Mon. Not. R. Astron. Soc.

326, 1533–1546 (2001)

A NICMOS imaging study of high-z quasar host galaxies

Marek J. Kukula,1P James S. Dunlop,1 Ross J. McLure,2 Lance Miller,2 Will J. Percival,1
Stefi A. Baum3 and Christopher P. O’Dea3
1
Institute for Astronomy, University of Edinburgh, Royal Observatory, Edinburgh EH9 3HJ
2
Nuclear and Astrophysics Laboratory, University of Oxford, Keble Road, Oxford OX1 3RH
3
Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA

Accepted 2001 May 23. Received 2001 May 17; in original form 2000 October 4

A B S T R AC T
We present the first results from a major Hubble Space Telescope programme designed to
investigate the cosmological evolution of quasar host galaxies from z . 2 to the present day.
Here we describe J and H-band NICMOS imaging of two quasar samples at redshifts of 0.9
and 1.9 respectively. Each sample contains equal numbers of radio-loud and radio-quiet
quasars, selected to lie within the same narrow range of optical absolute magnitude
ð224 $ M V $ 225Þ. Filter and target selection were designed to ensure that at each redshift
the images sample the same part of the object’s rest-frame spectrum, longwards of 4000 Å,
where starlight from the host galaxy is relatively prominent, but avoiding potential
contamination by [O III ]l5007 and Ha emission lines.
At z . 1 we have been able to establish host-galaxy luminosities and scalelengths with
sufficient accuracy to demonstrate that the hosts of both radio-loud and radio-quiet quasars lie
on the same Kormendy relation described by 3CR radio galaxies at comparable redshift.
Taken at face value the gap between the host luminosities of radio-loud and radio-quiet
objects appears to have widened from only . 0.4 mag at z . 0:2 to . 1 mag at z . 1, a
difference that cannot be a result of emission-line contamination, given the design of our
study. However, within current uncertainties, simple passive stellar evolution is sufficient to
link these galaxies with the elliptical hosts of low-redshift quasars of comparable nuclear
output, implying that the hosts are virtually fully assembled by z , 1.
At z . 2 the hosts have proved harder to characterize accurately, and for only two of the
nine z . 2 quasars observed has it proved possible to properly constrain the scalelength of the
host galaxy. However, the data are of sufficient quality to yield host-galaxy luminosities
accurate to within a factor of . 2. At this redshift the luminosity gap between radio-loud and
radio-quiet quasars appears to have widened further to . 1.5 mag. Thus while the hosts of
radio-loud quasars remain consistent with a formation epoch of z . 3, allowing for passive
evolution implies that the hosts of radio-quiet quasars are . 2–4 times less massive at z . 2
than at z . 0:2.
If the relationship between black hole and spheroid mass is unchanged out to redshift
z . 2, then our results rule out any model of quasar evolution which involves a substantial
component of luminosity evolution. Rather, this study indicates that at z . 2 there is a
substantial increase in the number density of active black holes, along with a moderate
increase in the fuelling efficiency of a typical observed quasar. The fact that this latter effect is
not displayed by the radio-loud objects in our sample might be explained by a selection effect
arising from the fact that powerful radio sources are only produced by the most massive black
holes.
Key words: galaxies: active – galaxies: evolution – quasars: general.

P
E-mail: [email protected]

q 2001 RAS
1534 M. J. Kukula et al.
galaxies. This result is consistent with measurements of the
1 INTRODUCTION
massive dark objects in the nuclei of nearby inactive galaxies,
Recent years have brought great advances in our understanding of which predict that only the largest spheroidal systems will harbour
the symbiotic relationship between active galactic nuclei (AGN) black holes of the requisite mass to produce luminous quasars
and the galaxies in which they occur. Some of the most impressive (Kormendy & Richstone 1995; Magorrian et al. 1998; van der
advances concern quasars – the most powerful known AGN – as Marel 1999), and also with the discovery by McLeod & Rieke
improvements in ground-based observing techniques and the (1995) of a lower limit to the mass of a quasar host which is
advent of the Hubble Space Telescope (HST ) have allowed the correlated with the luminosity of the quasar. The implication of
diffuse ‘fuzz’ of the underlying host galaxy to be reliably separated these studies is that massive elliptical galaxies are the parent
from the wings of the bright non-stellar nuclear point spread population of the quasar phenomenon. However, in a ground-based
function (PSF) for the first time. K-band study of luminous ð225 $ M V $ 227Þ RQQs, Percival
The problems inherent in observing quasar host galaxies from et al. (2001) found evidence that, on scales much larger than those
the ground are too well-known to require detailed explanation but, probed in optical HST images, some of the hosts are dominated by
considering the enormous difficulties involved in separating faint, a disc component. Clearly the issue of host morphology has yet to
diffuse galaxy light from the PSF of a bright quasar, it is perhaps be entirely resolved, but the finding that a quasar host requires at
surprising that so much progress has been made to date using the very least a massive spheroidal component seems secure.
ground-based techniques. Although such studies are effectively Spectral energy distributions (SEDs) also provide important
limited to z # 0:3, at these low redshifts a combination of ground- information on the nature of the host galaxies. McLure et al. (1999)
based and HST programmes is beginning to yield a coherent picture and Dunlop et al. (2001) found that the R 2 K colours of their
of the properties of quasar hosts in the local universe. quasar hosts were consistent with normal passively evolving
However, the local universe is not the most representative region elliptical galaxies with ages of , 12 Gyr, implying that the stellar
in which to study the quasar population: arguably the epoch of populations formed at high redshift ðz , 3Þ. Optical off-nuclear
greatest importance to quasar research occurred at redshifts of spectroscopy of the same objects (Hughes et al. 2000; Nolan et al.
z , 2 to 3, when quasars were 2–3 orders of magnitude more 2001) confirms that, longwards of the 4000 Å break in their rest-
numerous than they are today. Although simulations show that it is frame spectra, these hosts are dominated by light from an old, well-
very difficult to derive the properties of quasar hosts at such high z established population of stars. Similar results are obtained for
from the ground with any degree of confidence (Taylor et al. 1996), radio galaxies (e.g. de Vries et al. 2000).
numerous attempts have been made with (not surprisingly)
confusing results. In this paper we present the results of a study
1.2 High-redshift quasar hosts
using the Near-Infrared Camera and Multi-Object Spectrometer
(NICMOS) on HST to investigate how the luminosities, sizes and The fact that the host galaxies of low-redshift quasars have mature
morphologies of the hosts of both radio-loud and radio-quiet stellar populations and contain a massive bulge component implies
quasars (RLQs and RQQs) have evolved from the ‘golden era’ of that, like local inactive elliptical galaxies, they are arguably
quasar activity to the present day. consistent with the products of successive merger events. This
raises important questions about the potential redshift dependence
of the properties of quasar hosts. The dramatic cosmological
1.1 The host galaxies of low-redshift quasars
evolution exhibited by the quasar population itself, with a
At low redshifts, attention has been focused on determining the comoving number density which peaks at redshifts of 2–3 before
luminosities, scalelengths, morphologies and interaction histories undergoing a rapid decline to its present low value, has been known
of quasar host galaxies, and investigating the extent to which these for many years (e.g. Boyle, Shanks & Peterson 1988; Dunlop &
properties are correlated to the optical and radio luminosity of the Peacock 1990; Warren, Hewett & Osmer 1994) but the extent and
quasar (Véron-Cetty & Woltjer 1990; Dunlop et al. 1993; McLeod form of any evolution affecting quasar hosts remain as largely
& Rieke 1994a,b; Bahcall, Kirhakos & Schneider 1995a,b,1996; unknown quantities. The issue has been given renewed urgency by
Hutchings & Morris 1995; Disney et al. 1995; Taylor et al. 1996; continuing efforts to characterize the star formation history of the
Bahcall et al. 1997; Hooper, Impey & Foltz 1997; Boyce et al. universe, for which the redshift range of z ¼ 2–3 also appears to
1998; Carballo et al. 1998; McLeod, Rieke & Storrie-Lombardi correspond to an important epoch (Madau et al. 1996; Hughes et al.
1999; Schade, Boyle & Letawsky 2000; Hamilton, Casertano & 1998; Steidel et al. 1999).
Turnshek 2001; Márquez et al. 2001). For example, it has long been It is therefore important to determine how the luminosities,
known that low-luminosity AGN display marked preferences in scalelengths and the degree of morphological disturbance seen in
terms of host type, with (radio-quiet) Seyferts favouring spiral quasar host galaxies varies with cosmological epoch between the
hosts whilst (radio-loud) radio galaxies are exclusively associated peak of quasar activity at z . 2 and the present day, and whether
with massive ellipticals. However, recent studies have demon- the hosts of radio-loud and radio-quiet quasars differ in their
strated that this distinction breaks down at the higher redshifts and luminosity or morphological evolution. However, the difficulties
nuclear luminosities typical of radio-loud and radio-quiet quasars, involved in such studies are far more formidable than at low z and
and that powerful nuclear activity is predominantly associated with this is reflected in the confusion which surrounds the interpretation
bulge-dominated galaxies, regardless of radio luminosity. of ground-based attempts to observe high-redshift quasar hosts.
In our own HST study of low-redshift ð0:1 # z # 0:25Þ, low- Several groups have attempted such observations from the
luminosity ð223 # M 2 V # 226Þ RLQs and RQQs (McLure ground and have succeeded in detecting extensions around high-
et al. 1999; Dunlop et al. 2001) we found that for quasars brighter redshift quasars (Hintzen, Romanishin & Valdes 1991; Heckman
than M R , 224, the hosts are invariably massive elliptical et al. 1991; Lehnert et al. 1992; Aretxaga, Boyle & Terlevich 1995;
galaxies with L $ 2L*, with typical scalelengths of , 10 kpc, and Rönnback et al. 1996; Aretxaga, Terlevich & Boyle 1998). These
display a Kormendy relation identical to that of brightest cluster studies suggest that the host galaxies of quasars at z . 2 are , 2.5

q 2001 RAS, MNRAS 326, 1533–1546

 A NICMOS imaging study of quasar hosts 1535
to 3 mag brighter than those of low-redshift quasars, a result high-redshift ðz , 1, 2) samples, including redshifts, positions and
which is consistent with common scenarios for elliptical galaxy observing dates, are given in Table 1. It is these two samples which
evolution. However, the usefulness of any such direct comparison are the subject of the present paper.
is confused by the fact that these high-redshift quasars are typically
5 magnitudes more luminous than the low-redshift objects studied
to date. In addition, the observational situation is somewhat 2.1 Sample design and choice of filters
complex since other workers have failed to detect extended 2.1.1 Luminosity constraints
emission around high-redshift RQQs (Lowenthal et al. 1995).
Hutchings (1995) also concludes that the hosts of RQQs are We have confined our selection of high-redshift quasars to the
considerably fainter than those of RLQs at high redshift, a result absolute magnitude range 224 . M V . 225. This luminosity
which could however simply indicate that much of the light band is comparable to that of a subset of our low-redshift ð0:1 #
detected around high-redshift RLQs is not caused by stars, but z # 0:25Þ quasar sample observed with WFPC2 on HST (McLure
rather by processes associated with the extreme radio activity, as et al. 1999; Dunlop et al. 2001), allowing us to use this latter
is found for high-z radio galaxies (Tadhunter et al. 1992), and for sample as a low-redshift baseline against which to measure any
low-z RLQs (Stockton & MacKenty 1987). This emphasizes the redshift-dependent change in the properties of the hosts for a
importance of avoiding emission lines and sampling the rest-frame narrow range of quasar luminosities. Also, experience from low-z
spectrum longwards of the 4000 Å break. studies, as well as the results of simulations using artificial
The study described in this paper attempts to avoid the problems quasar/host combinations and our 2D modelling algorithm,
outlined above by observing test samples of quasars over a range of demonstrates that in order to reliably recover the luminosities
redshifts out to z ¼ 2 but limited to a narrow range of optical and scalelengths of their host galaxies it is currently desirable to
absolute magnitude ð224 # M V # 225Þ. Careful choice of filters limit the high-redshift samples to quasars of such moderate-to-low
ensures that our images always sample the same emission line free optical luminosities.
region of the object’s rest frame spectrum, thus avoiding any
regions of extended [O III ]l5007 and Ha emission, which may be 2.1.2 Filter/redshift combinations
associated with the active nucleus. Thus, at each redshift we are
always observing objects of roughly the same intrinsic luminosity, Having defined the luminosity range of the sample, the next
and in the same region (approximately V-band) of their rest-frame priority was to break the flux-density/redshift correlation which
spectrum. inevitably arises in samples derived from flux-limited surveys, by
The layout of this paper is as follows. In Section 2 we provide ensuring that we sample a comparable range of luminosities at
details of how our quasar samples were selected, and how the HST several different redshifts. However, to obtain a clean view of the
NICMOS observations were designed and implemented. The starlight from the host galaxies at all redshifts it is also necessary
process of data reduction is described in Section 3, including the to ensure that, as in our previous WFPC2 study at z . 0:2, the
details of how we chose to tackle the particular problems spectrum of the host galaxy is always observed at lrest . 4000A 
associated with producing NICMOS images of sufficient quality yet is uncontaminated by either [O III ] or Ha line emission. In the
for a study of this type. In Section 4 we explain how host-galaxy z . 0:2 study this entailed using the F675W filter on WFPC2. For
parameters were extracted from the data via 2D modelling, and the current study two NICMOS filters were selected: F110M
summarize the basic results of the image analysis. Then in Section (roughly J-band), which samples the desired rest-frame wave-
5 we explore the implications of combining these new results with lengths for redshifts 0:83 # z # 1:00 and F165M (roughly
the results of our Cycle 6 HST study of quasar hosts at much lower H-band), for 1:67 # z # 2:01. A third filter, F184W (roughly
redshift ðz . 0:2; McLure et al. 1999; Dunlop et al. 2001), and I-band) on WFPC2, was also chosen to bridge the gap between the
discuss our findings in the context of existing theoretical NICMOS observations and the McLure et al. (1999) low-redshift
predictions. Finally, our results are summarized in Section 6. For study by observing a third group of objects at 0:32 # z # 0:43.
ease of comparison with existing results/predictions, we assume (HST possesses slightly wider filters in all three of these
an Einstein – de Sitter cosmology with H 0 ¼ 50 km s21 Mpc21 wavelength regimes, but using these would have led to
throughout most of this paper. However, in the penultimate section contamination by [O III ] or Ha.) Thus the current study comprises
we focus on the implications of our results within the currently three samples at redshifts of z , 0:4, 1 and 2, using filters which
favoured flat cosmological model with Vm ¼ 0:3, VL ¼ 0:7 and approximate to standard I-, J- and H-bands respectively. At the
H 0 ¼ 65 km s21 Mpc21 . time of writing the WFPC2 observations of the sample at z , 0:4
were incomplete. Hence, the current paper deals only with the
NICMOS observations of the two high-z quasar samples.
2 T H E O B S E RVAT I O N S Because at each redshift our chosen filter corresponds to rest-
Previous observations of quasar hosts with HST have demonstrated frame V-band, this observing strategy also obviates the need for
the importance of careful sample construction, the matching of k-corrections when calculating host-galaxy luminosities. Absolute
samples and the selection of filter bandwidths to avoid the V-band magnitudes can be calculated directly from the observed R,
inclusion of strong emission lines. In this section we describe the I, J and H magnitudes without the need to assume any particular
selection criteria used in the construction of our quasar samples as spectral shape (and hence stellar population) for the hosts.
well as the considerations which shaped the final observing
strategy. The three samples resulting from these considerations, at
2.1.3 Radio loudness
z , 0:4, 1 and 2, are shown in Fig. 1. The observations of the
z , 0:4 sample, which use WFPC2 rather than NICMOS, were In order to attempt a meaningful comparison of the properties of
incomplete at the time of writing and discussion of the images of RLQ and RQQ hosts as a function of redshift, we selected five
these objects is deferred to a subsequent paper. Details of the two radio-loud and five radio-quiet quasars in each redshift regime,

q 2001 RAS, MNRAS 326, 1533–1546

1536 M. J. Kukula et al.

Figure 1. Left: absolute V magnitude versus redshift for the RQQs (filled circles) and RLQs (open circles) in the current study. The shaded vertical bars
represent the redshift regimes over which the indicated WFPC2 and NICMOS filters remain free from contamination by [O III ]l5007 and Hal6563 emission
lines. Also shown are the quasars from our previous HST imaging study using WFPC2/F675W (McLure et al. 1999; Dunlop et al. 2001). This low-z sample
spans almost 3 magnitudes in optical luminosity at a given redshift ðz . 0:2Þ whereas the three samples selected for the current study are confined to only 1 mag
in luminosity but span the bulk of the history of the Universe. Thus these complementary low-z objects provide the baseline against which to measure any
cosmological evolution in the high-z samples. Right: schematic showing typical rest-frame spectra of a quasar (upper panel) and an early-type galaxy (lower
panel). The two vertical lines delineate the region of the host-galaxy spectrum which is targeted by the redshift/filter combinations used in the observations
described in this paper. Note the avoidance of prominent quasar emission lines and the increase in galaxy flux density longwards of the break feature at 4000 Å.

Table 1. Summary of the quasar samples observed in the current NICMOS study. RQQs are derived from the UVX survey by Boyle et al. (1990) (SGP) and the
survey of Marshall et al. (1984) (BVF), and are confirmed as radio quiet ðP5 GHz , 1024:5 W Hz21 sr21 Þ by the Very Large Array (VLA) FIRST and NVSS
surveys. RLQs were selected from the Véron-Cetty & Véron (1993) quasar catalogue, with reported 5-GHz radio luminosities P5 GHz . 1025 W Hz21 sr21 and
steep radio spectra. V-band magnitudes are from various sources, and where necessary have been converted from B-band measurements. For the estimation of
luminosities at both radio and optical wavelengths we have assumed a quasar spectrum of the form f ðnÞ / n 20:5 .

J2000 Position Observing date Time on

Name Type RA (h m s) Dec (8 0 00 ) z V (dd/mm/yy) source (s)

J-band (F110M) sample ðz , 1Þ

SGP5:46 RQQ 00 52 22.76 227 30 02.8 0.955 18.88 25/08/97 1  2048
SGP2:47 RQQ 00 53 02.24 229 12 55.5 0.830 17.17 14/08/97 1  2048
BVF225 RQQ 13 04 10.47 135 36 50.8 0.910 18.54 16/02/98 1  2048
BVF247 RQQ 13 05 05.04 135 51 20.6 0.890 18.73 15/02/98 1  2048
BVF262 RQQ 13 05 30.91 135 17 13.5 0.970 18.57 18/02/98 1  2048
PKS0440 2 00 RLQ 04 42 38.64 200 17 43.4 0.844 18.41 11/08/97 1  2048
PKS0938118 RLQ 09 41 23.17 118 21 06.0 0.943 18.49 18/02/98 1  2048
3C422 RLQ 20 47 10.39 202 36 22.5 0.942 18.69 03/11/97 1  2048
MC21121172 RLQ 21 14 56.68 117 29 22.7 0.878 17.89 30/10/97 1  2048
4C02.54 RLQ 22 09 32.82 102 18 40.9 0.976 18.19 05/11/97 1  2048
SA107-626 PSF star 15 40 05.31 200 17 29.2 0 12.77 30/07/97 8  256
SA107-627 PSF star 15 40 07.45 200 17 23.0 0 13.47 17/02/98 8  256

H-band (F165M) sample ðz , 2Þ

SGP2:36 RQQ 00 51 14.32 229 05 19.7 1.756 19.62 20/11/97 3  2048
SGP2:25 RQQ 00 52 07.60 229 17 50.2 1.868 19.55 28/08/97 4  2048
SGP2:11 RQQ 00 52 38.47 228 51 12.9 1.976 19.77 02/09/97 4  2048
SGP3:39 RQQ 00 55 43.41 228 24 09.7 1.964 19.65 06/09/97 4  2048
SGP4:39 RQQ 00 59 08.88 227 51 24.7 1.716 19.64 30/08/97 3  2048
1148156W1 RLQ 11 50 44.8 156 32 56.0 1.782 19.69 30/06/97 4  2048
PKS1524 2 13 RLQ 15 26 59.44 213 51 01.3 1.687 19.69 31/07/97 4  2048
B2 2 2156 1 29 RLQ 21 58 42.0 129 59 08.0 1.759 19.29 05/11/97 3  2048
PKS2204 2 20 RLQ 22 07 33.94 220 38 34.9 1.923 19.49 04/09/97 4  2048
4C45.51 RLQ 23 54 22.27 145 53 05.2 1.992 19.79 14/11/97 4  2048
SA107-626 PSF star 15 40 05.31 200 17 29.2 0 12.77 17/02/98 4  256
SA107-627 PSF star 15 40 07.45 200 17 23.0 0 13.47 30/07/97 4  256

giving a total of 30 objects. In order to perform a clean comparison the optical luminosity distributions of the two types are well
of the hosts of RLQs and RQQs at each epoch it was important matched. This is not a trivial task. It requires that we select only
to ensure that the RLQs are genuinely radio-loud ðP5 GHz . RQQs which have been observed with the Very Large Array (VLA)
1025 W Hz21 sr21 Þ, that the RQQs are genuinely radio-quiet with high sensitivity but have not been detected, and requires that
ðP5 GHz , 1024:5 W Hz21 sr21 Þ, and that within each redshift bin we confine our RLQ sample to steep-spectrum objects whose

q 2001 RAS, MNRAS 326, 1533–1546

 A NICMOS imaging study of quasar hosts 1537
intrinsic radio luminosity is not being artificially boosted by known. This urgency is compounded by the fact that the PSF is a
relativistic beaming. complicated function of the filter used and the SED of the target as
The relatively small number of optically-faint steep-spectrum well as the position of the object within the aperture.
RLQs, along with our stringent definition of radio quietness (few On-orbit verification and calibration of the NICMOS instrument
RQQs have been observed with sufficient sensitivity in the radio to did not include the acquisition of empirical PSFs for every
meet our self-imposed luminosity requirement), means that the combination of filter and camera. Synthetic PSFs, although
resulting three quasar samples at z , 0:4, 1 and 2 (illustrated in normally an excellent match in the bright central regions, are often
Fig. 1) are almost uniquely defined. not so good at reproducing the faint outer wings of the structure
and usually fail to account for the effects of instrumental defects
and uncertainties. We therefore used two orbits of our allocated
2.2 Observing strategy HST time to obtain deep NICMOS stellar PSFs in both our chosen
2.2.1 Choice of detector filters, F110M and F165M. To safeguard against the possibility of
Vega-like circumstellar dust shells which would compromise the
NICMOS contains three HgCdTe arrays, NIC1, NIC2 and NIC3, PSF, we observed two different stars, SA 107-626 and SA 107-627.
each consisting of 256  256 pixels. Unlike CCDs, each pixel in The stars have absolute magnitudes, MV, of 13.47 and 13.34
the array is independent of its neighbours and can be read non- respectively, enabling us to obtain high-dynamic range images of
destructively, allowing multiple read-outs to be performed each star, tracing the PSF much further into the wings than the
throughout the duration of a single exposure. This ability can be quasar exposures, in under half an orbit. Observations with the
utilized for the recognition and removal of cosmic ray events United Kingdom Infrared Telescope (UKIRT) in 1997 April
during the course of the exposure. confirmed that the J 2 K colours of the stars were 0.60 and 0.48
We opted to use NIC1, the smallest of the three detectors, with a respectively – and thus a good match to the expected quasar SED
field of view 11  11 arcsec in extent. The NIC1 PSF is diffraction within our chosen filter bands.
limited for wavelengths $1 mm and its 0.043 arcsec pixels
guarantee critical sampling of the PSF over the wavelength range
of interest. 3 N I C M O S D ATA R E D U C T I O N
The data were calibrated using the standard NICMOS pipeline
2.2.2 Integration times software, CALNICA , together with the most recent calibration files.
However, certain peculiarities of the NICMOS Camera 1 detector
A total of 60 orbits of HST time was allocated for this project. were not adequately compensated for by the pipeline and had to be
From this total, 11 orbits were devoted to the WFPC2/F814W dealt with separately. The most important of these were the
observations of the 10 quasars in the z , 0:4 sample as well as a problems of cosmic ray persistence and the residual DC ‘pedestal’.
suitable PSF star, leaving 49 orbits for the two high-z NICMOS
samples. Simulations using the 2D modelling algorithm developed
for the analysis of our previous HST study of quasar hosts at 3.1 Cosmic ray persistence
z , 0:2, along with pre-launch predictions of the sensitivity of Because the declinations of our target quasars typically meant that
NICMOS through the F110M and F165M filters, indicated that in they would be visible for 3300 s orbit21, the period of visibility
order to detect host galaxies comparable in size and luminosity to remaining after the main MIF2048 exposure and associated read-
those found at low redshift we would require one orbit per object time was utilized for an additional 512-s exposure (using the pre-
for the sample at z , 1 and three to four orbits per object at z , 2 defined MIF512 sequence). This exposure was placed at the
(the final allocation of orbits is shown in Table 1). Two orbits were beginning of the orbit and was used to identify any ‘persistent’
reserved for observations of PSF stars through each of the two cosmic ray tracks that had not been completely flushed from
NICMOS filters (see below). the detector before the start of the observations. Cosmic ray
The NICMOS detectors allow several standard observing persistence is a particular problem in those observations which take
modes, taking advantage of the non-destructive readout capabili- place immediately after one of the spacecraft’s periodic passages
ties of the detector arrays. We chose to use one of the pre-defined through the South Atlantic Anomaly, but can also occur at any
‘multi-accumulate’ (MULTIACCUM ) sequences, with a total of point on the orbit of the HST. Unlike cosmic rays which impact
25 readouts at specific intervals during each integration, which during a science exposure, the persistent tracks cannot be removed
combine high sensitivity with large dynamic range and the ability by comparing adjacent readouts of a MULTIACCUM sequence
to process cosmic rays and reconstruct saturated regions of because they are present on the detector before the sequence
the image. The sequence used was MIF2048 which, with an begins. These ghost images generally decay on a time-scale of
integration time of 2048 s, was the longest sequence which would , 500 s, so by placing a short exposure at the beginning of the orbit
fit into a single orbit. we ensured that the problem was largely alleviated by the time the
longer science exposure began. Any remaining cosmic rays were
then identified by comparison of the MIF512 and MIF2048
2.2.3 Determination of the NICMOS point-spread function
exposures because, unlike ordinary cosmic ray tracks, they appear
Any attempt to determine the properties of a faint, diffuse object in both images.
surrounding a much brighter point source ultimately depends on
our ability to separate the emission which is genuinely spatially
3.2 DC offset
extended from light from the central source which has been
artificially spread out by the point-spread function (PSF) of the Of particular importance to the current study is the effect known as
instrument. Because of the extreme sensitivity of this process to the the ‘pedestal’. This is a time- and temperature-dependent DC offset
exact form of the PSF, it is vitally important that this is accurately arising from the design of the NICMOS3 array and amplifier

q 2001 RAS, MNRAS 326, 1533–1546

1538 M. J. Kukula et al.
system. Its properties are well understood (Rieke et al. 1993), and
4 D ATA A N A LY S I S
in principle the effects can be entirely removed as long as
contemporaneous dark frames, made at the same detector In this section we describe the process by which information about
temperature as the science exposures, are available. Where such the underlying host galaxies was extracted from the NICMOS
contemporaneous darks are not available – as in the present images of the quasars.
observations – a residual DC offset or ‘pedestal’ remains in the
NICMOS images after the subtraction of the dark reference file.
4.1 The z . 1 sample
This offset propagates throughout the subsequent calibration steps
and leads, ultimately, to a (small) multiple of the flat-field reference In most cases the presence of a significant component of extended
file being imprinted on to the final image. As the flat field contains emission underlying the nuclear point source can be inferred
spatial structure on scales of , 1 arcsec, similar to the expected simply from a visual inspection of the J-band quasar images. The
scalelengths of galaxies at redshifts of 1 to 2, removing the effects evidence is particularly obvious between radii of 0.1 and 0.2 arcsec
of the pedestal was a priority. from the nucleus, where the first minimum in the NICMOS PSF
The severity of the pedestal was reduced by an on-orbit occurs.
correction applied to all data taken after 1997 August 12, and by Simple subtraction of a scaled stellar PSF provides further
the construction of synthetic dark reference files which were often evidence for the presence of an extended component in the
better able to account for the remaining offset. However, the effect majority of the quasar images, and an example of a PSF-subtracted
is still present to varying degrees in all of the NICMOS images in image is shown in Fig. 2. However, in order to accurately describe
our study, and these corrections do not apply to the three quasars in the distribution of this extended light we applied the algorithm
our program which were imaged prior to 1997 August 12: SGP2:47 developed for the analysis of our previous HST/WFPC2 images of
ðz ¼ 0:83Þ, PKS0440-00 ðz , 0:844Þ and PKS1524-13 ðz , low-redshift quasars (McLure et al. 1999; Dunlop et al. 2001). Full
1:687Þ. The data quality for these objects is significantly worse details of this algorithm are given by McLure, Dunlop & Kukula
than for the rest of the sample. (2000). The algorithm uses x 2-minimization to match a synthetic
In order to alleviate this problem, staff at the Space Telescope quasar 1 host to the HST image, with nuclear luminosity, galaxy
Science Institute have devised a procedure called PEDTHERM which luminosity, galaxy scalelength, axial ratio and position angle as
can be used to estimate and remove the DC offset in each quadrant free parameters. The surface-brightness profile of the galaxy can be
of an image prior to flat-fielding in CALNICA . Recalibrating our set to either an exponentially decaying (disc) function or an r 1/4 de
data using this extra step led to significant reduction in the pedestal Vaucouleurs law (characteristic of an elliptical galaxy) and
effect, but failed to remove it entirely. the resulting fits compared to determine the morphology of the
In a final attempt to reduce the impact of the pedestal still underlying stellar population. In practice, in the majority of the
further, each of the calibrated NICMOS images was subjected to an NICMOS images the noise was too high to allow us to distinguish
additional corrective procedure devised by ourselves. This between a disc or an elliptical profile with statistical confidence. In
algorithm yielded a retrospective estimate of the height of the the case of the three most prominent host galaxies (BVF247,
remaining pedestal by insisting that the standard deviation across BVF262 and PKS0938118), the preference for a de Vaucouleurs
object-free regions of the final flat-fielded image was minimized. profile was more marked and we show the results of these fits in
Most of the images were noticeably improved by this final Fig. 3, but the formal significance of this preference is only
reduction step, though it proved impossible to remove the pedestal marginal.
entirely and its effects are particularly evident in images in which Table 2 lists the results of 2D modelling for all 10 of the F110M
the quasar is positioned close to the boundary of two quadrants (J-band) images, assuming an r 1/4 de Vaucouleurs surface-
(BVF225 and SGP2:11). brightness profile. Note that the software could not converge on an
For the z , 2 quasar sample, three or four separate 2048-s acceptable fit for the RQQ SGP2:47, the image with the worst
H-band exposures were obtained for each object. Each exposure residual pedestal problems. Despite the failure to unambiguously
was calibrated and treated for the pedestal effect separately before determine the morphologies of the z , 1 quasar hosts, in most
being co-added to produce the final image. other respects these galaxies do appear to be consistent with the
elliptical hosts uncovered for comparably luminous quasars at z .
0:2 by our R-band WFPC2 HST study (McLure et al. 1999; Dunlop
et al. 2001). Half-light radii range from , 4 to 20 kpc and nuclear-
to-host luminosity ratios are typically 1–2 (the RQQ BVF225
3.3 Other problems affecting image quality
appears to have an unusually high nuclear:host ratio, . 8, but the
Particulate contaminants (believed to be small flecks of paint) on image is unusually noisy and the resulting model fit is poorly
the surface of the NICMOS detector give rise to several regions of constrained). We note that the distribution of axial ratios of the
reduced sensitivity, each typically a few pixels in extent. The hosts peaks at a value of b/ a , 1:2, consistent with them being
positions of these contaminants are accurately known, allowing drawn from a population of elliptical galaxies (Sandage, Freeman
the affected pixels to be masked out. A further problem affects the & Stokes 1970; Ryden 1992).
upper left-hand quadrant of the NIC1 array, a significant area of
which has a sensitivity 2–3 times lower than the mean of the
4.2 The z . 2 sample
detector as a whole. The default pointing for NIC1 placed the
majority of our target quasars in the lower right-hand quadrant, Analysis of the H-band (F165M) images of the z , 2 quasars
thus avoiding this problem in most cases. Only the RLQ 4C45.51 proved far more problematic. One object, the RLQ 1148156W1,
from the z . 2 sample fell into the affected region and thus was not detected because of a combination of positional
suffers from a somewhat higher noise level than the rest of the uncertainty and the HST pointing errors. For the remaining nine
sample. quasars, despite the fact that the nominal sky background at H is

q 2001 RAS, MNRAS 326, 1533–1546

 A NICMOS imaging study of quasar hosts 1539

Figure 2. Results of PSF subtraction on the quasar BVF247. The left-hand panel shows the J-band image of the quasar after subtraction of a stellar PSF. Prior to
subtraction the PSF was re-centred and scaled such that the value of the central pixel was 0.84 times that of the corresponding pixel in the quasar image. Using a
larger PSF:quasar ratio leads to an over-subtracted image in which the flux ceases to rise monotonically towards the centre, causing a characteristic dip in the
values of the central few pixels. The image is 1:75  1:75 arcsec2 in size, with grey-scale levels extending from 0 per cent (white) to 25 per cent (black) of the
maximum flux in the image. The central panel shows the NICMOS J-band PSF, with an identical greyscale cut from 0 to 25 per cent of the peak. The PSF is
considerably more compact at the 25 per cent level than the subtracted quasar image, which clearly contains a significant component of extended emission
underlying the quasar point source. The right-hand panel shows a normalised encircled energy diagram for the unsubtracted quasar image. The plot shows the
cumulative flux in a circular aperture of increasing radius (with increments of one pixel), centred on the brightness peak and normalised at the radius at which
the background noise in the quasar image becomes significant (defined here to be the radius beyond which the enclosed flux ceases to increase monotonically).
The solid line shows the quasar point source plus underlying host galaxy whilst the dashed line shows the equivalent curve for a pure (stellar) PSF. The
shallower slope of the quasar profile is further evidence of an underlying extended flux component.

little different to that at J, by selecting objects of the same absolute preference for the smaller, 5-kpc model. Note that, though the
magnitudes as those at lower redshifts the apparent brightness of extended emission in SGP2:36 is clearly visible in the H-band
the targets had of course decreased substantially. image, the small extent of the galaxy relative to the wings of the
These factors, coupled with the wider H-band PSF, meant that H-band PSF was enough to frustrate the modelling software until
clear evidence for an extended component was immediately the number of degrees of freedom was reduced.
apparent in only two of the images (the RQQ SGP2:36 and the For the last four objects, neither model gave a significantly better
RLQ B222156129). In several objects, careful subtraction of a fit to the data, though both gave statistically acceptable results and
scaled PSF does reveal the presence of an underlying extended the flux of the host galaxy remained stable to within ^ 0.75 mag.
component although, as the example shown in Fig. 4 demonstrates, However, in view of the relatively small amount of extended flux, it
very little useful information on the quantity and extent of the seems unlikely that the galaxies are much larger than the FWHM of
emission can be derived from this relatively crude procedure. the NICMOS PSF (0.19 arcsec , 2 kpc). In Table 3 we therefore
In order to carry out a more rigorous search for the presence of adopt a tentative upper limit of 10 kpc.
underlying galaxies, we performed 2D modelling of the H-band
images using the algorithm described in the previous subsection.
5 DISCUSSION
As before, we assumed an r 1/4 (de Vaucouleurs) surface brightness
profile for the putative galaxy and used our stellar H-band image to Despite the difficulty in determining morphologies and, at z , 2,
represent the NICMOS PSF. The results of this 2D modelling are scalelengths for the quasar hosts, the information obtained from the
listed in Table 3. NICMOS images allows us to compare the galaxies with other
For two objects, the RLQs B222156129 and 4C45.51, the types of active and inactive galaxies at the same redshifts and to
model rapidly converged on a solution, finding luminous, extended investigate the link between quasar and galaxy evolution. We first
galaxies (half-light radii , 2 arcsec). For the remaining seven discuss the implications of our relatively robust results at z . 1,
objects, in order to obtain an adequate 2D fit to the images, the before considering the implications of our z . 2 results in the
algorithm once again found it necessary to include some extended context of theories of quasar/galaxy evolution.
emission in addition to the quasar PSF. However, the algorithm
experienced difficulty in converging on a unique solution for the
5.1 Comparison of RLQ and RQQ hosts at z . 1
underlying galaxy. The problem was identified as an inability to
determine the luminosity and scalelength of the galaxy In our study of low-redshift ðz , 0:2Þ radio-loud and radio-quiet
simultaneously, as a result of the fact that the amount of extended quasars (McLure et al. 1999; Dunlop et al. 2001; Hughes et al.
flux was generally low, and its distribution lay well within the 2000; Nolan et al. 2001), for quasars with M V # 224 we found no
wings of the H-band point spread function. statistically significant differences between the host galaxies of the
In order to encourage the algorithm to converge on a unique two types of quasar. Both RLQs and RQQs seem to lie in large
solution, we reduced the number of free parameters by using a ðr 1=2 , 10 kpcÞ, luminous ðL . L*Þ elliptical galaxies with colours
circular galaxy model and fixing the scalelength at either 5 or and ages consistent with those of inactive massive elliptical
10 kpc. By comparing the quality of the resulting fits, we could galaxies at the same redshift.
assess whether the ‘small’ or ‘large’ galaxy model gave a better At z , 1, although, we are unable to determine the morphology
match. In three cases, the RQQ SGP2:36 and the RLQs of the galaxies with confidence, we find that once again the
PKS1524-14 and PKS2204-20, this produced a significant properties of the radio-loud and radio-quiet hosts are statistically

q 2001 RAS, MNRAS 326, 1533–1546

1540 M. J. Kukula et al.

BVF 247 BVF 262 PKS 0938+18

Figure 3. Examples of azimuthally averaged surface-brightness profiles for three members of the z , 1 quasar sample observed through the F110M filter (from
left to right: the RQQs BVF247 and BVF262, and the RLQ PKS0938118). In each case the upper (solid) line shows the best-fitting model (galaxy plus quasar),
whilst the lower, dotted line shows the contribution from the nuclear (quasar) component only. In all cases the fit was obtained by assigning a de Vaucouleurs
(r 1/4) law to the host galaxy. Model parameters for each object are listed in Table 2.

Table 2. Results from the two-dimensional modelling of the z . 0:9 quasar sample.
The table gives the fits achieved using an r 1/4 (de Vaucouleurs) model for the galaxy’s
surface brightness profile, and assuming a cosmology with H 0 ¼ 50 km s21 Mpc and
Vm ¼ 1:0, VL ¼ 0:0. Column 3 lists the half-light radius, r1/2, of the host and column
4 the surface brightness, m1/2, at this radius (J mag arcsec22). Columns 5, 6 and 7 list
the apparent host and nuclear J-band magnitudes (with associated uncertainties of
, 0.4 and , 0.3 magnitudes respectively) and the ratio of nuclear to host-galaxy
luminosity, whilst column 8 gives the axial ratio of the host.

Source z r1/2/kpc m1/2 Jhost Jnuc Lnuc / Lhost b/ a

Radio-Quiet Quasars
SGP5:46 0.955 3.9 21.65 20.10 19.46 1.79 1.14
SGP2:47 0.830 – – – – – –
BVF225 0.910 17.1 23.78 20.11 17.91 7.61 3.20
BVF247 0.890 11.9 22.83 18.88 20.14 0.31 1.20
BVF262 0.970 4.6 21.56 19.85 19.24 1.76 1.36

Radio-Loud Quasars
PKS0440 2 00 0.844 13.0 22.71 18.84 18.47 1.41 1.64
PKS0938118 0.943 4.3 21.10 19.49 19.81 2.29 1.33
3C422 0.942 17.0 22.88 18.29 17.90 1.43 1.33
MC21121172 0.878 17.4 23.14 18.18 18.97 0.48 1.02
4C02.54 0.976 10.4 21.62 19.32 17.61 4.82 4.08

consistent with one another, although there is some evidence of a index (Taylor et al. 1996). This enabled us to compare the
trend towards a larger luminosity difference than was found at properties of the three main types of powerful active galaxy in the
z . 0:2. The mean J-band magnitude for the RLQ hosts is 18:8 ^ local universe, and to test models which claim to unify RLQs and
0:3 and that of the RQQs is 19:7 ^ 0:3, whilst the mean radio galaxies via viewing angle effects. We found that the radio
scalelengths are 12:4 ^ 2:4 kpc and 9:4 ^ 3:2 kpc respectively. A galaxies were indistinguishable from the hosts of RLQs and
similar tendency towards somewhat less massive hosts for RQQs luminous RQQs, having similar sizes, colours and luminosities and
was also seen in the z . 0:2 sample (McLure et al. 1999; Dunlop following a Kormendy relation identical to that of normal, inactive
et al. 2001), although once again the difference was not statistically massive elliptical galaxies (McLure et al. 1999; Dunlop et al.
significant. McLure et al. speculate that this might indicate that the 2001).
low-z RQQs contain smaller black holes than their radio-loud As a result of scheduling pressures it was not practical to include
counterparts, but are accreting gas with greater efficiency, to a comparison sample of radio galaxies in the current NICMOS
produce comparable luminosities. Larger samples will be required study. However, by utilizing the data in the HST archive we are able
at both redshifts in order to determine whether this apparently to construct a post hoc sample of radio galaxies which is roughly
small but persistent difference is in fact real. matched to the quasars at z , 0:9.
Of all the radio galaxy images in the archive the most useful in
terms of redshift, radio luminosity and observing waveband are the
5.2 Comparison of quasar hosts and radio galaxies at z . 1
I-band images of 3CR radio galaxies made with WFPC2 by Best,
In our previous HST study of quasar hosts at z , 0:2, as well as Longair & Röttgering (1997, 1998). These images have been
radio-loud and radio-quiet quasars we also included a sample of reanalysed by McLure & Dunlop (2000) with the same 2D
FRII radio galaxies which was carefully matched to the RLQ modelling software used to analyse the current NICMOS images.
sample in terms of extended 5-GHz luminosity and radio spectral By assuming an I 2 J colour of 0.8 for elliptical galaxies at z , 1

q 2001 RAS, MNRAS 326, 1533–1546

 A NICMOS imaging study of quasar hosts 1541

Figure 4. Results of PSF subtraction on the quasar SGP2:36. The left-hand panel shows the H-band image of the quasar after subtraction of a stellar PSF. Prior
to subtraction, the PSF was re-centred and scaled such that the value of the central pixel was 0.8 times that of the corresponding pixel in the quasar image. Using
a larger PSF:quasar ratio results in an over-subtracted image. The image is 1:1  1:1 arcsec2 in size, with grey-scale levels extending from 0 per cent (white) to
25 per cent (black) of the maximum flux in the image. The central panel shows the NICMOS H-band PSF, with an identical greyscale cut from 0 to 25 per cent
of the peak. Although the difference is less marked than in the J-band example of Fig. 2, the residual flux in the subtracted image is more extended than the PSF
and cannot be accounted for by a point source. The right-hand panel shows a normalized encircled energy diagram for the unsubtracted quasar image. The plot
shows the cumulative flux in a circular aperture of increasing radius (with increments of one pixel), centred on the brightness peak and normalized at the radius
at which the background noise in the quasar image becomes significant (defined here to be the radius beyond which the enclosed flux ceases to increase
monotonically). The solid line shows the quasar point source plus underlying host galaxy, whilst the dashed line shows the equivalent curve for a pure (stellar)
PSF. The slope of the quasar curve is less steep than that for the pure PSF, providing further evidence for an extended component of emission underlying the
quasar point source.

Table 3. Results from the two-dimensional modelling of the z . 2 quasar

sample. The table gives the fits achieved using an r 1/4 (de Vaucouleurs) model
for the galaxy’s surface brightness profile, and assuming a cosmology with
H 0 ¼ 50 km s21 Mpc and Vm ¼ 1:0, VL ¼ 0:0. Column 3 lists the half-light
radius, r1/2, of the host galaxy for those objects in which the model was able to
converge on a definite value. For SGP2:36, PKS1524 2 13 and
PKS2204 2 20, the model showed a preference for a small (,5 kpc) host,
but this value must be treated with caution. For the remaining objects, no
reliable estimate of the galaxy size could be obtained, although in view of the
low luminosity of the host, they are unlikely to be large. Columns 4, 5 and 6 list
the apparent host and nuclear H-band magnitudes (with estimated uncertainties
of 0.75 and 0.3 magnitudes respectively) and the ratio of nuclear to host-galaxy
luminosity, whilst column 7 gives the axial ratio of the host, where available.

Source z r1/2/kpc Hhost Hnuc Lnuc / Lhost b/ a

Radio-Quiet Quasars
SGP2:11 1.976 ? (,10) 20.65 18.97 4.70 –
SGP2:25 1.868 ? (,10) 19.89 19.60 1.31 –
SGP2:36 1.756 ,5 19.74 19.98 0.80 –
SGP3:39 1.964 ? (,10) 19.77 19.55 1.16 –
SGP4:39 1.716 ? (,10) 21.60 18.86 12.43 –

Radio-Loud Quasars
PKS1524 2 13 1.687 ,5 19.36 18.10 3.22 –
B2 2 2156 1 29 1.753 16.0 17.87 17.97 0.91 1.77
PKS2204 2 20 1.923 ,5 20.66 18.57 6.87 –
4C45.51 1.992 17.9 17.86 17.48 1.42 1.23

(Fasano et al. 1998) it becomes possible to make a comparison compatible, without the need to make surface-brightness correc-
between the quasar hosts imaged in J-band with NICMOS and the tions. Once again, we can convert the NICMOS J magnitudes for
I-band radio galaxy images. the quasars to I-band values by assuming an I 2 J colour of 0.8.
We find that the I-band properties of the quasar hosts are very The two sets of data are shown in Fig. 5, where it can be seen that
similar to those of the radio galaxies: mean scalelengths (with the Kormendy relations for both quasars and radio galaxies are
standard errors) are 12:6 ^ 2:3 (RGs) and 11:0 ^ 1:8 (quasar statistically indistinguishable. A least-squares fit to all the data
hosts), whilst the mean absolute I magnitudes are 224:62 ^ 0:13 points produces the relationship m1=2 ¼ 3:23^0:42 log r 1=2 1
(RGs) and 224:59 ^ 0:21 (quasar hosts). 19:82^0:4 (for the radio galaxies alone the fit has a slope of 3.5;
Given that the redshift distributions of the 10-object 3CR galaxy McLure & Dunlop 2000). Thus at these redshifts both the quasar
subsample and the new z , 1 quasar sample are similar, it is also hosts and radio galaxies appear to be derived from the same parent
possible to investigate whether their Kormendy relations are population. This result is consistent with the unification of RLQs

q 2001 RAS, MNRAS 326, 1533–1546

1542 M. J. Kukula et al.
well described by a de Vaucouleurs law. Thus, although our
analysis of the quasar hosts cannot distinguish between an elliptical
or a disc profile with any degree of confidence, their similarity in
all other respects to the 3CR radio galaxies provides strong
circumstantial evidence that, as at low redshifts, these quasars
inhabit massive elliptical galaxies. If this is the case it implies that,
just as at low redshift, an important pre-requisite for the presence of
a powerful AGN at z , 1 is a host galaxy with a massive spheroidal
component. Within the context of hierarchical clustering models
for galaxy formation, the similarity (but for passive stellar
evolution) between both radio galaxies and quasar hosts at z . 1
and their counterparts at z . 0:2 is most simply explained by a
cosmological model in which the massive elliptical galaxy
formation process is essentially complete by z . 1.

5.3 The nature of quasar hosts at z . 2

The strong cosmological evolution in comoving space density of
quasars has been known for nearly 40 years. Much work has gone
into empirical fits of the evolution (Schmidt & Green 1983; Boyle
et al. 1988; Dunlop & Peacock 1990; Hewett, Foltz & Chaffee
Figure 5. The apparent I-band Kormendy relation described by the z , 1
1993; Goldschmidt & Miller 1998; Goldschmidt et al. 1999) but
quasars (triangles) imaged with NICMOS through the F110M (.J ) filter.
this has not led to any physical understanding of its causes. Some
Conversion to I magnitudes has been carried out assuming I 2 J colours of
0.8, a typical value for elliptical galaxies at z , 1 (Fasano et al. 1998). Also recent theoretical work (Efstathiou & Rees 1988; Carlberg 1990;
shown are the z , 0:8 3CR radio galaxies from the sample of Best, Longair Haehnelt & Rees 1993; Percival & Miller 1999) has concentrated
& Röttgering (1997, 1998)(crosses). These objects were imaged in I-band on explaining the evolution as being driven by evolution in the rate
with WFPC2 on HST and the images reanalysed by McLure & Dunlop of galaxy merging, following the hypothesis that quasars display a
(2000) with the same algorithm used to model the NICMOS images in the cosmologically short-lived burst, or bursts, of activity following
current study. The solid line shows a least-squares fit to the combined such a merger. Kauffmann & Haehnelt (2000) have used a semi-
samples, and has a slope of 3:23 ^ 0:42 (cf. 3.5 for the radio galaxies analytic model to link the growth of supermassive black holes with
alone). the formation and evolution of galaxy dark matter haloes in a cold
dark matter (CDM) universe. They find that the evolution in merger
and radio galaxies via viewing-angle effects, a scenario which is rate of dark haloes alone cannot explain the large evolution in
already well supported at low redshifts. quasar space density, in agreement with the analytic work of
The lack of bright nuclear point sources in the radio galaxies Percival & Miller (1999). They suppose that accretion time-scale
means that their morphologies can be determined unambiguously; also varies with cosmic epoch, and thus they effectively add a
as expected all have radial surface brightness profiles which are component of luminosity evolution onto the basic model (as also

Figure 6. Host galaxy versus quasar (nuclear) absolute magnitudes at redshifts of 0.4, 1 and 2. The dashed line shows the value of LQ V for galaxies in the local
universe. The small grey points are the simulated data from the semi-analytic model of Kauffmann & Haehnelt (2000), showing the predicted evolution of the
relationship between MV of the host and MB of the active nucleus, and assuming the same cosmology used in the current paper. The large symbols show the
results of the current NICMOS study, with filled circles representing RQQs and open circles RLQs. As we currently lack data for objects at z ¼ 0:4, in the left-
hand panel we plot the subset of RLQs and RQQs with 224 $ M V ðtotalÞ $ 225 from our WFPC2 study of quasars at z , 0:2 (McLure et al. 1999; Dunlop
et al. 2001). The triangle in the left hand panel shows the locus of the quasar sample of McLeod et al. (1999). As at each redshift our chosen filter corresponds to
rest-frame V-band, the absolute V-band magnitudes can be calculated directly from the observed R, J and H magnitudes without the need to assume any
particular spectral shape for the hosts. For the quasars we have converted from MV to MB using a colour index of B 2 V ¼ 0:4 [equivalent to a spectrum of the
form f ðnÞ / n 20:2 Š. [Figure adapted from fig. 12 of Kauffmann & Haehnelt (2000).]

q 2001 RAS, MNRAS 326, 1533–1546

 A NICMOS imaging study of quasar hosts 1543

Figure 7. Mean absolute V-band magnitude versus mean redshift for the host galaxies of the RLQs (open circles) and RQQs (filled circles) in the current
NICMOS study. Also shown is the subset of five RLQs and seven RQQs from our WFPC2 study of quasars at z , 0:2 which have total ðhost 1 nuclearÞ
luminosities in the same range as the high-redshift samples ð224 $ M V $ 225Þ. Error bars show the standard error on the mean. The dotted lines show the
luminosity evolution of present day L*, 2L* and 4L* elliptical galaxies, assuming a formation epoch of z ¼ 5 with a single rapid burst of star formation
followed by passive evolution thereafter. Left-hand panel: assuming a cosmology with H 0 ¼ 50 km s21 Mpc21 , Vm ¼ 1:0 and VL ¼ 0:0. Right-hand panel:
H 0 ¼ 65 km s21 Mpc21 , Vm ¼ 0:3 and VL ¼ 0:7.

proposed by Haehnelt & Rees 1993) which can successfully cannot be explained away simply as the effect of the two unusually
reproduce the approximate luminosity evolution that is observed luminous RLQs in the z . 2 sample. [The V magnitudes quoted in
(Boyle et al. 2000). Within this framework, quasars of a given the Véron-Cetty & Véron (1993) catalogue are often merely
luminosity are powered by progressively lower-mass black holes extrapolations based on data in other wavebands, so it is perhaps
with increasing redshift. Hence if black hole mass is indeed related unsurprising that our sample should turn out to contain two objects
to the mass of the dark halo, as might be implied from the slightly more luminous than our nominal upper limit of
correlations between black hole mass and galaxy bulge luminosity M V ðtotalÞ ¼ 225:Š Even when these two objects are excluded,
(Kormendy & Richstone 1995; Magorrian et al. 1998; van der the remaining RLQ hosts at z . 2 show no evidence for a drop in
Marel 1999) and between black hole mass and galaxy velocity luminosity with redshift, and a tendency towards increasing
dispersion (Ferrarese & Merritt 2000; Gebhardt et al. 2000), the luminosity is clearly already present at z . 1.
Kauffmann & Haehnelt (2000) model predicts that quasars of a To clarify the trend of host luminosity with redshift seen in our
given luminosity should, on average, be found in progressively sample, we plot mean MV versus redshift in Fig. 7, for two
less-massive host galaxies with increasing redshift. In fact, alternative cosmologies, overlaid with the predictions for a
provided that a relationship between black hole mass and host passively evolving galaxy formed in a short-lived burst at high-
galaxy mass exists at all redshifts, then any luminosity-evolution redshift ðz ¼ 5Þ. Passive evolution commencing at high redshift
model will predict such an effect, and this prediction can now be appears to be the most reasonable starformation history to adopt,
tested. given the evidence from Nolan et al. (2001) and de Vries et al.
To date, ground-based studies have not strongly supported such a (2000) that low-z quasar hosts and radio galaxies are dominated by
picture, finding evidence for very luminous hosts around quasars at stellar populations of age . 12 Gyr.
redshifts of , 2.5. However, these observations have concentrated These plots demonstrate a number of potentially important
on the most luminous quasars, for which there are no counterparts points. It is clear that in either cosmology the hosts of RLQs
at low redshifts. Our current HST study thus offers the prospect of brighten with redshift in a manner which is perfectly consistent
the first, proper, unbiased view of how host-galaxy luminosity with pure passive evolution. Basically, the typical host of an RLQ
varies with redshift, in a manner which offers a direct and appears to be a passively evolving (present-day) 4 L* elliptical.
transparent test of theoretical predictions. This is not only because This result agrees well with what has been found for high-redshift
we have studied quasars of similar luminosity at z . 0:2, 1 and 2, radio galaxies. Specifically, if we compare our z . 2 results
but also because we have observed all the hosts at the same rest directly with H-band observations of radio galaxies at comparable
wavelength (removing the need for k-corrections), and because redshift, the average H-band magnitude for radio galaxies in the
host-galaxy luminosities have been derived using an identical redshift band 1:7–2:1 discussed by Lacy, Bunker & Ridgway
modelling technique at all redshifts. (2000) is 19:2 ^ 0:3, while the average H-band magnitude for the
In Fig. 6 we show the best-fitting host and nuclear luminosities RLQ hosts in our z . 1:9 sample is 18:9 ^ 0:7. Our results for
for our quasar samples at z . 0:2, 1 and 2, superimposed on to RLQ hosts are also consistent with the findings of a recent ground-
the scatterplot predictions for quasars/hosts at z . 0:4, 1 and 2 based imaging study of three RLQs at z , 1:5 by Falomo,
produced by Kauffmann & Haehnelt (2000). As can be seen, the Kotilainen & Treves (2001). Taken together, these results therefore
agreement between observation and prediction is excellent for the provide further support for a picture in which the formation of
low-redshift sample, but becomes progressively less convincing ellipticals with sufficiently massive black holes to support radio
within the higher redshift bins as the model predictions recede, activity (see McLure et al. 1999), is essentially complete before
while our derived host luminosities remain either roughly constant z . 2.
(in the case of the RQQs) or actually increase with redshift (in the The trend displayed by the hosts of RQQs does appear to be
case of the RLQs). We note that the behaviour of the RLQ hosts somewhat different however. While the error bars have also grown,

q 2001 RAS, MNRAS 326, 1533–1546

1544 M. J. Kukula et al.
it seems that the gap between the luminosities of the hosts of RQQs hierarchical models of Kauffmann & Haehnelt (2000), which still
and RLQs has grown still further to . 1.5 magnitudes, simply require a substantial (order of magnitude) component of luminosity
because the typical RQQ host in our sample appears to have the evolution to be invoked to explain the rapid evolution of the quasar
same absolute magnitude at all redshifts. luminosity function (e.g. Boyle et al. 2000).
The interpretation of this observation is dependent on choice of Our results are much more consistent with a picture in which the
cosmology. In the flat matter-dominated cosmology, the RQQ hosts increased availability of fuel (or, equivalently, increased frequency
are inconsistent with having had a constant mass in stars which has of mergers) at z . 2 results in a substantial increase in the number
evolved passively since z ¼ 5. Evaluation of any change in mass density of active black holes, along with a moderate increase in the
then depends on the choice of star-formation history, but is a factor fuelling efficiency of a typical observed quasar. An increase of a
of 4 less than the assumption of passive stellar evolution between factor of 2–3 in typical fuelling efficiency between z . 0:2 and
z . 2 and the present day. In the L-dominated cosmology, the z . 2 is certainly permitted by the results of detailed studies of
RQQ hosts are in fact consistent with no change in mass. Thus, black-hole masses in nearby RQQs (Dunlop et al. 2001; McLure
even accounting for the apparent difference between RLQ and et al. 2001), which suggest that on average RQQs at z . 0:2 are
RQQ hosts at z , 2, the large variation in typical host mass with radiating at only . 20 per cent of their Eddington limit. It is thus
redshift predicted by Kauffmann & Haehnelt (2000) is not not unreasonable to find that, in an era of greater fuel availability,
observed. our sample of z . 2 RQQs, matched in nuclear optical output to
Recent results from other studies of the hosts of high-redshift our low-redshift sample, could be largely produced by black holes
RQQs also indicate that they are less luminous than their radio- (and hence hosted by spheroids) with a characteristic mass a few
loud counterparts. In particular, Ridgway et al. (2001) report times smaller than in the comparison z . 0:2 sample. Importantly,
results from NICMOS imaging of five RQQs at z . 2–3, in which however, an observed trend towards moderately higher fuelling
they find host galaxies with typical luminosities of , L*. In fact, efficiencies with increasing z does not require one to invoke
two of their RQQs are rather fainter than those discussed here, and some redshift-dependent change in quasar black-hole accretion
if one confines attention to the three objects in their sample with mechanism. On the contrary, the observed, modest mass reduction
M B . 224, their PSF-subtracted host luminosities are . 2 L*. Any is in fact predicted by new models which attempt to explain
subsequent correction for PSF over-subtraction would only raise observed quasar evolution as a consequence of varying quasar
these values further, guaranteeing rather good agreement with the birthrate, combined with declining light-curves for individual
results shown in Fig. 7. quasars (Miller, Percival & Lambert, in preparation). In these
Other preliminary NICMOS-based results on the luminosities of models, the increased birthrate at z . 2 leads to a statistical bias
RQQ hosts at z . 2 have been presented by Rix et al. (1999), as an such that quasars of a given absolute magnitude are more likely to
interesting by-product of the CfA Arizona Gravitational Lens be observed earlier in (i.e. closer to the peak of) their declining
Survey (CASTLES). This study attempts to measure the host light curves, and therefore closer to maximum luminosity than at
luminosities of RQQ in strongly lensed systems, by correcting the low redshift, when quasar birth is relatively rare.
observed extended emission for the effects of the gravitational In answer to the second question, one possible explanation is the
lensing, and it has been claimed that their results mirror exactly the effect of the joint selection criteria of high radio luminosity and
predictions of Kauffmann & Haehnelt (2000). Such a conclusion moderate optical luminosity used in the definition of our RLQ sub-
may be premature, because it is hard to assess the surface samples. There is now growing evidence that luminous radio
brightness biases inherent in this method, but none the less it seems sources are only produced by a very massive subset of the black
unlikely that the derived RQQ host luminosities could be hole population that powers quasars in general. In particular, the
sufficiently underestimated (by , 2 mag) to be consistent with results of Dunlop et al. (2001) and McLure & Dunlop (2001)
the hosts of RLQs and radio galaxies. indicate that the RLQs in our z . 0:2 sample are powered by black-
It thus seems hard to avoid the conclusion that there exists a real holes which are typically three times more massive than the power
difference of . 1 mag between the hosts of radio-loud and radio- sources of RQQs, and are confined to the mass range M . 109 M( .
quiet AGN of comparable nuclear optical luminosity at z . 2. If this remains true at high redshift, and if the black hole spheroid
relationship remains basically unchanged, then radio-based
selection will effectively guarantee host galaxies of similar mass
5.4 Implications
at all redshifts. If, as we have done in this study, one also insists that
These results raise two fundamental questions. First, what are the the optical luminosities of the RLQs under study are comparable at
implications of the observed, rather modest, drop in RQQ host z . 2 and z . 0:2, then it would not be unexpected that such
mass as a function of redshift? Secondly why is a comparable drop subsamples of objects would show no evidence for the redshift-
in host mass not apparent for the RLQs? dependent change in fuelling efficiency found for the RQQs. This
The answer to the first question is that, provided the relationship does not of course mean that RLQs would not, on average, be more
between black hole and host-spheroid mass is basically unchanged efficiently fuelled at high redshift, but rather that the selection
out to z . 2, our results exclude any model of quasar evolution criteria used here would mitigate against us observing such an
which involves a substantial component of luminosity evolution. effect in our sample.
This is because the rather modest drop in RQQ host mass with
increasing redshift implies that a typical RQQ black hole power
6 SUMMARY
source at z . 2 is only radiating . 2–3 times more efficiently than
at low redshift. This is obviously completely at odds with pure We have presented the results of the first, major observational study
luminosity evolution models which, to explain the evolution of the designed to determine the properties of the hosts of both radio-loud
quasar optical luminosity function, require black holes to radiate and radio-quiet quasars from z . 2 to the present day in a
. 30 times more efficiently at z . 2. However, it is also in conflict genuinely unbiased manner. The key features of this study are: (i)
with more topical and apparently more realistic models, such as the sufficient HST-based angular resolution to allow a meaningful

q 2001 RAS, MNRAS 326, 1533–1546

 A NICMOS imaging study of quasar hosts 1545
attempt at determining galaxy scalelengths at all redshifts; (ii) the
AC K N O W L E D G M E N T S
use of quasar samples with the same characteristic absolute
magnitude at all redshifts; (iii) our insistence that the radio-quiet The authors would like to thank the referee, D. Hines, for many
quasars selected for study are known to lie below a definite radio- useful comments and suggestions, and E. Bergeron for help with
luminosity threshold; (iv) filter selection, which when coupled PEDTHERM . MJK, RJM & WJP acknowledge PPARC funding. JSD
with careful sample redshift constraints, guarantees line-free acknowledges the enhanced research time afforded by the award of
imaging at the same rest-wavelength (.5800 Å) (removing a PPARC Senior Fellowship. Support for this work was provided
concerns about emission-line contamination, and obviating the by NASA through grant numbers GO-06776.01-95A & GO-
need for k-corrections); (v) the extraction of host-galaxy 07447.01-96A from the Space Telescope Science Institute, which
parameters using an identical modelling approach at all redshifts, is operated by the Association of Universities for Research in
minimizing potential surface-brightness bias and concerns over Astronomy, Inc., under NASA contract NAS5-26555. Based on
aperture corrections; (vi) the use of properly sampled, high observations with the NASA/ESA Hubble Space Telescope,
dynamic range PSFs derived from observations of stars through the obtained at the Space Telescope Science Institute. This research
same filters, and on the same regions of the relevant detectors, as has made use of the NASA/IPAC Extragalactic Database (NED)
the quasar images. which is operated by the Jet Propulsion Laboratory, California
At z . 1 we have been able to determine host-galaxy parameters Institute of Technology, under contract with NASA.
with sufficient accuracy to demonstrate that the hosts of both RQQs
and RLQs lie on the same Kormendy relation as deduced for 3CR
radio galaxies at comparable redshift by McLure & Dunlop (2000). REFERENCES
The hosts of both RLQs and RQQs seem to have changed little, if Aretxaga I., Boyle B. J., Terlevich R. J., 1995, MNRAS, 275, L27
at all, in terms of size between z . 1 and z . 0:2. The typical Aretxaga I., Terlevich R. J., Boyle B. J., 1998, MNRAS, 296, 643
luminosity of the RLQ hosts has increased between z . 0:2 and Bahcall J. N., Kirhakos S., Schneider D. P., 1995a, ApJ, 447, L1
z . 1 by an amount that is perfectly consistent with pure, passive Bahcall J. N., Kirhakos S., Schneider D. P., 1995b, ApJ, 450, 486
evolution of a mature stellar population. The typical luminosity of Bahcall J. N., Kirhakos S., Schneider D. P., 1996, ApJ, 457, 557
the RQQ hosts is less enhanced, and indeed appears basically Bahcall J. N., Kirhakos S., Saxe D. H., Schneider D. P., 1997, ApJ, 479, 642
unchanged. There is thus a suspicion that the host masses of RQQs Best P. N., Longair M. S., Röttgering H. J. A., 1997, MNRAS, 292, 758
Best P. N., Longair M. S., Röttgering H. J. A., 1998, MNRAS, 295, 549
of comparable nuclear output are less massive at z . 1, but the
Boyce P. J. et al., 1998, MNRAS, 298, 121
significance of this result is marginal. Within a L-dominated Boyle B. J., Shanks T., Peterson B. A., 1988, MNRAS, 235, 935
cosmology our results are certainly consistent with the host masses Boyle B. J., Fong R., Shanks T., Peterson B. A., 1990, MNRAS, 243, 1
of both class of quasar being unchanged between z . 1 and Boyle B. J., Shanks T., Croom S. M., Smith R. J., Miller L., Loaring N.,
z . 0:2. Heymans C., 2000, MNRAS, 317, 1014
At z . 2, although the hosts are harder to detect and Carballo R., Sánchez S. F., González-Serrano J. I., Benn C. R., Vigotti M.,
consequently to model, we have found that host-galaxy magnitudes 1998, AJ, 115, 1234
can be extracted with sufficient confidence to demonstrate that the Carlberg R., 1990, ApJ, 350, 505
hosts of RLQs continue to brighten as expected under the de Vries W. H., O’Dea C. P., Barthel P. D., Fanti C., Fanti R., Lehnert M. D.,
hypothesis of pure passive evolution of hosts with non-evolving 2000, AJ, 120, 2300
Disney M. J. et al., 1995, Nat, 376, 150
mass. In contrast, the hosts of RQQs again seem little changed in
Dunlop J. S., Taylor G. L., Hughes D. H., Robson E. I., 1993, MNRAS, 264,
luminosity, and systematically smaller in size compared to at 455
z . 1. This apparently growing radio-loud/radio-quiet host-mass Dunlop J. S., McLure R. J., Kukula M. J., Baum S. A., O’Dea C. P., Hughes
gap is mirrored in other studies of radio galaxies and radio-quiet D. H., 2001, MNRAS, submitted
quasars at comparable redshift, lending further support to its Dunlop J. S., Peacock J. A., 1990, MNRAS, 247, 19
reality. Even after allowance for passive evolution, the inferred Efstathiou G., Rees M. J., 1988, MNRAS, 230, 5P
drop in RQQ host mass is relatively modest compared to the Falomo R., Kotilainen J., Treves A., 2001, ApJ, 547, 124
predictions of an order of magnitude drop in mass made by Fasano G., Cristiani S., Arnouts S., Filippi M., 1998, AJ, 115, 1400
Kauffmann & Haehnelt (2000). The mass drop is a factor of 4 in an Ferrarese L., Merritt D., 2000, ApJ, 539, L9
Einstein– de Sitter cosmology, assuming a passive evolution Gebhardt K. et al., 2000, ApJ, 539, L13
Goldschmidt P., Miller L., 1998, MNRAS, 293, 107
model, and a factor of 2 if Vm ¼ 0:3 and Vv ¼ 0:7.
Goldschmidt P., Kukula M. J., Miller L., Dunlop J. S., 1999, ApJ, 511, 612
The lack of strong evolution in host mass with redshift argues Haehnelt M. G., Rees M. J., 1993, MNRAS, 263, 168
against models of quasar evolution, such as the latter, which invoke Hamilton T. S., Casertano S., Turnshek D. A., 2001, (astro-ph/0011255)
some relationship between host mass and quasar luminosity Heckman T. M., Lehnert M. D., van Breugel W., Miley G. K., 1991, ApJ,
together with substantial luminosity evolution of the quasar 370, 78
component. Rather, this study indicates that the increased Hewett P. C., Foltz C. B., Chaffee F. H., 1993, ApJ, 406, 43
availability of fuel at z . 2 results in a substantial increase in the Hintzen P., Romanishin W., Valdes F., 1991, ApJ, 366, 7
number density of active black holes, along with a moderate Hooper E. J., Impey C. D., Foltz C. B., 1997, ApJ, 480, L95
increase in the fuelling efficiency of a typical observed quasar. This Hughes D. H. et al., 1998, Nat, 394, 241
type of moderate efficiency increase is certainly permitted by Hughes D. H., Kukula M. J., Dunlop J. S., Boroson T., 2000, MNRAS, 316,
204
studies of the fuelling efficiency of low-redshift quasars (Dunlop
Hutchings J. B., 1995, AJ, 110, 994
et al. 2001; McLure & Dunlop 2001), and indeed arises naturally as Hutchings J. B., Morris S. C., 1995, AJ, 109, 1541
a statistical effect predicted by new models which attempt to Kauffmann G., Haehnelt M., 2000, MNRAS, 311, 576
explain observed quasar evolution as a consequence of varying Kormendy J., Richstone D., 1995, ARA&A, 33, 581
quasar birthrate, combined with declining light-curves for Lacy M., Bunker A. J., Ridgway S. E., 2000, AJ, 120, 68
individual quasars (Miller, Percival & Lambert, in preparation). Laor A., 2000, ApJ, 543, L111

q 2001 RAS, MNRAS 326, 1533–1546

1546 M. J. Kukula et al.
Lehnert M. D., Heckman T., Chambers K. C., Miley G. K., 1992, ApJ, 393, Rieke M. J., Winters G. S., Cadien J., Rasche R., 1993, Proc. SPIE, 1946,
68 214
Lowenthal J. D., Heckman T. M., Lehnert M. D., Elias J. H., 1995, ApJ, Rix H.-W., Falco E., Impey C., Kochanek C., Lehar J., McLeod B., Muñoz
439, 588 J., Peng C., 1999, (astro-ph/9910190)
Madau P., Fergusson H. C., Dickinson M., Giavalisco M., Steidel C. C., Rönnback J., van Groningen E., Wanders I., Örndahl E., 1996, MNRAS,
Fruchter A. S., 1996, MNRAS, 283, 1388 283, 282
Magorrian J. et al., 1998, AJ, 115, 2285 Ryden S., 1992, ApJ, 396, 445
Márquez I., Petitjean P., Théodore B., Bremer M., Monnet G., Beuzit J.-L., Sandage A. R., Freeman K. C., Stokes N. R., 1970, ApJ, 268, 831
2001, A&A, 371, 97 Schade D., Boyle B. J., Letawsky M., 2000, MNRAS, 315, 498
Marshall H. L., Huchra J. P., Tananbaum H., Avni Y., Braccesi A., Zitelli V., Schmidt M., Green R. F., 1983, ApJ, 269, 352
Zamorani G., 1984, ApJ, 283, 50 Steidel C. C., Adelberger K. L., Giavalisco M., Dickinson M., Pettini M.,
McLeod K. K., Rieke G. H., 1994a, ApJ, 431, 137 1999, ApJ, 519, 1
McLeod K. K., Rieke G. H., 1994b, ApJ, 420, 58 Stockton A., MacKenty J. W., 1987, ApJ, 316, 584
McLeod K. K., Rieke G. H., 1995, ApJ, 441, 96 Tadhunter C. N., Scarrott S. M., Draper P., Rolph C., 1992, MNRAS, 256,
McLeod K. K., Rieke G. H., Storrie-Lombardi L. J., 1999, ApJ, 511, L67 53P
McLure R. J., Kukula M. J., Dunlop J. S., Baum S. A., O’Dea C. P., Hughes Taylor G. L., Dunlop J. S., Hughes D. H., Robson E. I., 1996, MNRAS, 283,
D. H., 1999, MNRAS, 308, 377 930
McLure R. J., Dunlop J. S., 2000, MNRAS, 317, 249 van der Marel R. P., 1999, AJ, 117, 744
McLure R. J., Dunlop J. S., 2001, MNRAS, in press, (astro-ph/0009406) Véron-Cetty M.-P., Woltjer L., 1990, A&A, 236, 69
McLure R. J., Dunlop J. S., Kukula M. J., 2000, MNRAS, 318, 693 Véron-Cetty M.-P., Véron P., 1993, ESO Sci. Rep. No. 13. ESO
Nolan L. A., Dunlop J. S., Kukula M. J., Hughes D. H., Boroson T., Jimenez Publications, Garching
R., 2001, MNRAS, 323, 308 Warren S. J., Hewett P. C., Osmer P. S., 1994, ApJ, 421, 412
Percival W. J., Miller L., 1999, MNRAS, 309, 823
Percival W. J., Miller L., McLure R. J., Dunlop J. S., 2001, MNRAS, 322,
843
Ridgway S. E., Heckman T. M., Calzetti D., Lehnert M., 2001, ApJ, 550,
122 This paper has been typeset from a TEX/LATEX file prepared by the author.

q 2001 RAS, MNRAS 326, 1533–1546