Radio-Quiet Quasars in the VIDEO Survey: Evidence for AGN-powered radio emission at S_1 ...

26 Pages·2015·3.86 MB·English

Mon.Not.R.Astron.Soc.000,1–??(2015) (MNLATEXstylefilev2.2)
Radio-Quiet Quasars in the VIDEO Survey: Evidence for
AGN-powered radio emission at S < 1 mJy
1.4GHz
5
1 Sarah V. White1⋆, Matt J. Jarvis1,2, Boris H¨außler1,3, Natasha Maddox4,5
0 1Astrophysics, University of Oxford, DenysWilkinson Building, Keble Road, Oxford, OX1 3RH, UK
2 2Department of Physics, University of the WesternCape, Bellville 7535, South Africa
n 3Centre for Astrophysics, Science & Technology Research Institute, Universityof Hertfordshire, Hatfield, Herts, AL10 9AB, UK
a 4Netherlands Institute for Radio Astronomy (ASTRON), PO Box 2, 7990 AA Dwingeloo, The Netherlands
J 5Astrophysics, Cosmology and Gravity Centre (ACGC), Universityof Cape Town, Private Bag X3, 7701 Rondebosch, South Africa

9
1

] ABSTRACT
A
Understandingtheinterplaybetweenblack-holeaccretionandstarformation,andhow
G to disentangle the two, is crucial to our understanding of galaxy formation and evo-
. lution. To investigate, we use a combination of optical and near-infrared photometry
h
to select a sample of 74 quasars from the VISTA Deep Extragalactic Observations
p 2
(VIDEO) Survey, over 1deg . The depth of VIDEO allows us to study very low ac-
-
o cretion rates and/or lower-mass black holes, and 26 per cent of the candidate quasar
r sample has been spectroscopically confirmed. We use a radio-stacking technique to
t
s sample below the nominal flux-density threshold using data from the Very Large Ar-
a ray at 1.4GHz and find, in agreement with other work, that a power-law fit to the
[
quasar-related radio source counts is inadequate at low flux density. By comparing

2 with a control sample of galaxies (where we match in terms of stellar mass), and by
v estimating the star formation rate, we suggest that this radio emission is predomi-
2 nantly caused by accretion activity rather than star-formation activity.
9
8 Key words: galaxies:active–galaxies:evolution– galaxies:high-redshift–quasars:
3 general – radio continuum: galaxies
.
0
1
4
1 1 INTRODUCTION (Croton et al. 2006) by AGN activity. Along with the peak
:
v at z ∼ 1 − 2 for both accretion (e.g. Wolf et al. 2003;
A galaxy is described as having an active galactic nucleus
i Uedaet al. 2003) and star-formation (e.g. Madau et al.
X (AGN)whenthereismaterialaccretingontoasupermassive
1996; Hopkins& Beacom 2006) activity, this is thought to
black hole at the centre. Extremely luminous, unobscured
r explain,forexample,whyblack-holemassiscorrelatedwith
a AGN appear point-like at high redshifts, and so are called
the bulge mass (Ferrarese & Merritt 2000; Gebhardt et al.
quasi-stellarobjects(QSOs)or‘quasars’.Someofthesemay
2000;Gu¨ltekin et al.2009).Muchworkoverthepastdecade
bedust-reddened,forwhich therearetwopopularexplana-
has revolved around understanding such feedback mecha-
tions: The orientation argument is that the dusty nuclear
nisms,withsemi-analyticmodelsbeingdevelopedtounder-
torussurroundingtheaccretiondiscprovidesalargeamount
standtheobservations(e.g.Benson et al.2003;Bower et al.
of extinction along certain lines of sight to the quasar (see
2006) and help inform high-resolution simulations (e.g.
Antonucci 1993 for a review). On the ‘evolution side’, the
Duboiset al.2011;Li & Bryan2014).However,debatecon-
quasarbeginslifehighly-obscured,followingagalaxymerger
tinues over how star-formation and accretion processes in-
andassociatedstarburst,butthengraduallyexpelsitsdusty
teract.
envelope(Sanders et al. 1988; Hopkinset al. 2008).
Such scenarios are difficult to decouple from other Galaxies that haveongoing star formation, oractively-
processes associated with either positive or negative feed- accreting black holes, will contain electrons that are accel-
back, where the star formation in the host galaxy can eratedinmagneticfields.Thisleadstoradioemissionbeing
either be stimulated (Silk & Mamon 2012) or truncated produced via synchrotron radiation (Kellermann & Owen
1988; Condon 1992), providing the means to measure both
starformationandAGNactivityat∼GHzfrequencies.How-
⋆ [email protected] ever, for objects without obvious jets, separating star for-
©2015RAS
2 White et al.
mation andaccretion componentsis difficult at radio wave- 2007; Silverman et al. 2009). This is investigated further
lengths. Therefore, multi-wavelength data is often used to by Kimball et al. (2011) and Condon et al. (2013), who
help disentangle thesetwo processes. study the radio emission from samples of optically-selected
quasars, leading them to propose that the radio emission
from these quasars is dueto star formation within the host
1.1 Radio emission from AGN
galaxy, rather than from the AGN.Complementary studies
Although quasars were discovered using radio observa- atotherwavelengthsalsosuggestthatthereisatleastsome
tions, only ∼ 10 per cent have significant radio emission ongoing star formation in quasar host galaxies. Using far-
(Hooper et al. 1996). The remaining radio-quiet quasars infrared observations from Herschel, Bonfield et al. (2011)
(RQQs), that do not show strong jets, need to be finda modest correlation between accretion luminosity and
better-studied at radio wavelengths in order to inves- starformation,andRosario et al.(2013)showthatthemean
tigate their contribution to the overall radio sources starformationrate(SFR)ofquasarhostsisconsistentwith
counts (e.g. Jarvis & Rawlings 2004; Simpson et al. 2006; typicalmassive star-forming galaxies.
Padovani et al.2014)andtheunderlyingphysicalprocesses On the other hand, Zakamska& Greene (2014) use
that may provide a different source of relativistic electrons emission-linekinematics,fromquasarsandtheirhostgalax-
(e.g. Fernandeset al. 2011; Condon et al. 2012). ies, to show that star formation in quasar hosts is insuffi-
The terms ‘radio-quiet’ and ‘radio-loud’ arise from the cient to explain the observed radio emission. Instead they
dichotomyinititallythoughttoexistintheAGNpopulation argue that the synchrotron emission is radiated from the
(Peacock et al.1986;Xu et al.1999).Howevertherearedif- shock fronts of quasar-drivenoutflows. Also, Simpson et al.
ferent definitions for radio loudness, historically this being (2006)citeradio-quietAGNasresponsibleforthechanging
a straight-forward boundary of 1025WHz−1 at radio fre- number counts, with their contribution towards the faint
quency 8.4GHz (Hooper et al. 1996). More common, and (1006S1.4GHz<300µJy)radiosourcepopulation thought
perhaps more physically meaningful, is the ratio between to be at least 20 per cent. However no distinction is made
radio luminosity and optical luminosity (Kellermann et al. with respect to the source of the radio emission in these
1989).However,thedistinctionremainsanarbitraryconcept objects.
asabimodality inradioloudnesshasnotbeenconvincingly
proven (e.g. Cirasuolo et al. 2003; Balokovi´c et al. 2012).
1.2 Paper outline
Whilst Ivezi´cet al. (2004) argue that the bimodality
in the radio-to-optical luminosity ratio is genuine, oth- InthispaperweextendthepreviousstudiesofKimball et al.
ers suggest it is a result of selection effects, with ob- (2011) and Condon et al. (2013), who use photometry and
jects actually found across the full range of radio pow- spectroscopy from the relatively shallow Sloan Digital Sky
ers (Lacy et al. 2001). There is also uncertainty surround- Survey (SDSS; York et al. 2000). We investigate the radio
ing the exact physical mechanism that turns energy from propertiesofradio-quietquasars,selectedfrommuchdeeper
accreted material into well-collimated jets, and why some andnarroweropticalandnear-infraredsurveys,todetermine
systems show no jets at all. Lin et al. (2010) find that whether there is evidence for star formation in the quasar
different radio morphologies are more dependent on the hosts, as a function of redshift and absolute i-band mag-
accretion rate than the galaxy’s structure, with highly- nitude. Section 2 describes the data used, and Section 3
extended, lobe-dominated radio galaxies having higher ac- explains how the sample is selected. Details of the spec-
cretion rates than their less-extended counterparts. Fur- troscopy,usedtoconfirmthequasars,aregiveninSection4.
thermore, the work of Fernandes et al. (2011) indicates We then use a template quasar spectrum to derive pho-
that there is a minimum accretion rate for a given ra- tometric redshifts and absolute i-band magnitudes, as de-
dio power, suggesting that there is a maximum efficiency scribedinSection5.InSection6weanalysetheradioemis-
with which the black hole is able to produce jets from in- sionfromthequasarsample,bothbystackingandbyadopt-
falling material. Meanwhile, the influence of the spin of ing a parametric description of the radio source counts for
the black hole is still an open question. Several theoreti- thequasars, and discusstheorigin oftheemission in radio-
cal and observational studies suggest that it may play a quietquasars.OurconclusionsareoutlinedinSection7.AB
role(e.g.Blandford & Znajek1977;Wilson & Colbert1995; magnitudes are used throughout this paper (see Table A1
McLure & Jarvis 2004; Volonteri et al. 2007; King et al. for conversions to Vega), and we use a ΛCDM cosmology,
2008; Fernandes et al. 2011), whilst van Velzen & Falcke with H =70kms−1Mpc−1, Ω =0.3, Ω =0.7.
0 m Λ
(2013) present evidence that it could beirrelevant.
Studies of the radio source populations as a func-
tion of radio flux density suggest that as we probe fainter
2 DATA
radio sources, the population changes from being AGN-
dominated (FRIs, FRIIs) to being dominated by star- Our aim is to study the radio properties of a sample of
forming galaxies (Hopkinset al. 2003; Wilman et al. 2008; quasars, with zero contamination from inactive galaxies, as
Condon et al. 2012). However, there are obviously cases afunctionofluminosityandredshift.(Hereafterwedescribe
in which AGN are hosted in galaxies with ongoing star such a sample as being ‘clean’.) For this we require multi-
formation (e.g. Canalizo & Stockton 2001; Netzer et al. wavelength observations for the initial selection and for de-
©2015RAS,MNRAS000,1–??
AGN-powered radio emission below 1mJy 3
slightly bluer, i′ filter were not used for the Deep Field
Table 1. Multi-wavelength data used over the XMM3 tile of 2
the VIDEO Survey, which is centred at RA(J2000) = 02:26:18, stacks.
DEC(J2000) =-04:44:00. TheVisible Multi-Object Spectrograph (VIMOS)VLT
CFHTLS–D1=Canada–France–HawaiiTelescopeLegacySurvey Deep Survey (VVDS)provides a catalogue of spectroscopic
Deepfield1,covering1deg2 withintheXMM3tile. redshifts selected with 17.5 6 i 6 24.75, with a spectral
VIDEO=VISTADeepExtragalacticObservations resolution of R≈230 in two deep fields, VVDS–CDFS and
SWIRE=SpitzerWide-areaInfraredExtragalactic VVDS–02h (Le F`evreet al. 2013). We use data from the
IRAC=InfraRedArrayCamera
VVDS–02h field, which overlaps entirely with the VIDEO–
XMM3 and CFHTLS–D1 fieldsover an area of 0.61deg2.
Surveyname Band Wavelength Point-sourcesensitivity
(µm) (ABmag,5σ)
CFHTLS–D1 u 0.38 27.4 2.3 Radio: VLA
CFHTLS–D1 g 0.48 27.9
To investigate the level of radio emission in the quasars we
CFHTLS–D1 r 0.63 27.6
use radio data, at 1.4GHz, from the VLA–VIRMOS Deep
CFHTLS–D1 i 0.77 27.4
CFHTLS–D1 z 0.89 26.1 Field(Bondi et al.2003).Theareacoveredis1deg2,centred
VIDEO Z 0.88 25.7 at RA(J2000) = 02:26:00, DEC(J2000) = -04:30:00, again
VIDEO Y 1.02 24.5 completelycontainedwithinwithVIDEO–XMM3field.The
VIDEO J 1.25 24.4 averagermsnoiseoftheradiomapis17.5µJyandtherestor-
VIDEO H 1.65 24.1 ing beam is a 6 × 6arcsec Gaussian. Bondi et al. (2003)
VIDEO KS 2.15 23.8 create a catalogue of all radio components with peak flux
SWIRE IRAC1 3.60 22.5
S >60µJy (∼3.5σ), resulting in 1054 radio sources, with
SWIRE IRAC2 4.50 22.1 P
19 thought to be multi-component. These data have previ-
SWIRE IRAC3 5.80 19.7
ously been cross-matched with the VIDEO survey data by
SWIRE IRAC4 8.00 20.0
McAlpine et al. (2012).
2.4 Mid-infrared: SWIRE
terminingphotometricredshifts.Wealsousefollow-upspec-
troscopy of a subsample of our quasars, as a check on the Tocheckourquasarselectionmethod,weuseSpitzer/IRAC
selection to low fluxes and photometric redshift accuracy. imaging from the SWIRE survey that completely overlaps
Thesedataaredescribedbelow,andasummaryoftheimag- with the VIDEO–XMM3 field. This allows dust, associated
ingdatausedinthispaperisprovidedinTable1.Thefinal with star-forming regions and AGN, to be traced out to
area over which we select the candidate quasars is 1deg2, z ∼ 3 (Lonsdale et al. 2003). For the IRAC data we follow
as determined by the extent of the CFHTLS–D1 field (see therecommendationtouse‘aperture2’,whichgivestheflux
Section 2.2). (in µJy) as measured with an apertureradius of 1.9arcsec.
2.1 Near-infrared: VIDEO
3 QUASAR SELECTION
Our primary selection relies on near-infrared photometry
Many studies have used multi-colour photometry to define
and we use the 5-band near-infrared ZYJHK data from
S
clean samples of quasars, free from contamination by stars
theVISTADeepExtragalactic Observations(VIDEO)Sur-
and galaxies. Mortlock et al. (2012) used optical and near-
vey (Jarvis et al. 2013), over the XMM3 tile, which spans
1.5deg2 inarea.TheVIDEOimaginghasatypicalseeingof infrareddata,whilstRichardset al.(2006)usedopticaland
mid-infrared data. In this paper we use a combination of
<0.9arcsec, and a 2arcsec diameter aperture was used for
opticalandnear-infraredforourselection,tofaintermagni-
themeasurements.
tudes than previous work, and check the robustness of our
selection using mid-infrared data.
Inassessingwhetherradioemissionfromquasarsisdue
2.2 Optical: CFHTLS and VVDS
to star formation in the host or from the accretion pro-
The Canada–France–Hawaii Telescope Legacy Survey cess, our key concern is to ensure a high reliability, at the
(CFHTLS) providesphotometryin ugriz over1deg2 of the expense of completeness. We therefore impose a series of
VIDEO–XMM3field(Gwyn2012).Thisdeepfield(labelled cutstomaximisethereliabilityofoursample.Maddox et al.
‘D1’) is centred at RA(J2000) = 02:25:59, DEC(J2000) = (2008) showed that theJ−K colour isvery effectiveat se-
-04:29:40, and again weuse a2arcsec aperturefor thepho- lectingquasars, with thefraction beingmissed duetodust-
tometry. Note that the MegaCam filters used are not iden- reddening determined to be less than 10 per cent. This is
tical to those of SDSS, and that the i filter needed to be because there is an excess in the J −K colour for quasars
replaced partway through the survey. The band referred to – even those reddened by dust – compared to that for fore-
in Table 1 is the original i′ filter, as images with the new, groundstars.Suchstarsmayhaveopticalcolours similar to
1
©2015RAS,MNRAS000,1–??
4 White et al.
thequasars,butnear-infrared colourallows themtobedis- distribution of r for the host galaxy over this regime has
eff
tinguished(Warren et al.2000).Toexploitthis,wetherefore noinfluenceon theK CLASS STARvalues.
startbyselectingallobjectsintheVIDEOK -bandbrighter To achieve > 90 per cent completeness over 21.5 6
S
than the 5σ limit of K = 23.8, as detailed in Jarvis et al. K 6 22.5, we find that the AGN’s contribution to the to-
S S
(2013). tal flux must rise again to at least 70 per cent. This means
thatweshouldbeawareofpossiblecontaminationfromthe
hostgalaxy,asaconsequenceofusingtheK CLASS STAR
3.1 K-band selection
>0.9cutacross thismagnituderange.Fortotalmagnitude
We expect the vast majority of quasars to outshine their KS > 22.4, no level of contribution from AGN to the to-
host galaxies in the KS-band, and therefore impose a mor- tal flux (fAGN/ftotal) allows us to attain even 50 per cent
phology cut based on the SExtractor K CLASS STAR completeness. Therefore we impose a cut of KS 6 22.4, in
parameter (Bertin & Arnouts 1996). This is a measure of addition to using K CLASS STAR > 0.9 as a selection cri-
thecompactnessofasourceonascaleof0to1,with1repre- terion,whichensuresthatthemorphologyofthecandidates
sentingapoint-likeappearance.Tochecktheeffectivenessof isreliable.Whilstthismeansthatwemaybebiasedtowards
K CLASS STARasamorphology indicator, weselect stars thebrightestquasarsandfaintest hostgalaxies, suchacon-
from the VIDEO catalogue using giJK colours (Fig. 1). servativeapproachissufficientforthepurposeofthispaper.
S
Following the method of Baldry et al. (2010), Jarvis et al. Indeed,itstrengthenstheconclusionswedrawregardingthe
(2013) show that the region of colour space defined as radio emission, as explained in Section 6.4.
J−K −f (g−i)<0.1 providesaclean stellar sample,
S locus
free from contamination bygalaxies. Fig. 2 shows thevalue
3.2 Photometric fitting
ofK CLASS STARasafunctionofK -bandmagnitudefor
S
theseobjects,andweseethatourpoint-sourceclassification As stated previously, optical bands alone have in the past
is robust down to K =22.8. Therefore imposing a restric- been used to definea colour space for quasar selection (e.g.
S
tion on the morphological parameter, K CLASS STAR > Richardset al. 2001). However their success in separating
0.9,effectivelymeansthatweshouldeliminateobjectswith theregionsoccupiedbystarsandquasarsdiminishesathigh
K >22.8. redshifts,andtheyareespeciallyaffectedbydust-reddening.
S
Asafurthercheck,sincewewillrelyonthismorphology Instead we follow the method of Maddox et al. (2008) and
parameter for quasars rather than stars, we create a set of use gJK colour space (Fig. 3). We also use full spectral
S
simulatedAGN.Thisisdonebyaddingapointsource,rep- energy distribution (SED) modelling, as inactive galaxies
resenting the output from the quasar nucleus, to the image may still be present amongst our potential quasar candi-
of a model host galaxy. The latter are simulated using the dates,contaminating oursample. FittingSED templatesto
methodology of H¨aussler et al. (2007). For numerous steps the VIDEO–CFHTLS–D1 data was carried out using Le
in the total K -band magnitude, different values of AGN PHARE (Ilbert et al. 2006).
S
magnitude (m ), galaxy magnitude (m ), and effec- There are 62 galaxy templates, based on the SEDs
AGN galaxy
tiveradiusforthehost(r )areassigned(seeAppendixB). compiled by Coleman et al. (1980) and those of starburst
eff
SExtractor is then used on the simulated images, out- galaxies. In combination with different extinction law val-
putting the K CLASS STAR value for each object. Next, ues, these produce 187 sets of model photometry that are
for each subset of the resulting catalogue – defined by a thenredshifted.Arangeofextinctionvaluesarealsousedin
particularcombinationoftotalmagnitude(K ),m and conjunction with the stellar library, consisting of 254 tem-
S AGN
m – the fraction of objects with K CLASS STAR > plates, and the 10 empirical SEDs that make up the AGN
galaxy
0.9 is calculated (Table B2). This provides an estimate of library.
our detection completeness, with respect to the AGN. TheseAGNtemplatesaretakenfrom theSWIREtem-
A simulated AGN that has been correctly classified as plate library (Polletta et al. 2007). Two of them, ‘Sey18’
point-likewillhaveaK CLASS STARvaluecloseto1.Our and ‘Sey2’, represent moderately-luminous Seyfert galax-
results show that for thebrightest objects, with K <19.0, ies. The ‘I19254’ template is from IRAS19254-7245, a star-
S
the AGN must account for at least 90 per cent of the total burst/ULIRG with Seyfert-2 galaxy properties. Another
flux if over 90 per cent are to haveK CLASS STAR > 0.9. ULIRG is included in the form of ‘Mrk231’, which has a
This corresponds to the AGN needing to outshine its host heavily-obscuredquasaratthecentre.Like‘I19254’andthe
galaxybyovertwoordersofmagnitude.Thereasonforthis ‘N6240’ AGN template, its SED is an AGN-starburst com-
is that the host galaxies belonging to such bright systems posite.
arethemselvesbright,andsotheirextendedappearancein- Hatziminaoglou et al. (2005) used an optically-selected
fluencestheK CLASS STARmeasurement.However,wedo sample of 35 SDSS/SWIRE Type-1 quasars to construct
notexpectthistogreatlybiasoursampleasveryfewquasars the three Type-1 quasar templates (‘QSO1’, ‘BQSO1’,
in our sample have K < 19.0. Over the more populated ‘TQSO1’), via a quasar composite spectrum in addition to
S
range of 19.0 6K 621.5 (due to the shape of the quasar rest-frame infrared data. The difference between them is
S
luminosity function and depth of our data), we are able to their IR/optical flux ratios, with ‘BQSO1’ representing the
recoverover90percentofthesimulatedAGNprovidedthat bottom quartile and ‘TQSO1’ the top quartile. Finally, the
theirfluxis>50percentofthetotalflux.Wefindthatthe ‘QSO2’ and ‘Torus’ templates are typical SEDs for Type-2
©2015RAS,MNRAS000,1–??
AGN-powered radio emission below 1mJy 5
2.0
1.5
)
i
−1.0
g
(
us
oc
fl
−
S
K
−
J 0.5
0.0
−0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
g i
−
Figure 1. The position of all VIDEO objects in giJKS colour space. Red contours (dashed, with interval of 2 objects) correspond to
the number of objects best fit by a star template, viathe Le PHARE code (Section 3.2). Green contours (solid, interval = 10) are for
thosebestfitbyagalaxytemplate,andbluecontours(dotted,interval=2)correspondtoanAGNtemplate.TheAGNandgalaxyloci
begintoblendintooneanother,whilsttheobjectsbelowtheblack(dash-dotted)lineareusedtocreateacleansampleofstars(Fig.2).
Overplottedareobjectsfromthe‘Gold’candidatequasarsubsample(blackdots,Section3.2),includingthosethatarecross-matchedwith
VVDS (magenta squares, Section 4.1). In addition are VIDEO objects cross-matched with VVDSbut lying outside the Gold selection
region, either withbroad emissionlines (red triangles) or without (yellow triangles). The definition of flocus(x), where x=g−i, is as
follows:forx60.3,flocus(x)=−0.6127;for0.3<x62.3,flocus(x)=−0.79+0.615x−0.13x2;forx>2.3,flocus(x)=−0.0632.
quasars, with the second of these having greater dust ob- We also show the broad region occupied by the galaxy
scuration. We determine the best fit for both redshift and tracksthatareusedbyLePHARE(shadedgreyinFig.3).
extinction using each of the SED libraries (‘star’, ‘galaxy’ Some of the AGN template-fitted data points are well-
and ‘AGN’). separatedfromthegalaxyregion,whilstothershavegreater
overlap. Those that remain ‘clear’ from the galaxy locus,
Fig.3showsthegJKScolourspaceforthepointsources withintheregiondemarcatedbyagreendashedline(defined
intheVIDEOKS-bandselectedcatalogue whereoneofthe bytheequationslistedinAppendixC),aresubsequentlyre-
tenAGNSEDsprovidesabetterfittothephotometrythan ferredtoasthe‘Goldcandidatequasarsample’.Weusethe
thegalaxyorstartemplates.Point-likeobjectsthatarebest- descriptor ‘Gold’ simply to distinguish the candidates that
fit by a stellar template are predominantly found in the we are confident of being quasars. (Additional samples, la-
light-blue, hatched region. Therefore the data points that belled‘Silver’forexample,couldbeconstructedusingmore-
lie within this region represent objects with stellar gJKS relaxed selection criteria.) The Gold sample comprises 75
colours, despite their photometry being better-modelled by quasars, with one later being removed as the result of an
anAGNSED.Theblue,dottedlineinFig.3istheevolution absolute i-band magnitude cut (Section 5.2). We deem the
track for a model quasar that has no host galaxy contam- slight overlap with thegalaxy tracks (shaded region) as ac-
ination, and follows a path that lies between the star and ceptable given the photometric uncertainties. The selected
galaxy regions.
©2015RAS,MNRAS000,1–??
6 White et al.
abiastowardsbrightquasarsresidinginfainthostgalaxies.
1.0
This is because a host that is bright in the K -band may
S
0.9
have an extended appearance in the imaging data, causing
0.8 the object to be eliminated by our morphological cut. This
potentialbiasis discussed inthecontextofourfinalresults
0.7
AR in Section 6.4.
T0.6
S
_
SS0.5
A
L 3.3 Mid-infrared colour
C0.4
_
K0.3 As an additional check on the validity of our selection, we
use Spitzer data to investigate where the Gold candidate
0.2
quasars lie in mid-infrared colour space. Dust extinction is
0.1 notaproblemwhenmeasuringemissioninthemid-infrared,
andsobothunobscuredandobscuredAGNcanbedetected.
0.0
16 18 K 20 22 24 However, we note that our Gold candidates are biased to-
S
wards the unobscured population, given the photometric fit
to Type-1 quasar SEDs. To check what proportion are in-
Figure2.ThevariationinK CLASS STAR,anindicatorofcom-
deedactive-galaxycandidates,weusetheworkofLacy et al.
pactness(1=point-like),foracleansampleofstars.Eachofthese
(2004) and Stern et al. (2005).
objects are KS-band selected (KS 6 23.8) in the VIDEO cata-
logue,andhavestellarcoloursusingthelocusshowninFig.1 Lacy et al. (2004) use the 8.0µm/4.5µm ratio versus
the5.8µm/3.6µmratiotoseparatestars,low-redshiftgalax-
.
ies, SDSS quasars and radio-selected quasars. Within this
colourspacetheyfindthattherearetwodistinctsequences,
the quasars belonging to one and the remaining objects to
objects are best-fitted by Type-1 quasar models: ‘QSO1’,
the other. We refer to the region that contains the quasar
‘BQSO1’ and ‘TQSO1’, and we later see where they lie in
sequenceas the ‘Lacy wedge’.
mid-infrared colour space (Section 3.3). Valuing reliability
ThetendencyofAGNtoberedderthangalaxiesinthe
over completeness, we chose not to extend the selection re-
mid-infrared is also exploited by Stern et al. (2005), who
giondownwardstoincludeafewextrapointsfittedbyType-
use a [3.6] − [4.5] versus [5.8] − [8.0] colour-colour space.
1 quasar models. This is because, with theirpositions lying
The combination of restricted [5.8]−[8.0] colours observed
well within the shaded region, there is a greater chance of
for powerful AGN (due to a lack of strong PAH emission)
their gJK colours being confused for galaxies. Similarly,
S withtheir[3.6]−[4.5] colour(muchredderthanthatoflow-
the selection region is not extended bluewards to include
redshiftgalaxies)leadstoanotherclearselectionregion.We
the BQSO1 objects around J −K = 0.5, g−J = 0.5, as
S call thisthe ‘Stern wedge’.
doing so may introducecontamination bystars.
We find that 62 of our 75 Gold quasar sample have
A similar method is used by Maddox et al. (2012) to
cross-matches in the SWIRE catalogue. Due to the depth
construct a sample of quasars, whose K-band magnitude
of the mid-infrared data, only 35 of these have detections
range is 15.8 6 K 6 18.4. They are able to achieve > 95
in each of the 4 bands, IRAC 1-4. Hence we use this as a
per cent efficiency and >95 per cent completeness with re-
consistencycheckonthebrightercandidates,ratherthanan
spect to known SDSS quasars. In addition, both their and
additional step in theselection. Fig. 4 shows that all of the
ourmethodsincorporateinformation from SEDfitting.Do-
Gold sample (with detections in all 4 bands) lie within the
ingsoenablesustobeconservativeinourselection,resulting
LacyandSternwedges.Thisprovidesfurtherevidencethat
in a clean sample of candidates. However, in contrast, our
ourquasar selection method is robust.
candidates are much fainter due to the depth of the pho-
tometry used, spanning 18.4 6 K 6 22.4. The light from
S
the host galaxy could lead to greater contamination as we
godeeper,butourselection (viamorphology andcolourin-
4 SPECTROSCOPY
formation) ensures that the quasar is dominating the total
fluxfrom thesystem. Inthissection weusespectroscopic datatocarryout afur-
As a result of our selection criteria, we actually have ther check of the robustness of our sample selection, in ad-
verygoodcompletenesswithrespecttounobscured quasars, dition to obtaining accurate redshift information. Existing
as themajority of Type-1sources lie in theselection region spectroscopy comes from the Visible Multi-Object Spectro-
(demarcated bythe green, dashed line in Fig 3). The SEDs graph(VIMOS)VLTDeepSurvey(VVDS),overtheXMM3
thatwehaveexcludedareforcompositesystemsorobscured tile within the VIDEO field. This is complemented by new
AGN. This means that our sample is not representative of spectroscopyforobjectsbelongingtoourGoldsample,using
theentirequasar population,and so weemphasise thatour the Southern African Large Telescope (SALT). We discuss
investigation accounts for the radio emission from Type-1 howthespectroscopicredshiftscomparetothephotometric
quasarsonly.Also,theuseofnear-infrareddatamayleadto redshifts obtained from the template fittingin Section 5.1.
©2015RAS,MNRAS000,1–??
AGN-powered radio emission below 1mJy 7
−1
0
1
2
N6240
J
− 3
g Sey2
QSO1
4 BQSO1
TQSO1
QSO2
5
Torus
Mrk231
6
I19254
Sey18
7
−1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 2.5
J K
− S
Figure 3. The positions of point-like objects (i.e. those with K CLASS STAR > 0.9) in gJKs colour space, following Maddoxetal.
(2008). Each of these data points has photometry that is best fit by one of 10 AGN templates (see legend, and their description in
Section3.2).Thelight-blue,hatchedregionrepresentsthestellarlocus,occupiedbyVIDEOobjectsforwhichastellartemplateprovides
thebestfittothephotometry.Thebluedottedtrackisamodelforhowaquasarwithnohostgalaxycontaminationevolveswithredshift,
lying between the stellar and galaxy loci (to the left and right, respectively). The redshift range of this track is 0.06z < 5.0, with a
blue star markingz =0.0 and blue ‘+’ markers every ∆z =0.2(with a boldmarker denoting z =4.8). The grey shaded region is the
areacoveredbyevolutiontracksforthephotometricfittingcode’slibraryofgalaxytemplates.Theredshiftrangeofthesegalaxytracks
is06z<6,withcoloursevolvedforeachstepof∆z=0.04taken.Theregionusedtoselect‘Gold’candidatequasarsisdemarcatedby
agreendashedline.
4.1 VVDS data to be quasars. However, inspection by eye reveals that this
is actually the case for only 5 of the objects, one of which
We cross-matched all of the objects that had best-fit SEDs is a Broad Absorption Line (BAL) quasar. Wealso confirm
of anAGN (obscured,unobscured,orSeyfert-like) with the that reliable redshifts have been obtained for each of these
VVDS, using a matching radius of 1arcsec. This identifies 5. They appear as red triangles in Fig. 1, with one lying
127 objects with spectroscopic redshifts over the 0.61deg2 beyond the plot range of the figure. Their colours are red-
covered by the VVDS–02h field, twelve of which are in our der than the Gold quasar candidates, and lie in the region
Gold sample. Each of these 12 objects have broad emission of gJK colour space covered by galaxy evolution tracks,
S
lines, again showing that we have a good basis for select- hencetheirexclusion from oursample. These extra quasars
ing quasars. They also lie in the quasar region of giJKS highlight the incompleteness of our Gold sample, although
colour space (magenta squares in Fig. 1), as identified by weemphasiseouraimforahighly-reliableselectionmethod
Baldry et al. (2010). instead.The110cross-matched objects thatshow nobroad
Theremaining115cross-matcheslieinthegalaxylocus emissionlines,yetarebest-fitbyanAGNtemplate,maybe
of both giJKS space (triangles in Fig. 1) and gJKS space obscured AGN.
(Fig. 3). According to the VVDS redshift flags, 7 of them
havebroademissionlines,meaningthattheyaremorelikely
©2015RAS,MNRAS000,1–??
8 White et al.
1.5 1.5
1.0
1.0
S]4.5 0.5 Vega)
og[S/108.0 0.0 6][4.5](− 0.5
l 3.
[
0.0
−0.5
−1.0
−0.5
−1.0 −0.5 0.0 0.5 1.0 −0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
log [S /S ] [5.8] [8.0](Vega)
10 5.8 3.6 −
Figure 4. Gold objects that have mid-infrareddata are shown, followingLacyetal. (2004) (left) and Sternetal. (2005) (right). The
dashedlinesoutlinetheregionsthattheseauthorsusetoselectactive-galaxycandidates.
4.2 SALT spectra (Ilbert et al. 2006). Such inaccuracies prompted us to seek
an alternative method for determining photometric red-
13 quasars from our Gold sample were observed between
shifts.
4thNovember2012and6thJanuary2013,usingtheRobert
We therefore used a different quasar template that has
Stobie Spectrograph (Kobulnickyet al. 2003) on SALT
no host galaxy contribution (Hewett et al., in prep.). The
(Buckley et al. 2006). We used the PG0900 grating with
the PC03850 filter, to obtain coverage over 4500–7500˚A. colours from this template were evolved over the redshift
A2arcsecslitprovidedadequateresolution (10˚A)formea- range0<z <7instepsof0.1,andwealsoincluded0.2mag
of uncertainty to the quasar template colours to account
suring redshifts from broad emission lines. Three exposures
for the variation in emission-line strength and the intrinsic
were made for each target to enable cosmic ray rejection.
quasarspectralshape.Thiswasaddedinquadraturetothe
The standard proceduresfor reducingSALTdata were
performed with the IRAF software. For flux calibration, measuredphotometricuncertainties.Thebestfitwasfound
through χ2 minimisation, and we name the corresponding
LTT1020wasusedasthestandardstar.Theresultingfully-
redshift ‘Colour-z’. Colour-z values calculated for the ob-
reduced spectra, extracted from the calibrated images, are
jects with spectroscopic redshifts are presented in Table 2.
presented in Appendix D. They correspond to the 8 Gold
candidates for which it was possible to determine reliable Fig. 5 shows the correlation between the spectroscopic
spectroscopic redshifts. These are discussed further in Sec- redshift, from both SALT and VVDS, and our photomet-
tion5.1.Strongemission linesarepresentinallofthespec- ric redshift. Two of the VVDS redshifts were highly dis-
tra, with Ciii] being the most common. Also present is crepant, quoted to be 0.1 and 0.2 (Table 2). Inspection of
1909
eitheraCiv1549 orMgii2799 broadline.Spectrafortheother their spectra showed that a Mgii2799 line had been mis-
5 objects observed by SALT were poor, due to being badly identified as a Hα6563 line. We therefore corrected the red-
affected byweather and technical difficulties. shifts and used these values for Fig. 5, which placed them
in line with their corresponding Colour-z value. Using all
of the objects with spectroscopic redshifts, the accuracy of
5 THE QUASAR SAMPLE Colour-z isσ∆z/(1+z) =0.046.Outofthe19objectswefind
there to be 3 catastrophic outliers (circled in Fig. 5). We
5.1 Photometric redshifts investigated whether confusion between the Mgii line
2799
The photometric redshifts, fitted using Le PHARE (Sec- and the Civ1549 line led to the catastrophic outliers with
tion 3.2), were compared to the spectroscopic redshifts of Colour-z = 0.7, and similarly whether a Ciii]1909 line was
theGold objects, obtained with theVVDSand SALT.The beingconfusedforaMgii2799 line,resultinginthethirdout-
majority areingood agreement(seeTable2),butthereare lier (with Colour-z = 2.4). However, in each case, the ab-
a few cases where the photometric fitting severely under- sence/presence of other expected emission lines confirmed
estimated the true redshift. Using the normalised median thepreviously-determined spectroscopic redshifts.
absolute deviation in ∆z/(1+z),where∆z=(z −z ),we Therefore our Colour-z photometric redshift estimates
s p
findσ =0.082andfourcatastrophicoutliers,where perform well for the majority of quasars observed by the
∆z/(1+z)
wedefineacatastrophicoutlierashaving|∆z|/(1+z)>0.15 VVDSandSALT.However,asinglespectrumisnotenough
©2015RAS,MNRAS000,1–??
AGN-powered radio emission below 1mJy 9
prompted us to generate 1000 Monte Carlo redshifts per
Table 2. Summary of redshifts for the QSOs spectroscopically
confirmedbytheVVDS,andthoseobserved bySALT (except 5 Gold quasar, according to the full PDFs calculated for the
forwhicharedshiftcouldnotbedetermined).*denotesaVVDS Colour-z redshifts. These redshifts are used for the distri-
object assigned an incorrect redshift; the redshift given is our butionin absolute i-bandmagnitudes(Section 5.2) and the
measurement(seetextfordetails). subsequent analysis of the radio emission associated with
the quasars. This therefore allows our final results to take
R.A. Dec. Le PHARE Colour VVDS/ theredshift uncertainties into account.
(hms) (dms) z z SALTz
02:25:25.68 -04:35:09.6 0.4 2.1 2.1
02:25:45.53 -04:34:45.6 1.7 1.7 1.7*
02:25:50.38 -04:33:24.6 2.8 2.5 2.7
5.2 Absolute i-band magnitudes
02:25:52.16 -04:05:16.1 1.5 1.4 1.4
02:26:09.62 -04:24:38.0 2.6 2.8 2.7 The same quasar spectrum (Hewett et al., in prep.), in ad-
02:26:12.64 -04:34:01.4 2.2 0.7 2.3 dition to the Colour-z values determined in the previous
02:26:18.59 -04:11:01.1 0.4 1.9 2.0
subsection, is used to calculate the K-corrected absolute i-
02:26:22.15 -04:22:21.8 2.0 1.8 2.0
band magnitudes (M ) of our quasar sample. We also use
02:26:24.64 -04:20:02.4 2.0 0.7 2.2 i
the redshift probability distributions to calculate the prob-
02:26:33.31 -04:29:47.8 2.4 2.4 2.1
ability distribution for M per quasar. We then impose the
02:26:39.83 -04:20:04.4 1.9 1.5 1.6 i
02:26:52.14 -04:05:57.1 1.4 1.4 1.4 criterion Mi 6 −22, a magnitude cut that approximately
02:27:07.54 -04:32:03.2 1.6 1.5 1.5* splits quasars from Seyfert galaxies (e.g. Schneideret al.
02:27:09.03 -04:55:10.1 3.0 2.7 2.7 2003). The objects that survive this cut are presented in
02:27:33.99 -04:25:23.5 1.8 1.5 1.6 Fig. 6 and used for the remainder of the paper. We find
02:27:36.93 -04:26:31.5 0.6 1.8 1.8 one quasar has no part of its redshift probability distribu-
02:27:38.99 -04:09:40.8 1.4 1.3 1.4 tionwithM <−22,andsothisobject isremovedfrom the
i
02:27:40.55 -04:02:51.1 2.4 2.6 2.6
sample. The properties of the final 74 quasars are given in
02:27:47.34 -04:27:53.7 0.2 2.4 1.1
Table E1.
61 of these 74 objects are beyond z = 1 and
brighter than M = −22, demonstrating the effectiveness
i
of VIDEO+CFHTLS in detecting very faint quasars. How-
3.0
VVDS ever,some oftheobjects may havean incorrect Mi,arising
SALT from an incorrect Colour-z. This is why it is important to
note that we take this into account via our sampling of the
2.5
photometric-redshift probability distribution. For example,
the fraction of simulated objects that have M 6 −22 is
z i
pic2.0 usedtoweight theGold quasarswhen analysingtheirradio
o
osc emission inthenextsection.InFig.7wealsoshowtheKS-
ctr band magnitude distribution of the final 74 quasars. Note
Spe1.5 thatthemajorityareindeedintherange19.06KS 621.5,
as previously mentioned in Section 3.1.
Tounderstandtheproportionofthetotalquasarpopu-
1.0
lation we haveselected with our criteria, we use the quasar
luminosity functions modelled by Croom et al. (2009)1.
These are integrated over the redshift range of the final
0.5
0.5 1.0 1.5 2.0 2.5 3.0
Photometricz(Colour z) Gold sample, 0.5 6 z 6 3.1 (Fig. 6), for objects brighter
−
than M (z = 2) = −22.8, which corresponds to our cut of
g
M (z=0)=−22.0.Usingtheluminositydependentdensity
Figure5.Spectroscopicredshifts,fromSALTandVVDS,show- i
evolution(LDDE)andluminosityevolution+densityevolu-
ingtheaccuracyofColour-zasaphotometricredshift,including
tion(LEDE)modelsaslowerandupperlimits,respectively,
thecatastrophic outliers(circled,seetextfordetails).
these luminosity functions predict there to be between 338
and381quasarsover1deg2.Wearethereforeapproximately
to describe the variety found in the Type-1 quasar popula- 20 per cent complete.
tion, and for this reason we still find a number of outliers.
As such, we cannot simply assume these best-fit values for
the remainder of the photometric sample, for which we do
not have spectroscopy. Instead we note that each of these
outliershaveadouble-peakedprobability distributionfunc- 1 Wenotethatthevaluegivenintable4ofCroometal.(2009)
tion (PDF) produced by the photometric fitting, with the shouldbe log (Φ∗)=−5.99±0.07 (S. Croom, private commu-
10 0
secondarypeakoverlappingthespectroscopicredshift.This nication).
©2015RAS,MNRAS000,1–??
10 White et al.
7 declination to an integer pixel number, the positional error
0
006 ofthequasarassociatedwiththechosenpixelis1.5arcsecfor
1
nt/5 both co-ordinates. This error is comparable to the angular
ou4 size of the objects in question, but deemed negligible given
erc3 the6arcsec resolution of thedata.
b2
m The mean and median radio flux-density values ob-
u1
N tained at the positions of the Gold objects are shown in
0
Table3,notingthattheradiomaphasanaveragermsnoise
-22.0 of 17.5µJy (Bondi et al. 2003). In order to take the noise
Mi
variation across the map into account, a random sample is
ude,-23.0 alsocreated,usingpixelsbetween17and42arcsecaway(in
nit a random direction) from each Gold quasar position. This
ag-24.0
m was to ensure that the new position was well outside the
d
an-25.0 beam covering the Gold candidate position (i.e. more than
b− twobeamsizesaway),andsoavoidanypossiblecorrelation
i
e-26.0 in the flux values. Also, using a fixed direction would have
ut
ol introducedasystematicerrorduetoartefactsarisingduring
Abs-27.0 theradio imaging process, caused by the6-order symmetry
of the VLA’s uv-coverage. Repeated 1000 times per Gold
-28.0
0.5 1.0 1.5 2.0 2.5 3.0 0 1 2 3 4 5 6 7 8 9 candidate, this method ensures that all of the random pix-
Redshift,z Numbercount/1000
els lie within annular loci, centred on a Gold position.
The fraction of negative flux density values is an in-
Figure 6. The distribution of photometric redshifts (Colour-z) dication of the level of non-detections, and is significantly
andK-correctedabsolutei-bandmagnitudesforthefinal74Gold
lowerfortheGoldpositionsthantherandompositions(Ta-
quasarswithMi6−22(contour interval=10),determinedusing ble3).SincetheGold quasarcandidatesarepart of aclean
thefullredshiftprobabilitydistributions.
sample, their median radio flux density being higher than
that for the random positions is expected. Fig. 8 shows
flux-density histograms for the quasars and the random
8000
positions, and clearly shows a positive tail in the case of
7000 the former, reinforcing the evidence for excess radio emis-
sion at the positions of the quasars. Within this tail are
6000 10 objects with detections at > 3σ, one of which is above
1mJyandappearsinboththeFIRSTandNVSScatalogues
unt5000 (Whiteet al. 1997; Condon et al. 1998). Assuming a typi-
o
c cal spectral index of α = −0.7 (where S ∝ να), this is
er4000 ν
mb theonlyquasarinthesamplethatis‘radio-loud’,according
Nu3000 to the definition of L8.4GHz > 1025WHz−1 (Hooper et al.
1996).Italsosatisfiesthecriteriaforradio-loudnessdefined
2000 by Kellermann et al. (1989) and Miller et al. (1990). How-
ever, we reiterate that these are arbitrary boundaries, and
1000
although the emission process may be physically different,
we continueto includethis object in our analysis.
0
18 19 20 21 22
K Lastly,although theaveragespresentedareinteresting,
S
it must be remembered that stacking leads to a loss of in-
formation and it isfar moreinformative tostudytheentire
Figure 7. The KS-band magnitude distribution forthe final 74
Goldquasars,forthosethathaveMi6−22generatedfromsam- flux-densitydistribution (e.g. Mitchell-Wynneet al. 2014).
plingtheredshiftprobabilitydistribution.
6.2 Statistical detections of radio emission
6 RADIO EMISSION FROM RADIO-QUIET The majority of objects from our quasar sample haveradio
QUASARS emission that is below the flux-density limit of the VLA
map.Thereforeweanalysetheradiofluxdistributionofthe
6.1 Radio flux-density measurement
quasars to determine whether or not we have a statistical
ToobtaintheradiopropertiesoftheRQQsinoursample,we detection of significant radio emission. Fig. 8 indicates that
initiallyperformedastackinganalysis.Weused1.4GHzflux we do, but to investigate this quantitatively a two-sample
densitiesextractedfromtheVLA–VIRMOSDeepFieldmap Kolmogorov–Smirnov (KS) test is used. This tests the null
(Bondi et al.2003)onasingle-pixelbasis,atthepositionsof hypothesisthattheGoldquasarsampleandtheflux-density
thequasars.Duetotheconversionfromrightascensionand measurementsatrandompositionsaredrawnfromthesame
©2015RAS,MNRAS000,1–??