin size of 0.1). Here, AGNs are defined as objects with log 1032 . To test whether this arbitrary threshold will affect our results,
𝐿 X /𝑀★ > 31. Galaxies are defined as objects not detected in the X- we further model the probability of finding an AGN as a function
ray. The composite spectra are created by normalizing each individual of 𝐿 X /𝑀★ in the 4XMM sample for all the bins in Figure 5. For the
spectrum in the subsample at rest-frame 4050 Å, and taking the COSMOS sample, this type of analysis is limited by the small number
median value. The presented composite spectra are smoothed with a of AGNs detected in each bin. We could see that this probability
boxcar with a width of 10 Å. We can see that the composite spectra of density distribution of a galaxy hosting an AGN as a function of
AGN look generally similar to those of galaxies, and do not exhibit 𝐿 X /𝑀★ as shown in Figure B1 follows a rough linear relation in
signs of any broad line. In Figures A3 and A4, we present the 4XMM the log-log space, similar to what has been found by Birchall et al.
sample AGN and galaxy composite spectra in different Dn 4000 bins (2022). Thus, the 𝐿 X /𝑀★ threshold adopted when calculating the
in different panels; in each panel, we also present spectra constituted AGN fraction would not materially affect the results.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 17

4XMM sample
Normalized F

4000 4500 5000 5500 6000

rest (Å)
Figure A1. Normalized composite spectra of galaxies or AGNs in different Dn 4000 bins (Dn 4000 = 1.1–2.0 with a step of 0.1 and a bin size of 0.1) for the
4XMM sample, represented by the blue or red lines; the deeper the color, the larger the Dn 4000 value. The SDSS quasar composite spectrum from Vanden Berk
et al. (2001) is shown as the black line for comparison.

COSMOS sample
Normalized F

3800 4000 4200 4400 4600

rest (Å)
Figure A2. Similar to Figure A1, but for galaxies or AGNs in different Dn 4000 bins (Dn 4000 = 1.3–1.7 with a step of 0.1 and a bin size of 0.1) in the COSMOS
sample.

APPENDIX C: AGN FRACTION (DEFINED WITH 𝐿 X ) AS A APPENDIX D: AGN FRACTION AS A FUNCTION OF

FUNCTION OF Dn 4000 WHEN CONTROLLING FOR 𝑀★, Dn 4000 WHEN CONTROLLING FOR MORPHOLOGY
SFR, AND 𝑍
We also tested whether host-galaxy morphology has an influence on
the observed trends with Dn 4000, as the dominant galaxy morpho-
logical type changes with the age of galaxies, and it is plausible that
different morphological types have different BH fuelling patterns.
For the 4XMM sample, we perform the elliptical/spiral morpholog-
ical classification with the Galaxy Zoo data (Lintott et al. 2008),
In Section 3.2, we examined how the fraction of log 𝐿 X /𝑀★ > 32 utilizing the CLEAN criterion developed by Land et al. (2008).
AGN changes with Dn 4000 when controlling for 𝑀★, SFR, and 𝑧. In For elliptical galaxies, we can see that the AGN fraction/BHAR
Figure C1/C2, we present how the fraction of log 𝐿 X > 42 AGN (in does not decrease significantly with Dn 4000 at Dn 4000 ∼ 1.8–2.0
the case, AGNs are defined with luminosity rather than with “specific when controlling for 𝑀★, SFR, and 𝑧 in Figure D1, similar to what
black hole accretion rate”) changes with Dn 4000 when controlling we observed in Figure 7. For spiral galaxies, there is a clear trend
for 𝑀★, SFR, and 𝑧 for the 4XMM/COSMOS sample. We could see that the AGN fraction/BHAR drops with Dn 4000 (see Figure D2).
that similar to the trends we observed in Figures 7 and 8, the log For the COSMOS sample, we adopt the bulge-dominated (BD) and
𝐿 X > 42 AGN fraction displays a consistent decreasing trend with non-bulge-dominated (non-BD) morphological classification from
Dn 4000 among galaxies in the 4XMM sample (except for the oldest Ni et al. (2021b) and perform the same analyses for BD galaxies and
galaxies) and the COSMOS sample. non-BD galaxies separately. We note that, among 449 BD galaxies

MNRAS 000, 1–21 (2015)

18
4XMM sample
AGN
Dn4000 = 1.1 galaxy

Dn4000 = 1.2

Dn4000 = 1.3
Normalized F

Dn4000 = 1.4

Dn4000 = 1.5

4000 4250 4500 4750 5000 5250 5500 5750

rest (Å)
Figure A3. In each panel, normalized composite spectra of galaxies or AGNs in the 4XMM sample at a given Dn 4000 bin (with a median Dn 4000 value as
labeled) are represented by the blue or red lines; the black lines represent synthetic spectra created by combining galaxy composite spectra that have stronger
Dn 4000 with the Vanden Berk et al. (2001) composite quasar spectrum in a proportion that mimics the observed Dn 4000 of the presented AGN composite
spectrum. These synthetic spectra show excess emission around the broad H𝛽 wings (indicated by the grey regions) compared to galaxy composite spectra. We
estimate the possible bias in the Dn 4000 measurements for AGNs in our sample by choosing the synthetic combination that best matches the observed excess
flux of the AGN composite spectrum over the galaxy composite spectrum in the grey regions. All the spectra presented are normalized at 4750 Å.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 19
4XMM sample
Dn4000 = 1.6 AGN
galaxy

Dn4000 = 1.7

Dn4000 = 1.8
Normalized F

Dn4000 = 1.9

Dn4000 = 2.0

4000 4250 4500 4750 5000 5250 5500 5750

rest (Å)
Figure A4. Figure A3 continued.

MNRAS 000, 1–21 (2015)

20
COSMOS sample

Dn4000 = 1.3

Dn4000 = 1.4

Dn4000 = 1.5
Normalized F

Dn4000 = 1.6

Dn4000 = 1.7

3800 3900 4000 4100 4200 4300 4400 4500 4600 4700
rest (Å)
Figure A5. Similar to Figure A3, but for galaxies or AGNs in the COSMOS sample. The synthetic spectra represented by black lines show excess emission
around the broad H𝛾 wings (indicated by the grey regions) compared to galaxy composite spectra. All the spectra presented are normalized at 4285 Å.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 21
ies could be seen in Figure D4. We can see that a decreasing trend
0.20 of AGN fraction and BHAR is present (though not very significant)
among BD galaxies and Non-BD galaxies separately. These results
0.15
Dn4000
suggest that the observed variation in AGN activity/BH growth with
Dn 4000 is unlikely due to pure morphological effects.
0.10 We also note that host-galaxy structural properties are unlikely to
cause the difference in AGN fraction/BHAR associated with Dn 4000
0.05 observed in our samples. In Ni et al. (2021b), it has been found that
BH growth is closely related to host-galaxy compactness (represented
0.00 by the projected mass density of the central 1 kpc, Σ1 ) among star-
forming galaxies; higher Σ1 values are associated with higher levels
2.00
Calibrated Dn4000

of AGN activity/BH growth. As older galaxies tend to be more com-

1.75 pact, the higher level of AGN activity/BH growth among younger
galaxies is unlikely a result of varying structural properties.
1.50
This paper has been typeset from a TEX/LATEX file prepared by the author.
1.25
1.00
1.0 1.2 1.4 1.6 1.8 2.0
Dn4000
Figure A6. Top: The underestimation of Dn 4000 (due to contamination from
underlying AGN emission) vs. observed Dn 4000 of AGNs in the 4XMM
sample, with error bars representing the 1𝜎 confidence intervals obtained
from bootstrapping. Bottom: “Calibrated” Dn 4000 values (when accounting
for the bias) vs. Dn 4000 values reported when assuming no AGN component
for AGNs in the 4XMM sample. We have verified that our results do not
change qualitatively when utilizing these “calibrated” Dn 4000 values for
AGNs.

0.15

0.10
Dn4000

0.05

0.00
1.8
Calibrated Dn4000

1.6

1.4

1.2
1.2 1.4 1.6 1.8
Dn4000
Figure A7. Similar to Figure A6, but for AGNs in the COSMOS sample.

in the COSMOS sample, only 7 of these galaxies have log 𝐿 X /𝑀★

> 32. Thus, for the analyses here, we do not adopt the stringent
𝐿 X /𝑀★ threshold as we did for the previous analyses. We directly
adopt the fraction of objects with log 𝐿 X > 42 as the AGN fraction
(without any sensitivity correction), and the results could be seen
in Figure D3. How BHAR varies with Dn 4000 when controlling for
𝑀★, SFR, and 𝑧 is also presented. The results for the non-BD galax-

MNRAS 000, 1–21 (2015)

 22
Probability density (dex 1)

10 1

10 2

10 3
Dn4000: 1.4 Dn4000: 1.6 Dn4000: 1.8 Dn4000: 1.9 Dn4000: 1.9 Dn4000: 2.0
logM : 10.5 logM : 10.8 logM : 10.8 logM : 10.9 logM : 10.9 logM : 11.1
10 4 logSFR: 0.3 logSFR: -0.3 logSFR: -1.0 logSFR: -1.2 logSFR: -1.2 logSFR: -1.2
30 32 30 32 30 32 30 32 30 32 30 32
log(LX/M ) log(LX/M ) log(LX/M ) log(LX/M ) log(LX/M ) log(LX/M )

Figure B1. Distributions showing the probability of finding an AGN as a function of specific black hole accretion rate. Different panels have different Dn 4000.

logM : 18 logM :
4
logSFR: 16 logSFR:
14
AGN fraction (%)

3 AGN fraction (%) 12

10
2 8
6
1 4
2
0 1.4 1.5 1.6 1.7 1.8 1.9 2.0 0 1.3
Dn4000 1.4 1.5 1.6 1.7 1.8
Dn4000
Figure C1. AGN fraction (defined by log 𝐿X > 42) as a function of Dn 4000 Figure C2. Similar to Figure C1, but for AGN fraction as a function of
among galaxies in the 4XMM sample when controlling for 𝑀★ , SFR, and 𝑧. Dn 4000 among galaxies in the COSMOS sample.
Different symbols and colors represent a set of subsamples with similar 𝑀★ ,
SFR, and 𝑧 values (as listed on top of the panel with the same color). The
horizontal position of each data point represents the median Dn 4000 of the
sources in each sample, with 𝑥-axis error bars demonstrating the 16th and
84th percentiles of the Dn 4000 values. The 𝑦-axis error bars represent the
1𝜎 confidence interval of AGN fraction from bootstrapping.

MNRAS 000, 1–21 (2015)

 AGNs in the lifecycle of galaxies 23

Elliptical galaxies in the 4XMM sample Spiral galaxies in the 4XMM sample
logM : 2.0 logM :
1.25 logSFR: logSFR:
AGN fraction (%)

AGN fraction (%)

1.00 1.5
0.75
1.0
0.50
0.25 0.5
0.00 0.0
1.7 1.8 1.9 2.0 2.1 1.4 1.5 1.6 1.7 1.8 1.9
Dn4000 Dn4000
logM : 2.5 logM :
1.0 logSFR: logSFR:
1.5
log BHAR (M /yr)

log BHAR (M /yr)

3.0
2.0
2.5 3.5
3.0
4.0
3.5
1.7 1.8 1.9 2.0 2.1 1.4 1.5 1.6 1.7 1.8 1.9
Dn4000 Dn4000
Figure D1. Top: AGN fraction as a function of Dn 4000 among elliptical Figure D2. Top: Similar to the top panel of Figure D1, but for spiral galaxies
galaxies in the 4XMM sample when controlling for 𝑀★ , SFR, and 𝑧. All the in the 4XMM sample. Bottom: Similar to the bottom panel of Figure D1, but
subsamples share similar 𝑀★ , SFR, and 𝑧 values. The horizontal position for spiral galaxies in the 4XMM sample.
of each data point represents the median Dn 4000 of the sources in each
subsample, with 𝑥-axis error bars demonstrating the 16th and 84th percentiles
of the Dn 4000 values. The 𝑦-axis error bars represent the 1𝜎 confidence
interval of AGN fraction from bootstrapping. Bottom: BHAR as a function of
Dn 4000 among elliptical galaxies in the 4XMM sample when controlling for
𝑀★ , SFR, and 𝑧. All the subsamples share similar 𝑀★ , SFR, and 𝑧 values.
The horizontal position of each data point represents the median Dn 4000 of
the sources in each subsample, with 𝑥-axis error bars demonstrating the 16th
and 84th percentiles of the Dn 4000 values. The 𝑦-axis error bars represent
the 1𝜎 confidence interval of BHAR from bootstrapping.

MNRAS 000, 1–21 (2015)

24

BD galaxies in the COSMOS sample Non-BD

15.0 galaxies in the COSMOS sample
logM : logM :
AGN fraction (%) 8 logSFR: 12.5 logSFR:

AGN fraction (%)

6 10.0

4 7.5
5.0
2
2.5
0 0.0
1.5 1.6 1.7 1.8 1.9 1.3 1.4 1.5 1.6
Dn4000 Dn4000
logM : 1.0 logM :
2.0 logSFR: logSFR:
2.5 1.5
log BHAR (M /yr)

log BHAR (M /yr)

3.0 2.0
3.5
2.5
4.0
3.0
4.5 1.5 1.6 1.7 1.8 1.9 1.3 1.4 1.5 1.6
Dn4000 Dn4000
Figure D3. Top: Similar to the top panel of Figure D1, but for BD galaxies Figure D4. Top: Similar to the top panel of Figure D1, but for Non-BD
in the COSMOS sample. Bottom: Similar to the bottom panel of Figure D1, galaxies in the COSMOS sample. Bottom: Similar to the bottom panel of
but for BD galaxies in the COSMOS sample. Figure D1, but for Non-BD galaxies in the COSMOS sample.

MNRAS 000, 1–21 (2015)