3

Astronomy & Astrophysics manuscript no.
main ©ESO 2022

August 18, 2022
The eROSITA Final Equatorial Depth Survey (eFEDS):

X-ray emission around star-forming and quiescent galaxies at 0.05 < z < 0.3
Johan Comparat1? , Nhut Truong2 , Andrea Merloni1 , Annalisa Pillepich2 , Gabriele Ponti3, 1 , Simon Driver4,5 , Sabine
Bellstedt4 , Joe Liske6 , James Aird7, 8 , Marcus Brüggen6 , Esra Bulbul1 , Luke Davies4 , Justo Antonio González
Villalba1 , Antonis Georgakakis9 , Frank Haberl1 , Teng Liu1 , Chandreyee Maitra1 , Kirpal Nandra1 , Paola Popesso10 ,
Peter Predehl1 , Aaron Robotham4 , Mara Salvato1 , Jessica E. Thorne4 , Yi Zhang1
1
Max-Planck-Institut für extraterrestrische Physik (MPE), Gießenbachstraße 1, D-85748 Garching bei München, Germany
arXiv:2201.05169v2 [astro-ph.GA] 17 Aug 2022
2
Max-Planck-Institut für Astronomie, Königstuhl 17, D-69117 Heidelberg, Germany
3
INAF-Osservatorio Astronomico di Brera, Via E. Bianchi 46, I-23807 Merate (LC), Italy
4
International Centre for Radio Astronomy Research (ICRAR), University of Western Australia, Perth, Western Australia, Aus-
tralia;
5
International Space Centre (ISC), University of Western Australia, Perth, Western Australia, Australia
6
Hamburger Sternwarte, University of Hamburg, Gojenbergsweg 112, D-21029 Hamburg, Germany
7
Institute for Astronomy, Royal Observatory, University of Edinburgh, Edinburgh EH9 3HJ, UK
8
School of Physics & Astronomy, University of Leicester, University Road, Leicester LE1 7RJ, UK
9
Institute for Astronomy and Astrophysics, National Observatory of Athens, V. Paulou & I. Metaxa, 11532, Greece
10
European Southern Observatory, D-85748 Garching bei München, Germany
August 18, 2022
ABSTRACT
Aims. The circumgalactic medium (CGM) plays an important role in galaxy evolution as the main interface between the star-forming
body of galaxies and the surrounding cosmic network of in- and out-flowing matter. In this work, we aim to characterize the hot phase
of the CGM in a large sample of galaxies using recent soft-X-ray observations made by SRG/eROSITA.
Methods. We stack X-ray events from the ‘eROSITA Final Equatorial Depth Survey’ (eFEDS) around central galaxies in the 9hr field
of the ‘GAlaxy and Mass Assembly’ (GAMA) survey to construct radially projected X-ray luminosity profiles in the 0.5–2 keV rest
frame energy band as a function of their stellar mass and specific star formation rate. We consider samples of quiescent (star-forming)
galaxies in the stellar mass range 2 × 1010 – 1012 M (3 × 109 – 6 × 1011 M ).
Results. For quiescent galaxies, the X-ray profiles are clearly extended throughout the available mass range; however, the measured
profile is likely biased high because of projection effects, as these galaxies tend to live in dense and hot environments. For the most
massive star-forming samples (≥ 1011 M ), there is a hint of detection of extended emission. On the other hand, for star-forming
galaxies with < 1011 M the X-ray stacked profiles are compatible with unresolved sources and are consistent with the expected
emission from faint active galactic nuclei (AGN) and X-ray binaries. We measure for the first time the mean relation between average
X-ray luminosity and stellar mass separately for quiescent and star-forming galaxies. We find that the relation is different for the
two galaxy populations: high-mass (≥ 1011 M ) star-forming or quiescent galaxies follow the expected scaling of virialized hot
haloes, while lower mass star-forming galaxies show a less prominent luminosity and a weaker dependence on stellar mass consistent
with empirical models of the population of weak AGN. When comparing our results with state-of-the-art numerical simulations
(IllustrisTNG and EAGLE), we find overall consistency on the average emission on large (> 80 kpc) scales at masses ≥ 1011 M , but
disagreement on the small scales, where brighter-than-observed compact cores are predicted. The simulations also do not predict the
clear differentiation that we observe between quiescent and star-forming galaxies in our samples.
Conclusions. This is a stepping stone towards a more profound understanding of the hot phase of the CGM, which holds a key
role in the regulation of star formation. Future analysis using eROSITA all-sky survey data, combined with future generation galaxy
evolution surveys, shall provide much enhanced quantitative measurements and mapping of the CGM and its hot phase(s).
Key words. X-ray, galaxies, circum-galactic medium
1. Introduction component of the intergalactic and halo gas, supposedly com-

prising 40% of the baryons, have been vastly unconstrained un-
A precise description of the different phases of the cosmic til now. In this regard, L∗ galaxies hosted by 1012 M haloes,
gas, from the intergalactic (IGM) to the circumgalactic medium where most of the stars and metals are formed (Moster et al.
(CGM) around galaxies, is the missing piece in the puzzle of the 2018; Behroozi et al. 2019), are of great interest. Soft-X-ray ob-
baryon budget in the Universe and currently prevents us from servations represent a particularly useful tool for investigating
obtaining a complete and accurate description (Tumlinson et al. the properties of the halo gas around galaxies, as the warm-hot
2017; Driver 2021). The physical properties of the warm-hot phase itself should reach virial temperatures such that most of
?
E-mail: comparat@mpe.mpg.de
Article number, page 1 of 23

A&A proofs: manuscript no. main
Fig. 1. Central galaxies used in this work. Top. Sky coverage of eFEDS data (gray points) and GAMA galaxies (magenta). Bottom panels.
Split between star-forming and quiescent galaxies (leftmost panel) formed by a boundary at log10 (sS FR) = −11. Stellar mass (log10 (M/M )) vs.
redshift 2D histograms (redshift bins have a width of 0.025 and stellar mass bins have a width of 0.25 dex) for the set of quiescent (middle panel)
and star-forming (right panel) galaxies available in the GAMA 9hr field.
the emission emerges in this band (Fukugita & Peebles 2004, signal from the CGM of Milky Way-mass (MW) galaxies and
2006). smaller (. 1010.7 M ).
Hot, X-ray-emitting haloes have been observed around indi-
vidual or small samples of galaxies in the past. This has been From a theoretical perspective, the extended soft-X-ray
achieved for mostly early-type massive galaxies (e.g., Gould- emission from the gaseous atmospheres of massive haloes has
ing et al. 2016; Bregman et al. 2018). The detection of X-ray- been predicted by galaxy formation models in the full cosmo-
emitting atmospheres around star-forming disk galaxies is rare logical context since the analytical models of White & Frenk
and limited to small samples of galaxies more massive than (1991). This has been confirmed, albeit with overall lower lu-
∼ 1011 M (e.g., Bogdán et al. 2013a,b, 2017; Anderson et al. minosity than previously expected, by the results of cosmologi-
2016; Li et al. 2016). cal hydrodynamical simulations, also around disk and MW-mass
Anderson et al. (2015) stacked X-ray photons from the galaxies (e.g., Crain et al. 2010; Bogdán et al. 2015; Kelly et al.
ROSAT all-sky survey around 250 000 massive galaxies from 2021, with the GIMIC, Illustris, and EAGLE simulations, re-
the Sloan Digital Sky Survey that are central within their dark- spectively). These works have shown that such X-ray emission
matter haloes. They reported a strong correlation between the is sensitive to the unfolding and the interplay of star formation,
mean X-ray luminosity of the volume-filling gas in the CGM feedback, and cooling processes, which can simultaneously re-
(i.e., in the range (0.15 − 1) × R500c 1 ) and the galaxy stellar mass move low-entropy gas by star formation, heat up the halo gas via
in the stellar mass range log M ∗ /M = 10.8 − 12. However, be- energy injections, and redistribute the gas via outflows, making
cause of the limited spatial resolution and the bright flux limit of gaseous haloes dilute. Recently, Truong et al. (2020) and Oppen-
ROSAT, it was not possible to firmly detect an X-ray emission heimer et al. (2020) showed that the state-of-the-art cosmologi-
cal galaxy simulations IllustrisTNG and EAGLE both predict an
1
R500c is the radius at which the density of the dark matter halo is 500 X-ray luminosity dichotomy at z ∼ 0: at the transitional stellar
times the critical density of the Universe at the redshift of the system. mass regime of 1010.5−11 M , simulated star-forming galaxies ex-
Comparat et al.: X-ray emission around GAMA galaxies in eROSITA/eFEDs
hibit somewhat higher soft-X-ray luminosity from the volume- structed wide spectral energy distribution (SED) and adjusted
filling gas within and around them than quiescent galaxies of the parameters of the stellar population constituting these galax-
the same mass, all the way out to galactocentric distances of ies (Robotham et al. 2020; Bellstedt et al. 2020, 2021). Of the
∼200 kpc. Despite the differences in feedback processes imple- GAMA fields, the 9hr field overlaps with the eFEDS observa-
mented therein, this has been shown to be a direct manifestation tions; see Fig. 1. In that field, for galaxies brighter than r < 19.8,
of the quenching mechanism in the simulations, with the reduc- the spectroscopic completeness is 98.67%.
tion of the gas mass within the haloes being due to super mas- From the 9hr field (SpectCatv27 and NQ> 2), we retrieve
sive black hole (SMBH)-driven outflows. Clearly, robust statisti- about 40 000 galaxies with a spectroscopic redshift in the range
cal constraints on the properties of the X-ray-emitting gas from 0.05≤ z ≤0.3 and with measured stellar mass and specific star
large samples of galaxies in this transitional mass regime may formation rate (sSFR) from Bellstedt et al. (2020, 2021). These
hold the key to further improving our understanding of the com- are derived using SED fitting (Robotham et al. 2020). They
plex physical processes shaping the gaseous atmosphere of their adopt a Chabrier initial mass function (IMF) and SFRs are av-
haloes. eraged over 100 Myr. Stellar masses are output in units of solar
In this article, we bridge the gap between past studies by at- mass (M ) and sSFR per year (yr−1 ). The stellar mass function
tempting to measure the faint, extended X-ray emission —that is, and cosmic SFR density are accurate and in excellent agreement
the so-called hot phase of the CGM— surrounding central galax- with the literature (Bellstedt et al. 2020; Koushan et al. 2021;
ies over a wide stellar mass range (≈ 109.6−11.8 M ) by stacking Driver et al. 2022).
SRG/eROSITA data, taking advantage of its high sensitivity in The GAMA 9hr galaxy sample only covers a fraction of the
soft X-rays, moderate spatial resolution, large grasp, and stable eFEDS area: 60 deg2 out of 140 deg2 ; see top panel of Fig. 1.
background (Predehl et al. 2021). To this end, we use data from Necessarily, we limit our X-ray analysis to this 60 deg2 area.
the eROSITA performance verification eFEDS field (Brunner The exact selection of the different galaxy subsamples adopted
et al. 2022), a 140 deg2 survey that partly overlaps with a highly in our analysis is detailed in Sect. 3.1.
complete spectroscopic sample of low-redshift (0.05 < z < 0.3)
galaxies (GAMA, Sloan Digital Sky Survey (SDSS), Liske et al.
2015; Ahumada et al. 2020). We can therefore stack X-ray data 3. Method
around galaxies in different bins of stellar mass, and distinguish By stacking large data sets, the noise decreases and features with
between star-forming and quiescent galaxies. weak signal may be unveiled (e.g., Zhu et al. 2015; Comparat
A brief description of the data used is presented in Sect. 2. et al. 2020b; Wolf et al. 2020). To do so, we stack X-ray events
The method, close to that adopted by Anderson et al. (2015), is around central galaxies with known spectroscopic redshift (Sect.
described in detail in Sect. 3. In Sect. 4, we discuss the measure- 3.1), after masking detected X-ray sources (Sect. 3.2).
ments obtained, while in Sect. 5 we compare the measurements We keep track of the angular distance between the detected
with hydrodynamical simulations. We discuss the possible im- X-ray events and the galaxies (and their redshifts) to build radial
plications of our results for galaxy evolution in Sect. 6. profiles. We record the event energies to build X-ray spectra. We
obtain a data cube where angular coordinates are converted to
proper distance (angle averaged) radii, and the energy (or wave-
2. Data length) vector is shifted to the galaxy redshift. The stacking pro-
2.1. eROSITA eFEDS data cedure is described in detail in Sect. 3.3.
In the measurement process, control on two systematic fea-
We use for this work the public Early Data Release (EDR) tures is key: the emission of the background and the instrumen-
eROSITA event file of the eFEDS field (Brunner et al. 2022)2 . tal signatures. To simplify our analysis, and in light of the ex-
It contains about 11 million events (X-ray photons) detected by pected SED of the signal we are interested in, we focus here
eROSITA over the 140 deg2 area of the eFEDS Performance on the rest-frame energy range between 0.5 and 2 keV. In do-
Verification survey. Each photon is assigned an exposure time ing so, we only consider the energy range where the background
using the vignetting-corrected exposure map. Photons close to emission is dominated by the well-understood diffuse emission
detected sources in the source catalog are flagged (see details in of the Milky Way and by the Cosmic X-ray Background (e.g.,
Sect. 3.2). These sources are cataloged as point-like or extended Predehl et al. 2021; Liu et al. 2022b, Ponti et al. in preparation),
based on their X-ray morphology (Brunner et al. 2022), and while the contribution of the particle (unvignetted) background
they are further classified (e.g., galactic, active galactic nuclei is negligible. Moreover, we avoid the complex response of the
(AGN), individual galaxies at redshift z < 0.05, galaxy group, lowest energy range (below ∼0.4 keV in the observed frame)
and cluster) using multi-wavelength information (Salvato et al. where both detector noise and the effects of the light leak on the
2022; Vulic et al. 2022; Liu et al. 2022a,b; Bulbul et al. 2022). TM5 and TM7 (see Predehl et al. 2021) could introduce yet un-
calibrated systematic effects. As in Brunner et al. (2022), we se-
lect good events from nominal field of view, exclude bad pixels,
2.2. GAMA galaxy catalog
and keep events with PATTERN≤15, which includes single, dou-
The Galaxy And Mass Assembly (GAMA) survey and its fourth ble, triple, and quadruple events. Also, given the relatively low
data release provides spectroscopic redshift for more than 95% signal-to-noise ratio (S/N) achieved from relatively small galaxy
of the galaxies brighter than ∼19.80 in the r-band (Liske et al. samples, we only focus our attention on broad-band photometric
2015; Driver et al. 2022). The released data and galaxy catalog measurements. Work is ongoing on the calibration of the low-
are described at length by Driver et al. (2022)3 . For each galaxy, energy response of eROSITA, and future works will explore the
using multi-wavelength observations covering the far-ultraviolet possibility of stacked spectral analysis also in the 0.15–0.4 keV
(FUV; ∼1500Å) to the far-infrared (FIR; ∼500µm), GAMA con- observed-frame energy range.
We apply a bootstrap procedure to reliably estimate the mean
2
https://erosita.mpe.mpg.de/edr expected background and its variance (Sect. 3.4). Finally, we es-
3
http://www.gama-survey.org/dr4/ timate the spatial extent of the point-source profile using an em-
Table 1. Samples considered in the analysis. We use the reported stellar masses and sSFRs to create sub-
samples of the galaxy population. In order to examine trends,
Sample stellar mass Ng Average we split the population into star-forming and quiescent galaxies
name min max M̄ z̄ assuming a boundary fixed at log10 (sS FR) = −11 (see discus-
ALL_M10.7 10.4 11.0 16142 10.7 0.22 sion by Davies et al. 2019; Thorne et al. 2021). For this study,
Fixed stellar mass selection in which we stack around a large number of galaxies, the exact
QU_M10.7 10.0 11.0 7267 10.72 0.2 boundary definition should have a minor impact. Figure 1 shows
SF_M10.7 10.4 11.0 9846 10.66 0.23 the distribution of galaxies in the redshift range of interest in the
Quiescent galaxies mass–sSFR plane.
QU_M11.71 11.616 11.973 50 11.71 0.27 To compare the star-forming and quiescent samples at fixed
QU_M11.58 11.556 11.616 50 11.58 0.27 stellar mass, we first adopt a stellar mass selection to obtain
QU_M11.54 11.523 11.556 50 11.54 0.26 two samples with the same mean stellar mass, different sSFR,
QU_M11.35 11.306 11.41 400 11.35 0.25 and a similar total number of galaxies. By taking objects within
QU_M11.2 11.138 11.269 1002 11.2 0.25 10 < log10 M ∗ < 11 for the quiescent and 10.4 < log10 M ∗ < 11
QU_M11.04 10.961 11.138 2000 11.04 0.24 for the star-forming galaxies, we obtain a mean stellar mass of
QU_M10.88 10.806 10.961 1999 10.88 0.23 5 × 1010 M for both, with a set of 7 267 and 9 846 galaxies, re-
QU_M10.73 10.641 10.806 1999 10.73 0.21 spectively.
QU_M10.53 10.362 10.641 2000 10.53 0.19 Each population, star-forming or quiescent, is then split in
star-forming galaxies a number of nonoverlapping stellar mass subsamples (see Table
SF_M11.25 11.17 11.754 400 11.25 0.27 1). As stellar mass should correlate with X-ray luminosity (An-
SF_M11.12 11.079 11.17 400 11.12 0.27 derson et al. 2015), in order to obtain a similar S/N from the
SF_M11.05 11.027 11.079 401 11.05 0.26 various subsamples, fewer galaxies are needed at higher mass
SF_M10.99 10.943 11.051 1000 10.99 0.26 than at lower mass. We therefore create subsamples of ∼2000
SF_M10.9 10.861 10.943 1001 10.9 0.25 at the low-mass end, then 1000, 400, and finally 50 galaxies at
SF_M10.86 10.795 10.943 2000 10.86 0.25 the high-mass end. Table 1 details the exact number of galaxies
SF_M10.74 10.68 10.795 2000 10.74 0.24 present in each subclass. There, we also report the mean redshift
SF_M10.63 10.574 10.68 2002 10.63 0.23 and mean stellar mass for each subsample defined in this way.
SF_M10.52 10.467 10.574 1998 10.52 0.22
SF_M10.41 10.358 10.467 2001 10.41 0.21 3.2. Masking approaches and possible sources of
SF_M10.3 10.241 10.358 2000 10.3 0.2 contamination
SF_M10.18 10.108 10.241 2001 10.18 0.19
SF_M10.03 9.947 10.108 1998 10.03 0.18 As we are looking for faint diffuse emission, it is vital to remove
SF_M9.86 9.761 9.947 2000 9.86 0.16 (mask out) as many sources of contamination as possible, that is,
SF_M9.64 9.491 9.761 2000 9.64 0.14 those produced by unresolved emission from compact sources
within galaxies. In this work, we investigate four possible mask-
Notes. SF stands for star-forming and QU for quiescent. We report the ing schemes (see Table 2):
number of galaxies present in each sample (Ng ) and their average prop-
erties: stellar mass ( M̄) and redshift (z̄). (i) ‘ALL’ mask: all detected X-ray sources are masked;
(ii) ‘M1’ mask: all detected sources are masked except for those
associated with a cluster or a group in the same redshift range
pirical point spread function (PSF) model based on the detected as the GAMA galaxies, as identified by Liu et al. (2022a) or
point sources in the same eFEDS field, as we describe in Sect. by Bulbul et al. (2022), taking CLUSTER_CLASS = 4 or 5,
3.5. (see Salvato et al. 2022);
(iii) ‘M2’ mask: all detected sources are masked except for point
3.1. Selecting galaxies sources in the redshift range of interest associated by Salvato
et al. (2022) to a GAMA galaxy;
We select central galaxies, most massive within their host dark (iv) ‘M3’ mask: all detected sources are masked except for those
matter halo, similarly to Planck Collaboration et al. (2013). For unmasked by the M1 or M2 mask. The signal obtained is to
each GAMA galaxy, we infer its host halo mass and corre- be interpreted as the sum of all emitting entities: AGN, X-
sponding virial radius with the stellar-to-halo-mass relation from ray binaries (XRBs), and hot gas augmented by systematic
Moster et al. (2013). If a galaxy lies within twice the virial ra- projection effects.
dius of a galaxy of higher stellar mass, it is considered a satellite
and is removed from the sample. The choice of two times the The masking radius for each detected source (with a de-
virial radius is a conservative one in order to account for the tection likelihood of greater than 6) is its radius of maximum
scatter in the stellar-mass to halo-mass relation. We treat the X- S/N, as determined while extracting the X-ray spectrum of each
ray-detected eFEDS clusters (Liu et al. 2022a) separately, for source (Liu et al. 2022b), augmented by 40%. By doing so, we
which we have individual measurements of R500c (Bahar et al. make sure there is no remaining correlation between the set of
2022). For them, we only remove satellites falling within one events outside of the mask and the source catalog (Comparat et
virial radius, taken as R500c /0.7. After this filtering, ∼ 10% of al. in prep.). The optimal masking radius, derived with limited
the galaxies are removed, and we obtain a sample of 35 521 cen- statistics on eFEDS data, suffers from uncertainties; augment-
tral galaxies. Thanks to the high completeness (∼ 98%) of the ing the masking radius by 30% (or 50%) is also reasonable and
GAMA sample, the sample of central galaxies should also be corresponds to masking 2% fewer (or more) events, as shown in
highly complete. We discuss limitations due to our sample defi- Table 2. This uncertainty on the total number of events directly
nition in Sect. 6.2 impacts the normalization of the profiles estimated. To account
Table 2. Fraction of masked events troscopic completeness achieved by GAMA at these magnitude
limits, and with the GAMA target selection (Baldry et al. 2010),
mask fraction of masked events masked area which uses a combination of criteria to exclude stars, while keep-
name % augmentation of masking radius fraction ing compact galaxies and quasi-stellar objects (QSOs). The re-
40 (30, 50) amask maining unmatched X-ray sources are typically fainter in the op-
ALL 0.2713 (0.2565, 0.2863) 0.069 tical than the GAMA limit, and are always masked out in the
M1 0.259 (0.245, 0.2732) 0.066 stacks.
M2 0.2542 (0.2394, 0.2696) 0.065
The point-source X-ray flux limit of eFEDS, of namely
M3 0.2414 (0.2275, 0.256) 0.061
∼ 6.5 × 10−15 erg s−1 cm−2 in the 0.5–2 keV band (Brunner
Notes. Fraction of masked events for a radius of maximum S/N aug- et al. 2022), corresponds to a rest-frame luminosity of between
mented by 30% to 50%, where 40% is the baseline used in the analysis. about 5×1040 and 2×1042 erg s−1 at the redshift of the GAMA
Masks are ordered by decreasing masked fraction. The percentage of sources we are interested in. We therefore detect essentially all
area masked corresponding to the 40% baseline is given in the last col- X-ray-bright AGN among GAMA galaxies; removing photons
umn. around all detected point-like X-ray sources (M1 mask) there-
fore removes the contamination from all the bright AGN from
the sample. However, within the GAMA galaxy catalogs, a frac-
for this, we add a systematic 2% uncertainty on the background tion of galaxies are expected to host fainter AGN, which remain
luminosity density (our normalization); see Sect. 3.4. undetected given the eROSITA/eFEDS flux limit. Aird et al.
We use the sensitivity map to generate a catalog of random (in preparation) study the point-source emission emerging from
points following Georgakakis et al. (2008). Armed with this, we GAMA galaxy stacks (as a function of stellar mass and redshift)
estimate that masking all X-ray sources removes 27.13% of the to determine the faint end of the AGN X-ray luminosity func-
events (and 6.9% of the area) in the 0.5–2 keV band. The least tion. These authors measure and model the average luminosity
stringent mask, ‘M3’, removes 24.14% of the events (and 6.1% and the fraction of galaxies hosting an X-ray AGN. For a stel-
of the area; see Table 2). lar mass (log10 (M ∗ /[M ])) of 9.75 (10.75, 11.75), they find an
The baseline mask used in this study is the M1 mask: average luminosity of log10 (LX /[erg s−1 ]) ≈ 40, (41, 42) and an
all sources are masked except for sources identified as galaxy occupation fraction of 0.1% (1%, 10%). We further discuss AGN
groups or clusters at 0.05 ≤ z ≤ 0.3. Indeed, masking these ex- contamination and compare these figures with our observations
tended sources would bias low the X-ray profiles of high-mass in Sect. 6.1.
galaxies. The other masks enable us to investigate systematic ef-
fects due to the masking procedure. In particular, we present a
detailed comparison with the results from the M3 mask, which 3.2.2. Expected X-ray binary signal
include the emission from all the GAMA sources detected by
eROSITA as point sources. The set of GAMA galaxies matched X-ray binaries, the end points of stellar evolution, are known
to X-ray point sources sample both the unobscured and the ob- contributors to the total X-ray luminosity of a galaxy (Tauris &
scured AGN loci. Their luminosity ranges from 5 × 1040 to van den Heuvel 2006). They are typically spatially distributed
2 × 1044 erg s−1 in the soft-X-ray band. Below, we discuss in following the stellar light, and therefore their emission would be
more detail the possible contamination due to faint, undetected unresolved by eROSITA at the redshift of interest here.
AGN or XRBs, and the relationship with the alternative masking We evaluate the possible contribution from these (unre-
approaches. solved) XRBs by taking advantage of the known scaling rela-
tion between their total X-ray luminosity and their host galaxy
properties. In particular, in order to predict the X-ray luminos-
3.2.1. Expected AGN signal ity emitted by each galaxy and attributable to these sources, we
X-ray emission from AGN is produced in a very compact (frac- use an analytical model based on Lehmer et al. (2016) and Aird
tion of a milli-parsec) region close to the central SMBHs in et al. (2017). These authors measured the dependence of the to-
the nuclei of galaxies, and thus represents a contamination to tal XRB luminosity (in the 2–10 keV energy band) on redshift,
the CGM signal. To ease the interpretation of the stacked pro- galaxy stellar mass, and SFR. To make sure our prediction is con-
files, we would therefore ideally remove as many active galaxies servative, we use the Aird et al. (2017) model, which produces a
from the sample as possible. However, given the moderate angu- 10%-20% brighter XRB luminosity for a given galaxy property.
lar resolution of eROSITA, this step is far from straightforward. We use their "model 5" with parameters given in their Table 3 to
In order to assess our ability to remove AGN contaminants, we predict the X-ray luminosity in the band 2-10 keV. We propagate
first discuss the completeness of GAMA towards X-ray AGN the uncertainties from this latter table to the prediction. We then
detected by eROSITA. convert (multiplication by 0.56) the luminosity to the 0.5–2 keV
Within the eFEDS X-ray point-source catalog, considering band assuming an absorbed (with nH column density set at mean
all those counterparts in the GAMA 9hr. field and in the redshift value of the field 4 × 1020 cm−2 ) power law with a photon index
range 0.05 < z < 0.3, using either spectroscopic or high-quality of 1.8 (as suggested by Basu-Zych et al. 2020). We note here
photometric redshifts (see Salvato et al. 2022), we obtain 619 X- that all the X-ray-detected point-like sources in eFEDS are sig-
ray sources. Of these, 474 (76.6%) are matched within 200 to a nificantly more luminous than predicted by the Aird et al. (2017)
galaxy present in the GAMA catalog. When limiting the X-ray model.
catalog to sources with an LS8 magnitude r< 19.8 (19), simi-
lar to the magnitude limit used in GAMA, 88.8% (90.2%) are 3.3. Stacking procedure
matched to GAMA galaxies. This implies that, at the magnitude
limit of GAMA, the galaxy catalog is nearly complete in terms We assume a flat ΛCDM cosmology with H0 = 67.74 km s−1
of counterparts of the bright X-ray point-source population de- Mpc−1 , Ωm (z = 0) = 0.3089 (Planck Collaboration et al. 2016).
tected in eFEDS. In turn, this is consistent with the known spec- Each galaxy is characterized by its position on the sky and its
redshift, as well as properties of its stellar population (mass, 3.4. Background estimation and its uncertainties
sSFR). We denote a galaxy with the vector G defined as
The projected radial profiles and integrated spectra obtained with
G = (GRA , G Dec , Gz , G M , G sS FR ). (1) the random cubes represent the null hypothesis of no signal, and
For each galaxy, we retrieve all the events within the angle are used to assess the robustness of any possible detection from
subtended by 3 Mpc at the galaxy redshift. We construct a "cube" the stacking samples. We repeat this measurement process at
of events surrounding each galaxy. For each event, we compute random positions 20 times (using different random points each
a rest-frame energy by multiplying the energy by one plus the time). From the outer shells of the radial background profiles
galaxy redshift: Er f = Eobs × (1 + Gz ). Therefore, each eROSITA (500 kpc to 3000 Mpc) of the 20 realizations, we estimate the
event is characterized by the following vectors: its position on mean background luminosity. It takes values of around 1.1×1037
the sky (R.A., Dec.), its rest frame and observed energy, the cor- erg s−1 kpc−2 . For each galaxy sample, the background value ob-
responding galaxy redshift, the exposure time, the on-axis tele- tained is different; indeed the total area covered and the masked
scope effective area as a function of energy (ARF) at the ob- area both vary from sample to sample.
served energy, and the projected distance (R p ) in proper kilopar- The uncertainty due to the source-masking procedure (see
secs (kpc) to the galaxy. We denote an event with the vector E Sect. 3.2) suggests that the total number of events is subject to
defined as a residual 2% systematic uncertainty at most. To be conserva-
tive, we therefore add a 2% systematic uncertainty to the mean
E = (R.A., Dec., Eobs , Er f , Gz , texp , ARF(Eobs ), R p ). (2) value of the background: σBG = 0.02 at all scales and ener-
The exposure times are obtained from the vignetted exposure gies. The uncertainty on the galaxy√ stack count rates follows
map (Brunner et al. 2022). Using the (angular) projected distance a Poisson distribution: σGAL = 1/ N GAL . The uncertainty on
induces projection effects which we discuss in Sect. 6.2. We re- the final background-subtracted measurement (i.e., galaxy mi-
peat this procedure with sets of random locations in the field, nus
p background) is thepquadratic sum of the two uncertainties:
replacing the galaxy positions with randomly drawn positions in (σGAL )2 + (σBG )2 = 1/N GAL + 0.022 .
the same area of the sky, taking advantage of the relatively uni- Additional sources of uncertainty may arise from the use of
form exposure of the eFEDS field (Brunner et al. 2022). inaccurate redshifts, source positions, or photon energies. Un-
Finally, in order to derive accurate correction to the measured certainties on centering and positions could artificially cause the
fluxes for masking and boundary effects (due to the reduction of PSF to broaden; uncertainties on energies could cause spectral
projected area), we repeat the above procedure with another two distortions; redshift uncertainties would cause both spectral and
sets of random events. A first set of random events uniformly spatial distortions. As we use spectroscopic redshifts, we con-
samples the area covered in the GAMA field (REG ). A second set sider the uncertainty coming from those to be negligible; how-
of events uniformly samples the GAMA area and an additional ever, one could imagine that a few catastrophically incorrect red-
1.5 degree wide stripe around (REW ). This allows us to account shifts are included. If, for example, these are additionally lo-
for boundary effects in the area normalization of the background cated in bright clusters (illustrative purpose), we would incor-
counts (see Sect. 3.4). rectly convert arcminutes to kpc, which could cause profiles to
We apply each selection defined in Table 1 to the galaxy sam- either be more concentrated (redshift is lower than true redshift)
ple and to the random galaxy samples. We concatenate the event or diluted (redshift measured is higher than true redshift). The
sets obtained. For each galaxy sample, we obtain five cubes of exact quantification of a possible systematic error arising from
events: galaxy events (data cube), random galaxy events (random catastrophic redshift, incorrect source positions, or photon ener-
cube), point-source galaxy events (point-source cube, detailed in gies is left for future studies.
Sect. 3.5), galaxies-REG cube, and galaxies-REW cube.
Each event in any of the cubes is weighted by the following
function: 3.5. Empirical point-source profile and validation against
AGN
Acorr (r) × 1.602177 × 10 −12
Er f 4πdL2 (Gz )
wi = , (3)
ARF(Eobs ) texp × Ng As we are interested in detecting extended CGM X-ray emission
around galaxies, a key prerequisite is an accurate characteriza-
where Ng is the number of galaxies in a sample (given in Ta- tion of the eROSITA point spread function (PSF) and its convo-
ble 1). Acorr is the area-correction term, which accounts for both lution with the redshift distribution of the galaxies.
boundary effects and masks:
To obtain an empirical point-source profile for comparison
REW (r) to each galaxy sample, we repeat the procedure described in
Acorr (r) = 1 + + amask , (4) Sect. 3.3 with sets of detected X-ray point sources in eFEDS
REG (r)
("point-source cube"). Each galaxy is artificially moved to the
where r is the proper projected separation in kpc. For the full sky position of an X-ray point source. To do that, we replace
W (r)
sample, the correction RE
REG (r) is 0.5% at 100 kpc, 1% at 300 kpc, the galaxy positions (on sky R.A., Dec.) with that of extra-
2.5% at 1000 kpc, and 5 % at 2 Mpc. For the M1 mask, amask = galactic point sources with moderately bright fluxes 10−14 <
0.066, and for other masks, values are given in Table 2. F0.2−2.3keV /[erg cm−2 s−1 ] < 10−12 and ERO_DET_LIKE > 20,
A surface luminosity projected profile in erg s−1 kpc−2 de- taken from the Brunner et al. (2022) catalog.
noted S X is obtained from the weighted (using wi ) histogram of In doing so, we convolve the eROSITA PSF with the redshift
projected separations to the galaxies (R p ) divided by the area in distribution of the galaxy sample, and we obtain the shape of the
kpc2 covered by each histogram bin: π(R11 −R20 )Acorr . Conversely, radial profile expected if all sources were bright and point-like in
an X-ray spectrum, in erg s−1 keV−1 in a given projected radial the eROSITA images. Figure 2 shows how these empirical PSF
aperture, is obtained with the weighted sum of the energies (Er f ) profiles (in kpc) evolve as a function of redshift. The higher the
of all events in a given energy bin divided by the width of that redshift, the broader the point-source profile. We do not stack
bin (in keV). beyond redshift 0.3 to avoid an overly wide PSF in kpc. At the
agreement with the mean of the measurements made on individ-

ual AGN.
4. Results
We discuss first the detection in the stacking experiment for the
full sample (Sect. 4.1). We consider to have a ‘detection’ when
the S/N is larger than 3, a ‘hint of detection’ if the S/N is between
1 and 3, and an ‘upper limit’ when the S/N is smaller than 1.
In Sections 4.2 and 4.3, we discuss the trends obtained when
splitting the sample according to its sSFR and stellar mass. The
comparison with theoretical predictions presented in Sect. 5 is
done on the binned samples, where the stellar population is best
controlled.
4.1. Detection in the complete stack

We first report the results of our stacking exercise applied to
the sample of 16 142 galaxies at a mean redshift of 0.22 and a
mean stellar mass of 10.7 (named ALL_M10.7). We focus here
on the results obtained by three possible masking procedures:
Fig. 2. eROSITA normalized point-source profiles as a function of red-
shift in bins of 20kpc. Each curve represents the average point-source ALL, M1, and M3 (see definitions in Sect. 3.2).
profile in a redshift bin of width of 0.05: 0.05-0.1, 0.1-0.15, 0.15-0.2, When applying the ALL mask, we obtain a detection above
0.2-0.25, 0.25-0.3. The labeled number gives the mean redshift of the the background; see top left panel of Fig. 3. The S/N is ∼3 within
bin. At the mean redshift of the sample (z ∼ 0.2), the half width at half R p < 80 kpc. At larger radii (> 80 kpc), the signal measured (ma-
maximum (HWHM) of the empirical PSF corresponds to about 60 kpc. genta crosses) is consistent with the background (green dashes).
The shape of the profile is marginally more extended than the
maximal point-source profile (gray step line).
mean redshift of the sample (z ∼ 0.2), the HWHM of the empir- The detection significance increases when using the M1
ical PSF corresponds to about 60 kpc. mask (Fig. 3, top middle panel), that is, when the galaxy clus-
We stress here that the purpose of this exercise is not to deter- ters and groups detected by eROSITA in the redshift range of
mine an accurate PSF profile for eROSITA (see e.g., Churazov the GAMA galaxies are not removed before stacking. The S/N
et al. 2021), but rather to have a term of reference with which accumulated within R < 80 kpc is about ∼5 (reported in Table
to assess the possible extended nature of the profiles measured 3). Compared to the ALL mask profile, the M1 is brighter and
around galaxies. The stacked profiles obtained here from the de- significantly deviates from the maximal point-source profile.
tected point sources are by construction clearly much brighter When using the M3 mask, that is, when the galaxy clusters,
than the stacked galaxy profiles (see Sect. 4 below). To ease groups, and point sources detected by eROSITA among GAMA
comparison, in each of the galaxy stacks we present below, we galaxies are not masked, the S/N within 80 kpc increases to ∼13;
re-scale the convolved PSF profile to match the central value of see Fig. 3 (top right panel), and Table 4. The overall stacked
the galaxy profile, creating a "maximal point source" (max PS) profile corresponds, qualitatively, to what is expected with the
term of comparison4 . addition of one (or multiple) bright unresolved source(s).
Finally, to measure the mean projected emission coming
3.6. Validation from around the galaxies, we subtract the background from
each stacked profile; see the bottom panels of Fig. 3. There,
To validate the stacking procedure, we apply it to known background-subtracted profiles are shown out to 300 kpc. The
(eFEDS-detected) AGN (with measured spectroscopic redshift). possible deviation from a point-source emission profile is made
We stack at the spectroscopic AGN redshift. Its integrated lumi- clearer by the comparison with the corresponding "max PS" pro-
nosity (in erg s−1 ) amounts to log10 (LX ) = 42.72 ± 0.08, while file. Using the background-subtracted profiles, we measure the
the mean luminosity of the same AGN set as determined by Liu integrated projected luminosity in an aperture R p (in kpc) as fol-
et al. (2022b) is log10 (LX ) = 42.75. The background-subtracted lows:
X-ray spectrum obtained is well fit by a power law with a photon Z r=R p
index of 2.05±0.05, compatible with the mean slope of 2.02 de- <R
termined on the same sample by Liu et al. (2022b). The projected LX p = [S X (r) − BG] 2π r dr , (5)
r=0
luminosity profile and stacked spectra are therefore in very good
where BG is the background level estimated as in Sect. 3.4. For
4
PSF profiles could be artificially broadened due to the clustering of each sample, the measured luminosity is reported for two aper-
the galaxies (Popesso et al. 2012). Complete simulations of the galaxy tures: 80 and 300 kpc in Tables 3 (for mask M1) and 4 (for mask
population and its X-ray emission would be needed over cosmologi- M3).
cal volumes to enable a quantitative assessment. Indeed, we need to
generate a model to populate the full sky with X-ray-emitting galax- The expected total (or radially integrated) XRB luminosity
ies together with their CGM (possibly following simulations of the gas for the ALL_M10.7 sample is 1.1×1040 erg s−1 . For the M1 (M3)
around galaxy clusters from Comparat et al. 2020a), which is beyond mask, the luminosity measured in the inner 80 kpc is 2.8±0.5 ×
the scope of this article. We defer the quantification of this effect to a 1040 (8.8±0.7 × 1040 ) erg s−1 ; therefore, the observed luminosity
future study. cannot come from XRBs alone.
Fig. 3. Measured X-ray radial projected luminosity profiles (0.5–2.0 keV rest-frame) for the ALL_M10.7 GAMA central galaxy sample (‘STACK’,
magenta crosses and shaded area). Each panel shows the result when a different mask is applied to the set of events: "ALL" (left), "M1" (middle),
and "M3" (right). We note the variation in the y-axis range in different panels. The green dashed line represents the background level, estimated as
discussed in Sect. 3.4. The profile shape expected if all sources stacked were point-like is shown with a gray line labeled "max PS". The bottom
series of panels shows the background-subtracted profiles with a linear radial scale extending to 300 kpc.
With the complete stack being constituted of a varied mix the conversion of baryons into stars is thought to be the most
of different galaxies, linking the detection to underlying physi- efficient (Behroozi et al. 2013; Moster et al. 2013). At this stellar
cal processes is complex. To further interpret the link between mass, Velander et al. (2014) found that the red galaxies reside
the detected emission and its possible sources (hot gas, XRBs, in denser environments than blue galaxies. These latter authors
faint AGN), we split the GAMA sample according to the phys- measured the host halo mass of blue and red ∼ M∗ galaxies,
ical properties of the galaxies, as we describe in the following and found that it was comprised between 1-3×1012 M for both.
sections. Based on these results, the samples QU_M10.7 and SF_10.7 are
hosted on average by similarly massive haloes, and, importantly,
reside in different environments.
4.2. Trend with sSFR at fixed stellar mass ∼ 5 × 1010 M
We split the sample into a quiescent subsample and a star- With the M1 mask
forming subsample. We set a maximum boundary in stellar
mass at 1011 M , which corresponds to haloes with a mass of In both star-forming and quiescent cases, there is an X-ray de-
∼ 5 × 1012 M (using the stellar-to-halo-mass relation from tection in the stacks obtained with the M1 mask; see Fig. 5 (left
Moster et al. 2013), well below the halo mass of groups and clus- panels). For the QU_M10.7 (SF_M10.7) sample, the cumulative
ters. We then search for the lower stellar mass boundary so that S/N within 80 and 300 kpc is of 7.3 (3.6) and 4.3 (0.9). In the
both samples have a mean mass of log10 (M ∗ /M ) ∼ 10.7, around case of the QU_M10.7 sample, the emission is clearly extended
the turn over (knee) of the stellar mass function, often called M∗ and not centrally peaked; see the red curves in the left panels
(Ilbert et al. 2013). We name these samples QU_M10.7 for the of Fig. 5. For the SF_M10.7 sample, the emission is centrally
quiescent galaxies and SF_M10.7 for the star-forming galaxies. peaked and consistent with a maximal PSF profile; see the blue
The mean redshift of each sample is close: 0.2 for QU_M10.7 curves in the left panels of Fig. 5. With a direct comparison of
and 0.23 for SF_M10.7, that is, a 0.3 Gyr difference. The sSFR of the background-subtracted profiles in Fig. 5 (bottom left panel),
QU_M10.7 is more than 100 times lower than that of SF_M10.7. the difference between QU_M10.7 and SF_M10.7 is made obvi-
Their redshift and stellar mass normalized cumulative distribu- ous. At scales larger than 100 kpc, the quiescent galaxy profile is
tions are shown in Fig. 4; the SF sample is biased towards higher at least two times brighter than that of the star-forming galaxies.
redshift and higher stellar mass compared to the QU sample. Our measurement, using a nearly complete galaxy catalog,
The galaxies in these samples typically reside in dark-matter provides firm observational evidence: at the same mean stellar
haloes of 5−50×1011 M with a mean at ∼ 16×1011 M , where mass of ∼ 5 × 1010 M , star-forming galaxies show significantly
Table 3. Cylindrical projected X-ray luminosity using the M1 mask.
sample log10 (Mvir ) S/N LX [1040 erg s−1 ]

name halo 80 300 XRB max PS R p <80 kpc R p <300 kpc
ALL_M10.7 12.2 5.2 2.3 1.1 +0.3
−0.2 2.8 2.8 ± 0.5 15.4 ± 6.6
Fixed stellar mass selection
QU_M10.7 12.2 7.3 4.3 0.6 +0.2
−0.1 3.1 4.0 ± 0.5 28.7 ± 6.6
SF_M10.7 12.2 3.6 0.9 1.4 +0.4
−0.3 2.6 1.9 ± 0.5 5.7 ± 6.7
Quiescent galaxies
QU_M11.71 15.0 10.9 10.3 7.1 +2.1−1.6 92.2 82.2 ± 7.6 440.7 ± 42.6
QU_M11.58 14.5 8.5 7.5 5.3 +1.6
−1.2 78.0 70.0 ± 8.2 314.3 ± 42.1
QU_M11.54 14.4 3.9 5.0 4.8 +1.4
−1.1 24.8 22.3 ± 5.7 174.1 ± 35.0
QU_M11.35 13.8 8.6 7.0 3.0 +0.9
−0.7 17.1 14.9 ± 1.7 87.5 ± 12.5
QU_M11.2 13.2 8.4 4.4 2.1 +0.6
−0.5 10.0 8.9 ± 1.1 40.5 ± 9.2
QU_M11.04 12.8 6.3 4.5 1.5 +0.4
−0.3 3.7 5.0 ± 0.8 36.7 ± 8.2
QU_M10.88 12.5 6.1 5.2 1.0 +0.3
−0.2 3.8 4.6 ± 0.8 41.5 ± 8.0
QU_M10.73 12.3 5.7 3.1 0.7 +0.2
−0.1 2.9 3.9 ± 0.7 22.4 ± 7.2
QU_M10.53 12.0 5.5 3.5 0.4 +0.1
−0.1 2.4 3.2 ± 0.6 22.5 ± 6.4
star-forming galaxies
SF_M11.25 13.4 3.3 1.7 4.0 +1.3
−1.0 10.3 5.7 ± 1.7 21.8 ± 12.7
SF_M11.12 13.0 2.5 1.5 3.1 +1.0
−0.7 7.9 4.1 ± 1.6 18.8 ± 12.2
SF_M11.05 12.8 2.4 1.7 2.6 +0.8
−0.6 4.8 3.7 ± 1.6 20.6 ± 11.9
SF_M10.99 12.7 3.5 1.3 2.4 +0.7
−0.6 5.3 3.7 ± 1.0 12.1 ± 9.1
SF_M10.9 12.5 3.2 1.2 2.0 +0.6
−0.5 3.3 3.1 ± 1.0 10.1 ± 8.7
SF_M10.86 12.4 3.0 1.1 1.9 +0.6
−0.5 3.5 2.3 ± 0.8 8.5 ± 7.7
SF_M10.74 12.3 2.8 0.6 1.6 +0.5
−0.4 3.3 2.1 ± 0.7 4.3 ± 7.5
SF_M10.63 12.1 2.8 0.7 1.3 +0.4
−0.3 3.1 2.0 ± 0.7 5.3 ± 7.3
SF_M10.52 12.0 1.8 0.6 1.2 +0.3
−0.3 0.8 1.2 ± 0.7 4.2 ± 7.0
SF_M10.41 11.9 2.5 0.8 1.0 +0.3
−0.2 1.5 1.6 ± 0.7 5.8 ± 6.9
SF_M10.3 11.8 1.9 0.5 0.9 +0.2
−0.2 1.8 1.2 ± 0.6 3.4 ± 6.5
SF_M10.18 11.8 2.9 -0.0 0.7 +0.2
−0.1 1.3 1.7 ± 0.6 -0.2 ± 6.2
SF_M10.03 11.6 2.1 0.5 0.6 +0.2
−0.1 1.4 1.1 ± 0.5 2.8 ± 5.9
SF_M9.86 11.5 -0.2 -0.4 0.4 +0.1
−0.1 0.3 -0.1 ± 0.4 -2.3 ± 5.3
SF_M9.64 11.4 1.3 0.3 0.3 +0.1
−0.1 0.3 0.5 ± 0.4 1.5 ± 4.8
Notes. Cylindrical projected luminosity in units of 1040 erg s−1 measured within projected distances of 80 and 300 kpc for each sample with the
M1 mask. In the column S/N we report the S/N of the measurement. We consider to have a "detection" when the S/N is larger than 3, a "hint of
detection" if the S/N is between 1 and 3, and an "upper limit" when the S/N is smaller than 1. These are compared with the XRB model prediction
of Aird et al. (2017). "max PS" is the luminosity obtained when integrating the point-source profiles (gray lines in the figures), which constitutes
an upper limit to the luminosity that can be attributed to point-source emission. log10 (Mvir ) is the estimated mean halo mass obtained using the
stellar to halo mass relation from Moster et al. (2013). Table 4 reports the same quantities as obtained when applying the ‘M3’ mask.
Fig. 4. Redshift and stellar mass cumulative normalized distributions of the QU_M10.7 and SF_M10.7 samples. Although the samples share a
similar mean stellar mass (10.66 and 10.72) and a similar mean redshift (0.2 and 0.23), the underlying distributions differ.

Table 4. Cylindrical projected X-ray luminosity as in Table 3, but using the M3 mask.
sample log10 (Mvir ) S/N LX [1040 erg s−1 ]

name halo 80 300 XRB max PS R p <80 kpc R p <300 kpc
ALL_M10.7 12.2 13.1 4.0 1.1 +0.3 −0.2 11.3 8.8 ± 0.7 29.2 ± 7.2
Fixed stellar mass selection
QU_M10.7 12.2 8.3 4.6 0.6 +0.2
−0.1 4.1 4.9 ± 0.6 31.8 ± 6.9
SF_M10.7 12.2 14.9 3.5 1.4 +0.4 −0.3 16.0 11.2 ± 0.7 25.5 ± 7.4
Quiescent galaxies
QU_M11.71 15.0 11.8 10.4 7.1 +2.1−1.6 111.4 97.4 ± 8.3 459.0 ± 44.0
QU_M11.58 14.5 9.2 7.6 5.3 +1.6
−1.2 83.9 74.2 ± 8.1 319.2 ± 42.0
QU_M11.54 14.4 6.3 5.5 4.8 +1.4
−1.1 57.9 42.3 ± 6.7 199.5 ± 36.6
QU_M11.35 13.8 11.0 7.0 3.0 +0.9 −0.7 25.0 19.5 ± 1.8 89.8 ± 12.8
QU_M11.2 13.2 8.9 4.6 2.1 +0.6
−0.5 11.4 9.9 ± 1.1 44.6 ± 9.7
QU_M11.04 12.8 8.3 4.7 1.5 +0.4
−0.3 5.5 7.0 ± 0.8 40.3 ± 8.6
QU_M10.88 12.5 7.1 5.6 1.0 +0.3
−0.2 5.1 5.8 ± 0.8 47.2 ± 8.5
QU_M10.73 12.3 6.6 3.3 0.7 +0.2
−0.1 4.2 4.8 ± 0.7 25.0 ± 7.5
QU_M10.53 12.0 5.5 3.6 0.4 +0.1
−0.1 2.5 3.3 ± 0.6 23.7 ± 6.7
star-forming galaxies
SF_M11.25 13.4 15.8 5.5 4.0 +1.3 −1.0 62.8 38.4 ± 2.4 78.0 ± 14.2
SF_M11.12 13.0 13.9 5.3 3.1 +1.0 −0.7 55.2 30.6 ± 2.2 73.5 ± 13.8
SF_M11.05 12.8 8.4 3.9 2.6 +0.8
−0.6 26.5 16.8 ± 2.0 51.1 ± 13.0
SF_M10.99 12.7 11.4 3.3 2.4 +0.7 −0.6 24.6 15.6 ± 1.4 33.4 ± 10.0
SF_M10.9 12.5 12.5 5.3 2.0 +0.6 −0.5 18.3 16.0 ± 1.3 52.3 ± 9.9
SF_M10.86 12.4 15.3 4.8 1.9 +0.6 −0.5 21.8 15.1 ± 1.0 41.6 ± 8.6
SF_M10.74 12.3 14.7 3.2 1.6 +0.5 −0.4 21.7 15.4 ± 1.0 26.0 ± 8.2
SF_M10.63 12.1 10.2 2.0 1.3 +0.4 −0.3 14.4 8.9 ± 0.9 15.3 ± 7.8
SF_M10.52 12.0 10.6 3.8 1.2 +0.3 −0.3 10.5 9.1 ± 0.9 30.2 ± 7.9
SF_M10.41 11.9 8.0 1.8 1.0 +0.3
−0.2 7.9 6.2 ± 0.8 13.3 ± 7.4
SF_M10.3 11.8 4.9 0.7 0.9 +0.2
−0.2 5.1 3.3 ± 0.7 4.8 ± 6.8
SF_M10.18 11.8 9.1 1.4 0.7 +0.2
−0.1 4.8 6.4 ± 0.7 9.5 ± 6.6
SF_M10.03 11.6 4.3 1.5 0.6 +0.2
−0.1 3.2 2.5 ± 0.6 9.5 ± 6.3
SF_M9.86 11.5 1.5 1.6 0.4 +0.1
−0.1 1.1 0.7 ± 0.5 9.4 ± 5.8
SF_M9.64 11.4 1.8 1.1 0.3 +0.1
−0.1 0.4 0.7 ± 0.4 5.7 ± 5.1
less projected X-ray emission on large scales in the 0.5–2 keV M1 mask, for QU_M10.7, the emission within 80 kpc is LX =
rest-frame energy range. The possibility of such a difference be- 4.0 ± 0.5 × 1040 erg s−1 , almost an order of magnitude greater
tween passive and star-forming ∼ M∗ galaxies was previously than the corresponding prediction for the unresolved XRB lumi-
suggested by Bregman et al. (2018), hinting at a difference in nosity from Aird et al. (2017) of LX = 6×1039 erg s−1 . Given this,
their evolutionary histories. and the fact that the profile of the QU_M10.7 galaxies is clearly
The shallow slope measured in the QU profile might come extended, we may conclude that this emission is mainly coming
from projection effects (see discussion in Sect. 6.2) due to the from both the hot gas in the CGM of the individual galaxies and
fact that quiescent galaxies tend to live in denser, hotter environ- from the projection of the large-scale environment around them.
ments, as mentioned above (e.g., Velander et al. 2014). Concurrently, the SF_M10.7 emission within 80 kpc amounts to
1.9 ± 0.5 × 1040 erg s−1 , which is compatible within 1σ with that
expected from XRBs (1.4+0.4 −0.3 × 10
40
erg s−1 ). The profile also
With the M3 mask does not appear extended. We are led to conclude that around
star-forming galaxies of a mean mass of log10 (M ∗ /M ) ∼ 10.7,
The right panels of Figure 5 show the projected luminosity pro-
an extended hot gas component is not significantly detected. We
files obtained with the M3 mask. When including the detected
refer the reader to Sect. 6.1 for a more comprehensive discussion
X-ray point sources, the central parts of the profiles increase sig-
of the AGN contamination.
nificantly for the SF profile, mainly because of the contribution
of bright AGN (see Sect. 3.2.1), while for the QU profile the
increase is less noticeable. As opposed to the M1 stack, there
is a hint of extended emission in the SF background-subtracted
profile; see the bottom right panel of Fig. 5. With the M3 mask, where the stack includes all X-ray-bright
detected AGN, the luminosity of the QU sample increases by
Projected luminosity within 80 and 300 kpc 20% compared to the M1 mask, while for the SF sample, it in-
creases by a factor of almost 6. There may be a hint of extended
The integrated projected luminosity (within 80 and 300 kpc emission in the SF profile measured on the larger scales between
apertures) is reported in Tables 3 (M1) and 4 (M3). With the 200 and 300 kpc.
Fig. 5. Measured X-ray radial projected luminosity profiles (0.5–2.0 keV rest-frame) for the quiescent sample "QU_M10.7" (red) and the star-
forming sample "SF_M10.7" (blue) of central galaxies, i.e., at similar median stellar mass around the scale of the Milky Way and Andromeda,
albeit in the range 0.05 < z < 0.3, using the M1 mask (which removes all X-ray-bright AGN in the samples; left column) and M3 mask (which
does not remove X-ray-bright AGN from the sample; right column). Both samples have the same mean stellar mass of ∼ 1010.7 M and mean
redshift of z ∼ 0.2 (but see the underlying distributions in Fig. 4). QU_M10.7 (M1 or M3) shows a clearly extended profile, while SF_M10.7 (M1
or M3) shows a profile compatible with that of a point source convolved with the eROSITA PSF (dashes). The background level for each stack is
given by the dotted lines. The bottom row shows the corresponding background-subtracted profiles.
4.3. Trends with stellar mass expectations: the higher the stellar mass (and therefore the host
halo mass), the brighter the emission and the higher the S/N (Ta-
We further investigate trends as a function of stellar mass and bles 3, 4). For the star-forming samples, with the M1 mask, all
sSFR with the set of samples specified in Table 1. For the qui- profiles except those at the highest mass end are broadly consis-
escent samples, we are limited by the total number of galaxies tent with one another on large scales and are dominated by noise.
available in the catalog and we are not able to create lower mass A difference between profiles arises in the first radial bins; for
bins. For the star-forming samples, we define samples down to example, the mean brightness of the central point source scales
stellar masses of 3 × 109 M . For the M1 mask, we report a de- with the stellar mass. For the star-forming samples with the M3
tection (S/N> 3 in 80 or 300 kpc) for all quiescent samples and mask, the amplitude correlates with stellar mass at all scales, as
only for a handful of the star-forming galaxy samples; see Table in the QU profiles. This follows the expectation that the mean
3. For the M3 mask, we report a detection for all samples ex- AGN luminosity is correlated with the host stellar mass (Aird
cept for the star-forming samples with a stellar mass lower than et al. 2017; Comparat et al. 2019; Georgakakis et al. 2019). Pos-
1010 M ; see Table 4. sible extended emission around star-forming galaxies remains to
The sets of background-subtracted projected luminosity pro- be significantly detected.
files obtained in the M1 and M3 masks are shown in the left
and right panels of Fig. 6, respectively. The qualitative trend ob-
served for the quiescent samples (M1 or M3 mask) is in line with
Fig. 6. Comparison of the background-subtracted projected luminosity profiles in the 0.5–2.0 keV rest-frame band (M1 mask, left panels; M3
mask, right panels) for quiescent (top) and star-forming (bottom) samples as a function of galaxy stellar mass. The star-forming profiles are
compatible with point-source emission profiles (dashes), although we note that the uncertainties are large, in particular for projected radii larger
than 100 kpc. The quiescent profiles appear extended in comparison.
4.4. Scaling between X-ray projected luminosity and stellar been detected in eFEDS. The simulations used start to be incom-
mass: M1 mask plete at stellar masses of 1010 M at z = 0.22, and we therefore
limit the prediction to above this mass.
Figure 7 shows the scaling measured between X-ray luminosity Still within 80 kpc (bottom left panel), only for the highest
and stellar mass within 300 kpc (main panel), in the inner 80 stellar mass quiescent galaxy samples, that is, for stellar mass
kpc (bottom left), and in the shell 80-300 kpc (bottom right), all > 2 × 1011 , corresponding to a halo mass & 5 × 1013 , do we mea-
obtained with the M1 mask applied. Overall, the S/N is highest sure a luminosity that is significantly brighter than the predicted
in the central 80kpc; see Table 3. It decreases when integrat- point-source emission. This is due to the large amount of hot gas
ing to 300kpc. Indeed extending the integration to larger scales, in projection present in galaxy groups and galaxy clusters.
the signal increases marginally while the noise increases much Within 300 kpc, the X-ray luminosity measured around
more, resulting in lower S/N. We find that X-ray luminosity cor- SF samples is consistent with the predicted average point-
relates with mean stellar mass. The trend for star-forming galax- source emission (combination of AGN plus XRBs, dominated
ies is different from that of quiescent galaxies. However, there by AGN emission). For stellar masses above 1011 , the emission
appear to be two regimes in the scaling between X-ray projected is marginally brighter than the expected point-source contribu-
luminosity and stellar mass. The emission from the inner parts tion.
is dominated by point sources (AGN and XRB), while that from
We now consider the results for the 80-300 kpc shell shown
the outer parts is dominated by CGM emission.
in the bottom right hand panel of Fig. 7. For the quiescent sam-
In particular, within 80 kpc, the slope of both SF and QU ple and stellar masses above log M ∗ ∼ 11.2, the measurements
galaxies is similar to (but offset from) that predicted for XRBs, are in good agreement with Anderson et al. (2015)5 . Below
and consistent with the predicted unresolved AGN population log M ∗ ∼ 11.2, the luminosity measured is significantly above
(orange shaded area, Comparat et al. 2019). The AGN popu- that of Anderson et al. (2015). We believe this is due to projec-
lation is predicted using eROSITA mock catalogs filtered on tion effects for the QU sample, which preferentially resides in
X-ray flux and optical magnitude to exclude the X-ray AGN dense and hot environment. We discuss this effect in Sect. 6.2.
that are optically brighter than the magnitude limit of GAMA:
F X < 6.5 × 10−15 erg cm−2 s−1 and r < 19.8. Those simulated 5
The slight discrepancy at the highest mass is likely due to the differ-
AGN could be hosted by GAMA galaxies but would not have ence in aperture: 500c is larger than 300 kpc for a 1015 M halo.

Still in the 80-300 kpc shell, for the star-forming samples, we range of the observational constraints both in terms of demo-
only measure upper limits to the extended emission, except for graphics and inner galaxy properties. For example, the TNG
the three highest stellar mass samples, where, on the other hand, simulations have been shown to accurately reproduce the low-
the error bars extend to a low-luminosity value, meaning only redshift results obtained from Sloan Digital Sky Survey data in
marginal detection, with S/N ∼1.3. relation to: the (g − r) color distributions across galaxy masses
(Nelson et al. 2018); the quiescent fractions of both centrals and
satellites as a function of stellar mass (Donnari et al. 2021);
4.5. Scaling between X-ray projected luminosity and stellar
and the small- and large-scale spatial clustering of galaxies, also
mass: M3 mask
when split by galaxy color (Springel et al. 2018). Third, in both
The relation obtained with the M3 mask is to be interpreted as cases, predictions for the X-ray emission of the gas within and
the sum of all emitting entities: AGN, XRBs, and hot gas aug- around galaxies at z ∼ 0 have already been extensively quanti-
mented by systematic projection effects. In that regard, there is fied across a wide range of masses, galactocentric distances, and
no need to split as a function of projected separation. Figure 8 for star-forming and quiescent galaxies separately (Truong et al.
shows the scaling measured in the inner 300 kpc with the M3 2020; Oppenheimer et al. 2020). For example, in the 0.5–2 keV
mask applied. band, the LX (< R500c ) versus M500c scaling relations of TNG
We predict the AGN, the galaxy group, and the galaxy clus- are within the observational constraints provided by for example
ter population using the eROSITA mock catalog methods (Com- Pratt et al. (2009); Vikhlinin et al. (2009), Sun (2012), Mehrtens
parat et al. 2019, 2020a; Liu et al. 2022c; Seppi et al. 2022). et al. (2012), Lovisari et al. (2015) throughout the 1013−15 M
For the AGN population, we select all X-ray AGN that are op- range (Pop et al. in prep.). As are those of EAGLE (Barnes et al.
tically brighter than the magnitude limit of GAMA: r < 19.8. 2017). Finally, TNG and EAGLE are publicly available (Nelson
These model AGN could be hosted by GAMA galaxies regard- et al. 2019; McAlpine et al. 2016).
less of whether or not they are detected in eFEDS. No filter is Haloes in the simulations are identified with the Friends-of-
applied for the cluster and group population. The black dashes Friends (FoF) algorithm both in the case of TNG and EAGLE:
represent the sum of the AGN and cluster contribution to this re- no a priori cuts are placed to their extent. Galaxies within haloes
lation. The sum of the two empirical models should correspond are identified by searching for gravitationally bound structures
to the relation measured when applying the M3 mask, that is, within the FoF haloes with the SUBFIND algorithm. All this is
when all detected sources are left in the stack. We find that the described in detail in the aforementioned release papers.
luminosity–stellar mass relation is in good agreement with the
models, which demonstrates that the models of Comparat et al.
(2019, 2020a), Seppi et al. (2022) accurately represent the large- 5.1. Extraction of the CGM observables from the simulated
scale structure seen in X-rays. data
At high mass, above 2 × 1011 M , the measurements are First, we construct simulated galaxy samples that are matched
slightly below the model. This is likely due to the fixed 300 to the observed ones by finding, for each galaxy in the GAMA
kpc aperture used, which for these masses is smaller than the set, its simulated equivalent in TNG and EAGLE. The details of
R500c used in the cluster model. For masses below 2 × 1010 M , this procedure are given in Appendix A. In practice, results are
measurements are consistent with the AGN model prediction, shown by averaging across 20 Monte Carlo samples of TNG100
meaning that a detection of CGM emission is unlikely. For stel- and EAGLE galaxies matched to the GAMA sample adopted in
lar masses between 2 × 1010 M and 2 × 1011 M , the positive this paper.
offset between the observations and models is likely related to Second, X-ray photons are not explicitly modeled by the
emission from the CGM and to projection effects. Given the un- TNG and EAGLE simulations. However, the X-ray intrinsic lu-
certainties on the measurement and the large scatter in the model minosity that would be emitted by the simulated galaxies can
prediction, the quantitative assessment of the difference between be derived given the physical properties of the gas (i.e., of the
the observation and the models is a complex undertaking. plasma) returned and predicted by the numerical model. In prac-
tice, here we rely on the mapping between observed eROSITA
5. Comparison with simulated galaxies photon count rates and X-ray fluxes adopted throughout and de-
scribed in Sect. 3. However, we only model the X-ray emission
We elect the IllustrisTNG (hereafter TNG, Pillepich et al. 2018; from the volume-filling gas, that is, we do not attempt to model
Nelson et al. 2018; Naiman et al. 2018; Marinacci et al. 2018; the contamination from XRBs or AGN.
Springel et al. 2018) and the EAGLE simulations (Schaye et al. For any gas cell or gas particle in the simulations, barring the
2015; Crain et al. 2015) as our reference points for the com- star-forming ones and with each one being characterized by a
parison of the results uncovered by eROSITA with the predic- density, temperature, and metallicity, we obtain the [0.5–2] keV
tions from current state-of-the-art cosmological hydrodynamical luminosity assuming a single-temperature Astrophysical Plasma
simulations of galaxies. The reasons are manifold. First, both Emission Code, APEC 3.0.9, as implemented in the XSPEC6
numerical projects provide flagship runs that encompass suffi- (Smith et al. 2001) package. There, we assume an optically thin
ciently large volumes and therefore sufficiently large numbers plasma in collisional ionization equilibrium. For element abun-
of galaxies for the construction of samples comparable to the dances, we employ the table provided by Anders & Grevesse
ones inspected in this paper — there are 6 478 and 3 557 galax- (1989) re-scaled by the overall metallicity of the gas cells7 .
ies with galaxy mass log10 M ∗ > 10 in the TNG100 (TNG) and
Ref-L0100N1504 (EAGLE) boxes, respectively, all at z = 0. 6
https://heasarc.gsfc.nasa.gov/xanadu/xspec/
This would not be the case with zoom-in projects, which for 7
We checked that, by using the individual abundances of nine ele-
massive galaxies are limited to examples of a few to a few tens ments tracked by TNG100 instead of the overall metallicity, the X-ray
of sources. Second, their outcomes have been contrasted to an profiles of 5 × 1010 M galaxies vary by about 0.1 dex and overall by
ever-increasing set of observables, with galaxy populations at negligible amounts in comparison to other systematic uncertainties (as
low and intermediate redshift that are well within (<1 dex) the described in Sect. 5.2).

Fig. 7. X-ray 0.5–2 keV projected luminosity around central galaxies as a function of galaxy stellar mass and split into star-forming (blue symbols
and annotations) and quiescent (red) samples, computed using the M1 mask. Each eFEDS+GAMA detection is indicated with circles, upper
limits with downwards arrows. In the Main panel, we show the luminosity integrated within 300 physical projected kpc. In the Bottom Left
Panel, we show the luminosity integrated within 80 physical projected kpc. In the Bottom Right Panel, the relation is shown for the outer shell
80-300 projected kpc. Gray squares are the measurements from Anderson et al. (2015), computed within R500c (main panel) and within 0.15–1
R500c (bottom right panel). The orange dashed line shows the prediction from the AGN population synthesis model (after excluding sources with
F X > 6.5 × 10−15 erg s−1 cm−2 , as per M1 mask) of Comparat et al. (2019). The orange solid line shows the prediction for the clusters and groups
using the model of Comparat et al. (2020a), i.e., the contribution of hot virialized haloes. The dotted line is the sum of the two.

Fig. 8. Same as Fig. 7 but for the M3 mask (i.e., including detected point sources). Predictions based on the empirical AGN and cluster models
from Comparat et al. (2019, 2020a) (now including sources with F X > 6.5 × 10−15 erg s−1 cm−2 , as per M3 mask) are shown as an orange dashed
line (AGN) and a solid orange line (groups and clusters) and its sum (thick dotted orange line). The agreement between model and observations is
remarkable. For stellar masses between 2 × 1010 M and 2 × 1011 M , the positive offset between the observations and models is likely related to a
combination of emission from the CGM and projection effects.
For each TNG100 and EAGLE galaxy matched to the

GAMA sample, we consider all the gas around it and that be-
longs to their FoF host halo, with no a priori cut to the spatial
extent of the gas. More specifically, we sum up the contribution
to the total X-ray luminosity along the line of sight in a given
projection by accounting for all the gas cells or particles within
the FoF selection: such a line-of-sight projection can span be-
tween many hundreds of kpc to several Mpc depending on the
halo mass. To obtain projected X-ray profiles, we take the min-
imum of the potential as the galaxy center and we determine,
for each individual galaxy, the X-ray luminosity as a function
of radius in a random projection, that is, in a random galaxy
orientation. We mimic the stacking signal of a specific galaxy
subsample by taking the average (mean) of the radially binned
X-ray luminosity values from all the simulated galaxies in the
matched subsample. As the eFEDS stacking profiles are de facto
weighted by the photon counts in each radial bin, we believe that
the mean profiles across individual simulated ones is the closest
approximation to observed stacked signals.
We convolve the mean simulation profiles with the eROSITA

PSF, but we do not remove —from the simulation data— those
unresolved sources that are indeed detected but then masked
Fig. 9. Left: X-ray luminosity projected radial profiles (0.5–2.0 keV rest-frame) for the quiescent QU_M10.7 and the star-forming SF_M10.7
samples (as in Fig. 5 left panel), i.e., for central galaxies with median galaxy stellar mass of about 5 × 1010 M and median redshift 0.20. Their
mass and redshift distributions are shown in Fig. 4. The profiles are compared to the results from the TNG100 (solid) and the EAGLE (dashed)
simulations, consistently matched in stellar mass, sSFR, and redshift. Observed stacking results are compared to the mean simulated profiles (thick
curves). For illustrative purposes, we convolve the mean predicted simulated profiles (thick curves) with the PSF (dashed lines in Fig. 5) and obtain
the thin curves (ideally this should be done on each individual simulated galaxy profile before taking the mean). This provides a rough idea of
how the flux in the inner regions is shifted to larger radii due to the PSF. Right: Complete set of profiles predicted by the simulation: mean (thick
lines) and median (thin lines). The median profiles show where the majority of the simulated profiles are located; the median curve is significantly
lower than the mean curve. Shaded areas represent the systematic uncertainties associated to the extraction of the mocked observable from the
simulations (specifically for TNG100); see text for details. We expect the possible systematic uncertainties to be similar for the EAGLE simulation,
but we refrain from adding shaded areas in order to avoid overcrowding the figure.
(M1) in the eFEDS+GAMA results8 . Moreover, we defer the ner SFR values from the simulations versus those averaged over
task of simulating eROSITA photons, as for example done for the last 100 Myr. In the profiles of Fig. 9, for example, these sys-
the tailored predictions of Oppenheimer et al. (2020), to a fu- tematic choices can amount to X-ray luminosity uncertainties of
ture dedicated paper. With such a full forward modeling into the about 0.5–0.7 dex at 200–300 kpc projected radii.
observational space, we will then also be in the position of repli-
cating, with the simulation data, the exact stacking procedure
adopted here for the eFEDS+GAMA data. This would help in Comparison between 80 and 300 projected kpc
particular to quantify the extent of projection effects along the Focusing on the CGM, at galactocentric distances & 80 kpc (be-
line of sight. yond the eROSITA PSF), the mean X-ray profiles of MW- and
M31-mass galaxies predicted by TNG100 (solid) and EAGLE
5.2. Results from IllustrisTNG and EAGLE (dashed curves) are very similar to one another, despite the dif-
ferent underlying galaxy physics models: they both fall within
Results for the TNG100 and EAGLE galaxies in comparison to approximately 1 dex from the observational results. Moreover,
eFEDS+GAMA inferences are shown in Figs. 9 and 10: there we the profiles of the simulated star-forming versus quiescent galax-
focus, respectively, on the projected radial profiles at the transi- ies are not significantly different from one another in the simu-
tional mass regime of 5 × 1010 M and on the integrated signal lations, with the simulated X-ray atmospheres around quiescent
from the CGM as a function of galaxy stellar mass, that is, inte- 1010.7 M galaxies being less luminous than what the observa-
grating the X-ray luminosity within various apertures: [0 − 80], tions imply in Fig. 9.
[0 − 300], and [80 − 300] projected kpc, with more emphasis The top panel in Fig. 10 shows a comparison between the
on the latter, that is, beyond the typical extent of the eROSITA observed extended (between 80 and 300 projected kpc) X-ray
PSF at the considered redshifts. In both figures, shaded areas luminosity as a function of stellar mass and the simulation pre-
around TNG100 results quantify the systematic uncertainties in dictions for the sample matched in redshift, mass, and sSFR to
the sample-matching procedure. Systematic uncertainties on the the GAMA sample. As for the case of the radial profiles, the
EAGLE simulation are expected to be similar. These are ob- CGM luminosity in the soft-X-ray band as a function of galaxy
tained by encompassing the 5th–95th percentile results when: stellar mass is not too dissimilar between TNG100 and EAGLE,
(i) marginalizing over the 20 Monte-Carlo sampling realizations with similar emission for quiescent and star-forming galaxies at
of the GAMA samples; (ii) using the total galaxy stellar masses fixed stellar mass in both models.
from the simulations versus those within smaller apertures: twice For the quiescent massive galaxies, TNG100 and EAGLE
half stellar mass radius; and (iii) using the instantaneous and in- are in good agreement with the observations, especially at &
8
The latter task requires modeling the X-ray emission not only from 2 × 1011 M . Star-forming galaxies at & 2 × 1011 M have lumi-
the gaseous component but also from AGN and X-ray binaries, which nosities below the TNG and EAGLE predictions: when consid-
is beyond the scope of the current paper. ering the systematic uncertainties due to the matching procedure
Fig. 10. Same as Fig. 7 but with the addition of the predictions from the TNG100 and EAGLE simulations of matched galaxies. Shaded regions
give the order of magnitude of the systematic uncertainty due to the process used to create mock observations. Shaded regions are shown for the
TNG simulation. For EAGLE, shaded regions should have a similar width, but they are not shown so as to not overcrowd the figure. We show the
observed and predicted soft X-ray luminosity from the central 80kpc (bottom left) from the full 300kpc (bottom right) and for the 80 < R p < 300
kpc range (main top panel) as a function of galaxy stellar mass.
(0.5-0.7 dex) and the uncertainties of the observations, the upper with the nondetection of extended emission around star-forming
limit of the measurement is at the limit of being consistent with galaxies, but is significantly lower than the observed luminosity
the lower (simulation) limit. For . 1011 M galaxies, simulations of the quiescent galaxies.
predict a lower luminosity than observed, which is consistent

We note that the simulations predict luminosities that are 6.1. AGN and XRB contamination
possibly greater than that observed by Anderson et al. (2015).
The difference between the simulated curves and the observa- The procedure to mask the event files according to a given cri-
tions gives a sense of the maximal amount of luminosity that terion has a significant impact on the results (see Sect. 4.1). Us-
could be imputed to projection effects for quiescent galaxies. ing ALL, M1, or M3 masks leads to drastically different mea-
surements with distinct physical meanings. In our case, when
all resolved sources are identified and classified, the M1 mask
Comparison below 80 projected kpc is then optimal, as it corresponds to minimization of the con-
tamination from known AGN to the inner projected luminosity
It is manifest from Fig. 9 that both TNG100 and EAGLE pre- profiles. However, the level of residual contamination from un-
dict much brighter atmospheres at small galactocentric distances detected faint AGN remains uncertain.
than what is found with eFEDS+GAMA: up to two orders of To further estimate the AGN contamination, we use the X-
magnitude brighter profiles at < 20 kpc. This discrepancy is ray AGN model described in Comparat et al. (2019) and vali-
also seen at other mass scales, as shown in the bottom panels of dated against eFEDS observations by Liu et al. (2022c). We use
Fig. 10, but to different degrees for star-forming and quiescent the X-ray group and cluster model from Comparat et al. (2020a),
galaxies: the integrated X-ray emission within 80 kpc (as well which was validated against eFEDs observations by Liu et al.
as within 300 kpc) of simulated quiescent galaxies appears to be (2022a), Bulbul et al. (2022). In particular, the simulations accu-
largely consistent with the observed values; on the other hand, rately reproduce the number density of sources as a function of
the simulated star-forming galaxies exhibit gaseous haloes that their flux (logN-logS) for each class separately. We limit these
are systematically brighter in X-rays than the observed counter- light cones to the redshift range 0.05 < z < 0.3 and flux in the
parts across the considered mass range (M∗ = 109.5−11.5 M ). band 0.5–2 keV to F X > 10−17 erg cm−2 s−1 . Figure 8 shows a
The simulations may over-predict the amount of hot gas in the comparison of the observations with the empirical models. The
central regions of the galaxies, in particular for star-forming observations are compatible with the two model lines: at low
galaxies. In fact, the latter possibility is partially at odds with stellar masses (log10 M ∗ < 11.) the total projected emission of
the conclusions of Truong et al. (2020, , their Figure 6) where both quiescent and star-forming galaxies must be contaminated
the X-ray luminosity within the stellar effective radii of TNG by faint AGN9 . The corresponding contamination from unre-
star-forming galaxies appears compatible with analogous mea- solved emission due to XRBs based on the adopted model for the
surements of individual star-forming galaxies in the local Uni- XRB population in galaxies (see discussion in Sect. 3.2.2; Aird
verse by Mineo et al. (2014), although less so with data from Li et al. 2017) is reported in Table 4 (M3 mask) and is shown as
& Wang (2013). It should be noted that the simulation signals colored shaded areas in Fig. 8. Clearly, the putative contribution
only come from the volume-filling gas (with no contributions from XRBs remains subdominant in our stacked subsamples.
from the hot ISM), whereas in the observations, part (if not all) While the comparison with the empirical models provides
of the signal comes from unresolved point sources. As model- a reasonable explanation for the observed projected luminos-
ing and accurately predicting the AGN and XRB X-ray emission ity of both star-forming galaxies (in terms of faint AGN) and
from the simulations is complex, we defer a thorough investi- high-mass quiescent galaxies (in terms of virialized hot haloes),
gation of this discrepancy to future studies, where we will also it does not satisfactorily explain the detection of a significant
replicate the whole analysis pipeline from mock simulation data, projected emission well beyond 80 kpc around low-mass quies-
including the masking process. cent galaxies, which is much more extended than the potential
contamination from point sources alone (see Fig. 2). Additional
sources of genuinely extended X-ray emission surrounding those
lower mass quiescent galaxies are likely needed. These may be
due to the CGM itself, to an incomplete removal of satellites of
6. Discussion luminous clusters and groups, or to projection effects. We dis-
cuss these latter two possibilities below.
The combination of eROSITA’s stable background and good sen-
sitivity at moderate spatial resolution with the availability of a
highly complete spectroscopic galaxy sample from GAMA al- 6.2. Central versus satellites and projection effects
lowed us to detect the faint X-ray emission around galaxies as a
function of their measured stellar masses and sSFRs. A well-defined central galaxy sample is key to enabling the in-
terpretation of our results. Galaxies and their haloes are not gen-
The work presented here shows a clear dichotomy in the erally isolated; they reside in clustered environments (Mandel-
average X-ray emission of star-forming and quiescent galaxies. baum et al. 2006; Gillis et al. 2013). For example, when measur-
While the former are only significantly detected on small scales, ing projected statistics, the signal coming from central galaxies
with a projected luminosity profile consistent with the eROSITA with high clustering (living in a clustered, dense environment)
PSF and an intensity compatible with the faint end of the AGN will be boosted compared to that of isolated galaxies. This is
population (with a possible contribution from XRBs), the latter because of neighboring central galaxies that are in projection
show clearly extended projected emission, with increasing inten- within 300kpc but in three dimensions within a few Mpc. With
sity for larger stellar masses (at least for log10 M ∗ > 11.2). its high completeness, namely of ∼ 98%, the GAMA spectro-
In this section, we first discuss possible systematic effects scopic survey allows an accurate distinction to be made between
that could affect the interpretations of our results, ranging from central and satellite galaxies in different environments; the re-
the estimate of the contribution from undetected faint AGN or moval of satellite galaxies is then straightforward, and this pro-
XRBs (Sect. 6.1) to the accuracy of the sample selection and the cedure should not induce systematic effects provided the GAMA
effect of the 2D projection of the large-scale emission surround-
ing the galaxies (Sect. 6.2). In Sect. 6.3, we move to a discussion 9
We note here that the AGN model prediction is in good agreement
of the comparison with the numerical simulation predictions. with the figures from Aird et al. (in prep) quoted in Sect. 3.2.1.

completeness is uniformly high irrespective of environment or

galaxy properties. Conversely, we also carried out the same mea-
surements in a sample where satellite galaxies are included and
found that the SF profiles remained unchanged, while the QU
profiles were systematically 25% brighter at all scales. This
is consistent with the fact that, on average, quiescent satellite
galaxies live in hot and dense environments (e.g., Velander et al.
2014; Hudson et al. 2015).
To illustrate the projection effects, we measured the clus-
tering (two-point correlation function) of the galaxies consid-
ered here using two statistics: w p (r p ) and ξ(s). ξ(s) is the angle-
averaged ("isotropic") 3D redshift space correlation function and
w p (r p ) is the line-of-sight projected correlation function (for de-
tailed definitions see e.g., Davis & Peebles 1983). The luminos-
ity profiles obtained are subject to projection, similarly to the
w p (r p ) clustering statistics; the environment in which galaxies
live has an effect on projected statistics. To investigate the impact
of projection effects, we measured the clustering of the galaxy
samples considered, before and after the central selection algo-
rithm is applied; see Fig. 11. In the 3D correlation function of
the central galaxies sample (ξ(s), top panel, red or blue squares),
there is a clear cut-off scale (∼ 400 kpc, identified by vertical
dashes) below which the clustering signal diminishes, while it
is still increasing in the complete sample (central plus satellite
galaxies, red and blue circles, respectively). This means that the
selection procedure for central galaxies works as expected.
When measuring the projected clustering w p (r p ), the cut-off
scale at ∼ 400 kpc is no longer visible; see the bottom panel of
Fig. 11. The projected clustering of the full sample and the cen-
tral sample both extend to very small scales. We clearly see a
projection effect here, which is due to the fact that galaxies live
in crowded, clustered environments. Therefore, when measuring
the projected luminosity profiles, the emission around individual
galaxies is indistinguishable from the emission from the envi-
ronment of the galaxies. Both star-forming and quiescent galaxy
samples are subject to a similar projection effect, as shown in
Fig. 11. However, from the difference seen between the star-
forming and quiescent stacked profiles, we infer that the environ-
ment of 5 × 1010 M (and lower) quiescent galaxies is dense and
hot, while that of star-forming galaxies needs to be either less
dense, cooler, or both. This is in agreement with the findings of
Velander et al. (2014). Accurately quantifying the strength of the Fig. 11. Clustering of the samples selected. In each panel we show the
projection effect and how much it biases the projected luminosity results for the quiescent central galaxies (CEN QU, red squares), the
profile as a function of scale is key to determining how much of star-forming centrals (SF CEN, blue squares), all the quiescent galaxies
the measured emission comes from the CGM. Future simulations (ALL QU, purple circles), and all the star-forming ones (ALL SF, dark
(with full fledged light cones) and further modeling in order to blue circles). Top Panel: Three-dimensional correlation function. The
jointly interpret halo occupation distributions (constrained with turn over due to the central selection function is clear, as illustrated by
clustering and galaxy-galaxy lensing measurements) and the lu- the vertical dashed line. Bottom Panel: Projected correlation function:
minosity profile will hopefully shed light on this uncertainty. no turnover is visible.
Because of this, it is unclear whether or not the CGM around
5×1010 M galaxies is detected in our sample. The eROSITA
eFEDS survey is not deep enough to provide a complete cen- files are significantly brighter than the simulated median pro-
sus of galaxy groups in the redshift range studied here, and so files: see Fig. 9, thick versus thin solid (TNG100) and dashed
applying the ALL mask will remove only a large part of the en- (EAGLE) curves. Conversely, this implies that the CGM signal
vironmental effects, but not all. of the median galaxy is not properly captured by the observed
mean stacks, which are instead biased high. By analyzing the
6.3. Insights from the comparison between observations and TNG100 results, we can quantify that the 10% most luminous
simulation results sources alone can bias high the X-ray signal at ∼ 100 kpc by al-
most 1 dex. This means that caution is necessary in the interpre-
The X-ray profiles and luminosities predicted at large galacto- tation of the results. Chiefly, given the small volumes considered
centric distances by TNG100 and EAGLE are in general agree- in this analysis, it is hard to tightly control the population of rare
ment with observations, and so this allows us to use the sim- objects in the same manner in the observations and simulations.
ulations to also gather insights that are not obtainable other- Samples collected over larger volumes will undoubtedly help in
wise. It is important to point out that the simulated mean pro- addressing this in the future. In the meanwhile, we notice that
in TNG100 and EAGLE, no matched galaxy that enters in this 7. Summary and outlook
analysis resides in a halo more massive than 3 × 1013 –1014 M ,
that is, throughout the whole sample and even though more mas- We quantified the X-ray emission around a large sample of qui-
sive haloes are present in the simulated volumes. On the other escent (star-forming) galaxies at 0.05 < z < 0.3 in the stellar
hand, there are at least 289 (X-ray-detected) galaxy clusters and mass range 2 × 1010 – 1012 M (3 × 109 – 6 × 1011 M ) (Fig. 1).
groups in the 60 deg2 of eFEDS and GAMA (Liu et al. 2022a). To do so, we stacked the eROSITA eFEDS events around cen-
tral GAMA galaxies to obtain projected luminosity profiles out
Also, when comparing results from the two simulations, to hundreds of project kpc (Figs. 3, 5, and 6). As anticipated, the
TNG100 and EAGLE, we find overall very consistent mean pro- stacking method is successful in overcoming the flux limit in the
files, whereas median profiles differ. The median discrepancy is X-ray observations 10 .
related to the different underlying galaxy physics models of the For quiescent (passive) galaxies, the X-ray profiles are
simulations. This highlights the fact that the median profiles are clearly extended throughout the available mass range (e.g., Fig.
sensitive to the galaxy evolution model, while the mean profiles 5); however, the measured profiles are likely biased high due to
are more sensitive to a sparse population of rare and luminous projection effects emanating from the fact that quiescent galax-
galaxies. For example, at the transitional mass regime of Fig. 9, it ies live in dense and hot environments (Fig. 11). Around star-
is evident that EAGLE atmospheres are brighter than TNG ones forming galaxies with < 1011 M , the X-ray-stacked profiles are
in X-rays, which is consistent with the notion that the SMBH compatible with unresolved sources and are consistent with the
feedback in TNG is more ejective than in EAGLE (Davies et al. expected emission from faint AGN and XRBs (Fig. 5). Only for
2020; Truong et al. 2021) and that EAGLE exhibits lower quies- the most massive star forming samples (≥ 1011 M ) is there a
cent fractions than TNG at this mass scale (Donnari et al. 2021). hint of detection of extended emission.
We measure for the first time the average relation between
At face value, in the 80-300kpc range, the X-ray luminosity mean projected X-ray luminosity within various apertures and
of the CGM around 1010.2−10.8 M star-forming galaxies is con- stellar mass separately for quiescent and star-forming galaxies
sistent with the upper limits inferred from eFEDS+GAMA data. (Fig. 7). We find that the relation is different for the two galaxy
The simulation results suggest that a detection of the hot CGM populations: high-mass (≥ 1011 M ) star-forming or quiescent
around star-forming MW/M31-mass galaxies may soon be ob- galaxies follow the expected scaling of virialized hot haloes (see
tained (see Sect. 7). orange solid line in Fig. 7). Lower mass star-forming galaxies
Importantly, it is also apparent from both Figs. 9 and 10 show a less prominent luminosity that is also more weakly de-
that TNG100 and EAGLE do not predict a significant difference pendent on stellar mass, consistent with empirical models of the
in the CGM X-ray signals between quiescent and star-forming population of weak AGN.
galaxies, as seen in the data. Instead, the mean profiles and In particular, when measuring the mean projected X-ray lu-
the CGM luminosities on large scales from the simulations fall minosity in a 300 kpc aperture while excluding X-ray-bright
closer to the observed SF eROSITA results. This is likely due to AGN detected as point sources (M1 mask, Fig. 7 main panel and
the aforementioned projection effects. Table 3), for quiescent galaxies with a stellar mass larger than
2 × 1010 M , we detect (S/N larger than 3) a faint X-ray emis-
Also, it should be noted that these results are not necessarily sion partly originating from hot gas. For star-forming galaxies
in contradiction with results reported in Section 1 which sug- with a stellar mass of larger than 6 × 1010 M , we report a hint
gest brighter X-ray atmospheres around simulated star-forming of detection (S/N between 1 and 3). For star-forming galaxies
galaxies than around quiescent ones (Truong et al. 2020; Op- with a stellar mass in the range 3 × 109 –6 × 1010 M we measure
penheimer et al. 2020). First, here we are comparing mean pro- upper limits (S/N smaller than 1). We find similar results and de-
files and not median properties of galaxies; second, we focus on tection levels when we measure the average projected X-ray lu-
somewhat higher redshift, z ∼ 0.2 versus z ∼ 0.1 (1.1 Gyr differ- minosity in a 80–300 kpc aperture (M1 mask, Fig. 7 bottom right
ence); and finally, the stacked profiles of Figure 9 are not from panel and Table 3), which hence characterize the extended emis-
volume-limited samples and reflect the mean results of galaxies sion beyond the galaxy itself: we detect X-ray extended emis-
in rather wide mass and redshift bins and with different stellar sion around quiescent galaxies at all probed masses, while for
mass and redshift distributions in the two star-forming and qui- star-forming galaxies we find upper limits in the 3 × 109 –1011
escent bins, with the former exhibiting a greater representation M range and a hint of detection at higher masses.
of lower mass and higher redshift galaxies (see Fig. 4). We additionally measure the average projected X-ray lumi-
nosity in a 300 kpc aperture while keeping all detected point
As pointed out above, for more accurate comparisons be-
sources (M3 mask, Fig. 8 and Table 4). We find good agreement
tween observations and simulations, a more thorough forward
with empirical models of the X-ray cosmic web of AGN and
model of the simulation data would be needed, for example us-
clusters from Comparat et al. (2019, 2020a). This reinforces the
ing sixte (Dauser et al. 2019). However, focusing on the dis-
notion that a robust assessment of the properties of hot haloes
crepancy at face value for the 5 × 1010 M quiescent galaxies
in Milky-Way-like galaxies (and smaller) requires accurate re-
(Fig. 9), we can argue that under-luminous atmospheres in the
moval of weak AGN contaminants.
simulations in comparison to data may indicate that (some of
When comparing our results with state-of-the-art numerical
the) simulated gaseous haloes may (a) be under-dense, (b) be
simulations (IllustrisTNG and EAGLE), we find overall consis-
characterized by lower temperatures, and/or (c) be less enriched
tency in the average emission at large (> 80 kpc) scales and
than (some of the) galaxies in the Universe, or more specifically
at masses ≥ 1011 M , but disagreement at smaller scales, where
in eFEDS. This discrepancy could be due to a SMBH feedback
brighter-than-observed compact cores are predicted (Fig. 9). The
implementation that is more ejective than in reality, at least in
some cases —more ejective SMBH feedback would imply more 10
For example, the QU_M10.53 sample, with a S/N of about 5.5, has
substantial outflows and hence a more substantial clearing of the a luminosity in the inner 80 kpc of 3.2×1040 erg s−1 at redshift 0.19,
halo of both hot and metal-enriched gas— or that is not effective corresponding to a flux of 3.1×10−16 erg cm−2 s−1 . This is about 20
enough at heating up the halo gas. times fainter than the flux limit of quoted by Brunner et al. (2022).

Table 5. Forecast S/N for future eROSITA-based experiments. and their interplay with the star formation and virialization pro-
cesses, as well as feedback processes from AGN.
Area Ng
t Acknowledgements. This work is based on data from eROSITA, the soft X-
[deg2 ] [106 ] exp
h texp,eFEDS i ξS/N ray instrument aboard SRG, a joint Russian-German science mission supported
eRASS1 + public 2021 21k 1.2 0.1 1.8 by the Russian Space Agency (Roskosmos), in the interests of the Russian
Academy of Sciences represented by its Space Research Institute (IKI), and
eRASS:8 + WAVES 1k 0.7 0.8 4.0 the Deutsches Zentrum für Luft- und Raumfahrt (DLR). The SRG spacecraft
eRASS:3 + DESI BGS 5k 7 0.3 7.7 was built by Lavochkin Association (NPOL) and its subcontractors, and is op-
eRASS:8 ‘Legacy’ 21k 20 0.8 21.2 erated by NPOL with support from the Max Planck Institute for Extraterrestrial
Physics (MPE). The development and construction of the eROSITA X-ray in-
Notes. Average S/N improvement ratio with respect to the strument was led by MPE, with contributions from the Dr. Karl Remeis Obser-
eFEDS+GAMA09 sample (ξSNR ) for four possible combinations of vatory Bamberg & ECAP (FAU Erlangen-Nürnberg), the University of Ham-
eROSITA all-sky survey depths and low-redshift galaxy spectroscopic burg Observatory, the Leibniz Institute for Astrophysics Potsdam (AIP), and the
Institute for Astronomy and Astrophysics of the University of Tübingen, with
samples. The Area (in square degrees) reported is the approximate over- the support of DLR and the Max Planck Society. The Argelander Institute for
lap between the German eROSITA all-sky survey (Sunyaev et al. 2021) Astronomy of the University of Bonn and the Ludwig Maximilians Universität
and the galaxy samples; Ng is the number of galaxies with z . 0.3 in Munich also participated in the science preparation for eROSITA. The eROSITA
texp
this area (in millions), while h texp,eFEDS i is the average ratio between the data shown here were processed using the eSASS software system developed
by the German eROSITA consortium. GAMA is a joint European-Australasian
corresponding eRASS exposure and the eFEDS one. project based around a spectroscopic campaign using the Anglo-Australian Tele-
scope. The GAMA input catalog is based on data taken from the Sloan Digital
Sky Survey and the UKIRT Infrared Deep Sky Survey. Complementary imag-
simulations also do not predict the clear differentiation that we ing of the GAMA regions is being obtained by a number of independent sur-
observe between quiescent and star-forming galaxies in our sam- vey programmes including GALEX MIS, VST KiDS, VISTA VIKING, WISE,
ples (Fig. 10). Herschel-ATLAS, GMRT and ASKAP providing UV to radio coverage. GAMA
is funded by the STFC (UK), the ARC (Australia), the AAO, and the partic-
This work is a stepping stone towards a better understand- ipating institutions. The GAMA website is http://www.gama-survey.org/ . This
ing of the hot phase in the CGM, which holds a key role in project acknowledges funding from the European Research Council (ERC) under
the regulation of star formation. In the next decade, by com- the European Union’s Horizon 2020 research and innovation programme (grant
bining eROSITA with upcoming spectroscopic galaxy surveys agreement No 865637).
(e.g., DESI, SDSS-V, 4MOST: DESI Collaboration et al. 2016;
Kollmeier et al. 2017; de Jong et al. 2019; Merloni et al. 2019;
Finoguenov et al. 2019), properties of the CGM and their rela- References
tion to AGN should be unraveled. The eROSITA all-sky survey
(eRASS) data cover more than 100 times more extragalactic sky Ahumada, R., Prieto, C. A., Almeida, A., et al. 2020, ApJS, 249, 3
Aird, J., Coil, A. L., & Georgakakis, A. 2017, MNRAS, 465, 3390
than eFEDS and therefore offer the opportunity to improve on Anders, E. & Grevesse, N. 1989, Geochim. Cosmochim. Acta, 53, 197
the analysis presented here. Anderson, M. E., Churazov, E., & Bregman, J. N. 2016, MNRAS, 455, 227
Table 5 provides a rough estimate of the average S/N im- Anderson, M. E., Gaspari, M., White, S. D. M., Wang, W., & Dai, X. 2015,
provement ratio (ξS/N ) based on a full-sample for a few se- MNRAS, 449, 3806
lected combinations of different eRASS depths (from the single- Bahar, Y. E., Bulbul, E., Clerc, N., et al. 2022, A&A, 661, A7
Baldry, I. K., Robotham, A. S. G., Hill, D. T., et al. 2010, MNRAS, 404, 86
pass eRASS1 to the full eight-pass eRASS:8) and extragalactic Barnes, D. J., Kay, S. T., Bahé, Y. M., et al. 2017, MNRAS, 471, 1088
spectroscopic surveys. We assume, for simplicity, that S/N ∝ Basu-Zych, A. R., Hornschemeier, A. E., Haberl, F., et al. 2020, MNRAS, 498,
1651
p
Ng texp , where Ng is the number of galaxies in the range
Behroozi, P., Wechsler, R. H., Hearin, A. P., & Conroy, C. 2019, MNRAS, 488,
z . 0.3 and texp is the average eROSITA exposure. We com- 3143
pute the ratio of the S/N, ξS/N , obtainable with these experi- Behroozi, P. S., Wechsler, R. H., & Conroy, C. 2013, ApJ, 770, 57
ments to the eFEDS+GAMA09 baseline one, for which we take Bellstedt, S., Robotham, A. S. G., Driver, S. P., et al. 2021, MNRAS, 503, 3309
Ng,eFEDS =35,521. We provide estimates for the following com- Bellstedt, S., Robotham, A. S. G., Driver, S. P., et al. 2020, MNRAS, 498, 5581
Bogdán, Á., Bourdin, H., Forman, W. R., et al. 2017, ApJ, 850, 98
binations: Bogdán, Á., Forman, W. R., Kraft, R. P., & Jones, C. 2013a, ApJ, 772, 98
Bogdán, Á., Forman, W. R., Vogelsberger, M., et al. 2013b, ApJ, 772, 97
– eRASS1 with a compilation of existing spectroscopic cata- Bogdán, Á., Vogelsberger, M., Kraft, R. P., et al. 2015, ApJ, 804, 72
logs (‘eRASS1 + public 2021’); Bregman, J. N., Anderson, M. E., Miller, M. J., et al. 2018, 862, 3
– An intermediate stage that combines a three-pass X-ray sur- Brunner, H., Liu, T., Lamer, G., et al. 2022, A&A, 661, A1
vey (eRASS:3) with the DESI ‘Bright Galaxy Survey’ (BGS, Bulbul, E., Liu, A., Pasini, T., et al. 2022, A&A, 661, A10
Ruiz-Macias et al. 2020); Churazov, E., Khabibullin, I., Lyskova, N., Sunyaev, R., & Bykov, A. M. 2021,
A&A, 651, A41
– The full-depth eRASS:8 combined with the 4MOST WAVES Comparat, J., Eckert, D., Finoguenov, A., et al. 2020a, The Open Journal of
wide survey (Driver et al. 2016); Astrophysics, 3, 13
– A putative ‘Legacy’ sample, combining eRASS:8 with all Comparat, J., Merloni, A., Dwelly, T., et al. 2020b, A&A, 636, A97
spectroscopic samples available in about a decade (early Comparat, J., Merloni, A., Salvato, M., et al. 2019, MNRAS, 487, 2005
2030s). Crain, R. A., McCarthy, I. G., Frenk, C. S., Theuns, T., & Schaye, J. 2010, MN-
RAS, 407, 1403
The average S/N improvement with respect to the sample Crain, R. A., Schaye, J., Bower, R. G., et al. 2015, MNRAS, 450, 1937
Dauser, T., Falkner, S., Lorenz, M., et al. 2019, A&A, 630, A66
analyzed here ranges from about a factor of 1.8 to more than a Davies, J. J., Crain, R. A., Oppenheimer, B. D., & Schaye, J. 2020, MNRAS,
factor of 20. Further improvements in the eROSITA data pro- 491, 4462
cessing, energy, and PSF calibration will likely also increase Davies, L. J. M., Lagos, C. d. P., Katsianis, A., et al. 2019, MNRAS, 483, 1881
the significance of the detection, and enable a combination of Davis, M. & Peebles, P. J. E. 1983, ApJ, 267, 465
spatial and spectral analysis. Thanks to those improvements, we de Jong, R. S., Agertz, O., Berbel, A. A., et al. 2019, The Messenger, 175, 3
DESI Collaboration, Aghamousa, A., Aguilar, J., et al. 2016, arXiv e-prints,
should be able to measure the temperature and metallicity of the arXiv:1611.00036
hot CGM, as well as its density and pressure profile. Hopefully, Donnari, M., Pillepich, A., Nelson, D., et al. 2021, MNRAS, 506, 4760
this will shed light on how hot haloes are created and energized, Driver, S. 2021, Nature Astronomy, 5, 852

Driver, S. P., Bellstedt, S., Robotham, A. S. G., et al. 2022, MN-

RAS[arXiv:2203.08539]
Driver, S. P., Davies, L. J., Meyer, M., et al. 2016, in Astrophysics and Space
Science Proceedings, Vol. 42, The Universe of Digital Sky Surveys, ed. N. R.
Napolitano, G. Longo, M. Marconi, M. Paolillo, & E. Iodice, 205
Finoguenov, A., Merloni, A., Comparat, J., et al. 2019, The Messenger, 175, 39
Fukugita, M. & Peebles, P. J. E. 2004, ApJ, 616, 643
Fukugita, M. & Peebles, P. J. E. 2006, ApJ, 639, 590
Georgakakis, A., Comparat, J., Merloni, A., et al. 2019, MNRAS, 487, 275
Georgakakis, A., Nandra, K., Laird, E. S., Aird, J., & Trichas, M. 2008, MNRAS,
388, 1205
Gillis, B. R., Hudson, M. J., Erben, T., et al. 2013, MNRAS, 431, 1439
Goulding, A. D., Greene, J. E., Ma, C.-P., et al. 2016, ApJ, 826, 167
Hudson, M. J., Gillis, B. R., Coupon, J., et al. 2015, MNRAS, 447, 298
Ilbert, O., McCracken, H. J., Le Fèvre, O., et al. 2013, A&A, 556, A55
Kelly, A. J., Jenkins, A., & Frenk, C. S. 2021, MNRAS, 502, 2934
Kollmeier, J. A., Zasowski, G., Rix, H.-W., et al. 2017, arXiv e-prints,
arXiv:1711.03234
Koushan, S., Driver, S. P., Bellstedt, S., et al. 2021, MNRAS, 503, 2033
Lehmer, B. D., Basu-Zych, A. R., Mineo, S., et al. 2016, ApJ, 825, 7
Li, J.-T., Bregman, J. N., Wang, Q. D., Crain, R. A., & Anderson, M. E. 2016,
ApJ, 830, 134
Li, J.-T. & Wang, Q. D. 2013, MNRAS, 428, 2085
Liske, J., Baldry, I. K., Driver, S. P., et al. 2015, MNRAS, 452, 2087
Liu, A., Bulbul, E., Ghirardini, V., et al. 2022a, A&A, 661, A2
Liu, T., Buchner, J., Nandra, K., et al. 2022b, A&A, 661, A5
Liu, T., Merloni, A., Comparat, J., et al. 2022c, A&A, 661, A27
Lovisari, L., Reiprich, T. H., & Schellenberger, G. 2015, A&A, 573, A118
Mandelbaum, R., Seljak, U., Kauffmann, G., Hirata, C. M., & Brinkmann, J.
2006, MNRAS, 368, 715
Marinacci, F., Vogelsberger, M., Pakmor, R., et al. 2018, MNRAS, 480, 5113
McAlpine, S., Helly, J. C., Schaller, M., et al. 2016, Astronomy and Computing,
15, 72
Mehrtens, N., Romer, A. K., Hilton, M., et al. 2012, MNRAS, 423, 1024
Merloni, A., Alexander, D. A., Banerji, M., et al. 2019, The Messenger, 175, 42
Mineo, S., Gilfanov, M., Lehmer, B. D., Morrison, G. E., & Sunyaev, R. 2014,
MNRAS, 437, 1698
Moster, B. P., Naab, T., & White, S. D. M. 2013, MNRAS, 428, 3121
Moster, B. P., Naab, T., & White, S. D. M. 2018, MNRAS, 477, 1822
Naiman, J. P., Pillepich, A., Springel, V., et al. 2018, MNRAS, 477, 1206
Nelson, D., Pillepich, A., Springel, V., et al. 2018, MNRAS, 475, 624
Nelson, D., Springel, V., Pillepich, A., et al. 2019, Computational Astrophysics
and Cosmology, 6, 2
Oppenheimer, B. D., Bogdán, Á., Crain, R. A., et al. 2020, ApJ, 893, L24
Pillepich, A., Nelson, D., Hernquist, L., et al. 2018, MNRAS, 475, 648
Planck Collaboration, Ade, P. A. R., Aghanim, N., et al. 2013, A&A, 557, A52
Planck Collaboration, Ade, P. A. R., Aghanim, N., et al. 2016, A&A, 594, A24
Popesso, P., Magnelli, B., Buttiglione, S., et al. 2012, arXiv e-prints,
arXiv:1211.4257
Pratt, G. W., Croston, J. H., Arnaud, M., & Böhringer, H. 2009, A&A, 498, 361
Predehl, P., Andritschke, R., Arefiev, V., et al. 2021, A&A, 647, A1
Robotham, A. S. G., Bellstedt, S., Lagos, C. d. P., et al. 2020, MNRAS, 495, 905
Ruiz-Macias, O., Zarrouk, P., Cole, S., et al. 2020, Research Notes of the Amer-
ican Astronomical Society, 4, 187
Salvato, M., Wolf, J., Dwelly, T., et al. 2022, A&A, 661, A3
Schaye, J., Crain, R. A., Bower, R. G., et al. 2015, MNRAS, 446, 521
Seppi, R., Comparat, J., Bulbul, E., et al. 2022, arXiv e-prints, arXiv:2207.09242
Smith, R. K., Brickhouse, N. S., Liedahl, D. A., & Raymond, J. C. 2001, ApJ,
556, L91
Springel, V., Pakmor, R., Pillepich, A., et al. 2018, MNRAS, 475, 676
Sun, M. 2012, New Journal of Physics, 14, 045004
Sunyaev, R., Arefiev, V., Babyshkin, V., et al. 2021, A&A, 656, A132
Tauris, T. M. & van den Heuvel, E. P. J. 2006, Formation and evolution of com-
pact stellar X-ray sources, Vol. 39, 623–665
Thorne, J. E., Robotham, A. S. G., Davies, L. J. M., et al. 2021, MNRAS, 505,
540
Truong, N., Pillepich, A., Nelson, D., Werner, N., & Hernquist, L. 2021, MN-
RAS, 508, 1563
Truong, N., Pillepich, A., Werner, N., et al. 2020, MN-
RAS[arXiv:1911.11165]
Tumlinson, J., Peeples, M. S., & Werk, J. K. 2017, ARA&A, 55, 389
Velander, M., van Uitert, E., Hoekstra, H., et al. 2014, MNRAS, 437, 2111
Vikhlinin, A., Burenin, R. A., Ebeling, H., et al. 2009, ApJ, 692, 1033
Vulic, N., Hornschemeier, A. E., Haberl, F., et al. 2022, A&A, 661, A16
White, S. D. M. & Frenk, C. S. 1991, ApJ, 379, 52
Wolf, J., Salvato, M., Coffey, D., et al. 2020, MNRAS, 492, 3580
Zhu, G. B., Comparat, J., Kneib, J.-P., et al. 2015, ApJ, 815, 48

Appendix A: Matching procedure between observed one; this occurs only for ∼ 400 (∼ 800) GAMA galax-
simulated and observed galaxy samples ies when matching to TNG (EAGLE). We repeat this procedure
for the overall sample 20 times for both TNG and EAGLE so as
To directly compare the results of the simulations for example to obtain 20 different Monte Carlo simulation samples matched
from Truong et al. 2020 and Oppenheimer et al. 2020 to those to the observational ones.
from our observations is a complex undertaking, first and fore- Quiescent and star-forming matched simulated galaxies are
most because they are given at different redshifts (z ∼ 0 vs. me- then divided with the fixed boundary cut at log10 (sSFR) = −11,
dian z ∼ 0.2), which correspond to a 2.5 Gyr difference in the as in Sect. 2.2. Bins in galaxy stellar mass as described in Table 1
age of the Universe. Second, the galaxy samples from the sim- are extracted from the overall matched samples.
ulations at any given epoch are volume limited whereas those As the TNG and EAGLE volumes (∼ 106 Mpc3 ) are smaller
in GAMA are magnitude limited, with progressively larger frac- than that of the galaxy sample considered here ( ∼ 107 Mpc3 ), the
tions of more luminous and massive galaxies at higher redshifts same simulated galaxies may be matched to multiple observed
(see Figure1). galaxies. In these cases, the direction from which the galaxy is
We therefore constructed simulated galaxy samples that are seen (when projected on sky) is changed to avoid repetitions
matched to the observed ones by finding, for each galaxy in the of the exact same X-ray profiles when simulating the stacking
GAMA set, its simulated equivalent in TNG and EAGLE. We procedure. In the matched TNG100 samples, about 30% of the
used simulated data from the TNG100 and Ref-L0100N1504 objects come from simulated galaxies that are not unique. This
runs, which both encompass approximately 100 comoving Mpc fraction increases to 43% for galaxies M∗ > 1011 M . The
a side. We consider only central galaxies, in line with the choices nonunique fraction is significantly higher for the EAGLE sim-
of Sect. 3.1. However, we do not replicate the methods adopted ulation (∼ 75%) because it has fewer snapshots (only 3) in the
for the observed data, but simply identify as centrals those galax- observed 0.05 < z < 0.3 range. Therefore, on average, an EA-
ies that occupy the lowest level of the gravitational potential of GLE galaxy could be matched to four or five observed galaxies.
the simulated haloes identified with the FoF algorithm. In TNG, This could be an issue in the case of outliers, that is, particularly
we exclude objects with SubhaloFlag ≡ 0 (see Nelson et al. bright systems, such as merging galaxies, because in such cases
2019, for more details). The analogous objects between the sim- their contribution to the predicted mean average X-ray emission
ulation outputs and the GAMA sample are found in the redshift, is artificially overestimated by many factors because of the du-
galaxy stellar mass, and sSFR parameter space. plication in the matching. In this regard, we noticed a merging
The match in redshift is done to the closest available simu- galaxy in EAGLE in the mass bin M∗ ∼ 1010.7 M that would
lation snapshot. In the 0.05 < z < 0.3 range considered here, have been matched in different projections to five different ob-
there are 18 data snapshots in TNG and 3 in EAGLE. As a fidu- served galaxies, causing the EAGLE mean surface brightness to
cial choice, we assume that the stellar mass of GAMA galax- be about one order of magnitude higher in the central regions
ies inferred by Bellstedt et al. (2020, 2021) is close to the total in comparison to the case when the system is omitted. Hence,
mass of a galaxy, i.e., is compatible with the sum of all the stel- to minimize the effect of the matching duplication in the fidu-
lar particles that are gravitationally bound to a simulated galaxy. cial EAGLE results presented throughout, we opt to count the
However, this is probably not the case for centrals in massive contribution of that merging galaxy only once.
haloes and so we bracket this uncertainty by also matching the
samples assuming that the GAMA stellar masses correspond to
the sum of the stellar particles that are gravitationally bound to a
simulated galaxy and within twice its stellar half-mass radius.
We also assume that the inferred SFR of the observed galax-
ies is equivalent to the instantaneous SFR of the gas cells of a
simulated galaxy, again within twice its stellar half-mass radius.
In fact, GAMA SFRs are averaged over 100 Myr: as shown by
Donnari et al. 2021, at least for TNG, whether SFRs are de facto
instantaneous or averaged across the past 1 Gyr and whether they
are measured across varying galaxy apertures should not be im-
portant for the main purposes of this paper, that is, for splitting
galaxies into quiescent and star-forming ones at low z and as
long as the rule for the grouping is the same; see below. How-
ever, even within the quiescent and star-forming samples, the X-
ray CGM signal may show trends with the actual level of SFR,
and hence we also show results adopting 100 Myr-averaged val-
ues measured throughout the galaxy body. We manually re-label
sSFR values below 10−15 yr−1 to exactly 10−15 yr−1 in both ob-
servation and simulations data in order to avoid being affected
by highly uncertain or numerical resolution-dependent SFR es-
timations.
For every GAMA galaxy at a given redshift (see Table 1), a
simulated analog is searched for in the stellar mass–sSFR plane
at the closest snapshot by randomly selecting a galaxy within
a rectangle whose widths are equivalent to the uncertainties as-
sociated to the measured values of stellar mass and SFR on a
galaxy-by-galaxy basis. If no simulated galaxy is found within
such limits, we simply take the closest simulated galaxy to the

3

Uploaded by

Copyright:

Available Formats

3

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3

Uploaded by

Copyright:

Available Formats

Astronomy & Astrophysics manuscript no.

main ©ESO 2022

The eROSITA Final Equatorial Depth Survey (eFEDS):

1. Introduction component of the intergalactic and halo gas, supposedly com-

Article number, page 1 of 23

agreement with the mean of the measurements made on individ-

4.1. Detection in the complete stack

Table 3. Cylindrical projected X-ray luminosity using the M1 mask.

sample log10 (Mvir ) S/N LX [1040 erg s−1 ]

Article number, page 9 of 23

sample log10 (Mvir ) S/N LX [1040 erg s−1 ]

Article number, page 12 of 23

Article number, page 13 of 23

Article number, page 14 of 23

For each TNG100 and EAGLE galaxy matched to the

We convolve the mean simulation profiles with the eROSITA

Article number, page 17 of 23

Article number, page 18 of 23

completeness is uniformly high irrespective of environment or

Article number, page 20 of 23

Article number, page 21 of 23

Driver, S. P., Bellstedt, S., Robotham, A. S. G., et al. 2022, MN-

Article number, page 22 of 23

You might also like