Disentangling star formation and AGN activity in powerful infrared luminous radio galaxies at 1 \&lt

26 Pages·2016·3.77 MB·English

A&A593,A109(2016) Astronomy
DOI:10.1051/0004-6361/201526880
&
(cid:13)c ESO2016 Astrophysics
Disentangling star formation and AGN activity in powerful infrared
luminous radio galaxies at 1 < z < 4
G.Drouart1,2,3,B.Rocca-Volmerange3,C.DeBreuck4,M.Fioc3,M.Lehnert3,N.Seymour2,D.Stern5,andJ.Vernet4
1 DepartmentofEarthandSpaceScience,ChalmersUniversityofTechnology,OnsalaSpaceObservatory,43992Onsala,Sweden
2 InternationalCentreforRadioAstronomyResearch,CurtinUniversity,Perth,Australia
e-mail:[email protected]
3 Institutd’AstrophysiquedeParis,98bisboulevardArago,75014Paris,France
4 EuropeanSouthernObservatory,KarlSchwarzschildStraße2,85748GarchingbeiMünchen,Germany
5 JetPropulsionLaboratory,CaliforniaInstituteofTechnology,MailStop169-221,Pasadena,CA91109,USA
Received3July2015/Accepted31May2016
ABSTRACT
High-redshiftradiogalaxiespresentsignsofbothstarformationandAGNactivity,makingthemidealcandidatestoinvestigatethe
connectionandcoevolutionofAGNandstarformationintheprogenitorsofpresent-daymassivegalaxies.Wemakeuseofasample
of11powerfulradiogalaxiesspanning1<z<4whichhavecompletecoverageoftheirspectralenergydistribution(SED)fromUV
toFIRwavelengths.UsingHerscheldata,wedisentangletherelativecontributionoftheAGNandstarformationbycombiningthe
galaxyevolutioncodePÉGASE.3withanAGNtorusmodel.Wefindthatthreecomponentsarenecessarytoreproducetheobserved
SEDs:anevolvedandmassivestellarcomponent,asubmmbrightyoungstarburst,andanAGNtorus.Wefindthatpowerfulradio
galaxiesformatveryhigh-redshift,butexperienceepisodicandimportantgrowthat1<z<4asthemassoftheassociatedstarburst
variesfrom5to50%ofthetotalmassofthesystem.Thepropertiesofstarformationdifferfromsourcetosource,indicatingno
generaltrendofthestarformationpropertiesinthemostinfraredluminoushigh-redshiftradiogalaxiesandnocorrelationwiththe
AGNbolometricluminosity.Moreover,wefindthatAGNscatteredlighthaveaverylimitedimpactonbroad-bandSEDfittingon
oursample.Finally,ouranalysisalsosuggestsawiderangeinoriginsfortheobservedstarformation,whichwepartiallyconstrain
forsomesources.
Keywords. galaxies:active–galaxies:evolution–galaxies:high-redshift–galaxies:starformation–quasars:general–
galaxies:starburst
1. Introduction mid-infrared (MIR) emission, which showed hot dust close to
the supermassive black hole (Haasetal. 2008; DeBreucketal.
High-redshift radio galaxies (HzRG) are among the most lu- 2010). In the orientation-based Unification scheme (e.g.
minous objects in the Universe and are the target of a consid- Antonucci 1993), the hot dust near central active galactic nu-
erable number of investigations aiming to understand galaxy clei (AGN) is completely obscured by material along the line
evolution(Miley&DeBreuck2008,forareview).Whilepow- of sight in type 2 AGN, and becomes gradually less obscured
erful radio emission betrays the presence of a supermassive when transitioning to type 1 AGN (e.g., Leipskietal. 2010;
black hole, near-infrared (NIR) images reveal that their host Drouartetal. 2012; Rawlingsetal. 2013). Spitzer observations
galaxies are amongst the brightest galaxies in the Universe also revealed that high-redshift radio galaxies are preferen-
at any redshift (the K-z relation, e.g., Lilly&Longair 1984; tiallyfoundindenserenvironments,consistentwithgalaxyclus-
Inskipetal. 2002). Modelling the K-band emission suggests ters in formation (e.g., Mayoetal. 2012; Galametzetal. 2012;
that radio galaxies are extremely massive galaxies, 1012M Wylezaleketal.2013a,b).
(cid:12)
(Rocca-Volmerangeetal.2004).Opticalimagesrevealthatmost
The availability of sensitive submm bolometer arrays, such
radio galaxies present a surprising alignment between the ul-
as SCUBA or LABOCA (Hollandetal. 1999; Siringoetal.
traviolet(UV)/optical emission and the radio jet axis (e.g.,
2009)allowedthefirstdetectionofdustystarformationathigh-
Chambersetal. 1987; McCarthyetal. 1987; McCarthy 1993;
redshift through the negative K-correction (Blainetal. 1999).
Pentericcietal.1999,2001).Moregenerally,thediscoverythat
Extensive surveys at z > 3 showed that radio galaxies exhibit
the mass of the central supermassive black hole is related to
high submm flux, indicating vigorous, on-going star formation
its host galaxy and dark matter halo (Magorrianetal. 1998;
(e.g., Archibaldetal. 2001; Stevensetal. 2003; Reulandetal.
Gebhardtetal. 2000; Ferrarese&Merritt 2000; Häring&Rix
2004), similar to submm galaxies (SMGs, e.g., Borysetal.
2004) implicates radio galaxies as important objects for under-
2003; Coppinetal. 2006; Weißetal. 2009) at similar redshift.
standingtheformationandevolutionofmassivegalaxies.
However, single-dish observations are limited in spatial reso-
The Spitzer Space Telescope (Werneretal. 2004) provided lution (typical 20(cid:48)(cid:48)), and only a handful of sources have been
host galaxy masses of these objects, that place them among observed with interferometric facilities (DeBreucketal. 2005;
the most massive galaxies in the Universe (Seymouretal. Ivisonetal.2012),whichmakesgeneralconclusionsontheen-
2007). Spitzer also allowed the first characterisation of their tiresampleofhigh-redshiftradiogalaxiesdifficult.
ArticlepublishedbyEDPSciences A109,page1of26
A&A593,A109(2016)
The Herschel Space Observatory (Pilbrattetal. 2010) cov- It is difficult to simultaneously characterise the UV-to-
ersthe60−700µmrange,allowingthefirstsensitiveexploration submmSEDwithAGNandgalaxyevolutioncodes:(i)themul-
oftheentireinfrared(IR)spectralenergydistribution(SED)of tiplecomponentsmeananincreasednumberoffreeparameters;
galaxies in the distant Universe. This part of the SED is of ut- (ii)anefficientfittingprocedureischallenging;(iii)asampleof
most importance because most of the energy of star-forming sourcesneedstobeconstrainedthatincludestheAGNandhost
galaxies is emitted at IR wavelengths. The IR comes from the contribution at similar levels; and (iv) homogeneous coverage
emission of dust that has absorbed a significant fraction of the inwavelengthforthesesamplesisnecessary.Forthesereasons,
UVandopticalemissionfromyoungstars(e.g.,Chary&Elbaz only a few attempts exist in the literature (although the num-
2001; Sanders&Mirabel 1996). AGN emission by dust also berisincreasingthankstotheavailabilityofnewtechniquesand
contributes mainly in the IR, which can make it difficult to de- data).Mostoftheseattemptssufferfromsimplifyingonecom-
termine the origin of the IR emission. While the submm emis- ponenttostudytheSED,eitherthehostgalaxyortheAGN,de-
sion is generally associated with cold dust and therefore with pendingonthesampleorthescientificgoalsofthestudy(e.g.,
star formation, the complete IR SED allows for a simultane- Trichasetal. 2012; Feltreetal. 2013; Rocca-Volmerangeetal.
ous characterisation of the AGN and star formation emission 2013; Banerjietal. 2014; Karouzosetal. 2014; Leipskietal.
(e.g., Tadhunteretal. 2007; Dickenetal. 2009; Mullaneyetal. 2014;Cieslaetal.2015).
2011;Feltreetal.2012;Rosarioetal.2012;Drouartetal.2014; This paper is structured as follows. We present the sample
Leipskietal. 2014). The transition regime between dust heated andthedatainSect.2.Section3describesthefittingprocedure
by the AGN and dust heated by star formation is at 40-60µm and the model used in this analysis. We present the results in
andisthereforeimportanttoobservebecauseitaffectsthetotal Sect.4anddiscusstheirimplicationsinSect.5.Throughoutthis
infraredemissionandcharacterisationofeachcomponent. paper,weassumethestandardconcordancecosmologicalmodel
As the wavelength coverage of galaxies increased over (H0 = 70kms−1Mpc−1, ΩΛ = 0.7, ΩM = 0.3, Spergeletal.
the past decade, more advanced models were developed to 2003).
predict the emission of galaxies and AGN for given evo-
lutionary scenarios and/or morphologies. The literature now
offers a large number of these models; we only mention
here the models using stellar libraries and scenarios of 2. Sampleanddata
evolution to predict the emission of the integrated emis-
sion of the galaxy over a wide wavelength range (e.g., To reliably separate the different emission components of ra-
Leitherer&Heckman 1995; Fioc&Rocca-Volmerange 1997; dio galaxies, a wide and homogeneous coverage of the SED
Maraston 1998; Bruzual&Charlot 2003). As the high-redshift is of prime importance. The HeRGÉ sample (Herschel Radio
infrared observations are becoming increasingly common, the Galaxy Evolution sample; Drouartetal. 2014) is ideally suited
models have extended their coverage to take the dust emis- to estimate the relative contributions of AGN and their host
sion into account (e.g. Leithereretal. 1999; Burgarellaetal. galaxies thanks to the existing Spitzer, Herschel and submm
2005;Maraston2005;daCunhaetal.2008;Grovesetal.2008; data. We briefly summarise here the criteria selected to build
Nolletal.2009, forarecentreview,seeConroy2013)Themain thissample.Theradiogalaxieswereselectedtocoverhomoge-
characteristicofPÉGASE.3 weuseherearetocoherentlypre- neouslytheradioluminosity-redshiftplane,applyingthecriteria
dictattenuatedmetal-dependentSEDsextendedtothedustemis- L3GHz > 1026WHz−1, where L3GHz is the total luminosity at a
sion in coherence with metal abundances. The code has been rest-frame frequency of 3GHz (Table 1; Seymouretal. 2007).
usedinRV13andwillbedetailedinFiocetal.(inprep.).Simul- To complete our SED coverage from UV to submm, we there-
taneously, numerous AGN torus models have been developed foredefinedasubsamplewiththefollowingcriteria.Eachsource
to characterise nearby AGN. These models predict the UV to (i) must have at least four Herschel detections; (ii) must have
far-infrared (FIR) SED as a function of torus geometry, solv- atleasttwobroad-bandphotometricobservationsbracketingthe
ing the radiative transfer equations. Different approaches have discontinuity at 4000 Å restframe; and (iii) the filter response
beenusedtodescribethetorusproperties,particularlytorepro- curves should avoid strong emission lines as much as possible.
duce the 9.7µm silicate feature. The two model types assume We aim with this selection to obtain more than ten data points
either a continuous dust distribution or a clumpy distribution from UV to submm wavelengths as homogeneously as possi-
(e.g., Pier&Krolik 1992; Hönigetal. 2006; Fritzetal. 2006; ble to allow for a reliable fit. In total, 11 sources remain out of
Nenkovaetal. 2008; Stalevskietal. 2012, for a recent review, 70 from the HeRGÉ sample (see Fig. 1). In addition, a polari-
seeAntonucci2012). sation measurement at restframe 1500 Å is beneficial, if avail-
In this paper, we present the first in-depth analysis of the able,tomeasurethescatteredAGNlight.Ofourelevensources,
different emission components present for a sample of power- seven have polarisation measurements (from broad-band pho-
ful radio galaxies, handling data continuously covering UV to tometry or spectroscopy, see Table 1 and the detailed discus-
submmwavelengths.Disentanglingthedifferentcomponentsof sion in Sect. 4.4). The sources span 1 < z < 4, with a median
a galaxy is particularly challenging in the UV-NIR domain be- redshift (cid:104)z(cid:105) = 2.5. The requirement of Herschel detections bi-
cause several components can contribute simultaneously to the ases this sample towards the brightest IR emitters, with LIR >
emissionandtohighlyvariableproportions.TheFIRbringsan 4×1012L (Drouartetal.2014).WhenLIRistakenintoaccount,
(cid:12)
essential piece of information as, thanks to the energy balance these radio galaxies cannot be distinguished from ULIRGs de-
between absorption and emission, constraints on each compo- tected by other samples (see Fig. 1). The two highest redshift
nent are added. We aim in this paper to take advantage of the sources,4C41.17andTNJ2007-1316areextensivelyanalysed
broadwavelengthcoverageofboththedataandmodelstoreli- in Rocca-Volmerangeetal. (2013, hereafter RV13). They had
ably disentangle the different components of a sample of high- been selected on the basis of their weak AGN contribution in
redshiftradiogalaxies.Toinvestigatethedifferentcomponents, theUVandopticaltotestthemethod,whichisfurtherimproved
we use two models, one to model the host galaxy and one to in this paper. We now extend the sample to 11 sources with a
modeltheAGN. widerrangeofAGNcontributions.
A109,page2of26
G.Drouartetal.:DisentanglingSFandAGNinluminousHzRGs
Table1.Summarisedcharacteristicsofoursample.
Name z Ref. RA Dec Morph.peculiarities Radiojetaxis logL Polarisation Ref.
3GHz
spec. [J2000] [J2000] [WHz−1] [%]
3C368 1.132 a 18:05:06.37 +11:01:33.1 twocomponents aligned 27.63 7.6±0.9(V) g
3C470 1.653 b 23:58:36.00 +44:04:45.0 twocomponents misaligned 28.24 ... ...
MRC0324-228 1.894 c 03:27:04.54 −22:39:42.1 oneclosecompanion ... 28.49 6.5+3.8(R) h
−3.3
PKS1138-262 2.156 c 11:40:48.38 −26:29:08.8 multiplecompanions aligned 28.14 ... ...
MRC0406-244 2.427 d 04:08:51.46 −24:18:16.4 multiplecomponents aligned 28.11 2.5+2.3(I) h
−2.5
MRC2104-242 2.491 e 21:06:58.28 −24:05:09.1 twocomponents misaligned 27.88 ... ...
USS0828+193 2.572 c 08:30:53.42 +19:13:15.7 multiplecomponents aligned 27.47 10.0±2.0(spec) i
4C28.58 2.891 c 23:51:59.20 +29:10:29.0 multiplecomponents aligned 27.89 ... ...
USS0943-242 2.923 c 09:45:32.73 −24:28:49.7 oneclosecompanion aligned 27.95 6.6±0.9(spec) i
4C41.17a 3.792 c 06:50:52.23 +41:30:30.1 multiplecomponents aligned 28.17 <2.4(spec) j
TNJ2007-1316a 3.840 f 20:07:53.26 −13:16:43.6 oneclosecompanion ... 27.79 ∼3(spec) k
Notes. Columns2and3indicatethespectroscopicallyconfirmedredshiftsandtheirreferences.Column6referstomorphologicalpeculiarities
at optical/NIR wavelengths: companion: source close to the object identified as the radio galaxy (few arcsecond) and components: potential
substructureswhendetectedintheimages.Column7indicatesthealignmentoftheradiojetaxiscomparedtotheoptical/NIRextensionwhen
detectable in the images. Column 8 lists the restframe 3GHz luminosity from DeBreucketal. (2010). The penultimate column indicates the
polarisationandtheband.
References. (a)Meisenheimer&Hippelein(1992);(b)Hewitt&Burbidge(1991);(c)Roettgeringetal.(1997);(d)McCarthyetal.(1996);(e)
McCarthyetal.(1990);(f)Bornancinietal.(2007);(g)diSeregoAlighierietal.(1989);(h)Buchard(2008);(i)Vernetetal.(2001);(j)Deyetal.
(1995);(k)Rocca-Volmerangeetal.(2013).
2.1. BuildingtheSEDfrombroad-bandphotometry
Weanalysedgalaxieswithmulti-wavelengthdata,thereforethe
consistency of the photometry throughout the entire spectral
rangeisessential.Specifically,thedatafromtheoptical/NIRand
FIR domain have very different spatial resolutions. The optical
datahavesub-arcsecresolution(HST),whiletheHerschelbeam
can reach ∼35(cid:48)(cid:48) (FWHM). For this reason, we report the total
fluxwhenwecompilephotometryfromtheliterature.Thetotal
fluxwascalculatedbyapplyingcorrectionstoaperturephotom-
etrybasedontheusermanualorprimaryreferenceofeachcor-
responding instrument, or by selecting a large aperture, that is, 0
>30kpc(>4(cid:48)(cid:48))or64kpc(∼8(cid:48)(cid:48)).
Basedonthefilterresponseandredshiftofthesource,strong
emission lines (i.e., Lyα, [OII], [OIII], Hα...) often fall in a
band, increasing the measured flux relative to the pure contin-
uumcontribution(byupto∼40%).Ifspectroscopicobservations
Fig. 1. Total infrared luminosity versus redshift for several samples
areavailableforthecontaminatedbands,wesubtractedtheesti-
availableintheliterature.ThelightgreendotsshowtheCOSMOSsam-
mated flux of the line based on the spectroscopy. All reported
plefromKartaltepeetal.(2010),makinguseofSpitzerdata.Theblue
fluxes in the tables are corrected and therefore correspond to squares denote the GOODS samples from Elbazetal. (2011) and the
pure continuum emission for the remainder of this paper (see darkgreendotsrepresenttheselectionfromSymeonidisetal.(2013),
Tables D.1−D.11). Only MRC 0406-244 does not have the in- both making use of Herschel data. The red circles indicate the sub-
formationonlinecontamination,andfluxesareusedasgiven. sample considered in this paper. The red diamonds show the sample
ofHzRGspresentedinDrouartetal.(2014).
2.2. Notesonindividualsources
The data set for each source is mainly rich and of high quality.
Especially,morphologicalpeculiaritiescanbeisolatedandhelp
to understand the overall picture in term of evolution in each
emission from specific components of the radio galaxy. The fi-
source. We report the notes on each source in the Appendix C.
nalgoalistodisentangletheAGNandstellaremissionthrough
Wedescribethecombinedhigh-resolutionimageandradiocon-
SED fitting. We made use of two models, the AGN model of
tours (when electronically available) in Sect. 5 along with the
Fritzetal.2006andthestellarpopulationmodelPÉGASE.3for
discussionontheevolutionarystatusofthegalaxy.
thehostgalaxy(Fiocetal.,inprep.).Wepresentthefittingpro-
cedure,consideredmodelsandtemplatelibrariesforeachcom-
ponent (AGN and galaxy) in the following sections. We also
3. Models,fittingprocedure,andtemplatelibraries
explain in detail the creation of the new templates and the as-
The broad wavelength coverage from UV to FIR allows us to sumptions made on both models to limit our total parameter
probemultipleregimesofenergyandthereforetobetterisolate space.
A109,page3of26
A&A593,A109(2016)
Table2.Fixedandfittedparametersforourlibraryofevolved,SB,andAGNtemplates.
Evolved SB AGN
Hubbletype[Sa,S0,E,E2] SSPscenario Openingangleθ=[40◦]
Fixedparameters IMF[Kroupa] IMF[Kroupa] Radialdensityprofile[α=0]
Azimuthaldensityprofile[γ=2]
Normalisation Normalisation log10fA2G0Nµm=[−3,−2,−1.7,−1,−0.7,0,0.7,1,2,3]
Fittedparameters Age,tevolved[<tUniverse,z] Age,tSB[<tevolved] Inclinationi=[40,50,60,70,80,90]
ColumndensityfactorK=[1,10] SizeY=[10,60,150]
Initialmet.Zinit=[0,0.0005,0.001,0.005,0.01] Opacityτ9.7µm=[0.1,1,10]
Notes.WecombinedtheAGNandSBlibrariestocreatehybridtemplates.Consideredrangesandvaluesinthefittingarereported.Inparticular,
theageofthestarburst,t ,mustbeyoungerthantheageoftheevolvedcomponent,t ,whichinturnmustbeyoungerthantheageofthe
SB evolved
Universeattheconsideredredshift.Thesinglestellarpopulation(SSP)iswithoutinfall,hasnogalacticwinds,andisinstantaneous.
3.1. Fittingprocedure stochasticheatingofgrains.ThecontinuoussyntheticSEDsthat
extend from the far-UV to submm wavelengths, were built by
ThefittingprocedureispresentedinRV13.Thecodeallowsfit- minimising the input parameter numbers to avoid large degen-
tingofuptotwostellarcomponentssimultaneouslybasedonχ2 eracies in the solutions. We characterised each star formation
minimisationbyautomaticallyexploringalibraryoftemplates. scenariowithonlyfourfreeparameters:starformationrate,ini-
ToadaptthefittingprocedurefromPÉGASE.2tothenewcode tialmassfunction(IMF),inflowsandoutflows.WerefertoFioc
PÉGASE.3,wemadethefollowingmodifications:(i)weadded et al. (in prep.) for a more detailed presentation of the code
anestimateoftheuncertaintiesonthefittedquantities;(ii)modi- PÉGASE.3 and its detailed documentation, which is available
fiedthegraphicaldisplayingproceduretoincludetheAGNcom- onthePÉGASEwebsite1.Inthepresentversion,usedhereafter
ponent; (iii) extended the filter database; and (iv) updated the asinRV13,thestarformationparametersbyspectraltypesare
compatibilitywiththenewoutputsofPÉGASE.3. taken from LeBorgne&Rocca-Volmerange (2002), fitting the
Adding the AGN component was challenging in the con- HubblesequencegalaxypropertiesinthelocalUniverse.These
textofthetwo-componentfittingprocedurewithoutmodifyinga parameters are set to reproduce observed local galaxies from
largepartofthecoreprocedure.Wesolvedthisdifficultybycre- SDSS (Tsalmantzaetal. 2009, 2012), including evolution with
atinghybridtemplatescomposedofthesumoftheevolvingstar- aformationatzform =10.Wefocusedourlibraryonfourgalaxy
burst (SB) and AGN components (see Sect. 3.2). The principal types, E, E2, S0, and Sa to test our sensitivity to the galaxy
problem in adding new components stems from the degenera- typewhenmodellingtheradiogalaxyhost.Despitetheirnames,
cies induced by the increase in the number of free parameters. thesetemplatesrefertodifferentstarformationhistoriesandnot
We addressed this specific problem by combining a broad and to specific morphologies, see LeBorgne&Rocca-Volmerange
homogeneousSEDcoverage(with12−19datapoints)andcare- (2002).
fullychoosingmodelsandparameters:theself-consistenttreat- Thestarburst(SB)templatesaredefineddifferentlythanthe
mentoftheopticalandIRemissionoftheAGNandPÉGASE.3 evolved, main component. They are single stellar populations
models(calculatedwithradiativetransfercodes),enabledusto (SSPs), assuming a short formation in one step of time, that
leave a significant number of parameters free with only lim- is, 1 Myr (δ function). SSPs do not take into account any in-
ited parameter degeneracies (see Sect. 5 for a discussion of the fall or galactic winds. We discuss the effect of this 1 Myr for-
uncertainties). mation assumption in detail in Sect. 4.2. We adopted a Kroupa
IMF(Kroupa2001)andarangeofinitialmetallicitiesofthegas
Our approach uses grid-model fitting and requires brute-
(Z ). RV13 showed that the column density plays a large role
forcecalculationthroughtheparameterspace.Wethereforeset init
intheobservedSED.
some parameters to physically justified values and took repre-
Weexploredthesamequantitiesfortheevolvedpopulation
sentativemodelstoensurethatthecoderemainedexecutableon
with the SB component, that is, the age and mass of the stel-
a desktop machine in a reasonable computing time. More ex-
lar populations, but we also added the initial metallicity and
tensiveexplorationoftheparameterspacewouldrequireanew
the column density factor K (see RV13, Sect. 3). Table 2 re-
fitting algorithm using a Bayesian approach and Monte Carlo
portstherangeofvaluesconsideredinourlibrary.Wehereex-
samplingoftheparameterspace,butthisisbeyondthescopeof
plore the age and mass of the starburst to determine whether
this paper because it requires a thourough redesign of the core
the ten times higher column density preferentially found in ra-
procedure.Althoughnotcomplete,thecoverageoftheparame-
ter space is still satisfying, with >107 templates tested for each diogalaxiesbyRV13isconfirmedforalargersample.Wealso
note that the fitting procedure requires the evolved component
source.
to be younger than the age of the Universe at a given redshift
(t <t )andtheyoungcomponenttobeyoungerthan
evolved Universe,z
3.1.1. PÉGASEmodelandlibrary theevolvedcomponent(tSB <tevolved;seealsoTable2).
We used the PÉGASE.3 code to generate a library of galaxies
3.1.2. AGNmodelandlibrary
and starbursts.The newcode PÉGASE.3is acoherent spectro-
chemical evolution model predicting simultaneously the metal- The AGN component was incorporated using the AGN model
licityenrichment,thecorrespondingSED,andtheattenuationor fromFritzetal.(2006).Tobrieflydescribethismodel:acentral,
emission by dust grain models calculated with radiative trans-
ferMonteCarlosimulations,takingintoaccountscatteringand 1 www2.iap.fr/pegase
A109,page4of26
G.Drouartetal.:DisentanglingSFandAGNinluminousHzRGs
point-like source emits an SED composed of a sum of power Table 3. Best χ2 for one (evolved only), two (evolved and starburst),
laws.Thetorusisdefinedinsphericalcoordinatesbyitsgeom- andthree(evolved,starburst,andAGN)components.
etry (size and opening angle), dust profile density (radial and
azimuthal),opacityat9.7µmandinclinationwithrespecttothe Name 1comp. 2comp. 3comp.
observer. The torus is divided into cells assuming an axisym- #freeparameters 2 6 10
metric geometry and the radiative transfer equations are solved
3C368 3615[17] 101[13] 73[9]
usingtheΛ-iterationtechnique(seeFritzetal.2006,foracom-
3C470 4000[12] 71[8] 32[4]
pletedescriptionofthemodel).Feltreetal.(2012)showedthat
MRC0324-228 2152[11] 57[7] 25[3]
theFritzetal.(2006)modelisdirectlycomparabletotheclumpy
PKS1138-262 14896[14] 834[10] 79[6]
modelfromNenkovaetal.(2008)foracertainsetofparameters
MRC0406-244 1208[14] 65[10] 19[6]
(onlypartiallycoveredinourcurrentlibrary).
MRC2104-242 167[10] 20[6] 9[2]
We adopted the Fritzetal. (2006) model for the present USS0828+193 5235[12] 619[8] 36[4]
study, leaving the comparison with clumpy models for further
4C28.58 247[13] 87[9] 29[5]
publications. Moreover, a precise characterisation of the torus
USS0943-242 250[13] 16[9] 5[5]
propertiesisbeyondthescopeofthispaper,becauseitrequires
4C41.17 491[16] 39[12] 11[8]
spectroscopyaroundthesilicatefeatureat9.7µmrestframe.We
TNJ2007-1316 148[13] 73[9] 9[5]
focusonthelargestimpactparameters,thatis,thesize,theopac-
ity, and the inclination of the torus. The size, Y, is the ratio Notes. We also indicate the number of degrees of freedom (d.o.f.) in
Rmax/Rmin, where Rmin is set as the sublimation radius of the brackets,d.o.f.=N−p−1whereNisthenumberofdatapointsandp
dust (which, in turn depends on the luminosity of the central isthenumberoffreeparameters.
source).Theopacity,τ ,correspondstotheintegratedopac-
9.7µm
ityat9.7 µmalongtheequatorialaxis(i = 90◦).Thedustden-
brute-force calculation through our parameter space, therefore
sity(radialandazimuthal)ofthetorusiscalculatedwithrespect
each fit possesses a χ2 value, that indicates the goodness of fit.
tothesize,opacityandprofilechosenbytheuser.Thesevalues
We can therefore consider the χ2 distribution to gain insights
were set to their default values here, radially constant and az-
into the confidence interval of the adopted free parameters. We
imuthallydecreasingasthesquareofthealtitude(theyalsohave
a limited effect on the SED). The inclination, i, is the orienta- used the description from Numerical recipes (Pressetal. 1992,
Chap. 15) to estimate the uncertainties on the parameters, us-
tion of the torus with respect to the observer. An edge-on view
corresponds to i = 90◦, and i = 0◦, corresponds to a face-on ing the complementary incomplete gamma function to calcu-
view. The opening angle, θ, was set to θ = 40◦, as suggested late the corresponding ∆χ2 corresponding to a 68.3% confi-
dence interval. For our ten free parameters, this corresponds to
by previous studies (e.g., Barthel 1989; Drouartetal. 2012). In
practice, all inclinations i < 50◦present a line of sight free of ∆χ2 = 11.5. We defined the confidence interval as χ2min +∆χ2
(seeFigs.D.1−D.10andTable4).
dustandthereforesimilarSEDs.Tosavecomputingtimebutin-
cludethepossibilityofatype1AGN,weonlyusedi = 40◦ as Theuncertaintiesonstellarmassesandluminositieswerees-
timatedwitha differenttechniquebecausethey aredirectlyex-
atype1AGNtemplate.Table2presentsalltheassumedvalues
pressedfromthenormalisationofthedifferentcomponents.We
for the fixed and fitted parameters. The objective of this selec-
considered the extreme values of each mass within the allowed
tion is to explore typical torus configurations and evaluate how
68%confidenceintervalofallotherparametersanddefinedthis
ourmethodhandlestheadditionofthisthird(AGN)component.
asthelowerandupperuncertainties.FortheAGNluminositywe
addedtheuncertaintyfromtheAGNfraction(f20µm)inquadra-
3.2. HybridAGN-SBcomponent tureandthenormalisationofthehybridcompoAnGenNt.Finally,we
add a special caution of interpreting these uncertainties. They
Wecreatedhybridtemplates,correspondingtothesumoftheSB
onlyrepresentthestatisticaluncertaintiesontheparametersin-
templatesproducedbyPÉGASE.3andanAGNtorus.Foreach
duced from the noise in the observations and do not take into
evolvingSBtemplate,wenormalisedanAGNtorus(foragiven
accountpotentialsystematicsassociatedwiththemodelchoices
inclinationi,sizeY,opacityτ )toafractionofthetotalflux
9.7µm and assumptions (see Appendix A for a more extensive discus-
at20µmfollowingtheequation:
sionofuncertaintiesandparametercalculation).
F (t)= F (t)+F (i,Y,τ )× f20µm, (1)
hybrid SB AGN 9.7µm AGN
where F isthesumoftheAGNandtheSBtemplates, F
hybrid SB 4. Results
istheSBtemplate, F istheAGNtemplateand f20µm isthe
AGN AGN WeperformedtheSEDfittingdescribedintheprevioussections
relativefractionoftheAGNfluxcomparedtotheSBtemplateat
20µm2(seeTable2).Weperformedastep-wiseχ2-minimisation on the 11 sources in our sample, taking one (evolved compo-
nentonly),two(evolvedandstarburst),andthree(evolved,star-
thatpermittingdeterminingofintermediatesolutionsandcheck-
ingfordegeneracies.Intotal,>107templateswerefittedforeach burst, and AGN) components into account. Table 3 shows the
best χ2 depending on the number of components. The addition
source.
of extra components clearly dramatically increases the quality
of the fit. This improvement is particularly strong from one to
3.3. Uncertaintiesofparameters,68%confidenceintervals two components and is still significant from two to three com-
ponents(ameliorationbyafactorsuperiortotwo).Wetherefore
Calculating uncertainties on the parameters can be challenging
only consider the three-component fit in the remainder of this
for a multi-dimensional analysis. However, as we performed
paper. Table 4 summarises the results of the fitting, including
2 Weevaluated f20µmwithanidealisedfiltercentredon20µmof4µm theirassociated(68%)uncertainties.Figure2showsanexample
bandwidthtominiAmGNiseeffectsfromsharpgradientsand/oremissionand of the best-fit model for 4C 41.17, while the results for the re-
absorptionlines(i.e.PAHs). mainingthesamplearepresentedinFigs.D.1−D.10.Wepresent
A109,page5of26
A&A593,A109(2016)
4C 41.17, z=3.792
Fig. 2. Best fit for 4C 41.17 (z = 3.792). The orange, blue and green dashed lines symbolise the evolved, starburst, and AGN components,
respectively.Thesumofthecomponentsisthedarkline,theblackdiamondscorrespondtobroad-bandphotometry,theverticallinestothe1σ
uncertainties and horizontal lines to the FWHM of the filters. The downward triangles represent the 3σ upper limits. The insets show the χ2
distributionforsevenofthefreeparametersofthefit.Fromtoplefttobottomright:ageoftheevolvedcomponent(orangelineintheSED),age
ofthestarburst(bluelineintheSED),inclinationoftheAGNtorus(greenlineintheSED),fractionoftheAGNat20µm,equatorialopacityof
thetorus,sizeofthetorusandinitialmetallicityofthestarburst.Black,blueandredindicatethethreedifferentapproachesweusedtoestimate
the polarisation effect, without, with the lowest and the highest contamination, respectively. We note that in this case the lowest polarisation
contribution is zero, hence the black and blue lines are superposed. The horizontal coloured lines correspond to the 68% confidence interval
describedinSect.3.3.
here the overall trends on the different components for the en- mass distribution seen in large field galaxy samples (e.g.,
tire sample. Overall, the fittings are satisfying for all sources Marchesinietal. 2009; Ilbertetal. 2013). Interestingly, these
even if some discrepancies are still observed (see Sect. 5.1). It masses are also comparable to SMGs at similar redshift
is important to note that three components are necessary to re- (e.g., Borysetal. 2003). This suggests that powerful ra-
producetheSEDofthepowerfulradiogalaxiesinthissample, dio galaxies are drawn from the same sample of massive
and each appears to be dominating at a particular wavelength galaxies, formed at high-redshift (z > 6) on a relatively
range. short timescale. Similarly as in RV13, the evolved compo-
nent clearly dominates the Spitzer photometry, consistent with
the shift of the 1µm peak of the most evolved early-type
4.1. Evolvedcomponent
galaxies.
We call the oldest stellar component the evolved component TherightpanelofFig.3confirmsthisresultbecauseallred
(>0.5Gyr), which can be defined as the radio galaxy host. points are clustered along the black lines (plain and dashed),
Figure 3 shows the mass of the evolved stellar population as consistentwithanearlyformationepoch.Wealsonotethatthe
a function of redshift (red points). The host galaxy appears fourHubbletypesusedinouranalysisprovidesimilarχ2values
massive at z = 4, and this evolved massive component is within the 68% confidence ranges. We therefore conclude that
present in every source throughout our redshift range (M > low-resolution broad-band photometry alone cannot constrain
stel
1011M , except for MRC 0324-228 with 1010.7M ). Previ- theexactstarformationhistoryofthegalaxies,butstillfavoursa
(cid:12) (cid:12)
ous studies have shown that the most powerful radio galax- formationatearlytimesintheUniverse.Weemphasisethatthe
ies present exceptionally high masses (e.g. DeBreucketal. ages determined by our method are well constrained, as illus-
2002;Rocca-Volmerangeetal.2004;Seymouretal.2007).This tratedinFigs.D.1−D.10wherethebestχ2 solutionsarelocated
places powerful radio galaxies at the top of the galaxy indeepminima.
A109,page6of26
G.Drouartetal.:DisentanglingSFandAGNinluminousHzRGs
Fig.3. Twostellarpopulations,youngandevolved,shownasbluecirclesandreddiamonds,respectively.Theverticallinesindicatethe68%
confidenceintervalsfromthefit.Whenpolarisationdataareavailable,wereportthefitwithlowestandhighestcontributionfromscatteredAGN
light as the downward and upward empty triangles respectively. For clarity, we do not add the 68% confidence intervals for the polarisation-
subtractedfits.Left:massesofthetwostellarpopulationsversusredshift.Right:agesofthetwostellarpopulationswheretheblacklineshows
theoldestuniverseageasafunctionofredshiftassumingthestandardcosmologicalconcordancemodel(Sect.1).Thedashedlinerepresentsthe
ageofagalaxyforz =10.Wenotethatthe68%confidenceintervalcanbesmallerthanpointsizeandpolarisation-subtractedfitsaremainly
form
containedinthe68%confidenceintervalswhenscatteredAGNlightisignored.
4.2. Intenseandmassivestarburst andlocalisedminima.Moderatelyhigh-resolutionspectroscopic
observationsarenecessarytofurtherexplorethispartofthepa-
Figure 3 and Table 4 show two remarkable features of the
rameter space because the emission lines are the result of the
youngcomponentrequiredtoreproducethestrongHerscheland
ISMphotoionisationfromboththeyoungstarsandtheAGN.
submmemissioninoursample.Firstly,thestarburstisextremely
We make a final note about the star formation rate related
massive,M >1010M .Ineachcase,itrepresentsasignificant
stel (cid:12) tothestarburstcomponent.Bydefinition(inthemodel),allthe
fractionofthemassofthesystem(5%to50%).Thesehigh-mass
star formation is taking place in a single time step (we discuss
fractions indicate that even if the bulk of the stars has already
thedurationofSBinSect.5).Whenthefitconvergestoanage
been formed at high-redshift (Sect. 4.1), the radio galaxies in
>1Myr, the related star formation is therefore null. Moreover,
our sample still experience vigorous star formation3. Secondly,
theabsenceofinfraredimagingatarcsecresolutionandtheun-
Fig. 3 also shows that this starburst spans a wide range in age,
certainties about the duration of the burst make deriving a star
5Myr < tSB < 100Myr, indicating that star formation is on- formationratefromourfittingdifficult.Wethereforeusedsimple
goingorintheaftermathofaviolentevent.
physicalassumptionsandempiricalrelationstoestimatestarfor-
It is also interesting to note that all but one (MRC 2104-
mationratesfromthestarburstcomponent(seeSect.5.2).
242)ofthegalaxieshaveacolumndensityfactor K = 10.This
indicatesthatstarburstsassociatedwithradiogalaxieshaveanon
average ten times higher column density than the Milky Way4. 4.3. BolometricallyluminousAGNwithanopaquetorus
This indicates that high-redshift radio galaxies, or at least the
most IR-luminous radio galaxies, favour higher covering dust OurSEDfittingallowsustoderivesomeoftheAGNandtorus
fractions,indicatingahighobscuration,similartoSMGsinthe properties of HzRGs, such as the size, opacity and inclination.
sameredshiftrange(e.g.,Chenetal.2015). WithEq.(1)wecalculatedthecorrespondingintrinsicluminos-
ity of the AGN (Table 4, last column). All our AGN are lumi-
About half of the solutions favour a high initial metallic-
ity (see Table 4). This could be expected for a starburst form- nous,withLAinGtN >∼1012 L(cid:12),mostwithLAinGtN ≈1013−14 L(cid:12).Un-
fortunately,thesmallspanofpropertiesofoursample(interms
ing from IGM gas, which is expected to be metal-rich at high-
ofluminosityormass),usuallyonlycoveringoneorderofmag-
redshift(e.g.,Pierietal.2014).Interestingly,theotherhalftends
nitude,doesnotallowustoidentifyanycorrelationbetweenthe
tohavealow(ornull)initialmetallicity,indicatingamorepris-
AGN and host properties. Interestingly, these values are sim-
tine gas source. The evolution of the starburst implies ISM en-
ilar to optically luminous quasars observed at similar redshift
richmentwhenSNexplodeandthereforetheobservedmetallic-
(e.g.,Kollmeieretal.2006).Thisisconsistentwithradiogalax-
ity will be higher than the one reported here (initial metallicity
ies experiencing a strong radiation-dominated accretion phase
ofthegas).Giventhatweusedonlybroad-bandphotometry,the
withcopiousaccretionontothecentralsupermassiveblackhole.
derivedinitialmetallicitiesshouldbetreatedwithagrainofcau-
InFigs.2andD.1−D.10,weshowthattherelativecontribution
tion, even though they do provide some clues about the overall
at20µmandthereforetheAGNluminosityiswellconstrained.
metallicity of the initial gas of the starburst (null, sub-solar, or
super-solar).ThebroadtrendiswellshowninFigs.D.1−D.10, Except for 3C368, the tori in our sample appear opaque,
wherethebest-fitsolutionsareshownnottobelocatedindeep with τ9.7µm ≥ 1.0. This opacity corresponds to AV > 18 using
theDraine(2003)conversion.Theobscurationisalsorelatedto
3 We note that our sample selection requires bright IR emission (≥4 theinclination.Theinclinationrepresentstheamountofduston
Herschel detections), implying the sample is likely biased towards thelineofsight.Mostofourbestfitsconvergetosolutionswith
strongstar-formingsystems. i>40◦,indicatingType2AGN(edge-onview),asexpectedfor
4 N ∼6.8×1021atomscm−2isthestandardgalacticcolumndensity radiogalaxies.Theoneexceptionis3C368,thesourcewiththe
HI
inPÉGASE.3. lowest redshift in our sample, with i = 40◦. The best solutions
A109,page7of26
A&A593,A109(2016)
TN3aCbalme36e48.SummHartuyySbpaobeflethelofi1gt1Mt1i.n03a[g+−sEM00srve..10(cid:12)so]ullvt4se.5d[0MAc0ogy+−me52r00]p00o0ne0n.0Zt1∗12 lo1g1L1.60to[+−tL01..(cid:12)23] d light [%]10800 M4RCC T 40N 14J.021607-0274-4131M6RUCS S0 3302C9 443-362-822842USS 0828+193
3MCR4C700324 ES2a 1110..77+−+000...113 31500000+−+012000000 00..00029624 1121..08+−+000...370 arise 60
PKS1138 S0 12.3−+10..21 2000−+450000 0.0142 12.7−+00..02 pol
−0.0 −200 −1.0 n
MMRRCC02410064 EE22 1111..10+−+000...143 11220000++−14030000 00..00225599 1111..87+−+000...330 n of u 40
USS0828 E2 11.50+.00.0 140−02+000 0.0257 12.20+.00.3 tio
4C2858 S0 11.60+.00.1 2000+−00 0.0142 12.0−+00..02 rac 20
USS0943 E 11.3−+00..01 500+−120000 0.0114 12.1−+10..12 F θ=45° θ=90°
−0.4 −100 −0.2
4TCNJ41210707 SE0 1111..59+−+000...211 1700000+−[phone]..00015478 1122..34+−+000...262 00 5 10 15 20 25 30
−0.1 −100 −0.7 Observed Polarisation [%]
SBcomponent∗ Fig. 4. Highest and lowest polarisation diagram at 1500 Å. The grey
shaded area represents the allowed fraction of unpolarised light de-
Name Mass Age Zinit K Ltot pendingontheinclination(seetext).Polarisationmeasurementsofour
log10[M(cid:12)] [Myr] log10[L(cid:12)] sourcesarereportedasverticallines withouttheuncertaintiesforvis-
3C368 10.1+0.1 7+1 0.01 10 12.7+0.3 ibility(i.e.theuncertaintiesincreasetheallowedrangeofunpolarised
−0.2 −2 −0.4 lightaccordingly).
3C470 10.0+0.4 5+3 0.01 10 12.9+0.1
−0.2 −1 −0.5
MRC0324 10.8+−00..32 18+−74 0.01 10 12.9+−00..15 vary from Y = 10 to Y = 150 for the size. As we calculated
PKS1138 10.7+0.3 20+15 0.0005 10 13.2+0.1 theintrinsicluminosity,wecanestimatethephysicalsizeofthe
−0.3 −6 −0.4
MRC0406 11.3+0.4 45+45 0.0 10 13.1+0.1 torus (see Eq. (1) of Fritzetal. 2006). Taking the median of
−0.2 −15 −0.4
MUSRSC02812084 1100..63+−+000...873 906+−[phone]..001 110 1123..92+−+000...711 oanudrsRammapxle∼, L4Ai1nGtpNc,∼∼22.456×pc1,01a3nLd(cid:12)∼t6ra1n6splactefsorinYto=Rm1in0,∼604,.1anpdc
4C2858 10.8−+00..13 30−+130 0.005 10 13.3−+00..60 150, respectively. We note the difficulties in constraining the
−0.5 −12 −0.1 toruspropertiesareexpectedbecauseevenwithdedicatedsam-
USS0943 11.4+0.5 90+110 0.01 10 13.0+0.3
−0.8 −60 −0.1 ples in the nearby Universe, some parameters are only loosely
4C4117 11.4+−00..02 30+−55 0.001 10 13.4+−00..35 constrained (e.g., AsensioRamos&RamosAlmeida 2009). In-
TNJ2007 11.0+0.4 45+95 0.001 10 13.0+0.2 terestingly, even after chosing the most relevant parameters for
−1.0 −27 −0.1
the AGN torus, the properties of the torus vary from source to
AGNcomponent
source, with no clear trend on the entire sample (see Figs. 2
Name Y τ i Ltot andD.1−D.10).
9.7µm
[degrees] log [L ]
10 (cid:12) 4.4. ModerateeffectofAGN-scatteredlightinbroad-band
3C368 150+0 0.1+0.0 40+0 11.7+0.3 SEDfitting
−0 −0.0 −0 −0.4
3C470 60+90 10.0+0.0 50+10 13.2+0.5
−0 −0.0 −0 −0.4 The AGN emission is partly absorbed by the dusty torus or es-
MRC0324 60−+900 1.0+−00..09 90+−010 12.3+−00..42 capes along the open part of the torus. Escaping photons can
PKS1138 10+0 1.0+0.0 50+10 13.5+0.5 bescatteredbydustpresentintheionisationconetowardtheob-
−0 −0.0 −0 −0.3
MRC0406 10+0 10.0+0.0 50+20 13.3+0.5 server,andtherebycontaminatetheoverallUVandopticallight.
MRC2104 150−+00 10.0−+00..00 80−+010 13.7−+00..49 Thiscontributionespeciallyaffectstype2AGN(seenedge-on),
USS0828 10+−090 1.0+−00.0.0 70−+3100 13.5−+00..15 andcanbestrongintheUVandopticalrest-frame(upto100%–
4C2858 60−+00 10.0−+00.0.0 90−+00 14.3−+00..67 diSeregoAlighierietal. 1996; Cimattietal. 1998; Vernetetal.
−0 −0.0 −0 −0.1 2001). It is important to quantify this effect in our SED fitting.
USS0943 150+0 10.0+0.0 70+10 13.4+0.6
−140 −0.0 −20 −0.0 WeusedthetechniquefromVernetetal.(2001)toestimatethe
4C4117 10−+1040 10.0+−00..00 80+−1100 13.9+−00..35 range of possible contributions of the optical AGN scattered
TNJ2007 10+0 1.0+0.0 50+20 13.0+0.8 light, as illustrated in Fig. 4. We stress that this range relies
−0 −0.0 −0 −0.1
on several assumptions regarding the geometry and inclination
Notes. Theχ2 isgiveninthelastcolumnofTable3.Thetableisdi- oftheobscuringdustdistribution,whichareconsistentwiththe
vided in three parts corresponding to each fitted component. For the adoptedtorusmodels(θ=40◦,inclinationi).
evolvedcomponent:Z referstotheaveragemetallicitylockinstarsfor
∗ One fundamental property assumed here is the grey be-
thegivenage.FortheSBcomponent:Z referstotheinitialmetallicity
init haviour of polarisation, that is, the fact that polarisation from
ofthegas,K isthefactorrelatingtothecolumndensity.FortheAGN
dust scattering is approximately independent of wavelength
component:Yreferstothesizeratiobetweentheouterandinnerradius
(Vernetetal. 2001). This allows us to estimate the range of
ofthetorus,τ theopacityat9.7µm,anditotheinclinationwith
9.7µm
respecttotheobserver.Asymmetricuncertaintiesofeachparameterare AGN contribution in the UV and optical domain from a sin-
providedasexplainedinSect.3.3.Wenotethatzerovaluesoftheun- gle polarisation measurement at 1500Å. When a polarisation
certaintiesareduetothelackofcoverageintheparameterspaceand measurementwasavailable,wethereforeconsideredthefollow-
arecontainedinthe68%uncertainties,refertoFigs.D.1−D.10. ingthreecases:(i)withoutpolarisation;(ii)withlowestpossible
A109,page8of26
G.Drouartetal.:DisentanglingSFandAGNinluminousHzRGs
AGN contamination; and (iii) with the highest possible AGN
contamination. We applied the following correction to the UV
USS 0828+193, z=2.572
andopticalphotometrybeforethefitting:wescaledthe1500Å
contributionfromFig.4andsubtractedthecorrespondingfrac-
tion from our broad-band photometry5 with a typical UV and
opticalcontinuumquasarspectrumfromCristiani&Vio(1990).
The latter represents a typical UV and optical SED of a type
1 AGN (i.e. without obscuration). We recall that these are the
boundary cases and that the true contamination is located be-
tweenthesetwoextremes.
The most polarised source of our sample, USS 0828+193,
with P = 10.0±2.0% is predicted to have a contribution from
unpolarised light of between 0% and ∼65%. The remaining
AGNcontributionisthusbetween∼35and100%(illustratedin
Fig.5).Subtractingthelowest(35%)AGNcontributionhasal-
mostnoeffectonthefit(χ2 = 35.7orχ2 = 37.1,respectively).
This suggests either that (i) the addition of this component is
notparticularlyjustified,orthat(ii)thetruecontaminationisnot USS 0828+193, z=2.572
close to the lowest 35% contribution of the AGN. As the high-
estpossibleAGNcontributionforthissourceis100%intheUV,
thiswouldleavenoroomforanystellarcontribution.Suchanex-
ceptionalcase(onlyoneinoursample)cannotbeproperlytaken
intoaccountinourfittingprocedure.Wethereforeconsiderthe
lowest possible AGN contribution as an upper limit on the ef-
fect on USS 0828+193. For this lowest contribution, the effect
onthefittingislimitedtoafactorof∼2forthemassand∼2for
the age of the stellar components (see Fig. 3). For the remain-
ing six sources with polarisation measurements, the difference
inducedbypolarisationistypicallycontainedinour68%confi-
denceintervals(seeemptysymbolsinFig.3).Theeffectonthe
otherparametersispresentedintheinsetsintheSEDsinFigs.2
and D.1−D.10 (blue and red points for the lowest and highest Fig.5.SEDsofUSS0828+193showingtheimpactofthepolarisation
contamination,respectively). onourfitting.ThecolourcodingisthesamethanFig.2.Thetoppanelis
We conclude that the effect of polarisation for our broad- withoutpolarisation,whilethebottompanelistheminimalsubtraction
bandSEDfittingisweakerthanourglobaluncertaintiesoneach case.Themaximalcaseofa100%subtractionisnotreported(seetext).
parameter. Polarisation therefore has a negligible effect on our
main conclusion, and we ignore this in the remainder of this
paper.
themorphologyofoursourcesmakinguseofacombinationof
multi-wavelengthdata.
5. Discussion
5.1. Modellinguncertainties,limitations,
Our fitting procedure enables us to carefully disentangle the
andmulti-wavelengthdataset
threemainspectralcomponentsofoursources:anevolvedstel-
lar component, an SB component and an AGN. We focus on
The self-consistency in our models is a key point of our mod-
the new information provided by the Herschel data in the far-
elling:forinstance,thecomputedFIRemissionisdependenton
infrared(FIR),allowingustodisentangletheSBfromtheAGN
thedustpropertiesdefinedbytheevolutionaryscenarioandpro-
contribution. We find that the two stellar components are mas-
duced from the stellar population evolution (ISM enrichment,
sive(>1010M )andthatboththeAGNandtheSBareverylu-
(cid:12) see Sect. 3.1.1). We discuss here the limitations and degenera-
minous (>1010L ). The properties of the evolved stellar pop-
(cid:12) cieswithinourlibrariesandfromourobservationsonthefitting
ulation confirm a very high formation redshift of the galaxy
procedure. We refer to Sect. 3 and Appendix A for the discus-
host, leading to local massive early-type galaxies as previously
sion of the model assumptions and systematics introduced by
proposedinliterature(e.g.Lilly&Longair1984).Interestingly,
ourchoiceofcodeandlibraries.
our sources have similar properties as quasars and SMGs in
ThePÉGASE.3modelonlyexploresalimitednumberofstar
thesameredshiftrange(e.g.,Alexanderetal.2005;Wangetal.
formationhistories.Fortheevolvedcomponent,thestarforma-
2011; Leipskietal. 2014, see Sect. 4). We discuss the limita-
tionhistoryissetbytheHubbletype,althoughthedatapresented
tionsofourmodellinginafirststepandthenexploretheprop-
hereareunabletopreferasingletypewithintheconfidencein-
erties of star formation making use of physical arguments and
tervals (Sect. 3.3). However, the effect on the age and mass of
empiricallawsfromSEDfitting.Finally,weclassifyanddiscuss
stellar populations is limited, within a factor of 2 (see RV13
for better details). For the starburst, the star formation is con-
5 Weareinterestedonlyinthecontinuumemission,thereforewere-
sidered instantaneous (in one time step). Forming ∼1010M in
moved the strong emission lines from the templates by linearly inter- (cid:12)
polatingthecontinuumbelowtheline.Inaddition,linepolarisationis suchashorttimeappearshighlyunlikely.Nevertheless,wenote
typicallyclosetonullbecausethephysicalprocessesinvolvedinline thatonlyashortburstcanreproducethestrongsubmmemission
emissionaredifferentfromthecontinuum. in our sample. We stress that star formation is not necessarily
A109,page9of26
A&A593,A109(2016)
spatially concentrated, but is most probably temporally con- V = σ/0.6(e.g.,Rixetal.1997),wecalculatethedynamical
circ
nected. Therefore, the SB component may be interpreted as a timeast =2r /σ .
dyn SB SB
collectionofsmallerclumps,formedinashortperiod.Inaddi- The projected star formation per surface unit is a power-
tion,RV13havediscussedtheeffectofthisδfunction,showing ful proxy to estimate the physical conditions of the starburst
thattheagefromthefittingismeasuredwithinafactorof∼2for (e.g., Lehnert&Heckman 1996). Based on the physical sizes
constantstarformationoveralongerperiodoftime. and timescales derived above, we calculate the star formation
TheFritzetal.(2006)AGNmodelisoneofseveralmodels density, Σ , with the following formula Σ = SFR/(2πr2 )
SFR SFR SB
availableintheliterature.Directcomparisonsbetweendifferent withSFR=M /t .
SB dyn
modelsaredifficultbecauseofthevariationinthenumberoffree Itisinterestingtonotethattheoverallpropertiesfromthese
parametersandbecauseoftheassumptionsonthetorusproper- empirical calculations indicate (i) a high efficiency of forming
ties itself (smooth, clumpy, bi-phased). More investigations are starswithansSFR>20Gyr−1inallcases,consistentwithmea-
reserved for further papers, taking different AGN models into surements from Drouartetal. (2014); (ii) a dense star-forming
account(e.g.Podigachoskietal.,inprep.).Wethereforedonot region Σ > 5M yr−1 kpc−2, similar to the local starburst
SFR (cid:12)
furtherdiscusstheimplicationsofthesmoothtorusmodelonour (e.g.Bigieletal.2008)andlargelyexceedingthelimitforstar-
analysis, but we warn that the results provided here are depen- burst driven winds (e.g. Lehnert&Heckman 1996); and (iii)
dentonthischoice.Wenotethatpartofourbestfithasdifficul- half (5/11 sources) of our sample presents an SB mass of the
ties to reproduce the 30µm restframe data (e.g. PKS 1138-262 same order as the host component (Table 5), strongly suggest-
andUSS0828+193).Themostlikelyexplanationisthelackof ing merging activity. These estimates have important implica-
exploration in the parameter space, especially on the size pa- tions in terms of star formation properties because it suggests
rameter.Alargertorus,havingdustatlargerradii,wouldbetter that the star formation is very efficient in very confined re-
reproducethiscoldemission. gions. However, the large projected radius (>1kpc), suggests
Finally,observationsthemselvescanbeaffectedbysystem- thatthestarformationarisesinseverallocationswithinthesame
atics. This is particularly true for our set of multi-wavelength galaxy.
observationsthatwereobtainedwitharangeofinstrumentsand
telescopes, each presenting their own calibration uncertainties,
5.3. Localisationofthestarformationfromtheimages
especiallyatmoderatesignal-to-noiseratio.TheFIRandsubmm
domain are particularly challenging observations and the pho- ThelimitedspatialresolutionofHerschelmakesitdifficulttoan-
tometry estimation method can have a significant effect on the swerthekeyquestionsofthelocationofstarformationandofthe
final measurement, especially in the <5σ range (by a factor of process that leads to the observed star formation. Complemen-
∼20%; Popessoetal. 2012). A good illustration is the 160µm tarytothespectralinformation,spatialinformationcanbeused
photometryin4C28.58(Fig.D.8),wherethe160µmfluxmight to understand the localisation of star formation and the overall
beexpectedtobehigher.Similarly,sky-subtractionstrategy(on- evolutionarystatusoftheradiogalaxiesinoursample.Wemade
offversusraster)mayaffectthe850µmphotometry,possiblyex- use of the high-resolution HST and radio imaging, along with
plainingthedifferencesinthecolddustpartoftheSED.These the moderate resolution of the NIR data, to investigate the rel-
uncertainties are mitigated by the good sampling of our SEDs, ative location of the emission at different wavelengths6. We vi-
however. suallyclassifiedoursourcesintothreenon-exclusivecategories
as follows (and report their likelihood in Table 5 as none, low,
medium,andhigh):
5.2. StarformationpropertiesfromtheSEDfitting
Intheprevioussectionsweestimatedseveralcrucialproperties 1. “exsitustarformation”,wherestarformationoccursoutside
relatedtothestarformationinoursample,suchastheage,aver- thegalaxy,at>5kpcoftheradiogalaxy(late-stagemergers
agecolumndensity,massand,tosomeextent,initialmetallicity. andstar-formingcompanionsfallintothiscategory);
Evenifthestarformationmostlikelyoccurssimultaneouslyin 2. “in situ star formation”, which refers to star formation that
different clumps, we treat this component as a single idealised occurswithintheradiogalaxy,at<5kpc(rotatingstructure
starburst (completely localised and isolated). This provided us andnuclearstarformationfallintothiscategory);
withinformationusefulforcomparingoursampletoothergalax- 3. “AGN-driven star formation” corresponding to star forma-
ies.Assumingthestarbursthasaprojectedcirculargeometry(at tion linked to AGN activity, independently of spatial scale
constantdensity),weestimatethesizeofthestarburstregionas (bothjet-inducedandshock-inducedstarformationfallinto
follows: thiscategory).
(cid:32) (cid:33)1
rSB ∝ πMKNSB 2 , (2) TtichaellaimnditNoIfR57kdpocmisaidnrsivaenndbtyhethteypreicsaollustiizoenoofbataminaesdsivinetghaelaoxpy-
HI
atz>1(Fanetal.2010).
where r is the projected radius, M is the mass of the star-
SB SB
burst,KisthecolumndensityfactorandN thecolumndensity
HI
(here N = 6.8×1021 atomscm−2). These values are reported 5.3.1. Exsitustarformation
HI
in the third column of Table 5. With an estimated size, we can
Ex situ star formation can be divided into two subcategories, a
defineatypicalvelocitydispersionassumingequilibrium:
massive star-forming companion or a multiple system of less
(cid:114)
GM massive sub-units, which can be interpreted as a signature of
σ ∼ SB, (3)
SB r
SB 6 The HST data are downloaded from the HLA. The Herschel data
where σ is the velocity dispersion of the gas, and G is the
SB are unresolved when compared to optical or NIR images and are not
gravitational constant. We recall that when the system is col- reportedforsimplicity(butavailablein Drouartetal.2014).
lapsing,thisvalueisanupperlimit,whileforanexpandingsys- 7 Werecallthatatz > 1,1(cid:48)(cid:48) correspondsroughlyto8kpc,assuming
tem, this value becomes a lower limit. Assuming the relation thestandardcosmologicalmodel.
A109,page10of26