Mon. Not. R. Astron. Soc.

000, 000000 (0000)

Printed 25 September 2014

(MN LATEX style file v2.2)

Unifying X-ray Scaling Relations from Galaxies to Clusters

Michael E. Anderson1? , Massimo Gaspari1 , Simon D. M. White1 , Wenting Wang2 ,
Xinyu
Dai3
1
Max-Planck Institute for Astrophysics, Garching bei Muenchen, Germany
for Computational Cosmology, Department of Physics, University of Durham, South Road, Durham, DH1 3LE, UK
3 Homer L. Dodge Department of Physics and Astronomy, University of Oklahoma, Norman, OK 73019, USA

arXiv:1409.6965v1 [astro-ph.CO] 24 Sep 2014

2 Institute

25 September 2014

ABSTRACT

We examine a sample of 250000 locally brightest galaxies selected from the Sloan
Digital Sky Survey to be central galaxies within their dark matter halos. These were
previously stacked by the Planck Collaboration to measure the Sunyaev-Zeldovich
signal as a function of central galaxy stellar mass. Here, we stack the X-ray emission
from these halos using data from the ROSAT All-Sky Survey. We detect emission
across almost our entire sample, including emission which we attribute to hot gas
around galaxies spanning a range of 1.2 dex in stellar mass (corresponding to nearly
two orders of magnitude in halo mass) down to M = 1010.8 M (M500 1012.6 M ).
Over this range, the X-ray luminosity can be fit by a power-law, either of stellar mass
or of halo mass. A single unified scaling relation between mass and LX applies for
galaxies, groups, and clusters. This relation has a steeper slope than expected for
self-similarity, in contrast to the YSZ -M500 relation, showing the importance of nongravitational heating. If this heating is predominantly due to AGN feedback, the lack
of a break in our relation suggests that AGN feedback is tightly self-regulated and
fairly gentle, in agreement with recent high-resolution simulations. Our results are
consistent with earlier measurements of the LX -LK relation for elliptical galaxies and
of the LX -M500 relation for optically-selected galaxy clusters. However, our LX -M500
relation lies more than a factor of two below most previous relations based on X-rayselected cluster samples. We argue that optical selection offers a less biased view of
the LX -M500 relation.
Key words: galaxies: clusters: general, galaxies: groups: general, galaxies: haloes,
X-rays: galaxies, X-rays: galaxies: clusters

INTRODUCTION

Galaxy cluster scaling relations are fundamental tools for

connecting cluster astrophysics to cosmology. These scaling relations are typically expressed within the framework
of the self-similar model (Kaiser 1986), which predicts that
power-law relations should exist between basic properties of
clusters (mass, luminosity, temperature, etc.) unless some
physical process occurs to produce a characteristic scale and
break the self-similarity.
In this work the scaling relation of particular interest is
the LX -M relation. This relation connects the X-ray luminosity of the hot gas in a galaxy cluster (ideally the bolometric luminosity) to the total mass M of the cluster. The
self-similar prediction for this relation is that LX M 4/3

email: [email protected]

c 0000 RAS

(Sarazin 1986) assuming the X-ray luminosity is dominated

by thermal bremsstrahlung.
A number of studies have estimated the LX -M relation for galaxy clusters (e.g. Stanek et al. 2006, Maughan
2007, Rykoff et al. 2008, Vikhlinin et al. 2009, Pratt et al.
2009, Mantz et al. 2010, Planck Collaboration 2011, Wang
et al. 2014). They typically find slopes in the range of 1.62.0, which is significantly steeper than the predicted value
of 4/3. Some of this discrepancy may be due to redshift evolution and radial temperature profiles (Vikhlinin et al. 2009,
Kravtsov & Borgani 2012), but the majority is usually attributed to non-gravitational heating, either by changing the
entropy of the gas (Evrard 1990) or through its indirect impact on the baryon fraction of the halo (Mushotzky & Scharf
1997).
Non-gravitational heating is also thought to play an important role in the formation and evolution of galaxies and

Anderson et al.

galaxy groups, and there are a number of theoretical predictions for the behavior of the X-ray luminosity of these
lower-mass systems1 . One common prediction from largescale cosmological simulations (e.g. Sijacki et al. 2007, Puchwein et al. 2008, Fabjan et al. 2010, McCarthy et al. 2010) is
a steepening of the decline in LX as the halo mass decreases,
below T 1 keV (M500 1013.5 M ). This break is caused
by active galactic nucleus (AGN) feedback. Specifically, as
pointed out by Planelles et al. (2014) and Gaspari et al.
(2014a), the thermal blast prescription for AGN feedback in
these simulations raises the cooling time of the intragroup
gas above the Hubble time, converting these galaxy groups
into non-cool-core objects. In contrast, gentler self-regulated
mechanical feedback (acting through outflows which induce
X-ray buoyant bubbles, weak shocks, and the uplift of lowentropy gas and metals) preserves the cool core and therefore
produces no break (Gaspari et al. 2011, Gaspari et al. 2012).
We can therefore learn more about AGN feedback if we
can extend observations of cluster X-ray scaling relations
down to the regime of galaxies and galaxy groups. This has
posed a formidable challenge, however, due to the lower Xray luminosity of these less massive systems. Previous observational studies of galaxy groups disagree about whether
a 1 keV break exists in the LX -T relation (Ponman et al.
1996, Helsdon & Ponman 2000a, Mulchaey 2000, Osmond
& Ponman 2004, Sun et al. 2009), and the LX -M relation
has only recently begun to be explored at these scales (e.g.
Bharadwaj et al. 2014, Lovisari et al. 2014). So far, systematic studies of the LX -M relation in low-mass groups and
isolated galaxies have proven beyond the reach of current
X-ray telescopes.
Other X-ray properties and scaling relations have been
studied in galaxy-mass halos, however. The closest analogue
to the LX -M relation is probably the LX -LB relation (or for
more modern observations, the LX -LK relation) in elliptical galaxies, which relates the stellar content of an elliptical
galaxy to the X-ray properties of its hot gaseous halo. The
slope of this relation is also interesting, since it gives clues
about the processes which govern the hot gaseous halo.
Unfortunately, the slope of this relation is difficult to
measure either as a function of LB or LK . It seems to depend
somewhat on the environment of the galaxies (Mulchaey &
Jeltema 2010) as well as on the degree of rotational support
of the galaxy (Sarzi et al. 2013). Even worse, the scatter
in both LX -LB and LX -LK is extremely large (an order
of magnitude at 1011 L ; Boroson et al. 2011) which makes
sample selection a particularly difficult issue. Sample selection is also a major issue for the LX -M relation, since nearly
every study (with the notable exceptions of Dai et al. 2007,
Rykoff et al. 2008, and Wang et al. 2014) relies on X-rayselected clusters. This raises the issue of Malmquist bias2 ,
which can have a very significant effect on the inferred LX M relation (Stanek et al. 2006).

Generally these studies focus on the LX -T relation instead of

the LX -M relation, but temperature and mass are tightly linked
in hydrostatic equilibrium, so these relations have roughly the
same qualitative behavior.
2 Malmquist bias is the tendency to overestimate the average
luminosity of a population when observing a flux-limited sample,
since the more luminous objects are easier to detect.

In this study, we take a significant step towards alleviating all of the above issues. We examine an optically-selected
sample of central galaxies, which is sensitive to different selection effects than X-ray selected samples. The sample contains 250000 galaxies, each selected to be the most luminous galaxy in its dark matter halo, spanning halos from
intermediate-mass clusters down to galaxies about half the
mass of the Milky Way. The selection criteria and basic properties of our sample are described in Section 2.
With our sample of uniformly-selected central galaxies,
we can make the a uniform comparison of the X-ray luminosities of halos across galaxies, galaxy groups, and galaxy
clusters for the first time. In order to make this comparison
most effectively, and to detect the X-ray emission in lowermass halos, we employ a stacking technique. This technique
is detailed in Section 3. We present the results of our stacking
in Section 4. Finally, in Section 5 we measure the LX -M500
relation and in Section 6 we measure the LX -M relation.
For both relations, a single power-law describes the data
from galaxy clusters all the way down to Milky-Way-mass
halos. This suggests that the two relations are actually the
same. We discuss the implications of this result in Section
7.

SAMPLE

In this paper we examine the sample of locally brightest

galaxies (LBGs) introduced in Planck Collaboration (2013)
(hereafter P13). Full details of the sample selection are presented in that work, but we briefly summarize them here.
The LBGs are chosen from the New York University
Value-Added Galaxy Catalog (Blanton et al. 2005) based on
Data Release 7 from the Sloan Digital Sky Survey (SDSS).
The goal is to select a population of central galaxies in
dark matter halos. To do this, P13 selected galaxies with
extinction-corrected Petrosian r-magnitude r < 17.7 and
redshift z > 0.03 which are brighter in r than any other
galaxy within 1 projected Mpc and 1000 km s1 . To account for potential satellites without spectroscopic redshifts,
locally brightest galaxies were also required not to have any
galaxies in the SDSS photometric redshift 2 catalog (Cunha
et al. 2009) with brighter r-magnitudes projected within
1 Mpc and having a photometric redshift with a greater
than 10% chance of being consistent with the redshift of the
locally brightest galaxy. The total sample contains 259567
LBGs.
For each LBG, stellar masses have been estimated from
the SDSS photometry by Blanton & Roweis (2007). Following P13, we divided the sample into 20 bins, logarithmically
spaced in stellar mass from log (M /M ) = 10.0 to log
(M /M ) = 12.0 (there are 9660 LBGs which lie outside
this range in stellar mass and are discarded). The distribution of galaxies across stellar mass is presented in Figure 1.
The sample is approximately flux-limited: SDSS photometric completeness for galaxies at r = 17.7 is essentially 100%,
although spectroscopic completeness is lower due to fiber
collisions where two or more galaxies fall within the same
SDSS fiber. Checking the FGOT parameter from the ValueAdded Galaxy Catalog, we find approximately 91% completeness for the locally brightest galaxy sample. We also
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

30000

all
z cut
z+R cuts

25000

Ngal

20000
15000
10000
5000
10.0

10.5

11.0
log M*

11.5

12.0

Figure 1. Stellar mass distribution of locally brightest galaxies.

The red line shows all 249907 LBGs within our stellar mass limits.
The blue line shows the 239389 LBGs within our redshift limits.
The black line shows the 201011 LBGs that fall within our redshift
limits and do not overlap with the edge of one of the slices in the
RASS.

note that the selection criteria may introduce subtle and

complex biases compared to a purely flux-limited sample.
In order to better understand these biases, we employ
a catalog of simulated locally brightest galaxies originally
generated for P13. The procedure for generating this catalog starts with the Millennium simulation (Springel et al.
2005) which follows the evolution of cosmic structure within
a box of comoving side length 500 h1 Mpc. Halo merger
trees are complete for subhalos above a mass resolution of
1.7 1010 h1 M . The original Millennium simulation is
based on WMAP1 cosmology, and the rescaling technique
of Angulo & White (2010) has been adopted to convert it
to the WMAP7 cosmology. A semi-analytic galaxy formation prescription (Guo et al. 2011) has been applied to these
simulated halos. The galaxy formation parameters have been
adjusted to fit several statistical observables at z = 0, such
as the luminosity, stellar mass and correlation functions of
galaxies. We project the simulation box along the z-axis,
and assign every galaxy an artificial redshift based on its
line-of-sight distance and peculiar velocity, i.e., parallel to
the z-axis. In this way we can select a sample of galaxies
from the simulation using isolation criteria exactly analogous to those used for the locally brightest galaxy sample
from SDSS.
These simulations allowed P13 to match the stellar content of galaxies to dark matter halos from the Millennium
Simulation, as well as to estimate the fraction of satellite
galaxies or other failure modes in the sample. They find
that the fraction of locally brightest galaxies which are centrals is over 83% across the sample. About 2/3 of the satellite LBGs are in fact brighter than the central galaxy, while
the other 1/3 are either more than 1 Mpc from the central
or are offset in velocity by more than 1000 km s1 . They
also are able to estimate the relationship between M and
halo mass for galaxies in each stellar mass bin. They find
that the satellite contamination biases the mean halo mass
upwards fairly significantly in the lower-mass bins, but in
c 0000 RAS, MNRAS 000, 000000

general these satellite LBGs are significantly offset from the

center of their massive halos and therefore do not appreciably bias the stacked signal. We therefore treat the terms
locally brightest galaxy and central galaxy as approximately interchangeable throughout this work.
This simulated catalog also allows estimation of effective halo masses for each stellar mass bin down to log
M = 10.8 10.9. P13 computed halo masses by assuming
self-similar scaling in the YSZ -M500 relation and assuming
the Arnaud et al. (2010) pressure profile for the hot gas.
Since X-ray luminosity scales with the projected squared
density instead of the projected density, the effective halo
masses are slightly higher for X-ray emission. Also, one of
our conclusions is that the LX -M relation is not self-similar,
so we instead use our best-fitting relation when computing the effective halo masses. In Appendix A, we detail the
derivation of our effective halo masses and compare with the
masses used in P13; our effective halo masses are similar to
those in P13, and the halo masses seem to be fairly robust.
The stellar masses, halo masses, and other derived
quantities in both P13 and this work rely on an assumed
WMAP7 cosmology (Komatsu et al. 2011), with m =
0.272, = 0.728, and H0 = 70.4 km s1 Mpc1 . We
generally work with halo masses expressed as M500 , the
mass within a radius R500 which encloses a mean density
500 times the critical density. We estimate R500 from M500
using the relation
4
3
R500
c
(1)
3
The critical density of the Universe, c , has a redshift
dependence, and we compute it at the mean distance dL of
the galaxies in the bin, which is defined in section 3.1. This
redshift effect is small, since the highest effective redshift of
any bin is 0.29. For each stellar mass bin, the effective values
of M500 and R500 we adopt are listed below in Table 1, along
with dL and several other properties.
M500 500

STACKING

We examine stacked images of these galaxies using data from

the ROSAT All-Sky Survey (RASS). The dataset is a shallow all-sky survey in the soft X-rays. The stacking software
was originally generated by Dai et al. (2007), and the stacking procedure here is very similar to that used in Anderson
et al. (2013) (hereafter ABD13). In brief, for each of the 20
samples of LBGs binned by stellar mass (Table 1), we follow
this procedure:
1. Extract a RASS image of each LBG and its surroundings, extending in physical space out to Rextract . Also extract
the corresponding RASS exposure map for each field.
2. Exclude bright point sources from each image, using
a combination of cross-matching with the ROSAT Bright
Source Catalog and Faint Source Catalog (Voges et al. 1999)
and automatic flagging of pixels with large count rates.
3. Add together each of the RASS images to produce a
stacked image in physical space. Also add together each of
the exposure maps, weighting the individual exposure maps
by a factor proportional to the angular area covered by the
field.
4. Construct an empirical point spread function (psf)

Anderson et al.
Table 1. Parameters for Locally Brightest Galaxies in Stellar Mass Bins
log M
(M )

log M500
(M )

R500
(kpc)

Rextract
(kpc)

zmin

zmax

Ngal

Nstacked

dL
(Mpc)

11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9
10.7-10.8
10.6-10.7
10.5-10.6
10.4-10.5
10.3-10.4
10.2-10.3
10.1-10.2
10.0-10.1

14.56
14.41
14.29
14.08
13.90
13.70
13.51
13.29
13.09
12.91
12.75
12.60
12.34
12.20
12.09
11.99
11.90
11.82
11.75
11.69

938
838
772
665
584
504
437
372
320
280
248
222
182
164
150
140
131
124
117
111

3000
3000
2500
2000
2000
1500
1000
1000
1000
1000
1000
1000
1000
1000
1000
1000
1000
1000
1000
1000

0.15
0.15
0.10
0.10
0.10
0.10
0.07
0.07
0.06
0.06
0.05
0.05
0.05
0.05
0.05
0.05
0.04
0.04
0.04
0.04

0.40
0.40
0.35
0.35
0.30
0.30
0.25
0.25
0.25
0.25
0.20
0.20
0.18
0.18
0.18
0.18
0.14
0.14
0.12
0.12

44
145
573
1624
3664
7160
11615
16871
22085
26026
28325
27866
25309
21619
17328
13221
9862
7499
5223
3848

36
114
455
1326
2967
5970
9615
14194
18430
21583
22689
22490
20041
17168
13729
10353
7425
5693
3821
2912

1492.7
1433.7
1275.7
1088.5
1009.8
915.6
788.6
714.6
633.9
592.0
523.1
485.3
455.4
428.2
407.0
386.3
339.1
325.1
308.0
298.6

Properties of LBGs in each of our 20 stellar mass bins. Stellar masses are measured from
SDSS photometry, halo masses are estimated from simulations as described in Appendix A,
and R500 is estimated from the halo mass using equation 1. Rextract , zmin , and zmax are
defined in section 3.1. Ngal is the total number of LBGs in the bin, and Nstacked is the
number which pass our additional selection criteria in section 3.1 and are included in the
stacks. The mean distance of the stacked galaxies, dL , is described in section 3.1 as well.

for the stacked image, by stacking at least 10000 known

point sources from the ROSAT Faint Source Catalog and
ROSAT Bright Source Catalog using the same apertures and
weighting scheme as the LBGs.
5. Perform aperture photometry on the stacked image
in order to estimate average background-subtracted count
rates. We use two apertures: a circle with radius corresponding to the average R500 for the LBGs in the stack, and an
annulus extending from (0.15 1) R500 .
6. Convert the measured count rates into an average
X-ray luminosity for each stack.
Stacking in physical space instead of angular space is
a natural choice for this project, since it allows us to study
the galactic and circumgalactic emission as a function of radius around locally brightest galaxies. However, this choice
does complicate the analysis, in three major ways. First, it
means the aperture will vary in angular size from galaxy to
galaxy, which will naturally over-weight the nearest galaxies, since their apertures will be the largest and therefore
enclose the most photons. Second, the point spread function becomes more complicated, since the psf is typically
defined in angular units. Finally, the varying apertures also
introduce complications for the nearest and the most distant
galaxies in each stellar mass bin. In the rest of this section,
we describe our procedure in more detail, including how we
account for the complications listed above.

3.1

Additional Details

The first issue in our procedure, similar to ABD13, is selection of minimum and maximum redshifts zmin and zmax for

locally brightest galaxies in each sample. We choose these

redshifts to enclose as much of the sample as possible, with
the additional requirement that zmin to be about a third of
zmax . This excludes the handful of galaxies with the lowest
redshifts, preventing them from dominating the overall signal due to the larger apertures necessary to enclose these
galaxies. A handful of the most distant galaxies are also excluded, although their apertures are so small that they add
comparatively little signal. The redshift distributions of the
LBG samples and the locations of zmin and zmax can be seen
in Figure 2.
The next issue is the selection of Rextract . The natural
choice for Rextract is 1000 kpc, since this is the projected
radius within which the isolation criterion is defined. We
choose this radius for most stellar mass bins, but the largest
LBGs have values of R500 close to 1000 kpc, so we extend
Rextract to larger values for these systems. As Rextract increases, we become increasingly limited by the size of the
individual fields (6.4 6.4 ) which comprise the RASS. In
ABD13 we were able to model the so-called effective vignetting introduced when the extraction region extended
beyond the edge of one of these fields, but for our simpler
aperture photometry in this work, we exclude any galaxy
from our sample if the Rextract aperture overlaps with the
nearest edge of its RASS field. The effect of this exclusion
can be seen in the blue histogram in Figure 1.
With zmin , zmax , and Rextract defined, we can now extract RASS images and exposure maps for each LBG. We
extract images in the 0.5-2.0 keV band, and compute the
angular size of Rextract for each galaxy using its SDSS redshift and assuming WMAP7 cosmology. We create images
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

Figure 2. Redshift distributions of locally brightest galaxies in

each of our 20 stellar mass bins. In each panel, the range of log
M* is indicated, and the redshift cuts are denoted by the vertical
red lines. The total number of LBGs in each bin varies, as shown
in Figure 1.

which are 200 200 pixels in size. This means that, for the
most distant galaxies in each bin, the effective pixel size is
often a bit smaller than the ROSAT 1 pointing accuracy
of 6. In practice this is not a major concern: the most distant galaxies contribute the fewest photons, our empirical
psf technique should account for this effect where it exists,
and we analyze the images using aperture photometry with
large apertures so that single-pixel accuracy is not particularly important.
The next issue is how to treat bright point sources that
randomly lie within the images. We use a similar approach as
ABD13, masking out any portion of an image with a source
listed in the ROSAT Bright Source catalog (BSC) or the
ROSAT Faint Source catalog (FSC). Many locally brightest
galaxies are listed in these catalogs, however, so for each bin
we impose a minimum count rate (using the count rates for
these sources listed in the catalogs) for sources to be excluded. These count rates are tabulated in Appendix B, and
correspond to luminosities (if the source were at the mean
distance of the LBGs in each bin) which are more than an
order of magnitude above the mean LBG luminosity. Additionally, we exclude any observation that has more than 10
counts in a single pixel, although this too has little effect
on the results. We explore the effect of changing the minimum count rate and removing the 10 count threshold in
Appendix B, and find that these changes have little effect
on the results.
To generate the stacked images, we add the individual
images without any weighting. Thus the stacked images contain integer numbers of photons and are subject to the usual
Poisson statistics. We generate a composite exposure map
as well, by stacking the individual exposure maps, but we
weight the exposure maps in the stack in order to account
for the differences in aperture sizes due to the different distances of the LBGs. In general, locally brightest galaxies are
either undetected or marginally detected in the RASS, so
c 0000 RAS, MNRAS 000, 000000

the images are background-dominated and the appropriate

weighting is therefore proportional to the area subtended by
the image3 . In Figure 3, we present the stacked images in
each of our 20 stellar mass bins.
Next, we construct empirical psfs to match each stacked
image. To do this, we stack at least 104 point sources from
the combination of the ROSAT Faint Source Catalog and the
ROSAT Bright Source Catalog (Voges et al. 1999) matched
in count rate distribution to the expected count rate distribution of the galaxies in each stack, and we assign apertures
to each source matching the aperture distribution used for
the corresponding stack as well. For more details on this process, see ABD13. These psfs are used as part of the aperture
photometry, as described below.
In Figure 4, we present radial surface brightness profiles
for each of our bins, centered on the locally brightest galaxy.
The empirical psf is also indicated, normalized to the count
rate in the center of the image. We also indicate the fit to the
background with the horizontal line and the adopted value
of R500 with the vertical line in each image.
We analyze these images using aperture photometry.
We specify two different source apertures: the total aperture is a circle extending out to R500 and the circumgalactic or CGM aperture is an annulus extending from
(0.15 1) R500 . The background is determined from an
annulus extending from 1.5 R500 to Rextract . As Figure 4
shows, the psf is much more compact than R500 , so we apply no psf correction to the total aperture, but psf effects
can be important when performing the CGM photometry.
For this region, we compute the count rate within the CGM
annulus and subtract from it the count rate within 0.15R500
scaled by the fraction of the power in the psf which falls
within the CGM annulus.
We also allow for uncertainty in R500 , which stems primarily from uncertainty in Mhalo for each bin. We estimate
this uncertainty using a catalog of simulated locally brightest galaxies in dark matter halos, described in more detail
in section 4. For the eight lowest-mass bins, which are not
included in the simulated catalog, we use the abundancematching relation of Moster et al. (2010) to compute M500 ,
and we propagate the uncertainties in this relation through
into the estimation of R500 .
Finally, we estimate the average X-ray luminosity of the
LBGs in each bin. We start with the results of the aperture
photometry, which are background-subtracted fluxes within
two apertures, in the observer-frame 0.5-2.0 keV band, in
units of counts/second/galaxy. We then k-correct this count
rate into the rest-frame 0.5-2.0 keV band, using the average
(aperture-area-weighted) luminosity distance of the LBGs
in each bin in order to estimate the effective redshift. These
effective redshifts are small (maximum z = 0.29) and so the
k-corrections are also small. To compute the k-correction,
we estimate the virial temperature of the LBGs in the bin
based on the M500 -T relation in Sun et al. (2009). We use
their relation calibrated across the full sample of galaxy
groups and clusters (Tier 1 + 2 + clusters). This relation

This assumes a spatially uniform X-ray background, which is an

acceptable approximation for this analysis since any variations in
the background are averaged out over the large number of images
in each stack.

Anderson et al.
11.9-12.0

1.5 Mpc

0.2

0.4

0.6

0.8

10

12

19

20

21

22

24

10

16

11

12

13

14

1.2

12 2

10

18

10

12

14

16

12

26

27

28

17

18

19

20

21

10.3-10.4

18

15

16

17

22

10

11

14

20

0.5

1.5

22

23

24

12

25

18

19

14

20

3.5

21

14

15

16

17

15 5

18

23

20

10

24

21

22

11

10

11.2-11.3

11

12

13

10.8-10.9

0.5 Mpc

25 17 18 19 20 21 22 23 24 25 26 27 28

23

10.4-10.5

0.5 Mpc

11

12

13

14

15

10.1-10.2

10

0.5 Mpc

10.5-10.6

0.5 Mpc

10.9-11.0

22

19

4.5

11.6-11.7

1.0 Mpc

11.3-11.4

0.5 Mpc

10.2-10.3

13

0.5 Mpc

17

26

2.5

0.5 Mpc

10.6-10.7

0.5 Mpc

1.6

11.0-11.1

0.5 Mpc

25

1.4

11.7-11.8

1.25 Mpc

11.4-11.5

10.7-10.8

0.5 Mpc

10

0.8

0.5 Mpc

14

23

0.6

11.1-11.2

0.5 Mpc

18

0.4

0.75 Mpc

0.5 Mpc

0.2

11.5-11.6

1.0 Mpc

1.2 0

11.8-11.9

1.5 Mpc

16

17

18

19

4.5

5.5

6.5

20

21

10.0-10.1

0.5 Mpc

7.5

8.5

Figure 3. Stacked 0.5-2.0 keV RASS images of locally brightest galaxies in each of our 20 stellar mass bins. In each image, log M is
noted at the top right and R500 is indicated with the black circle. Note that the physical scale and colorbar vary across these images.
Each image has been smoothed with a Gaussian 3-pixel kernel.

c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

Figure 4. Azimuthally averaged surface brightness profiles of the stacked images in Figure 3. In each image, log M is noted at the
top right and R500 is indicated with the black dashed vertical line. The black dashed horizontal line indicates the fit to the background.
The red line is the shape of the empirical psf in each figure, normalized to match the value of the central bin, and is shown here for
illustration of the extended nature of many of these profiles.

gives an average value for the temperature with the core

of the group/cluster excluded, but we neglect the difference
between this value and the emission-weighted temperature
with the core included. We then estimate the k-correction
assuming an APEC model (Smith et al. 2001) with that temperature for the X-ray emission4 . We then convert the count
rate into flux by multiplying by temperature-dependent con-

version factor computed from the APEC model using the

WEBPIMMS5 tool. In the bins with log M < 10.8, Xray binaries can be expected to contribute significantly to
the emission as well, so we set this conversion factor to
1.2 1011 erg count1 cm2 , which we showed in ABD13
is fairly insensitive to the contribution of X-ray binaries to
the hot gas emission. Finally, the luminosity is given by

In the less-massive halos, X-ray binaries will also contribute

significantly to the emission, but here the redshifts are so small
that the k-correction is already negligible.
c 0000 RAS, MNRAS 000, 000000

http://heasarc.gsfc.nasa.gov/cgibin/Tools/w3pimms/w3pimms.pl

Anderson et al.

Table 2. Conversion Factors for Computing the Luminosity of

Locally Brightest Galaxies
log M
(M )

kT
(keV)

11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9
10.7-10.8
10.6-10.7
10.5-10.6
10.4-10.5
10.3-10.4
10.2-10.3
10.1-10.2
10.0-10.1

5.0
4.0
3.4
2.5
1.9
1.5
1.1
0.8
0.6
0.5
0.4
0.3
0.2
0.2
0.1
0.1
0.1
0.1
0.1
0.1

0.88
0.90
0.91
0.95
0.96
0.98
1.00
1.02
1.03
1.05
1.06
1.06
1.07
1.07
1.06
1.06
1.06
1.06
1.05
1.05

cflux
(1011 erg count1 cm2 )

Cbolo

1.2
1.2
1.2
1.1
1.1
1.1
1.0
1.0
0.9
1.0
1.0
1.1
1.2
1.2
1.2
1.2
1.2
1.2
1.2
1.2

3.2
2.8
2.5
2.1
1.8
1.4
1.3
1.2
1.1
1.1
1.1
1.2
1.2

Conversion factors and derived average hot gas temperatures

for locally brightest galaxies. The mean temperature of the
hot gas, kT , is described in section 3.1. The factor k is the
estimated k-correction factor to convert the observed 0.5-2.0 keV
flux into the rest frame, and the factor cflux is the estimated
(temperature-dependent) conversion factor between counts and
erg cm2 . The bolometric correction Cbolo is used in sections 4
and 5 to study scaling relations; this analysis is only performed
on the uppermost 12 bins so no bolometric correction is ever
applied to the lowest 8 bins.
2

LX = 4dL fX . In Table 2 we present all of these correction

factors for each observation.
In Appendix C, we perform null tests to verify that our
stacking procedure has no intrinsic bias, and in Appendix
D we stack simulated data in order to verify that we can
recover injected LX -M500 relations correctly.

RESULTS

Here we present the results of the stacking procedure detailed in Section 3. In Table 3 and Figure 5, we present the
measured luminosities. The power-law in Ltotal seems unbroken down to stellar masses as low as log M = 10.7 10.8.
The flattening below this mass is analogous to the flattening
observed in P13, and is largely due to the X-ray signal becoming too faint to distinguish from the background. However, unlike with the Sunyaev-Zeldovich (SZ) effect in P13,
there are additional sources of X-ray emission which become
important in low-mass galaxies. While bright AGNs are not
a major concern (see Appendix B), we do have to account
for low-mass X-ray binaries (LMXBs) and high-mass X-ray
binaries (HMXBs).
To do this, we make use of established scaling relations.
LMXB emission is correlated with stellar mass (or equivalently with K-band stellar luminosity), so we can relate log
M to the expected LMXB signal. Converting the Boroson

et al. (2011) relation to the 0.5-2.0 keV band, the expected

LMXB luminosity is approximately

LLMXB 3 1028

LK
erg s1
L

(2)

For this figure we assume a very conservative K-band

M/L ratio of 0.8 (the M/L ratio has a very weak dependence
on stellar mass and on morphology, which we neglect) to
convert this relation into a function of M .
HMXB emission is correlated with star formation rate
(SFR). To estimate the SFR, we use the B30 indicator
from the NYU Value-Added Galaxy Catalog (Blanton et al.
2005, Blanton & Roweis 2007) which estimates the fraction
of the galaxys stellar mass which has formed in the past 300
Myr. We multiply this by M for each galaxy and divide
by 300 Myr to get a SFR. Blanton and Roweis note that
their B300 indicator underestimates the true star formation
rate; in particular they show a 0.7 dex offset between results
inferred using their indicator and the results of Hopkins et al.
(2003). We therefore correct the inferred SFR upwards by
a factor of 0.7 dex to account for this offset. We then use
the Mineo et al. (2012) relation (shifted to the 0.5-2.0 keV
band):

LHMXB 1.4 1039

M
erg s1
1M yr1

(3)

We can get a very crude estimate of the expected

HMXB signal by averaging together LHMXB for all the galaxies in each bin. The fraction of blue galaxies increases quickly
as stellar mass decreases, so this actually leads to a prediction that LHMXB should increase towards less-massive galaxies in Figure 5.
Note that essentially all the observed X-ray emission in
the lowest 8 bins can be explained as the sum of LMXB
and HMXB emission. On the other hand, the highest 12
bins are barely contaminated by this emission. The most
contaminated bin, log M = 10.8 10.9, has an estimated
21% of its emission due to XRBs.
This is also consistent with the results for LCGM . In
seven of the lowest eight bins, the measured luminosity is
consistent with or less than zero. This is exactly what we
would expect if the signal were dominated by LMXBs and
HMXBs from within the galaxy, since the emission from
the galaxy is not included in our CGM annulus. The uppermost 11 bins all show secure detections of extended emission
which is almost certainly hot gas. In the log M = 10.810.9
bin, the measured luminosity lies on the same power-law,
but the detection of hot extended emission is less secure
(see next paragraph).
We also can estimate the intrinsic variation in LX
within each bin from the sample error on the mean. We
do this using bootstrapping. For each bin, we generate 100
bootstrapped samples (with replacement) and stack each
sample, applying the same analysis as the real data, and
deriving a measurement for LX . The standard deviation of
these 100 bootstrapped measurements is plotted for each
bin with thin error bars. The sample error on the mean decreases as the number of LBGs in each bin increases, but it
still gives some sense of the intrinsic scatter in LX among
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

kT (keV)

0.2 0.3 0.4 0.6

1.0

2.0 3.0 4.0 5.0

44

log LX

43
42
41
40
39
10.0

10.5

11.5

11.0

log M

12.0

(a) Total

kT (keV)

0.2 0.3 0.4 0.6

1.0

2.0 3.0 4.0 5.0

44

log LX

43
42
41
40
39
10.0

10.5

11.0

log M

11.5

12.0

(b) CGM
Figure 5. Average 0.5-2.0 keV luminosities of stacked locally brightest galaxies. Panel (a) shows Ltotal , the average luminosity projected
within R500 , and panel (b) shows LCGM , the average luminosity projected between (0.15 1) R500 . In both plots, the thick error bar
shows the 1 measurement error from photon counting statistics and the thin error bar shows the 1 uncertainty in the mean value
as determined from bootstrapping analysis. Both Ltotal and LCGM obey simple power-law relations with no breaks down to the lowest
luminosities where emission can be distinguished from the background. In (a), we also include lines showing the approximate expected
contribution from low-mass X-ray binaries (red) and high-mass X-ray binaries (blue) in each bin. In both plots, the upper X-axis shows
the approximate average gas temperatures, as described in section 3.1.

c 0000 RAS, MNRAS 000, 000000

10

Anderson et al.
Table 3. Luminosities of Locally Brightest Galaxies
log M
(M )

log Ltotal
(erg s1 )

m Ltotal
(dex)

b Ltotal
(dex)

log LCGM
(erg s1 )

m LCGM
(dex)

b LCGM
(dex)

11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9
10.7-10.8
10.6-10.7
10.5-10.6
10.4-10.5
10.3-10.4
10.2-10.3
10.1-10.2
10.0-10.1

43.82
43.46
43.39
42.98
42.64
42.34
41.80
41.52
41.29
40.97
40.58
40.40
39.96
40.10
39.60
38.96
39.94
40.00
39.60
<0

0.03
0.02
0.01
0.01
0.01
0.01
0.02
0.02
0.02
0.04
0.07
0.09
0.27
0.19
0.97
0.86
0.21
0.19
0.97

0.21
0.11
0.09
0.06
0.05
0.06
0.06
0.05
0.07
0.11
0.10
0.19
0.46
0.63
0.78
0.83
0.28
0.47
0.86
0.40

43.51
43.25
43.18
42.77
42.47
42.07
41.58
41.26
40.99
40.55
40.28
39.28
<0
39.91
<0
<0
39.63
39.07
39.76
<0

0.06
0.02
0.01
0.01
0.01
0.02
0.02
0.02
0.02
0.04
0.05
0.44

0.10

0.08
0.73
0.09

0.17
0.11
0.09
0.05
0.06
0.06
0.09
0.08
0.11
0.53
0.48
0.93
0.30
0.80
0.72
0.68
0.81
0.86
0.85
0.42

Measured 0.5-2.0 keV average luminosities of locally brightest galaxies. For each luminosity, two 1 uncertainties are quoted. The first is the measurement uncertainty
(incorporating Poisson uncertainty in the source region, uncertainty in the value of
R500 , and Poisson uncertainty in the level of the background) and the second is the
sample error on the mean as estimated from bootstrapping analysis. 1 upper limits
which are negative are denoted as < 0.

LBGs in each bin. The scatter is discussed more in section

5.2 and in Appendix F.

THE LX -M500 RELATION

In this section we derive a simple LX -M500 relation from our

results for the uppermost 12 bins. As is typical for studies
of galaxy clusters, we assume a parametric form for this
relation with a power-law dependence on M500 and a factor
of E(z)7/3 to allow for self-similar redshift evolution6 :
LX, bolometric = E(z)7/3 L0, bolo

M500
M0


(4)

However, we do not measure the bolometric luminosity. We

have rest-frame 0.5-2.0 keV luminosities, so we can divide
L0, bolo by a temperature-dependent bolometric correction
Cbolo and express this relation in terms of the 0.5-2.0 keV
luminosity:
1
LX, 0.5-2.0 keV = E(z)7/3 L0, bolo Cbolo

M500
M0


(5)

We set M0 = 4 1014 M . This equation then has

two free parameters: L0, bolo (the normalization) and (the
slope). We fit for these parameters using forward-modeling.
We create a grid of combinations of these two parameters.
6

E(z) is the dimensionless

Hubble parameter, approximated
p
here as E(z) = m (1 + z)3 + where m and refer to
their z = 0 values.

For each value of , we compute the effective halo mass using

the procedure described in Appendix A. With the effective
halo mass, we can invert the LX -M500 relation to get an updated value of LX . We then compare the model prediction
of LX for each combination of parameters to the observed
luminosities. The observed values of LX also depend slightly
on the effective halo mass (through the R500 aperture size),
so we recompute the observed luminosities for each stellar
mass bin using this new value of R500 as well (this yields very
small changes in the luminosity, of order a few hundredths
of a dex). To minimize the effect of variations caused by the
finite number of galaxies in the true catalog (i.e. the dispersion between different simulations of the same parameters,
visible in Figure D.1) we perform 10 simulations of each bin
for each combination of parameters and average the results.
We compare each combination of parameters to the observed luminosities using the 2 goodness of fit parameter.
The uncertainty we use is the sum, in quadrature, of: the
sample error on the mean for the data (i.e. the thin error
bars in Figure 5, which are always larger than the thick error bars), the standard deviation of the best-fit luminosities
in the ten simulated stacks, and a factor of 10% of the measured luminosity to account for uncertainties in the countsto-flux conversion and the k-correction. We also include in
quadrature a factor of 10% in the bolometric correction applied to each bin; for the higher-mass objects this factor
(with larger bolometric corrections) this factor dominates
the uncertainty budget. Our assumed values for the bolometric correction are listed in Table 2.
The resulting best-fit values are shown in the contour
plot in Figure 6, along with 1, 2, and 3 uncertainty
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

2.2
2.1
2.0

M07

1.9

P09

1.8

D07

1.7
1.6
1.5

W14

R08

P11
V09

M10

S06

1.4
0

L0 (1044 erg s1 , bolometric)

10

Figure 6. Best-fit parameters for the LX -M500 relation, assuming the functional form of eq. (2) and normalizing to M0 =
4 1014 M . Contours indicate 1, 2, and 3 confidence intervals. The best-fit combination of parameters is = 1.85,
L0, bolo = 1.4 1044 erg s1 (indicated with the asterisk). The
best-fit individual values, marginalized over the other parameter,
are = 1.84, L0, bolo = 1.4 1044 erg s1 (indicated with the
green X symbol). For comparison, the blue and green points are
adapted from other published works, as described in section 5
(blue corresponds to X-ray-selected samples, green to optical or
near-IR selection).

regions (defined as 2 < 11.54,18.61,26.90, respectively).

The best-fit combination of parameters is = 1.85+0.15
0.16 ,
L0, bolo = 1.4 0.4 1044 erg s1 . The best-fit individual
values, marginalized over the other parameter, are = 1.84,
L0, bolo = 1.4 1044 erg s1 .
In Appendix E, we cross-check these results by fitting
equation 5 to our binned data directly (using the effective
value of M500 for the bin) instead of forward-modeling the
LX -M500 relation through our simulated catalog. We find
agreement at the 2 level, although the forward-modeled
relation prefers a slightly lower slope. We also checked the results using the measurement errors (i.e. the thick error bars
in Figure 5) instead of the sample error on the mean, and
find essentially the same best-fit values for and L0, bolo ,
though with less acceptable 2 values.

5.1

Other Measurements of the LX -M500 Relation

There have been a number of previous estimates of the LX M500 relation, all of which have been restricted to galaxy
clusters and massive groups. In Figure 6 we have plotted a
number of the more recent results, for comparison with our
own results. To do this, we use a fiducial redshift of z =
0.30 (which is just above the mean redshift of our highestmass bin). We evaluate each relation at M500 = 4 1014 M
(which is the pivot point we use in our relation, and lies
just above our highest-mass bin), and divide the result by
E(z)7/3 1.39 to match our assumed parametrization.
The S06 data point corresponds to Stanek et al. (2006),
who give relations for m = 0.24 and m = 0.30 cosmology.
We compute the slope and normalization of each relation
at our pivot point and take their mean to get the result
c 0000 RAS, MNRAS 000, 000000

11

for the WMAP7 cosmology. The S06 relations use M200 as

the independent variable instead of M500 ; we adopt a value
of 5.7 1014 M for M200 , which assumes an NFW profile
(Navarro et al. 1997) with a concentration of 5.1 (Prada et al.
2012). We also apply a bolometric correction to convert the
normalization from their 0.1-2.4 keV band into bolometric
luminosity.
The M07 data point is Maughan (2007), which provides a bolometric spectroscopic luminosity measured within
R500 , so no conversions are necessary. The P09 data point is
Pratt et al. (2009), from which we use their Malmquist biascorrected relation evaluated using BCES regression. The
V09 data point is Vikhlinin et al. (2009), and we use the
relation given in equation 22 from that paper. This relation
has a pivot point at 1M , so extrapolating the uncertainties
in the slope up to 4 1014 M yields apparent uncertainties
in the normalization which are orders of magnitude in size;
we therefore neglect this uncertainty when computing the
uncertainty in the normalization at our pivot point. For this
relation we apply a bolometric correction to convert from
0.5-2.0 keV into bolometric luminosity. The M10 data point
is Mantz et al. (2010), from which we use the bolometric LM relation for the full dataset. The P11 data point is Planck
Collaboration (2011), from which we use their fiducial relation, which has no correction for Malmquist bias.
Finally, we also examine three samples of clusters identified from optical or near-IR photometry, unlike all the X-ray
flux-limited relations listed above. The D07 data point is Dai
et al. (2007), which uses a stacking approach very similar to
ours, applied to galaxy clusters identified in the 2MASS catalog. We use the bolometric luminosities and mean redshifts
for their five mass bins, and the estimates of M500 from
Dai et al. 2010, and fit relations of the form of equation 4
to these five data points in order to estimate the slope and
normalization at our pivot point. R08 is Rykoff et al. (2008),
which uses 17000 galaxy clusters from the maxBCG catalog
(Koester et al. 2007) selected from SDSS. They also measure the X-ray luminosities from ROSAT stacking, and use
optical richness (calibrating with weak lensing) as a mass
proxy. We shift their relation to our pivot point, and we
also convert M200 into M500 and apply a bolometric correction just as with Stanek et al. (2006). Finally, W14 is Wang
et al. (2014), which also uses an optically-selected sample of
clusters from SDSS. We use their LX -M200 relation (their
equation 11), and again convert M200 into M500 and applying a bolometric correction.
The most striking feature of these relations is the huge
variation among them. Most of these relations are not consistent with one another within the 1 uncertainties. Some
of this disagreement is due to differences in assumptions
about E(z) and the different redshift ranges examined by
each study (we apply a bolometric correction assuming a
mean redshift of 0.3, but deviations from this may change
the result by 10%). However, Rozo et al. (2014) have studied the discrepancies in X-ray scaling relations in a more
systematic way and showed that there are persistent disagreements between various LX -M relations which are not
a function of either of these effects. One significant effect is
the assumed M -fgas relation, which is responsible for the difference in normalization between M10 and P09/V09. Other
systematic offsets in the X-ray observables are also implicated, and several correction factors are required in order to

12

Anderson et al.

bring these three relations into agreement. Even after applying these offsets, Rozo et al. (2014) note that extending
these relations to lower masses (which is the area of interest
for our analysis) will lead to additional divergences.
5.2

Normalization and Scatter in the LX -M500

relation

The other striking feature of Figure 6 is the difference in

normalization between the optically-selected samples (ours,
D07, R08, and W14) and the X-ray selected samples (S06,
M07, P09, V09, M10). Except for S06, the X-ray-selected
samples all have higher normalizations than the opticallyselected samples. The obvious culprit is Malmquist bias:
there is intrinsic scatter in LX at a given M500 , and any
sample of clusters which includes X-ray flux in the selection
criterion (as all of the above X-ray samples do) will systematically overestimate the normalization.
However, each of the above X-ray studies has attempted
to account for Malmquist bias. Stanek et al. (2006) infer an
intrinsic scatter in LX at a fixed halo mass of ln L = 0.68.
This is much larger than the intrinsic scatter inferred by
Maughan (2007) (ln L = 0.17 0.39), Pratt et al. (2009)
(ln L = 0.38), Vikhlinin et al. (2009) (ln L = 0.39), or
Mantz et al. (2010) (ln L = 0.19). It is notable that the
inferred normalization of each relation is inversely proportional to the inferred value of ln L , and it is conceivable that
the correction for Malmquist bias is the dominant cause of
the differences in normalizations in Figure 6.
In theory, a logarithmic scatter ln L 0.4 should produce a natural logarithmic bias of order 0.42 0.3,
corresponding to an offset of about 30%. This could explain
some of the observed offset between the optically-selected
samples and the X-ray selected samples, but probably not
all of it. On the other hand, if the Stanek value of 0.68 is
correct, then the bias rises to about 80%, which brings most
of the relations within 1 of our results.
There are several reasons to suspect the intrinsic bias
may be underestimated in many of the X-ray-selected studies. Simulations by Rasia et al. (2012) predict underestimates in the mass of order 25-35% from X-ray measurements, and Sereno & Ettori (2014) report biases of 40% in
mass between hydrostatic X-ray estimates and weak lensing.
Moreover, a study by Hicks et al. (2013) finds that optically
selected clusters tend to be have more disturbed X-ray morphologies than the clusters in typical X-ray samples, which
suggests that typical X-ray samples may be underestimating the scatter in representative samples of clusters. Angulo
et al. (2012) recently attempted to estimate the expected intrinsic scatter in the LX -M500 relation, predicting a scatter
in log10 M at fixed LX of 0.28. Given their predicted slope
of 1.5, this translates to ln L 0.98. Finally, P13 find a Y M relation with a normalization 20% lower than is found in
X-ray-selected samples, even for the most massive clusters.
We can convert our measurements of the standard error on the mean, derived from bootstrapping analysis, into
estimates of the standard deviation in LX for individual
galaxies by multiplying the standard error by the square
root of the number of galaxies in each bin. These values are
shown in Appendix F. We find a very significant amount of
intrinsic scatter among our sample, which is comparable to
the Angulo et al. (2012) prediction at the high-mass end,

and further increases as the mass decreases towards galaxy

groups and galaxies.
Finally, as an independent way of checking the importance of selection effects, in Appendix G we impose an Xray flux limit of 1 1012 erg s1 cm2 on our sample. This
yields a much noisier and poorer subsample, but the inferred
normalization of the LX -M500 rises by more than a factor of
two, which makes it consistent with other X-ray flux-limited
studies.

5.2.1

Malmquist Bias in Locally Brightest Galaxies?

The preceding discussion is premised upon the claim that

Malmquist bias in the LX -M500 relation should be less
significant for optically-selected samples than for X-rayselected samples. This is plausible, since scatter in X-ray
flux is much more likely to be correlated with X-ray luminosity than scatter in optical flux of the central galaxy.
However, we can also show this explicitly by generating a
volume-weighted stack of our locally brightest galaxies. To
do this, we take the k-corrected r-band magnitude for each
locally brightest galaxy, and weight each RASS image and
each exposure map by the maximum volume within which
this galaxy would be detected. The results are nearly identical to the values in Table 3 for our fiducial run; the median
offset between the Malmquist-corrected luminosities and the
fiducial (unewighted) luminosities is just 0.02 dex and the
largest offset is 0.14 dex (in the bin with log M /M =
10.1-10.2; in the upper twelve bins the largest offset is just
0.08 dex). These offsets are comparable to the 1 uncertainties on LX (see Table 3). This justifies our claim that our
optically-selected sample is much less significantly affected
by Malmquist bias than the X-ray-selected samples.

5.3

Implications for AGN Feedback

If we assume the steepness of the observed slope is due to

nongravitational heating, then AGN feedback is likely an
important contributor. In this section we assume that AGN
feedback controls the behavior of the hot gas in these systems, and we discuss what conclusions can be drawn from
the data.
A useful framework for this discussion is the distinction
between two classes of AGN feedback studied in Gaspari
et al. (2014a) using 3D high-resolution hydrodynamic simulations. In self-regulated models, thermal instabilities due to
radiative cooling in the hot halo lead to steady condensation
onto the central black hole, which boosts the accretion rate
up to 100 times over the Bondi rate, re-heating the central
region and restarting the cycle. The other class (thermal
blast models) is a much more extreme version of this cycle,
with steady cold gas inflow triggering quasar-mode feedback
near the Eddington limit.
Gaspari et al. (2014a) showed that thermal blast models
generically produce a break in the LX -T relation at T 1
keV, below which point the thermal heating overcomes the
binding energy of the hot gas. This type of feedback irreparably destroys the cool core and overheats the hot gas halo,
increasing the cooling time significantly above the Hubble
time. On the other hand, self-regulated AGN outflow feedback does not generate a break since the heating and cooling
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

remain roughly balanced at all mass scales for several Gyr.
In order to compare our data to these simulations, we have
converted the effective M500 values into effective temperatures, using the Sun et al. (2009) M500 -T relation (these
values are also listed in Table 2).
Figure 7 shows the results of this comparison, for both
Ltotal and LCGM . Lacking measurements of the temperature,
we assume the same temperatures for both regions, which
is an oversimplification but in most cases the difference is
only a few percent (see Gaspari et al. 2014a). The difference
between the X-ray flux-limited sample and the full sample
is discussed in Appendix G. Since this plot is essentially
a rescaled version of the LX -M500 plot the same offset in
normalization is visible for our optically-selected sample.
The slope of our LX -T relation is steeper than the selfsimilar prediction (a slope of 2 when thermal bremsstrahlung
dominates the emission) for the total sample. This shows the
same effects of non-gravitational heating as the LX -M500 relation. There is also clearly no break in the LX -T relation,
which is strong evidence for AGN feedback operating as gentle self-regulated mechanical feedback instead of more powerful thermal blasts. As suggested by Mulchaey (2000) and
others, the break which is sometimes seen in other samples
is likely due to the low sensitivity of current instruments for
observing single groups, and our stacking can largely overcome this.
We can make this comparison more quantitative as
well. In Figure 7a, the observed values of Ltotal are compared to the predictions for self-regulated feedback from
Gaspari et al. (2014a). The simulations7 consider gas-rich
non-merging X-ray bright systems with active AGN feedback, so the X-ray flux-limited sample is the more relevant
sample for comparison. The consistency between observations and self-regulated simulations is good across the entire
mass range. The key result is that self-regulated mechanical AGN feedback preserves the LX -T relation (excising the
core or not) without any major break for several Gyr. In
order to preserve the slope at all masses, the simulations require tight self-regulation based on the cooling of the central
gas instead of inefficient hot Bondi-like accretion (Gaspari
et al. 2013, Gaspari et al. 2014b). As in our observed sample, the scatter increases in the regime of less massive groups
(Appendix F).
Figure 7b compares the observed values of LCGM to the
Gaspari et al. (2014a) predictions for thermal blast feedback
(all systems start with fgas = 0.15).This class of AGN feedback creates a strong break in the scaling relations below 1
keV. The system is nearly emptied and even the core-excised
luminosity LCGM is seen to be two or three dex below the
measured values. This discrepancy is a defining characteristic of thermal blast models: the central cooling time is
raised well above the Hubble time when a thermal blast occurs, transforming the systems into non-cool-core objects.
In flux-limited samples, this is not observed: the majority of

The self-regulated simulations (left panel) have been rescaled

adopting realistic gas fractions 0.14, 0.10, and 0.07, for the poor
cluster, massive group, and compact group, respectively. The original setup used fgas ' 0.15 for all systems to test if self-regulated
feedback could provide any strong evacuation ab initio.
c 0000 RAS, MNRAS 000, 000000

13

(a) Ltotal

(b) LCGM
Figure 7. Luminosity-temperature relation for galaxies, groups,
and clusters. The upper plot shows Ltotal measured for our full
sample of LBGs (open data points) as well as the best-fit relation to the X-ray flux-limited sample of LBGs (black line). The
lower plot shows our LCGM results plotted in the same way, using the same assumed temperatures for the CGM annulus as for
the total system. In both plots, the data are compared to the
indicated AGN feedback models as simulated by Gaspari et al.
(2014a); for each system 5 Gyr of evolution are plotted, separated into 500 Myr time steps (starting with the point outlined
in magenta). The self-regulated feedback models (upper panel)
have been rescaled by realistic gas fractions described in the text.
For comparison, measurements of individual objects are also included, from Helsdon & Ponman (2000a,b, cyan), Mulchaey et al.
(2003) and Osmond & Ponman (2004, magenta), Sun et al. (2009,
blue), Pratt et al. (2009, green), and Maughan et al. (2012, red).
The predictions for self-regulated feedback match the flux-limited
observations well, while thermal blasts predict a break around 1
keV which is not observed in our data.

Anderson et al.

detected systems harbor a weak or strong cool core (Mittal

et al. 2009, Sun et al. 2009).
Overall, the power-law nature of our observed L-T relations is very difficult to explain with thermal blast feedback,
and it indicates that self-regulated mechanical feedback preserves the large-scale scaling relations from massive clusters
to L* galaxies. More detailed comparisons between observations and new simulations (e.g., varying fgas ) will be carried
out in future work.

5.0
4.5
MJ10

4.0
3.5

14

3.0

CGM

total

2.0

GALAXY-SCALE HALOS

In the regime of massive (elliptical) galaxies, it is common

to examine scaling relations between LX and a stellar mass
proxy such as LK or LB , rather than relating LX to the halo
mass. This is a much more straightforward relation to fit to
our data than the LX -M500 relation, since our independent
variable is already M . Here we explore linear fits to the
data in Fig.4 for relations with the form
1
L0,* bolo
LX, 0.5-2.0 keV = Cbolo

M
M,0


(6)

as well as relations without the bolometric correction, i.e.


LX, 0.5-2.0 keV = L0,*

J08

B11

2.5

M
M,0

E06

1.5
1.0

40.0

40.5

41.0

log L0, (bolometric)

41.5

Figure 8. Best-fit values for the free parameters in the M -LX

relation (equation 6), including both the stellar mass of the central galaxy and the satellite galaxies within M . Results are fit
both to Ltotal (which has the higher normalization) and LCGM
(which has the lower normalization). Contours indicate 1, 2,
and 3 uncertainty regions for each relation. Approximate comparisons are also made with results from the literature, as described in the text.


(7)

Since we are able to recover input X-ray luminosities well, there is no need to forward-model these relations
through the simulated galaxies in this case. The forward
modeling allows us to constrain uncertainties in the matching between M and M500 , but these relations use M as
the independent variable, so we can just fit to the observed
data directly. The error budget is identical to the budget
in Section 4. For the relations with a bolometric correction,
we include a 10% uncertainty in this factor as well, and this
increases the magnitude of the uncertainty and reduces the
2 .
We also explore the effect of adding the stellar mass
of the satellite galaxies into the relation as well. To estimate this, we use the conditional mass function inferred from
the abundance-matching simulations of Moster et al. (2010).
The satellite galaxies contain 13% as much stellar mass as
the central galaxy in the lowest-mass bin (log M = 10.810.9), rising smoothly up to 232% as much stellar mass as
the central galaxy in the highest-mass bin.
In Table 4 we present each of these best-fit relations.
The normalization of these relations is barely affected by
the inclusion of satellite galaxies, since our pivot point is at
a total stellar mass of 1011 M where satellite galaxies only
contain 24% as much stellar mass on average as the cental galaxy. Accounting for satellite galaxies does lower the
slope significantly, though. Excluding the core region lowers
the normalization but leaves the slope of the relation largely
unaffected. Finally, converting to the bolometric band steepens the inferred slope. The best-fit regions for the bolometric
form of these relations (including stellar mass in subhalos)
are shown in Figure 8.
Note that these formulations assume all the emission
we observe is from hot gas. We showed in section 3 that this
is likely to be true for log M > 10.8, but we can also test

this by excluding the lowest two data points and re-fitting

the other 10 data points. We find that the best-fit values are
unchanged within the 1 uncertainties, suggesting that our
results are not strongly affected by contamination in these
two bins.
These results can be compared to existing LX -LK relations for individual elliptical galaxies (e.g. Ellis & OSullivan
2006, Jeltema et al. 2008, Mulchaey & Jeltema 2010, Boroson et al. 2011). In order to make this comparison, it is
necessary to assume a parametric form for the mass-to-light
ratio as a function of either LK or M . We use the De Lucia
& Blaizot (2007) semi-analytic prescription to estimate the
K-band M/L ratio for central galaxies at z = 0.1 with stellar
masses around 1011 M , and infer a M/L ratio of 0.41. We
then evaluate each of the above relations at a luminosity of
LK = 2.4 1011 L (corresponding to M = 1011 M ) to
estimate their normalization at our pivot point.
For Ellis & OSullivan (2006), we use their KS -band
relation fit with the expectation maximization algorithm to
their largest combined dataset. For Jeltema et al. (2008),
we use their relation for group galaxies, excluding nondetections. We use the equation listed in section 3 for the
Mulchaey & Jeltema (2010) relation, which is fit to field
galaxies. Both Jeltema et al. (2008) and Mulchaey & Jeltema (2010) are measured in the 0.5-2.0 keV band, so we
apply our bolomteric correction to convert to bolometirc luminosities. Finally, for Boroson et al. (2011) we use their
quoted slope of 2.6 0.4, but they do not quote a normalization so we estimate the normalization from their Figure
5a (the uncertainties on this value span most of the plot,
so the exact value is not important). We also apply a small
bolometric correction from their 0.3-8.0 band.
There are still some discrepancies between different LX M relations, but these discrepancies can largely be attributed to different sample selection criteria (e.g. environc 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

15

Table 4. LX - M relations
Relation

log L0
(erg s1 )

2 / d.o.f.

40.75
40.63
40.45
40.32
40.71
40.60
40.38
40.26

3.34
2.21
3.46
2.29
3.81
2.53
4.02
2.64

1.19
1.16
1.44
1.72
1.27
0.68
0.74
0.66

Ltotal , 0.5-2.0 keV, central only

Ltotal , 0.5-2.0 keV, with satellites
LCGM , 0.5-2.0 keV, central only
LCGM , 0.5-2.0 keV, with satellites
Ltotal , Bolometric, central only
Ltotal , Bolometric, with satellites
LCGM , Bolometric, central only
LCGM , Bolometric, with satellites

Best-fit parameters for the Ltotal - M relation for massive galaxies

(eqs. 4 and 5). We examine eight different relations, described in
section 6. The reduced 2 of the best fit is also indicated (each fit has
10 degrees of freedom). In general the fit is improved by the inclusion
of stellar mass from satellite galaxies, and the bolometric relations
also have better fits due to extra uncertainty from the bolometric
corrections.

ment). Our relations seem completely consistent with most

previous observations.

UNIFYING X-RAY AND SZ ANALYSES

P13 showed that the stacked effective Compton Y-parameter

Y500 , as measured through the SZ effect, also has an unbroken power-law dependence on M for locally brightest
galaxies. Matching each value of M to an effective value
of M500 , P13 find a Y500 -M500 relation with a slope close
to the self-similar prediction of 5/3. As discussed in Section
5, it is probably more appropriate to compute the slope of
the Y500 -M500 using forward-modeling. P13 note that fixing the slope of their relation at 5/3 yields an unacceptable
reduced 2 of 3, so it would be interesting to repeat their
analysis using forward-modeling to understand the robustness of their result. Appendix E suggests that the results do
not change dramatically if we use their simpler technique
instead, although the slope changes a bit and the 2 of the
fit is poorer; however this may not necessarily remain true
for an SZ analysis. In this section, we take the P13 result at
face value, and offer brief comments on the implications.
The P13 result implies self-similar scaling in the integrated gas pressure around central galaxies, while we find
steeper scaling in the X-ray luminosity. Both analyses see
no evidence of a break in the relation, down to at least log
M = 11.2 for P13 and down to at least log M = 10.8 in
this work. If both of these results are correct, it implies that
the density profile of the hot gas within the halo is systematically flattening as the stellar mass of the central galaxy
decreases. Thus, galaxy-sized halos would contain the cosmic fraction of baryons, just like galaxy clusters, but they
would be less concentrated and therefore emit less X-ray
radiation than their larger counterparts.
This result can naturally be achieved with AGN feedback and/or with pre-heating, but definitive conclusions
would require a joint X-ray and SZ analysis of these galaxies,
which may be performed in future work.
c 0000 RAS, MNRAS 000, 000000

CONCLUSIONS

In this paper we have presented a stacking analysis of 201011

locally brightest galaxies, which are overwhelmingly centrals
in dark matter halos. The masses of these dark matter halos
range from below the mass of the Milky Way up to mediumsized galaxy clusters. For 20 logarithmically spaced bins in
central galaxy stellar mass, we have computed the effective
M500 and R500 , and stacked the X-ray emission from these
locally brightest galaxies as seen in the ROSAT All-Sky Survey.
Our analysis is novel in a few ways. It systematically
examines the X-ray properties of halos across nearly three
orders of magnitude in halo mass, generating a uniformlycalibrated dataset of X-ray luminosity across this whole
range. Unlike most other studies of these X-ray properties, it
relies on an optically-selected sample of central galaxies, and
is therefore not subject to the Malmquist bias which typically plagues these analyses. Finally, it relies on the exact
same galaxies as examined by P13 in the Sunyaev-Zeldovich
effect, which facilitates a comparison with these results.
The major results of our analysis can be summarized as
follows:
1. Extremely bright soft X-ray sources (i.e. AGN) are
very uncommon in locally brightest galaxies, comprising no
more than about 0.2% of the sample (see Appendix B for
more details).
2. In halos with central galaxies with log M = 10.812.0, the 0.5-2.0 keV observed X-ray luminosity (Ltotal ) can
be related to the stellar mass of the central galaxy with a
simple power-law function, with a slope of 3.34. This powerlaw holds for galaxies, galaxy groups, and galaxy clusters.
3. A simple power-law relation is also found after applying bolometric corrections to the data, and/or by adding additional stellar mass to each bin to account for stars in satellite galaxies. The bolometric correction tends to increase the
slope, while accounting for satellite galaxies tends to decrease it. A power-law relation is also found if we exclude
the emission from the central 15% of R500 . These relations
are consistent with previous measurements of the LX -LK
relation in elliptical galaxies.
4. We clearly detect X-ray emission in the (0.15 1)

16

Anderson et al.

R500 annulus in galaxies with log M = 10.9 11.0, which is

likely to be the signature of a hot gaseous halo, and we tentatively detect a similar signature in this annulus for galaxies
with log M = 10.8 10.9.
5. We find statistically acceptable power-law fits for the
LX -M500 relation for halos with central galaxies with log M
= 10.8-12.0. The best-fit slope (marginalized over the normalization) of this relation is 1.84, which is steeper than the
self-similar prediction of 4/3. This slope is consistent with
other measurements, which have been restricted to galaxy
clusters and massive groups instead of the much larger mass
range studied here.
6. The inferred normalization of the LX -M500 relation
from our data is lower than the normalizations inferred from
X-ray-selected studies. Our normalization is consistent with
the normalization inferred from a recent optically-selected
sample, however, and we hypothesize that the X-ray-selected
samples are not adequately accounting for the effects of
Malmquist bias.
7. To test this hypothesis, we show that the observed
scatter in LX at a given mass is probably much larger than
inferred by X-ray-flux selected studies. We show other evidence which also points to larger scatter and to increasing
scatter for lower-mass halos. Finally, we stack an X-ray fluxlimited subsample of locally brightest galaxies, and recover
a higher normalization consistent with X-ray flux-selected
samples.
8. The slope of our inferred LX -M500 relation is steeper
than self-similar, showing the influence of non-gravitational
heating (likely AGN feedback). Comparing our results to
hydrodynamical simulations of AGN feedback performed by
Gaspari et al. (2014a), we find excellent agreement with the
predictions of self-regulated feedback models, and significant disagreement with models using thermal blast feedback.
AGN feedback does not seem to break either the LX -M500
relation or the L-T relation.
9. Our results are an important complement to the
results of P13. While the YSZ -M500 relation may be selfsimilar, the LX -M500 relation is steeper than self-similar.
These results may be reconciled by smoothly changing the
density profile of the hot gas as a function of halo mass,
although future work is needed to understand this in more
detail.

ACKNOWLEDGEMENTS

The authors would like to thank J.A.R. Martin and the

Planck Collaboration for graciously sharing their catalog
of locally brightest galaxies. We would also like to thank
J. Bregman, T. Davis, J.C. Hill, G. H
utsi, R. Khatri, and
R. Sunyaev for helpful discussions and advice during the
production of this manuscript. This research has made use
of data, software and/or web tools obtained from NASAs
High Energy Astrophysics Science Archive Research Center (HEASARC), a service of Goddard Space Flight Center
and the Smithsonian Astrophysical Observatory. This research has made use of NASAs Astrophysics Data System.
This research made use of Astropy, a community-developed
core Python package for Astronomy (Astropy Collaboration
2013).

REFERENCES
Anderson, M. E., Bregman, J. N., & Dai, X. 2013, ApJ,
762, 106
Angulo, R. E., Springel, V., White, S. D. M., et al. 2012,
MNRAS, 426, 2046
Angulo, R. E., & White, S. D. M. 2010, MNRAS, 405, 143
Arnaud, M., Pratt, G. W., Piffaretti, R., et al. 2010, A&A,
517, A92
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J.,
et al. 2013, A&A, 558, 33
Bharadwaj, V., Reiprich, T. H., Schellenberger, G., et al.
2014, ArXiv e-prints, arXiv:1402.0868
Blanton, M. R., & Roweis, S. 2007, AJ, 133, 734
Blanton, M. R., Schlegel, D. J., Strauss, M. A., et al. 2005,
AJ, 129, 2562
B
ohringer, H., Schuecker, P., Guzzo, L., et al. 2001, A&A,
369, 826
Boroson, B., Kim, D.-W., & Fabbiano, G. 2011, ApJ, 729,
12
Cavaliere, A., & Fusco-Femiano, R. 1976, A&A, 49, 137
Cunha, C. E., Lima, M., Oyaizu, H., Frieman, J., & Lin,
H. 2009, MNRAS, 396, 2379
Dai, X., Bregman, J. N., Kochanek, C. S., & Rasia, E. 2010,
ApJ, 719, 119
Dai, X., Kochanek, C. S., & Morgan, N. D. 2007, ApJ, 658,
917
De Lucia, G., & Blaizot, J. 2007, MNRAS, 375, 2
Ellis, S. C., & OSullivan, E. 2006, MNRAS, 367, 627
Evrard, A. E. 1990, ApJ, 363, 349
Fabjan D., Borgani S., Tornatore L., Saro A., Murante G.,
Dolag K., 2010, MNRAS, 401, 1670
Gaspari M., Brighenti F., DErcole A., Melioli C., 2011,
MNRAS, 415, 1549
Gaspari M., Brighenti F., Temi P., 2012, MNRAS, 424, 190
Gaspari M., Ruszkowski M., Oh S. P., 2013, MNRAS, 432,
3401
Gaspari, M., Brighenti, F., Temi, P., & Ettori, S. 2014a,
ApJL, 783, L10
Gaspari, M., Ruszkowski, M., Oh, S. P., Brighenti, F., &
Temi, P. 2014b, ArXiv e-prints, arXiv:1407.7531
Grimm, H.-J., Gilfanov, M., & Sunyaev, R. 2002, A&A,
391, 923
Guo, Q., White, S., Boylan-Kolchin, M., et al. 2011, MNRAS, 413, 101
Helsdon, S. F., & Ponman, T. J. 2000a, MNRAS, 319, 933
. 2000b, MNRAS, 315, 356
Hicks, A. K., Pratt, G. W., Donahue, M., et al. 2013, MNRAS, 431, 2542
Hopkins, A. M., Miller, C. J., Nichol, R. C., et al. 2003,
ApJ, 599, 971
Jeltema, T. E., Binder, B., & Mulchaey, J. S. 2008, ApJ,
679, 1162
Kaiser, N. 1986, MNRAS, 222, 323
Koester B. P., McKay T. A., Annis J., Wechsler R. H.,
Evrard A., Bleem L., Becker M., Johnston D., Sheldon
E., Nichol R., Miller C., Scranton R., Bahcall N., Barentine J., Brewington H., Brinkmann J., Harvanek M.,
Kleinman S., Krzesinski J., Long D., Nitta A., Schneider
D. P., Sneddin S., Voges W., York D., 2007, ApJ, 660, 239
Komatsu, E., Smith, K. M., Dunkley, J., et al. 2011, ApJs,
192, 18
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

Kravtsov, A. V., & Borgani, S. 2012, ARA&A, 50, 353
Lovisari L., Reiprich T., Schellenberger G., 2014, ArXiv
e-prints
Mantz, A., Allen, S. W., Ebeling, H., Rapetti, D., & DrlicaWagner, A. 2010, MNRAS, 406, 1773
Maughan, B. J. 2007, ApJ, 668, 772
Maughan, B. J., Giles, P. A., Randall, S. W., Jones, C., &
Forman, W. R. 2012, MNRAS, 421, 1583
McCarthy I. G., Schaye J., Ponman T. J., Bower R. G.,
Booth C. M., Dalla Vecchia C., Crain R. A., Springel V.,
Theuns T., Wiersma R. P. C., 2010, MNRAS, 406, 822
Mineo, S., Gilfanov, M., & Sunyaev, R. 2012, MNRAS, 419,
2095
Mittal R., Hudson D. S., Reiprich T. H., Clarke T., 2009,
A&A, 501, 835
Moster, B. P., Somerville, R. S., Maulbetsch, C., et al. 2010,
ApJ, 710, 903
Mulchaey J. S., 2000, ARA&A, 38, 289
Mulchaey, J. S., Davis, D. S., Mushotzky, R. F., &
Burstein, D. 2003, ApJs, 145, 39
Mulchaey, J. S., & Jeltema, T. E. 2010, ApJL, 715, L1
Mushotzky, R. F., & Scharf, C. A. 1997, ApJL, 482, L13
Navarro, J. F., Frenk, C. S., & White, S. D. M. 1997, ApJ,
490, 493
Osmond, J. P. F., & Ponman, T. J. 2004, MNRAS, 350,
1511
Planck Collaboration, Aghanim, N., Arnaud, M., et al.
2011, A&A, 536, A10
Planck Collaboration, Ade, P. A. R., Aghanim, N., et al.
2013, A&A, 557, A52
Planelles S., Borgani S., Fabjan D., Killedar M., Murante
G., Granato G. L., Ragone-Figueroa C., Dolag K., 2014,
MNRAS, 438, 195
Ponman T. J., Bourner P. D. J., Ebeling H., B
ohringer H.,
1996, MNRAS, 283, 690
Prada, F., Klypin, A. A., Cuesta, A. J., Betancort-Rijo,
J. E., & Primack, J. 2012, MNRAS, 423, 3018
Pratt, G. W., Croston, J. H., Arnaud, M., & B
ohringer, H.
2009, A&A, 498, 361
Puchwein E., Sijacki D., Springel V., 2008, ApJL, 687, L53
Rasia, E., Meneghetti, M., Martino, R., et al. 2012, New
Journal of Physics, 14, 055018
Rozo, E., Evrard, A. E., Rykoff, E. S., & Bartlett, J. G.
2014, MNRAS, 438, 62
Rykoff E. S., Evrard A. E., McKay T. A., Becker M. R.,
Johnston D. E., Koester B. P., Nord B., Rozo E., Sheldon
E. S., Stanek R., Wechsler R. H., 2008, MNRAS, 387, L28
Sarazin, C. L. 1986, Reviews of Modern Physics, 58, 1
Sarzi, M., Alatalo, K., Blitz, L., et al. 2013, MNRAS, 432,
1845
Sereno, M., & Ettori, S. 2014, ArXiv e-prints,
arXiv:1407.7868
Sijacki D., Springel V., Di Matteo T., Hernquist L., 2007,
MNRAS, 380, 877
Smith, R. K., Brickhouse, N. S., Liedahl, D. A., & Raymond, J. C. 2001, ApJL, 556, L91
Springel, V., White, S. D. M., Jenkins, A., et al. 2005,
Nature, 435, 629
Stanek, R., Evrard, A. E., B
ohringer, H., Schuecker, P., &
Nord, B. 2006, ApJ, 648, 956
Sun, M., Voit, G. M., Donahue, M., et al. 2009, ApJ, 693,
1142
c 0000 RAS, MNRAS 000, 000000

17

Vikhlinin, A., Burenin, R. A., Ebeling, H., et al. 2009,

ApJL, 692, 1033
Voges, W., Aschenbach, B., Boller, T., et al. 1999, A&A,
349, 389
Wang, L., Yang, X., Shen, S., et al. 2014, MNRAS, 439,
611

APPENDIX A: ESTIMATING EFFECTIVE

HALO MASSES
The independent variable in most of our analysis is the stellar mass of the locally brightest galaxy. This parameter can
be used for analysis in some cases (such as the LX -LK relation, see section 6) but for many purposes it is useful to
relate the stellar mass of the LBG to the halo mass. We
do this by defining an effective halo mass for each stellar
mass bin, in a similar fashion to P13.
For the twelve stellar mass bins which are dominated
by hot gas, we calculate the effective halo mass through an
iterative method. We assume a power-law LX -M500 relation
(equation 5) which has two free parameters. In section 5 we
explore a grid of parameters in order to find the best-fit values; for the rest of this paper we use the best-fit values which
are = 1.72, L0, bolo = 1.8 1044 erg s1 . In each stellar
mass bin, we generate 10 simulated stacked images by randomly drawing galaxies in the same stellar mass bin from
the simulated catalog (described in section 3) and populating each galaxy with an X-ray halo using the halo mass and
position within the halo of the simulated galaxy and the assumed LX -M500 relation (see Appendix D for more details).
Each simulated stack contains the same number of simulated
galaxies as are stacked in the observed catalog (these numbers are listed in Table 1). Each simulated galaxy is assigned
the redshift of one of the observed galaxies. We stack these
simulated galaxies and then measure (using aperture photometry) the average LX within R500 for each bin (assuming
the effective halo mass used by P13 as an initial guess for
R500 ). We then invert the assumed LX -M500 relation in order to derive an effective M500 for the bin. We explored the
effect of iterating this procedure, since the initial guess for
R500 (used to measure LX ) can differ from the value implied
by the inferred M500 , but we found that this difference is too
small to have a significant effect on our results (especially
since R500 only depends on the cube root of M500 ). This
process yields converged estimates of M500 for each stellar
mass bin. We take the effective M500 to be the mean value of
M500 across the ten realizations of this process for each stellar mass bin. The uncertainty on M500 is the the standard
deviation of these ten realizations.
The assumption of a single power-law LX -M500 relation
in estimating the masses has little effect on our results. The
normalization of the relation cancels out and has no effect,
and the slope only has a slight effect through the relative
weighting of the different halo masses within each bin (see
Table A1). Moreover, the use of a single power-law is justified by the results in Figure 5, which strongly suggest a
single power-law relation applies for the uppermost 12 bins.
For the other eight (low-mass) stellar mass bins, the
emission is dominated by X-ray binaries instead of hot gas.
It is therefore not appropriate to use the LX -M500 relation
to estimate the effective halo mass. Since the emission is

18

Anderson et al.

Table A1. Effective Halo Masses from Simulated Stacks

log M
11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9

Meff,

bf

14.56
14.41
14.29
14.08
13.90
13.70
13.51
13.29
13.09
12.91
12.75
12.60

Meff,

ss

14.50
14.33
14.22
14.01
13.83
13.62
13.41
13.19
12.98
12.80
12.64
12.47

Meff,

Table B1. Parameters for Masking Sources Listed in the ROSAT

Bright and Faint Source Catalogs

P13

14.54
14.34
14.20
13.99
13.84
13.63
13.41
13.21
12.97
12.71
12.62
12.40

Effective halo masses computed using the simulated catalog of locally brightest galaxies. Meff, bf are the values used in this work,
and are computed using our best-fit LX -M500 relation. Meff, ss
are computed assuming a self-similar ( = 4/3) LX -M500 relation; these values are not very different from our adopted values,
which shows the relative insensitivity of the effective halo mass
to the slope of the LX -M500 relation. For comparison, we also
list the effective halo masses computed by P13 (Meff, P13 ) for
pressure-weighted observations instead of emission-weighted observations, assuming a self-similar scaling and using the Arnaud
et al. (2010) pressure profile.

concentrated in the galaxy itself, we have no significant detections of X-ray emission in the CGM annulus, so the exact value of R500 is not especially important. We therefore
just invert the abundance matching relation of Moster et al.
(2010) in order to compute the effective halo mass for each
of these stellar mass bins. We propagate the uncertainties
in this relation in order to estimate the uncertainties in the
effective halo mass.

APPENDIX B: CONTAMINATION FROM

BRIGHT SOURCES
Stacking analysis is a measure of the mean properties of a
sample, so like any mean estimator it is fairly sensitive to
outliers. In this section we examine our criteria for masking
bright sources (point-like and extended) from our images to
prevent them from biasing our results.
We use essentially the same technique as in ABD13.
This technique employs two separate filters. First, we crossmatch each of our fields with the ROSAT Bright Source
Catalog and Faint Source Catalog (Voges et al. 1999) and
mask any region which contains a bright source above a
minimum count rate. We repopulate the masked region with
photons that match the flux observed elsewhere in the field.
Second, as a backup measure we discard any observation if
it contains a pixel with more than 10 photons in it.
For the first filter, if we set the minimum count rate
to zero, we end up masking out many of the locally brightest galaxies themselves, especially at the high end where
these galaxies lie at the centers of moderately luminous
galaxy clusters or groups. We instead use a fairly conservative threshold, which is listed for each bin in Table B1.
The corresponding luminosity is also listed for each threshold, if the sources to be masked were at the mean distance of

log M

min cps
(count s1 )

log Leff
(erg s1 )

N500

frac

11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9
10.7-10.8
10.6-10.7
10.5-10.6
10.4-10.5
10.3-10.4
10.2-10.3
10.1-10.2
10.0-10.1

2.8 101
1.5 101
1.0 101
7.7 102
4.5 102
2.8 102
2.2 102
1.3 102
9.5 103
5.0 103
3.2 103
1.7 103
1.1 103
1.3 103
1.4 103
1.6 103
2.0 103
2.2 103
2.4 103
2.6 103

45.0
44.7
44.4
44.1
43.8
43.5
43.2
42.9
42.6
42.3
42.0
41.7
41.5
41.5
41.5
41.5
41.5
41.5
41.5
41.5

1
3
7
13
28
48
54
68
45
27
25
23
13
18
8
2
2
2
0
0

0.0278
0.0263
0.0154
0.00980
0.00944
0.00804
0.00562
0.00479
0.00244
0.00125
0.00110
0.00102
0.00065
0.00105
0.00060
0.00019
0.00027
0.00035
0
0

Numbers used for masking sources listed in the ROSAT Bright

and Faint source catalogs. The min cps column lists the minimum
count rates (as listed in the catalogs) for each bin; any source
with a listed count rate higher than this value is masked if it
falls anywhere within the stacked fields. The Leff column shows
the luminosity of a source at the mean distance of the LBGs
in each bin, if it had the minimum count rate required to be
masked. These luminosities are much larger than plausible hot
gas luminosities for each bin. N500 is the number of LBGs in
each bin with masked source that have a centroid within R500 ,
and frac is the fraction of the stacked LBGs in each bin with such
a source.

the LBGs in that bin. These luminosities are more than an

order of magnitude larger than our measured luminosities,
and it is difficult to imagine any physically plausible way
to produce these luminosities from hot gas. We therefore
expect that the sources masked by this filter are a combination of bright foreground objects, background quasars, and
AGN associated with the locally brightest galaxies themselves. The latter category is scientifically interesting, so for
each bin we count the number of LBGs with a source above
our minimum count rate with a centroid within R500 . This
number is also listed in Table B1, as well as the fraction of
the total sample to which this number corresponds. In total
there are 387 LBGs with masked sources within R500 , which
is 0.2% of the total.
Note that masking these sources could slightly reduce
the average inferred luminosity of our stacks, by a factor
roughly proportional to the fractions listed in table B1, but
these fractions are very low so the bias is insignificant.
We also want to test the effect of these two methods on
our result. We therefore generate an alternative stack where
we multiply the minimum count rate by two, a stack where
we remove the second filter, and a stack where we make
both changes at once. Table B2 shows the average inferred
luminosities for each case.
As Table B1 shows, above log M = 10.7-10.8, the difc 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

19

Table B2. Average LX for Different Point Source Masking Techniques

44
log LX
(fiducial)

log LX
(2 min cps)

log LX
(filter B off)

log LX
(both)

43

11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9
10.7-10.8
10.6-10.7
10.5-10.6
10.4-10.5
10.3-10.4
10.2-10.3
10.1-10.2
10.0-10.1

43.82
43.46
43.39
42.98
42.64
42.34
41.80
41.52
41.29
40.97
40.58
40.40
39.96
40.10
39.60
38.96
39.93
40.00
39.60
<0

43.82
43.46
43.39
42.98
42.65
42.35
41.80
41.52
41.29
40.97
40.58
40.39
39.96
40.11
39.59
39.03
39.93
40.02
39.60
<0

43.68
43.46
43.39
43.98
42.64
42.34
41.80
41.52
41.27
40.92
40.53
40.40
39.74
40.07
39.60
<0
39.91
39.76
39.46
<0

43.68
43.46
43.39
42.98
42.65
42.35
41.80
41.52
41.27
40.91
40.53
40.39
39.73
40.08
39.59
<0
39.90
39.79
39.46
<0

42

Inferred Ltotal for each bin, using variations on our fiducial point
source masking technique. The fiducial technique masks sources
from the ROSAT Bright and Faint Source Catalogs above a minimum count rate and excludes any observation containing a pixel
with more than 10 counts (Filter B). The effects of modifying
one or both of these filters are shown in the final three columns.
For the upper 12 bins where we attribute the X-ray emission to
hot gas, our results are quite robust to the details of the masking
technique.

ferences in the average luminosity are insignificant between

the different techniques, except for a small difference in the
uppermost bin. In fact, if we turn off one or both filters, the
luminosity is about as likely to decrease as to increase, which
suggests that about half the bright sources we are masking
lie outside R500 and are contributing their photons to the
background region instead of the source region. Below log
M = 10.8-10.9, the differences in the average luminosity
become more important, suggesting that a significant portion of the inferred signal is coming from individual objects.
The most extreme example of this is the log M = 10.4-10.5
bin, where the signal disappears entirely when we remove filter B. This suggests that an individual bright source in the
background is sufficient to raise the background enough to
drown out the very weak signal from this bin.

APPENDIX C: NULL TESTS

Here we repeat the analysis in section 3, but instead of stacking locally brightest galaxies we stack random positions on
the sky. Within each bin, we use the same number of random positions as the number of locally brightest galaxies in
the bin, and we assign each random position the redshift of
one of the locally brightest galaxies. The results are shown
in Figure C1. The 1 uncertainties are consistent with zero
in 18/20 cases, the exceptions being the log M = 10.3-10.4
c 0000 RAS, MNRAS 000, 000000

log LX

log M

41
40
39
10.0

10.5

11.0

log M

11.5

12.0

Figure C.1. Null tests showing the measured values of Ltotal

for stacks of random fields using the same assumed redshifts and
numbers of galaxies as the locally brightest galaxies. Error bars
show the measured 1 uncertainties; the uncertainties are almost
always consistent with zero, as expected.

bin, which is a 3.0 detection, and the log M = 11.4-11.5

bin, which is a 1.3 detection.

APPENDIX D: SIMULATED STACKS

As described in Section 3, we have adapted our stacking
analysis from ABD13 for use on locally brightest galaxies.
In ABD13 we extensively tested our procedure on simulated
data, but in this work we are studying more distant galaxies,
and instead of modelling the image we are using the simpler
technique of aperture photometry. In light of these changes,
it is appropriate to verify that we can recover the correct
L500 with simulated data.
For each stellar mass bin, we draw galaxies from the
simulated catalog (described in section 3), noting their stellar mass, halo mass, and the offset (in kpc) between the
galaxy and the center of the halo. In most cases this offset is
near zero, the primary failure mode turns out to be galaxies which lie more than 1 Mpc from the center of a galaxy
cluster. When this occurs, the locally brightest galaxy is
actually a satellite of a much more massive halo, but since
the galaxy lies near the outskirts of the cluster this failure
mode turns out not to be very serious.
We draw the same number of galaxies from the simulated catalog as we have in the observed catalog, and we
randomly assign each simulated galaxy the redshift of one of
the observed galaxies. We assume an arbitrary LX -M500 relation and populate each simulated halo with X-ray luminosity distributed over a -model (Cavaliere & Fusco-Femiano
1976) with = 0.6. We assume the core radius is equal to
the NFW scale radius, which we compute for each halo from
its mass and redshift using the mass-concentration relation
of Prada et al. (2012). We then add a uniform background
of 3 104 count s1 arcmin2 to each simulated halo (with
an assumed integration time of 400 s), stack them together
in physical space, and convolve the resulting image with the
empirical psf which we determined for that stellar mass bin

20

Anderson et al.
APPENDIX F: TOTAL, EXTRINSIC, AND
INTRINSIC SCATTER IN LX

2.2
2.1
2.0

1.9
1.8
1.7
1.6
1.5
1.4
0

L0 (1044 erg s1 , bolometric)

10

Figure E.1. Best-fit parameters for the LX -M500 relation, as

measured by forward-modeling (gray contours) and by fitting to
the effective M500 values for each bin (red contour). The latter method gives less acceptable results (the best-fit relation has
2 /d.o.f. = 12.0/10), so only the 2 and 3 contours can be plotted for this method. The 2 contour regions for both methods
overlap nicely, although the forward-modeled relation prefers a
slightly lower slope.

(see section 3.1). We run our aperture photometry analysis

on these simulated images, and compare the results to the
input LX . This comparison can be seen in Figure D1 for
a variety of assumed LX -M500 relations which straddle the
parameter space we search in the following section.
As Figure D1 shows, we can generally recover LX within
the measurement errors across the entire range of M . The
largest deviations occur at the high end where the intrinsic variation in LX is largest (due to the steep slope of the
M*-M500 relation) at this end. There is also a very small
systematic bias, but the magnitude of the bias is not significant and it seems most prominent for relation A, which has
much higher normalization than our observed relation.

APPENDIX E: CROSS-CHECKING OUR

LX -M500 RELATION RESULTS
Since our results in section 5 rely on extensive forwardmodeling of the LX -M500 through a simulated catalog, it is
appropriate to check how important the forward-modeling
is for our results. To do this, we use the simpler technique
of P13, and fit a linear relation to the Ltotal results as a
function of the effective M500 in each bin. We fit relations of
the form of equation 5, using the mean temperature (from
Table 2) to compute the mean bolometric correction and the
mean redshift (from the distance in Table 1) to compute the
mean E(z) in each bin.
The resulting contours are shown in Figure E1, along
with the contours from our forward-modeling for comparison. The best-fit linear relation yields a 2 /d.o.f. of 12.0/10,
which is unacceptable at 1 but still allows for a 2 and a
3 contour region. The 2 contour regions for both methods overlap nicely, although the forward-modeled relation
prefers a slightly lower slope.

In Table F1 we present the measured scatter in Ltotal and

LCGM , and we distinguish between sources of extrinsic and
intrinsic scatter. The measured scatter is estimated from
the sample error on the mean as estimated from bootstrapping analysis, multiplied by the square root of the number of
locally brightest galaxies in each bin. Since the mean number of photons per locally brightest galaxy is low (11.8 and
6.5 in our two highest-mass bins, decreasing down to 0.01
in the log M = 10.1 10.2 bin), we expect a significant
contribution to this measured scatter from Poisson noise;
this scatter is quantified in the Poisson columns. Another
extrinsic source of scatter comes from the dispersion in the
objects stacked within a given M bin. This includes the
dispersion in the M500 -M relation, as well as the offsets
between locally brightest galaxies and the centers of their
dark matter halos. We estimate this scatter (which we denote as stacking scatter in Table F1) using our 10 different
realizations of the simulated stacking analysis. We measure
Ltotal for each object in our simulated stacks, compute the
standard deviation of Ltotal in each stellar mass bin for each
simulation, and take the mean of these ten realizations to
estimate the stacking scatter.
After accounting for these two forms of scatter, the remaining scatter (which we denote as other in Table F1)
is due to scatter around the LX -M500 relation (intrinsic
scatter) along with some unknown contribution from X-ray
binaries and low-luminosity AGN activity, both of which
likely become increasingly important at lower X-ray luminosities. The other scatter is therefore an upper limit on
the amount of intrinsic scatter, although at the high-mass
end we expect it to be close to the true intrinsic scatter. Towards the low-mass end the other scatter increases significantly. This is likely physical, reflecting increasing intrinsic
variation in the X-ray properties of galaxies, and is probably
compounded by the increasing contribution of blue galaxies
to the sample as the stellar mass decreases.

APPENDIX G: IMPOSING AN X-RAY FLUX

LIMIT
In order to test our hypothesis that X-ray flux selection is
responsible for the previous overestimates of the normalization of the LX -M500 relation, in this Appendix we impose an X-ray flux limit on our sample before stacking. We
only stack galaxies if they have a 0.5-2.0 keV observed-frame
background-subtracted X-ray flux of at least 1 1012 erg
s1 cm2 within R500 . This flux limit is lower than the limit
used by REFLEX (B
ohringer et al. 2001). However, we use
a fixed aperture of size R500 while they employed an adaptive aperture, which allows them to maximize the S/N of
each cluster. Our flux limit still removes almost all of our
sources: only 161 locally brightest galaxies remain, and the
uppermost two bins have 2 and 6 objects respectively. Contamination from AGN and X-ray binaries also becomes more
important, since the stacks only contain a handful of objects (or in some of the low-mass bins, single objects). We
therefore eschew forward-modeling, and just fit a power-law
relation with the form of equation 5 to the data, using the
c 0000 RAS, MNRAS 000, 000000

Unifying X-ray Scaling Relations

21

45
44

log LX (erg s1 )

43
42
41

Rel A
Rel B
Rel C
Rel D

40
39
10.8

11.0

11.2

11.4
log M*

11.6

11.8

12.0

Figure D.1. Simulated emission from locally brightest galaxies, assuming four different power-law relations for the LX -M500 relation.
For each relation, the data points represent the input relation using the effective Mhalo for each bin, and the error bars represent recovered
values for LX after simulating realistic distributions of locally brightest galaxies, adding a background, stacking the galaxies, convolving
with the psf, and performing aperture photometry. The recovered values match the input parameters very well, with deviations visible
only for points which are several times fainter than the true values (Figure 5). The relations in this plot have the have same form as
equation (6) and use the same values for E(z), Cbolo , and M0 as the true data. The values of (L0 , ) for relations A, B, C, and D
respectively are (1045 , 4/3), (1045 , 2), (1044 , 4/3), and (1044 , 2).

effective M500 as the independent variable (the same technique as Appendix E).
We are unable to get a statistically acceptable fit to
the uppermost 12 bins (best-fit 2 = 137.3 for 10 d.o.f.),
although this is not surprising given the paucity of sources.
The inferred LX for each bin is significantly higher, however,
with a median increase of 1.2 dex in the upper 12 bins. If we
restrict our analysis to just the galaxy cluster regime, like
other flux-limited studies using ROSAT, then the data do
obey a simple power-law (with 2 = 0.8 for 2 d.o.f.; we use
the bins with kT > 2 keV only). The best-fit slope is 1.75
and the best-fit normalization is 4.5 1044 erg s1 , which
is more than twice the inferred normalization for the full
sample, and much closer to the other X-ray flux-selected
samples. These relations are also plotted in Figure 7 (the
solid lines).

c 0000 RAS, MNRAS 000, 000000

22

Anderson et al.
Table F1. Total, Extrinsic, and Intrinsic Scatter
log M

Total
Measured

Total
Poisson

Total
Stacking

Total
Other

CGM
Measured

CGM
Poisson

CGM
Stacking

CGM
Other

11.9-12.0
11.8-11.9
11.7-11.8
11.6-11.7
11.5-11.6
11.4-11.5
11.3-11.4
11.2-11.3
11.1-11.2
11.0-11.1
10.9-11.0
10.8-10.9
10.7-10.8
10.6-10.7
10.5-10.6
10.4-10.5
10.3-10.4
10.2-10.3
10.1-10.2
10.0-10.1

1.31
1.16
1.63
1.75
1.79
2.42
2.70
2.63
3.24
3.76
3.70
4.42
5.59
6.05
6.38
6.38
4.35
5.01
5.95
4.38

0.26
0.33
0.34
0.43
0.55
0.67
0.87
1.03
1.09
1.35
1.59
1.74
2.17
1.99
2.46
3.13
1.96
1.87
2.23

0.72
0.66
0.73
0.70
0.96
1.04
1.02
1.34
1.44
1.78
1.95
2.63

1.06
0.89
1.41
1.54
1.41
2.08
2.34
2.02
2.69
3.03
2.72
3.10
5.16
5.71
5.89
5.56
3.89
4.64
5.52

1.04
1.16
1.64
1.52
2.07
2.46
3.12
3.17
3.67
5.85
5.71
7.02
4.93
6.54
6.22
5.96
6.16
6.15
5.92
4.38

0.35
0.41
0.42
0.52
0.64
0.83
1.03
1.23
1.32
1.72
1.87
2.89

2.37

4.47

0.71
0.66
0.72
0.70
0.99
1.08
1.03
1.40
1.49
1.86
2.02
2.71

0.67
0.86
1.41
1.24
1.70
2.05
2.75
2.56
3.08
5.28
5.00
5.80

5.69

Logarithmic scatter ln L per object as measured from our stacking analysis. See text for definitions
of each column. Dashes in the stacking column refer to central galaxy stellar masses for which our
assumed halo mass - X-ray luminosity relation is not an appropriate model for the observed emission.
Dashes are indicated for the Poisson scatter in some bins because the net counts are negative.

c 0000 RAS, MNRAS 000, 000000