Herschel-ATLAS: The link between accretion luminosity and star formation in quasar host galaxies

Abstract

We use the science demonstration field data of the Herschel-ATLAS to study
how star formation, traced by the far-infrared Herschel data, is related to
both the accretion luminosity and redshift of quasars selected from the Sloan
Digital Sky Survey and the 2SLAQ survey. By developing a maximum likelihood
estimator to investigate the presence of correlations between the far-infrared
and optical luminosities we find evidence that the star-formation in quasar
hosts is correlated with both redshift and quasar accretion luminosity.
Assuming a relationship of the form L_IR \propto L_QSO^{\theta} (1 +
z)^{\zeta}, we find {\theta} = 0.22 +/- 0.08 and {\zeta} = 1.6 +/- 0.4,
although there is substantial additional uncertainty in {\zeta} of order +/- 1,
due to uncertainties in the host galaxy dust temperature. We find evidence for
a large intrinsic dispersion in the redshift dependence, but no evidence for
intrinsic dispersion in the correlation between L_QSO and L_IR, suggesting that
the latter may be due to a direct physical connection between star formation
and black hole accretion. This is consistent with the idea that both the quasar
activity and star formation are dependent on the same reservoir of cold gas, so
that they are both affected by the influx of cold gas during mergers or heating
of gas via feedback processes.

Full text

Page 1

arXiv:1103.3905v2 [astro-ph.CO] 31 Mar 2011
Mon. Not. R. Astron. Soc. 000, 1–10 (2010)Printed 1 April 2011(MN LATEX style file v2.2)
Herschel-ATLAS: The link between accretion
luminosity and star formation in quasar host galaxies⋆
D. G. Bonfield1†, M. J. Jarvis1, M. J. Hardcastle1, A. Cooray2, E. Hatziminaoglou3,
R. J. Ivison4,5, M. J. Page6, J. A. Stevens1, G. de Zotti7,8, R. Auld9, M. Baes10,
S. Buttiglione7, A. Cava11, A. Dariush9,12, J. S. Dunlop5, L. Dunne13, S. Dye9,
S. Eales9, J. Fritz10, R. Hopwood14, E. Ibar4, S. J. Maddox13, M. J. Micha? lowski5,
E. Pascale9, M. Pohlen9, E. E. Rigby13, G. Rodighiero7, S. Serjeant14,
D. J. B. Smith13, P. Temi15, P. van der Werf5,16
1Centre for Astrophysics Research, Science & Technology Research Institute, University of Hertfordshire, Hatfield, AL10 9AB, UK
2Dept. of Physics & Astronomy, University of California, Irvine, CA 92697, USA
3ESO, Karl-Schwarzschild-Str. 2, 85748 Garching bei M¨ unchen, Germany
4UK Astronomy Technology Centre, Royal Observatory, Edinburgh, EH9 3HJ, UK
5Scottish Universities Physics Alliance, Institute for Astronomy, University of Edinburgh, Royal Observatory, Edinburgh, EH9 3HJ, UK
6Mullard Space Science Laboratory, University College London, Holmbury St. Mary, Dorking, Surrey RH5 6NT, UK
7INAF – Osservatorio Astronomico di Padova, Vicolo Osservatorio 5, I-35122, Padova, Italy
8SISSA, Via Bonomea 265, I-34136 Trieste, Italy
9School of Physics & Astronomy, Cardiff University, Queen’s Buildings, The Parade, Cardiff, CF24 3AA, UK
10Sterrenkundig Observatorium, Universiteit Gent, Krijgslaan 281 S9, B-9000 Gent, Belgium
11Instituto de Astrof´ ısica de Canarias (IAC) and Departamento de Astrof´ ısica de La Laguna (ULL), La Laguna, Tenerife, Spain
12School of Astronomy, Institute for Research in Fundamental Sciences (IPM), PO Box 19395-5746, Tehran, Iran
13Centre for Astronomy and Particle Theory, The School of Physics & Astronomy, Nottingham University, University Park Campus,
Nottingham, NG7 1HR, UK
14Department of Physics & Astronomy, The Open University, Walton Hall, Milton Keynes, MK7 6AA, UK
15Astrophysics Branch, NASA/Ames Research Center, MS 245-6, Moffett Field, CA 94035, USA
16Leiden Observatory, Leiden University, P.O. Box 9513, NL-2300 RA Leiden, The Netherlands
Received Month dd, yyyy; accepted Month dd, yyyy
ABSTRACT
We use the science demonstration field data of the Herschel-ATLAS to study how
star formation, traced by the far-infrared Herschel data, is related to both the
accretion luminosity and redshift of quasars selected from the Sloan Digital Sky
Survey and the 2SLAQ survey. By developing a maximum likelihood estimator
to investigate the presence of correlations between the far-infrared and optical
luminosities we find evidence that the star-formation in quasar hosts is correlated
with both redshift and quasar accretion luminosity. Assuming a relationship
of the form LIR ∝ Lθ
0.4, although there is substantial additional uncertainty in ζ of order ±1, due
to uncertainties in the host galaxy dust temperature. We find evidence for a
large intrinsic dispersion in the redshift dependence, but no evidence for intrinsic
dispersion in the correlation between LQSOand LIR, suggesting that the latter
may be due to a direct physical connection between star formation and black
hole accretion. This is consistent with the idea that both the quasar activity
and star formation are dependent on the same reservoir of cold gas, so that they
are both affected by the influx of cold gas during mergers or heating of gas via
feedback processes.
QSO(1 + z)ζ, we find θ = 0.22 ± 0.08 and ζ = 1.6 ±
Key words: galaxies: active – galaxies: high-redshift – quasars: general
⋆Herschel is an ESA space observatory with science instru-ments provided by European-led Principal Investigator con-
sortia and with important participation from NASA
c ? 2010 RAS

Page 2

2 Bonfield et al.
† Email (DGB): d.bonfield@herts.ac.uk
1INTRODUCTION
AGN activity is now widely believed to be an important
phase in the evolution of every massive galaxy in the
Universe. This belief stems firstly from the discovery of a
relatively tight correlation between the mass of a galaxy’s
bulge and the mass of its central supermassive black hole
(SMBH; e.g. Magorrian et al. 1998; Ferrarese & Merritt
2000; Gebhardt et al. 2000), which implies that build-
up of the stellar mass and the SMBH mass are causally
connected (although see also Jahnke & Macci` o 2010, who
show that averaging of parameters by mergers can also
produce such a correlation).
From a more theoretical perspective, AGN have
become a key ingredient for semi-analytic models of
galaxy formation (e.g. Benson et al. 2003; Cole et al.
2000; Guiderdoni et al. 1998; Granato et al. 2001). The
over-production of stars at the bright end of the galaxy
luminosity function by these models has led to the inclu-
sion of an extra source of heating or “AGN feedback” (see
e.g. Bower et al. 2006; Croton et al. 2006; Granato et al.
2004). However, it is still unclear what kind of AGN-
driven feedback is the most important, e.g. two
plementary mechanisms are described by Croton et al.
(2006): “quasar-mode” feedback, proposed to be effec-
tive during episodes of efficient accretion of cold gas, and
“radio-mode” feedback, caused by the relatively ineffi-
cient accretion of hot gas (in the absence of cold gas)
when an AGN would have lower optical luminosity, but
would also be generating radio jets. Thus, it is now clear
that if we are to understand galaxy formation and evo-
lution, we must also obtain a clear picture of AGN, their
environments and the co-evolution of SMBHs and their
host galaxies.
It is difficult to study the host galaxies of dis-
tant quasars, due to the nuclear quasar emission over-
whelming the stellar emission at optical and near-
infrared wavelengths. However, Hubble Space Telescope
and high spatial-resolution ground-based imaging have
enabled the investigation of the stellar populations of
quasar hosts (e.g. Bahcall et al. 1994; McLure et al. 1999;
Dunlop et al. 2003; Kotilainen et al. 2009), demonstrat-
ing that most luminous quasars reside in massive el-
liptical galaxies.A number of studies have attempted
to determine the star-formation properties of quasar
host galaxies, with some using optical colours (e.g.
S´ anchez et al. 2004) or spectroscopy (e.g. Nolan et al.
2001; Silverman et al. 2009; Trichas et al. 2010). Some
information on the host galaxies of AGN has been
gleaned from powerful radio galaxies, where the quasar
nucleus is obscured due to the putative dusty torus (e.g.
Antonucci 1993). Again these studies show that pow-
erful AGN reside in massive galaxies (e.g. Jarvis et al.
2001; McLure et al. 2004; Seymour et al. 2007) and there
appears to be a trend where the most powerful exhibit
signs of recent star formation (e.g. Baldi & Capetti 2008;
Herbert et al. 2010). However, optical colours and emis-
sion lines can be contaminated by the quasar, so that
its contribution must be modelled to estimate the star-
formation rate, and all of these studies suffer from the
uncertainty arising from having very little information
about how much star formation is obscured by dust, and
com-
c ? 2010 RAS, MNRAS 000, 1–10

Page 3

H-ATLAS: quasar hosts with Herschel3
thus how much of the emission due to young stars is re-
processed to far-infrared and sub-millimetre wavelengths.
Spectroscopy with the Spitzer Space Telescope has
enabled the determination of star-formation rates in in-
dividual AGN via mid-infrared emission lines, which
are much less susceptible to dust, (e.g. Schweitzer et al.
2006; Netzer et al. 2007; Mart´ ınez-Sansigre et al. 2008;
Lutz et al. 2008; Shi et al. 2009; Trichas et al. 2009;
Lutz et al. 2010). However, the relatively low sensitivity
of Spitzer’s infrared spectrograph (IRS) has meant that
only small, severely flux-limited samples of objects could
be investigated, making the precise relationship between
star-formation rate and quasar luminosity difficult to ro-
bustly determine.
(Sub)millimetre instruments such as SCUBA and
MAMBO have provided important information on repro-
cessed emission from star formation in distant quasars
and radio galaxies, e.g. photometric surveys revealing lu-
minous (∼ 1013L⊙) far-infrared emission from a number
of radio galaxies and quasar hosts, making it clear that
major (SFR ∼ 1000M⊙yr−1) episodes of star formation
can coexist with powerful AGN
2001; Priddey et al. 2003; Omont et al. 2003; Page et al.
2004; Mart´ ınez-Sansigre et al. 2009). Furthermore, sub-
mm imaging has revealed overdensities of sources in the
vicinity of individual high-z AGN, plausibly belonging
to the same large-scale structures (Stevens et al. 2003;
Priddey et al. 2007; Stevens et al. 2010).
cross-correlating SCUBA data with deep X-ray surveys,
Alexander et al. (2005) also show a large fraction of
sub-mm galaxies contain buried AGN, while Laird et al.
(2010) find that the X-ray AGN fraction in sub-mm
galaxies is 20–29 %.
The main drawback of the sub-mm studies to date
is that, due to the high background associated with
ground-based observing and the small number of pix-
els in existing instruments, it has not been possible to
obtain very large samples of objects. A recent study by
Serjeant & Hatziminaoglou (2009, hereafter SH09)
tempted to overcome this shortfall by combining data
from a wealth of Spitzer and ISO observations at mid-
and far-infrared wavelengths to investigate how obscured
star formation in the hosts of quasars evolved over cos-
mic time and also how this may be related to the bolo-
metric output of the AGN. They found evidence for an
increase in far-infrared luminosity as a function of both
cosmic epoch and quasar luminosity. However, they mea-
sured the correlation between far-infrared and quasar lu-
minosity using wide redshift bins, which can potentially
confuse the effect of redshift with that of quasar luminos-
ity, since the latter evolves strongly with redshift in flux-
density-limited samples. In addition, their work extrapo-
lated the far-infrared flux from mid-infrared wavelengths
using SED templates, and it would clearly be preferable
to measure the far-infrared flux directly.
The recently-launched ESA Herschel space observa-
tory (Pilbratt 2010) offers an unprecedented combination
of sensitive detectors, large (3.5-metre diameter) collect-
ing area, and low background, which makes it possible,
for the first time, to conduct deep surveys of large areas
of the sky at far-infrared/sub-mm wavelengths.
In this paper we present a study of the far-infrared
(e.g. Archibald et al.
Conversely,
at-
fluxes of known quasars in science demonstration obser-
vations of the Herschel Astrophysical Terahertz Large
Area Survey (H-ATLAS; Eales et al. 2010). This is the
first single sample of quasars measured in the sub-mm
which is large and diverse enough to directly determine a
relationship between far-infrared (cool dust) luminosity
and optical (quasar) luminosity, without using infrared
information at wavelengths short enough to be possibly
contaminated by warm dust in the AGN torus. While
Serjeant et al. (2010) determine a relationship between
optical and far-infrared luminosity using this data in com-
bination with infrared luminosities derived from shorter-
wavelength Spitzer and ISO data, here we use only the
H-ATLAS data.
Section 2 describes the H-ATLAS observations used.
Section 3 explains how we construct our quasar sam-
ple, determine far-infrared fluxes for them, and find a
maximum-likelihood fit to the relationship between far-
infrared luminosity, optical luminosity, and redshift. The
results of this fitting are discussed in Section 4 and con-
clusions are presented in Section5.
2OBSERVATIONS
When completed, H-ATLAS will provide photometry for
550 square degrees of the sky in five far-infrared/sub-
mm bands. The 5-σ depths at wavelengths of 100, 160,
250, 350, and 500 µm are 132, 126, 32, 36 and 45 mJy
respectively.
In this paper, we use only data from Herschel’s
SPIRE instrument (Griffin et al. 2010), at 250, 350,
and 500 µm, since the PACS 100 and 160 µm data
(Poglitsch et al. 2010) are not deep enough to provide
useful constraints on the infrared luminosities of the
quasars we are studying (Ibar et al. 2010). The longer
SPIRE wavelengths are also less likely to be contami-
nated by direct emission from the warm dust of the AGN
torus. This means that the emission can be more reli-
ably attributed to cool dust heated by star-formation
processes, and its spectral energy distribution (SED) ap-
proximated by a simple greybody.
This paper is also restricted in area to the H-ATLAS
science demonstration (SD) field, which covers approxi-
mately 16 square degrees – around 3 per cent of the total
survey area. This SD data has been publicly released, and
both the maps and catalogues can be retrieved from the
H-ATLAS webpage1.
3METHOD
Our quasar sample consists of all objects in the field
which are identified as quasars in the seventh data re-
lease of the SDSS Quasar Catalog (Schneider et al. 2010),
or the 2dF-SDSS LRG (luminous red galaxy) and Quasar
spectroscopic catalogue (2SLAQ; Croom et al. 2009). Af-
ter removing the known blazar HB0906+015, since its
far-infrared emission will not be dominated by star for-
mation (see e.g. Gonz´ alez-Nuevo et al. 2010), we find 372
1http://www.h-atlas.org
c ? 2010 RAS, MNRAS 000, 1–10

Page 4

4Bonfield et al.
quasars in the whole SV field, with a reasonably conser-
vative cut on the area (by eye) to avoid using the noisier
data at the edges of the field.
Of the 372 quasars, only 29 have matches to ob-
jects in the 5-σ H-ATLAS catalogue (Rigby et al. 2010;
Smith et al. 2010). To improve our statistics, we mea-
sure fluxes directly from the SPIRE maps described by
Pascale et al. (2010) at the position of each quasar. To
do this, we take a point measurement of the flux in the
nearest pixel to the SDSS position on each PSF-filtered
and background-subtracted map.
To account for source confusion, the most likely flux
from the object (which is used to produce Figure 2) is ob-
tained by subtracting the mean pixel value in the map.
However, when we perform a maximum-likelihood fit for
the coefficients of Equation 2, as described below, there
is an implicit assumption that the measurement noise is
symmetric about zero. The confusion noise is very sig-
nificantly skewed, and can cause false correlations to be
found in random data, so for the fluxes used in these fits
we symmetrise the noise distribution by subtracting the
flux in a different randomly selected pixel from each flux
measurement. This increases the overall noise level, so we
average the likelihoods from a number of random back-
ground realisations in order to reduce the noise level in
the final result.
We also construct a comparison sample of fluxes
drawn from random positions in the maps, with their
backgrounds subtracted in precisely the same way as
the real fluxes, and are then used to produce maximum-
likelihood estimates in the same way as the real data.
Comparing these estimates with those from the real data
allows us to increase our confidence that any trend de-
tected in the data is due to the influence of the quasars,
and not merely a result of the noise properties of the
maps, which have a significant impact due to our inclu-
sion of individual noisy objects instead of using binned
data.
To compare the AGN luminosity of the quasars with
that from star formation, we make the assumption that
the optical light is dominated by the AGN, while the far-
infrared/sub-mm is dominated by the cool dust heated
by star formation. This latter assumption is consistent
with the results of Hatziminaoglou et al. (2010), who
fit models with AGN and star-formation contributions
to the mid- and far-infrared emission from AGN hosts
and find that star formation dominates the far-infrared
fluxes. Their result depends on the assumption that the
dusty torus heated directly by an AGN has a fairly sim-
ple, approximately toroidal geometry, as described by
Fritz et al. (2006). The infrared SEDs of AGN can, how-
ever, be fully reproduced by more complex clumpy torus
models (e.g. Nenkova et al. 2008), so, at least until these
models are tested by high-spatial-resolution imaging of
AGN tori, we cannot completely rule out the possibil-
ity of a contribution to far-infrared fluxes from dust
heated directly by the AGN. However, the fact that
Hatziminaoglou et al. (2010) find that the SPIRE colours
of AGN are indistinguishable from those of normal galax-
ies suggests that the AGN does not make a strong direct
contribution at these wavelengths.
All of the quasars have spectroscopic redshifts from
Figure 1. Absolute i-band magnitudes (calculated as de-
scribed in the text) for our sample of spectroscopically con-
firmed quasars, plotted as a function of redshift. Objects with
spectra from SDSS are shown as black diamonds; those with
2SLAQ spectra have red squares.
their respective surveys, and optical photometry in the
ugriz bands from the SDSS (York et al. 2000). We use
these data to estimate the quasar luminosity, follow-
ing Schneider et al. and others in using the absolute i-
band magnitude, Mi, as a proxy for the luminosity of
the central engine. This is calculated from the observed
i magnitude (after the subtraction of Galactic extinc-
tion) and the redshift, by assuming a power-law SED
(Sν ∝ να) with spectral index α = −0.5 and assum-
ing a ΛCDM cosmology with Ωm = 0.3, ΩΛ = 0.7, and
H0 = 70kms−1Mpc−1. Figure 1 shows the distribution
of optical luminosities of our sample as a function of
redshift, indicating the source of the spectroscopic red-
shift (SDSS or 2SLAQ, or both). From this we can see
that the 2SLAQ measurements in this field allow the
inclusion of quasars at least one magnitude fainter (at
z < 3) than the SDSS spectroscopic survey probes. This
greater range in optical luminosity is crucial for decou-
pling optical luminosity and redshift, and gives our study
an advantage over similar work by Hatziminaoglou et al.
(2010). We note, however, that the range of bolometric
luminosities spanned by our quasar sample is 1045erg
s−1< Lbol< 1048erg s−1, so all of the objects are firmly
in the quasar (as opposed to Seyfert) regime. Thus, if
there are two “modes” of accretion with different feed-
back properties, this study will only be relevant to un-
derstanding the “quasar mode”, making it usefully com-
plementary to the work of Shao et al. (2010), which uses
different Herschel data to probe the star-formation rates
in lower-luminosity X-ray-selected AGN.
As mentioned earlier, the SPIRE photometry at 250,
350, and 500 µm is unlikely to be contaminated by the
AGN directly for even the higher redshift quasars in the
sample, and is deeper than the H-ATLAS PACS photom-
c ? 2010 RAS, MNRAS 000, 1–10

Page 5

H-ATLAS: quasar hosts with Herschel5
Figure 2. Estimated total infrared luminosities (calculated
by integrating the best-fitting greybody with T = 30 K and
β = 1.5 over the 8–1000 µm wavelength range) for our sample
of spectroscopically confirmed quasars, plotted as a function of
redshift. Objects with spectra from SDSS are shown as black
diamonds; those with 2SLAQ spectra have red squares. Error
bars are plotted where LIR is greater than the uncertainty
in the estimate, otherwise 1-σ upper limits are shown using
arrows.
etry (Ibar et al. 2010), so we use only the SPIRE fluxes to
constrain the overall infrared luminosity. To make the re-
sults directly comparable with each other, we fit a simple
isothermal grey dust model, with flux density
Sν ∝ ν(3+β)/(ehν/kT)− 1)(1)
where h and k are Planck’s and Boltzmann’s constants,
ν is rest-frame frequency, and where we assume fixed val-
ues for both temperature T = 30K and β = 1.5. We then
integrate the greybody curve between rest-frame wave-
lengths of 8 and 1000 µm to obtain an estimate of LIR.
(This wavelength range is chosen for convenience and for
comparison with other studies, and is not intended to
imply that the cold dust represented by the greybody
dominates the SED at wavelengths as short as 8 µm.)
We estimate the uncertainty on LIR by increasing it un-
til the χ2value for the fit to the data increases by unity.
Since we include far-infrared flux measurements at
our quasar positions regardless of whether there is a sig-
nificant detection in the far-infrared, fitting the normali-
sation of a fixed template is the only robust measurement
possible. This procedure does not account for the possi-
bility of dust temperature or emissivity evolution either
with redshift or quasar luminosity, which may well be im-
portant effects to consider. However, we are able to use
the well-detected objects in our sample to address the
suitability of our choice of T and β.
To check our choice of greybody temperature, we fit
greybody SEDs to the SPIRE fluxes of individual quasars
with good detections (signal-to-noise > 3) in all three
SPIRE bands, with β fixed at 1.5 and T allowed to vary.
The best-fitting temperatures for eight out of the nine
well-detected quasars were very similar, with mean 32.4
K and a standard deviation of 2.8 K, while one outlier
fits a temperature of 58.5+14.0
a redshift range of 1.1 < z < 2.3 and a range in infrared
luminosity (estimated by integrating the greybody be-
tween 8 and 1000 µm) of 6×1012< LIR/L⊙< 2×1013.
They are consistent with the temperatures of galaxies as
a whole as determined by Amblard et al. (2010), who find
a mean temperature of 28 ± 8 K for galaxies detected at
a significance of 3-σ in the H-ATLAS SDP field, and a
mean temperature of 30±9 K for a compilation of galax-
ies in H-ATLAS and a number of other far-infrared sur-
veys. This consistency is unsurprising, since Elbaz et al.
(2010) also found (using deeper data from Herschel) that
the dust temperatures of radio-quiet AGN hosts were not
significantly different from the normal galaxy population.
Thus we consider that a temperature T = 30 K is likely
to be a reasonable choice for our quasar sample, but is
not completely certain, and we return to this point later.
The outlying object has substantial uncertainties
on its temperature, so we do not consider it an indi-
cation of a serious problem with our use of a single
temperature for all objects. However, the fact that it
is at a higher redshift (z = 3.4) and slightly higher
luminosity (LIR/L⊙= 3 × 1013) than the other ob-
jects means that it is broadly consistent with the evo-
lution of dust temperature, to higher temperatures at
higher redshifts and luminosities, seen in several pre-
vious studies (Chapman et al. 2005; Kov´ acs et al. 2006;
Micha? lowski et al. 2010; Amblard et al. 2010) and others.
We do not attempt to use a luminosity- or redshift-
dependent dust temperature in this work. Increasing the
dust temperature serves to shift the greybody peak to
shorter wavelengths, which would result in a higher lu-
minosity fit to our data (which are mostly longward of
the peak). Thus, if the temperature of the dust in quasar
hosts does increase with redshift and/or luminosity, then
we will systematically underestimate LIR for objects at
high redshift and LIR. However, since the trend in mean
temperature found by Amblard et al. is weak compared
with the dispersion around the relation, we do not expect
this to be a very significant effect.
To get an indication of the effect of temperature (and
its uncertainty) on the parameters we derive, we perform
fits to all objects using both T = 30 K and T = 45 K,
where we have selected 45 K because it has a similar peak
wavelength to the average z > 4 quasar SED constructed
by Priddey & McMahon (2001) (who find parameters of
T = 41 ± 5 with β = 1.95 ± 0.3). We also note that,
while β may not in fact be 1.5, the luminosity estimates
obtained using β = 2 (using the best-fitting temperature
at this β) differ by less than 1 per cent, so we consider
that the selection of a value of β is broadly degenerate
with the choice of T, at least for the purpose of estimating
infrared luminosity from SPIRE fluxes.
Following SH09, we assume that the dependence of
the infrared luminosity on the optical quasar luminosity
and redshift can be expressed as power laws, i.e.:
−7.5K. The eight quasars span
LIR,model= ALθ
QSO(1 + z)ζ
(2)
c ? 2010 RAS, MNRAS 000, 1–10

Page 6

6Bonfield et al.
with intrinsic (Gaussian) dispersion in θ and ζ of ∆θ and
∆ζ respectively.
Unlike Serjeant & Hatziminaoglou, who stack in-
frared data to look for trends in bins of redshift and
optical luminosity, we fit for the power law coefficients
of equation (2), and the normalisation A, by finding the
peak in the likelihood distribution calculated over all data
simultaneously. Our approach has the substantial disad-
vantage that variations in the power laws as a function
of redshift or luminosity will be hidden from us. How-
ever, under the assumption that the values of ζ and θ are
the same at all redshifts and luminosities, our method
enables us to make the most sensitive possible determi-
nation of these parameters, since we do not lose informa-
tion through binning. Additionally, we are immune to the
potentially confusing effects of correlations within bins.
We incorporate the dispersions ∆θ and ∆ζ by com-
bining them with the uncertainty σi on each data-
point. The likelihood, Prob(data|θ,ζ,∆θ,∆ζ,A) at a
given point in parameter space is then given by
Prob(data|θ,ζ,∆θ,∆ζ,A) =
?
i
e−χ2
i/2
?2π(σ2
i+ σ2
θ,i+ σ2
ζ,i)(3)
where
χ2
i=(LIR,data,i− LIR,model,i)2
σ2
i+ σ2
θ,i+ σ2
ζ,i
,(4)
σθ,i= LIR,model,ilnLQSO,i∆θ , and(5)
σζ,i= LIR,model,iln(1 + z)∆ζ .(6)
The prior space on the parameters θ, ζ, ∆θ, ∆ζ, and
A, over which the likelihood was evaluated, was chosen
iteratively to enclose more than 99.999 per cent of the
likelihood. The priors were linear in θ, ζ, ∆θ, and ∆ζ,
and logarithmic in A, with the limits ∆θ ? 0 and ∆ζ ? 0.
4RESULTS AND DISCUSSION
Figure 3 shows the results of the likelihood evaluations,
marginalised over all parameters except the two shown in
each panel. This is the result of averaging the likelihoods
obtained in 17 random realisations of the background, as
described in Section 2. Contours labelled 1-σ, 2-σ, 3-σ,
and 4-σ enclose 68.3, 95.4, 99.73, and 99.994 per cent of
the total likelihood. We show the results of fits assuming
both T = 30 K and T = 45 K to give an indication of
the way in which our uncertainty in T should contribute
to the uncertainty in the derived parameters.
This figure shows that, assuming the dust in quasars
has a temperature T = 30 K, there is good evidence for
correlations of LIR with both LQSO and z, since the max-
imum likelihood values of both θ and ζ are significantly
non-zero, and that while higher temperatures can mimic
the effect of the correlation with z, the correlation with
LQSO is robust to simple changes in the overall dust tem-
perature. For the T = 30 K case, the best-fitting values of
each parameter, with their 68.3 per cent confidence lim-
its (marginalising over all other parameters in each case)
are:
θ=0.22 ± 0.08(7)
ζ=1.6 ± 0.4(8)
∆θ=0.08 ±
0.48
0.08
0.9
0.7
(9)
∆ζ=3.4 ±
(10)
Although, as can be seen in the lower left panel of
Figure 3, there is some degeneracy between θ and ζ, our
data exclude θ = 0 at approximately the 2.5-σ level. The
(flat) prior on θ allowed it to take values between -1.5
and 1 (i.e. a much larger range than we show in Figure
3), so we are confident that this is not simply an artefact
of reaching the edge of the prior range.
The other panels of Figure 3 show that there is not
a strong degeneracy between any other parameters, so
it is not possible to reproduce the effect of a correlation
of LIR with either LQSO or z by simply increasing the
intrinsic dispersion in the other parameter; indeed these
panels suggest that approximately the same result would
be obtained if dispersion in the power law indices was
ignored altogether.
While the figure shows that the detected evolution
of LIR with redshift is very strongly dependent on the
assumed value of T (although a temperature as high as
45 K for the whole sample seems unlikely), the correlation
with LQSOseems to be robust to this uncertainty. We find
a slightly weaker index, θ, for the correlation with LQSO
than found by SH09, whose average result (θ = 0.44 ±
0.07) is shown in figure 3 as the point with errorbars.
This may in part be due to the relatively large redshift
bins (e.g. 0.5 < z < 1, 1 < z < 2) which were used in
SH09, since, because of the strong degeneracy between
θ and ζ, these could allow trends with redshift to affect
objects within a single bin. However, the disagreement is
not very significant.
In contrast,performing
likelihood fit to data with far-infrared fluxes drawn from
random positions in the maps (with the same position
used at each wavelength, and the dust-fitting performed
assuming a temperature of 30 K) we obtain the results
shown in Figure 4. The important thing to note is that
the best-fitting value of θ is zero for the random flux case,
indicating that the result we have obtained is not simply
due to the noise properties of the far-infrared maps, but
rather is due to far-infrared flux genuinely associated with
the quasars. (Note that the non-zero value for ζ derived
from the fit to random data is not unexpected, since the
real quasar luminosities and redshifts were still used in
this case, and these are themselves correlated.)
Using deeper far-infrared data from Herschel, but
with a narrower range in optical luminosity (because
only SDSS, and not 2SLAQ, quasars are available in
the field), Hatziminaoglou et al. (2010) find a very sim-
ilar result, θ = 0.35, for objects at redshifts z > 2.
Serjeant et al. (2010) use a compilation of H-ATLAS
and other data to find a weakly declining correlation
between LIR and quasar luminosity, parametrised as
θ = (0.5875 ± 0.045) − (0.09275 ± 0.012) z for redshifts
0 < z < 4, which is a slightly higher value of θ than we
find, but broadly consistent within the errors and at the
redshifts we are sensitive to. Our value of θ is also consis-
thesamemaximum-
c ? 2010 RAS, MNRAS 000, 1–10

Page 7

H-ATLAS: quasar hosts with Herschel7
Figure 3. Likelihood contours for pairs of the parameters θ and ζ (from equation 2) and their intrinsic dispersions ∆θ and ∆ζ,
marginalised over the parameters not displayed in each panel. Contours labelled 1-σ, 2-σ, 3-σ, and 4-σ enclose 68.3, 95.4, 99.73,
and 99.994 per cent of the total likelihood. The point with error bars in the lower left-hand panel shows the average values of
θ and ζ, and their 1-σ uncertainties, determined by SH09 via stacking in the mid-infrared. Solid contours show a fit to objects
assuming T = 30 K; dotted contours show a fit to the objects assuming T = 45 K.
tent with the work by Silverman et al. (2009), who find
θ = 0.28 ± 0.22 for the slope of the power law relating
[Oii] luminosity (tracing star formation after correcting
for AGN contamination) to X-ray luminosity (tracing the
AGN). Silverman et al. study a sample of AGN over a
narrower range of redshifts (0.5 < z < 1) and with some-
what lower luminosities than most of our sample (1044.3
erg s−1<
rection Lbol/L2−10keV ∼ 30); the consistency of their
result with ours suggests that the relationship between
LIR and LQSO may not vary with LQSO as suggested by
Lutz et al. (2010) and Shao et al. (2010). Their steeper
power law, with a slope of 0.8, for luminous quasars is
∼Lbol<
∼1045.8erg s−1, assuming a bolometric cor-
not directly comparable with our result as it does not
separate out the independent effect of redshift on LIR.
However, we would also note that the quasar part of
this relationship is based on samples presented by Netzer
(2009, and earlier works), which ignore AGN without de-
tections in the far-infrared or mm, so that the observed
correlation may be partly due to far-infrared flux limits.
To investigate whether our fit could be affected by a
redshift-dependence in θ, we perform the maximum like-
lihood fitting again, using objects in restricted redshift
ranges where we have enough objects to make a robust
measurement, and marginalise over all parameters except
θ. The results of this are shown in Figure 5; the derived
c ? 2010 RAS, MNRAS 000, 1–10

Page 8

8 Bonfield et al.
Figure 4. As Figure 3 but with infrared fluxes drawn from random positions in the maps rather than from the quasar positions,
as a check that the noise properties in the maps are not responsible for the correlation we detect between LIRand LQSO.
values of θ are not significantly different from each other
for redshifts 0.5 < z < 2.5, although the small number
of objects in the redshift bins means we cannot make a
strong statement for or against evolution of θ as a func-
tion of redshift. However, the 0.5 < z < 1 range shows a
detection of non-zero θ at better than the 2-σ level, which
is complementary to the result of Hatziminaoglou et al.
(2010). While we do not have enough data to detect any
significant decline in θ as a function of redshift, the data-
points in Figure 5 are consistent with the relation found
by Serjeant et al. (2010), and with our smaller H-ATLAS-
only dataset we cannot rule out the existence of such a
weak redshift dependence.
Since the far-infrared luminosity LIRhas been shown
to be due to star formation (Hatziminaoglou et al. 2010),
the physical interpretation of these results is that star-
formation rate is higher both as redshift increases and
as quasar luminosity increases. As was also suggested by
SH09, and as has been incorporated into the “quasar-
mode” feedback recipes of semi-analytic models (e.g.
Croton et al. 2006), one of the simplest ways to inter-
pret this is if both black hole accretion and star for-
mation depend on a common supply of cold gas, as
might be supplied during a galaxy–galaxy merger or cold-
accretion event (Dekel et al. 2009; Dijkstra & Loeb 2009;
Smith & Jarvis 2007; Smith et al. 2008).
tion of the non-unity-slope is that the dependence of ei-
ther star-formation or accretion luminosity (or both) on
the gas supply cannot simply be linear with gas mass, as
one might na¨ ıvely suppose, but is instead more complex.
For both star formation and black-hole accretion,
such gas must be sufficiently cool that pressure support
The implica-
c ? 2010 RAS, MNRAS 000, 1–10

Page 9

H-ATLAS: quasar hosts with Herschel9
Figure 5. Best-fitting values of θ based on maximum-
likelihood fits using data in the redshift ranges denoted by
the widths of the errorbars. Black, solid errorbars are 68.3 per
cent confidence limits; blue, dashed errorbars are 95.4 per cent
confidence limits.
does not prevent it from being accreted by the AGN or
from collapsing under its own gravity to form stars, so an
alternative way to make this connection is via a mecha-
nism (such as AGN or supernova feedback) which is ca-
pable of heating gas at all scales within a galaxy.
5 CONCLUSIONS
Using far-infrared measurements (at 250, 350 and
500 µm) of SDSS and 2SLAQ quasars in the H-ATLAS
science demonstration field, we find evidence for corre-
lations between the infrared luminosities of quasar hosts
with both redshift and quasar luminosity. While the best-
fit power-law index for the correlation with redshift can
vary significantly depending on the assumed tempera-
ture of the dust in the quasar hosts, the correlation with
quasar luminosity is robust to this effect, and is non-
zero at the 2.5-σ level when we combine all objects in a
maximum-likelihood fit.
We find a power-law index θ = 0.22 ± 0.08 for the
relationship between LIR and Lqso. Our data show that
θ is non-zero at the 2-σ level in the 0.5 < z < 1 redshift
range alone, but we have too few objects to find evidence
for evolution in θ over the range 0.5 < z < 2.5.
We find evidence for a large intrinsic dispersion in the
redshift dependence, but no evidence for intrinsic disper-
sion in the correlation between LQSOand LIR, suggesting
that the latter is due to a direct physical connection be-
tween star formation and black hole accretion. One pos-
sible interpretation of this is that both the quasar activity
and star formation are dependent on the same reservoir
of cold gas, and are thus both affected by influx of gas
during mergers or cold accretion, or heating of gas via
feedback processes.
With the full 550 square degrees of the H-ATLAS
survey, we will be able to substantially increase the sam-
ple of quasars and improve these constraints. The larger
area will allow us to obtain representative samples of the
objects with the lowest density on the sky, i.e. quasars
with high luminosities at low redshift, which will increase
the sensitivity of our test at low redshifts and permit us to
remove the assumption of power-law correlations and test
more general models. More importantly, a larger sample
of objects at all redshifts will make it possible to measure
the temperatures of dust in quasar hosts as a function of
Lqso and redshift by stacking the far-infrared fluxes; this
in turn will remove the largest source of uncertainty in
the relationship between LIR and z.
ACKNOWLEDGEMENTS
MJJ acknowledges support from an RCUK fellowship.
MJH thanks the Royal Society for a Research Fellowship.
The Herschel-ATLAS is a project with Herschel,
which is an ESA space observatory with science instru-
ments provided by European-led Principal Investigator
consortia and with important participation from NASA.
The H-ATLAS website is http://www.h-atlas.org/.
Funding for the SDSS and SDSS-II has been pro-
vided by the Alfred P. Sloan Foundation, the Partic-
ipating Institutions, the National Science Foundation,
the U.S. Department of Energy, the National Aeronau-
tics and Space Administration, the Japanese Monbuka-
gakusho, the Max Planck Society, and the Higher Educa-
tion Funding Council for England. The SDSS Web Site
is http://www.sdss.org/.
The SDSS is managed by the Astrophysical Research
Consortium for the Participating Institutions. The Par-
ticipating Institutions are the American Museum of Nat-
ural History, Astrophysical Institute Potsdam, Univer-
sity of Basel, University of Cambridge, Case Western
Reserve University, University of Chicago, Drexel Uni-
versity, Fermilab, the Institute for Advanced Study, the
Japan Participation Group, Johns Hopkins University,
the Joint Institute for Nuclear Astrophysics, the Kavli
Institute for Particle Astrophysics and Cosmology, the
Korean Scientist Group, the Chinese Academy of Sci-
ences (LAMOST), Los Alamos National Laboratory, the
Max-Planck-Institute for Astronomy (MPIA), the Max-
Planck-Institute for Astrophysics (MPA), New Mexico
State University, Ohio State University, University of
Pittsburgh, University of Portsmouth, Princeton Univer-
sity, the United States Naval Observatory, and the Uni-
versity of Washington.
REFERENCES
Alexander D. M., et al., 2005, ApJ, 632, 736
Amblard A., et al., 2010, A&A, 518, L9+
Antonucci R., 1993, ARA&A, 31, 473
Archibald E. N., Dunlop J. S., Hughes D. H., Rawlings
S., Eales S. A., Ivison R. J., 2001, MNRAS, 323, 417
c ? 2010 RAS, MNRAS 000, 1–10

Page 10

10Bonfield et al.
Bahcall J. N., Kirhakos S., Schneider D. P., 1994, ApJ,
435, L11
Baldi R. D., Capetti A., 2008, A&A, 489, 989
Benson A. J., Bower R. G., Frenk C. S., Lacey C. G.,
Baugh C. M., Cole S., 2003, ApJ, 599, 38
Bower R. G., et al., 2006, MNRAS, 370, 645
Chapman S. C., Blain A. W., Smail I., Ivison R. J.,
2005, ApJ, 622, 772
Cole S., Lacey C. G., Baugh C. M., Frenk C. S., 2000,
MNRAS, 319, 168
Croom S. M., et al., 2009, MNRAS, 392, 19
Croton D. J., et al., 2006, MNRAS, 365, 11
Dekel A., et al., 2009, Nature, 457, 451
Dijkstra M., Loeb A., 2009, MNRAS, 400, 1109
Dunlop J. S., McLure R. J., Kukula M. J., Baum S. A.,
O’Dea C. P., Hughes D. H., 2003, MNRAS, 340, 1095
Eales S., et al., 2010, PASP, 122, 499
Elbaz D., et al., 2010, A&A, 518, L29+
Ferrarese L., Merritt D., 2000, ApJ, 539, L9
Fritz J., Franceschini A., Hatziminaoglou E., 2006, MN-
RAS, 366, 767
Gebhardt K., et al., 2000, ApJ, 539, L13
Gonz´ alez-Nuevo J., et al., 2010, A&A, 518, L38+
Granato G. L., De Zotti G., Silva L., Bressan A., Danese
L., 2004, ApJ, 600, 580
Granato G. L., et al., 2001, MNRAS, 324, 757
Griffin M. J., et al., 2010, A&A, 518, L3+
Guiderdoni B., Hivon E., Bouchet F. R., Maffei B., 1998,
MNRAS, 295, 877
Hatziminaoglou E., et al., 2010, A&A, 518, L33+
Herbert P. D., et al., 2010, MNRAS, 406, 1841
Ibar E., et al., 2010, MNRAS, accepted, arXiv:1009.0262
Jahnke K.,Macci` o A.,
arXiv:1006.0482
Jarvis M. J., et al., 2001, MNRAS, 326, 1585
Kotilainen J. K., Falomo R., Decarli R., Treves A.,
Uslenghi M., Scarpa R., 2009, ApJ, 703, 1663
Kov´ acs A., et al., 2006, ApJ, 650, 592
Laird E. S., Nandra K., Pope A., Scott D., 2010, MN-
RAS, 401, 2763
Lutz D., et al., 2008, ApJ, 684, 853
—, 2010, ApJ, 712, 1287
Magorrian J., et al., 1998, AJ, 115, 2285
Mart´ ınez-Sansigre A., Lacy M., Sajina A., Rawlings S.,
2008, ApJ, 674, 676
Mart´ ınez-Sansigre A., et al., 2009, ApJ, 706, 184
McLure R. J., Kukula M. J., Dunlop J. S., Baum S. A.,
O’Dea C. P., Hughes D. H., 1999, MNRAS, 308, 377
McLure R. J., et al., 2004, MNRAS, 351, 347
Micha? lowski M., Hjorth J., Watson D., 2010, A&A, 514,
A67+
Nenkova M., Sirocky M. M., Nikutta R., Ivezi´ cˇZ.,
Elitzur M., 2008, ApJ, 685, 160
Netzer H., 2009, MNRAS, 399, 1907
Netzer H., et al., 2007, ApJ, 666, 806
Nolan L. A., Dunlop J. S., Kukula M. J., Hughes D. H.,
Boroson T., Jimenez R., 2001, MNRAS, 323, 308
Omont A., et al., 2003, A&A, 398, 857
Page M. J., Stevens J. A., Ivison R. J., Carrera F. J.,
2004, ApJ, 611, L85
Pascale E., et al., 2010, MNRAS, submitted
Pilbratt G. L. a., 2010, A&A, 518, L1+
2010,ApJL, submitted,
Poglitsch A., et al., 2010, A&A, 518, L2+
Priddey R. S., Gallagher S. C., Isaak K. G., Sharp R. G.,
McMahon R. G., Butner H. M., 2007, MNRAS, 374,
867
Priddey R. S., Isaak K. G., McMahon R. G., Robson
E. I., Pearson C. P., 2003, MNRAS, 344, L74
Priddey R. S., McMahon R. G., 2001, MNRAS, 324, L17
Rigby E., et al., 2010, MNRAS, submitted
S´ anchez S. F., et al., 2004, ApJ, 614, 586
Schneider D. P., et al., 2010, AJ, 139, 2360
Schweitzer M., et al., 2006, ApJ, 649, 79
Serjeant S., Hatziminaoglou E., 2009, MNRAS, 397, 265
Serjeant S., et al., 2010, A&A, 518, L7+
Seymour N., et al., 2007, ApJS, 171, 353
Shao L., et al., 2010, A&A, 518, L26+
Shi Y., Rieke G. H., Ogle P., Jiang L., Diamond-Stanic
A. M., 2009, ApJ, 703, 1107
Silverman J. D., et al., 2009, ApJ, 696, 396
Smith D.,etal.,2010,
arXiv:1007.5260
Smith D. J. B., Jarvis M. J., 2007, MNRAS, 378, L49
Smith D. J. B., Jarvis M. J., Lacy M., Mart´ ınez-Sansigre
A., 2008, MNRAS, 389, 799
Stevens J. A., et al., 2003, Nature, 425, 264
—, 2010, MNRAS, 405, 2623
Trichas M., Georgakak is A., Rowan-Robinson M., Nan-
dra K., Clements D., Vaccari M., 2009, MNRAS, 399,
663
Trichas M., et al., 2010, MNRAS, 405, 2243
York D. G., et al., 2000, AJ, 120, 1579
MNRAS, submitted,
This paper has been typeset from a TEX/ LATEX file pre-
pared by the author.
c ? 2010 RAS, MNRAS 000, 1–10