Table Of Content

MNRASinpress,1–11(2015) Preprint5thJanuary2016 CompiledusingMNRASLATEXstylefilev3.0 10C continued: a deeper radio survey at 15.7 GHz I. H. Whittam1,2(cid:63), J. M. Riley2, D. A. Green2, M. L. Davies2, T. M. O. Franzen3, C. Rumsey2, M. P. Schammel and E. M. Waldram2. 1PhysicsDepartment,UniversityoftheWesternCape,Bellville7535,SouthAfrica 2AstrophysicsGroup,CavendishLaboratory,19J.J.ThomsonAvenue,CambridgeCB30HE 3InternationalCentreforRadioAstronomyResearch(ICRAR),CurtinUniversity,Bentley,WA6102,Australia Accepted—;received—;inoriginalform— 6 1 0 ABSTRACT 2 Wepresentdeep15.7-GHzobservationsmadewiththeArcminuteMicrokelvinImagerLarge ArrayintwofieldspreviouslyobservedaspartoftheTenthCambridge(10C)survey.These n observationsallowthesourcecountstobecalculateddownto0.1mJy,afactoroffivedeeper a J thanachievedbythe10Csurvey.Thenewsourcecountsareconsistentwiththeextrapolated fittothe10Csourcecount,anddisplaynoevidenceforeithersteepeningorflatteningofthe 3 counts.Thereisthusnoevidencefortheemergenceofasignificantnewpopulationofsources ] (e.g.starforming)at15.7GHzfluxdensitiesabove0.1mJy,thefluxdensitylevelatwhichwe A expectstarforminggalaxiestobegintocontribute.ComparisonswiththedeZottietal.model G and the SKADS Simulated Sky show that they both underestimate the observed number of sources by a factor of two at this flux density level. We suggest that this is due to the flat- . h spectrumcoresofradiogalaxiescontributingmoresignificantlytothecountsthanpredicted p bythemodels. - o Keywords: galaxies:active–radiocontinuum:galaxies–catalogues–surveys r t s a [ 1 INTRODUCTION 2008)at15.7GHz.Theobservations,mappingandsourceextrac- 1 tionaredescribedinAMIConsortium:Franzenetal.(2011)(here- v The high-frequency radio sky (ν (cid:38) 10 GHz) has been much less 2 after Paper F) and the source counts and spectral properties are widelystudiedthanthepopulationatlowerradiofrequencies(e.g. 8 presentedinAMIConsortium:Daviesetal.(2011)(hereafterPa- 2 1.4GHz),mainlyduetotheincreasedtimerequiredtosurveyan perD).The10Csurveyiscompleteto1mJyintendifferentfields area to an equivalent depth at higher frequencies. In recent years 0 coveringatotalof≈27deg2;deepareascovering≈12deg2,con- several high-frequency surveys have been conducted, albeit with 0 tainedwithinthesefields,arecompleteto0.5mJy,makingthe10C 1. higherfluxdensitylimitsthansurveysatlowerfrequencies.Wal- surveythedeepesthigh-frequencyradiosurveypublishedtodate. dram et al. (2003, 2010) carried out the Ninth Cambridge (9C) 0 survey at 15 GHz using the Ryle Telescope. This survey covers Whittam et al. (2013) used data at a range of frequencies to 6 520 deg2 to a completeness limit of ≈ 25 mJy, and was the first study the spectral indices of 10C sources in the Lockman Hole. 1 high-frequencysurveytocoverasignificantproportionofthesky. Theyfoundasignificantchangeinspectralindexwithfluxdens- : v A series of deeper regions were also observed (Waldram et al. ity;themedianspectralindexα(whereS ∝ ν−α forasourcewith Xi 2010),with115deg2completeto≈10mJyand29deg2complete fluxdensityS atfrequencyν)calculatedbetween1.4and15.7GHz to≈5.5mJy. changesfrom0.75forsourceswithS15.7GHz >1.5mJyto0.08for r a ThewholeSouthernskyhasbeensurveyedbytheAustralia sourceswithS15.7GHz < 0.8mJy.Thisshowsthatthereisapopu- Telescope20GHz(AT20G)survey(Massardietal.2011),which lationofflat-spectrumsourcesemergingbelow1mJy;Whittamet has a flux density limit of 40 mJy and is 93 per cent complete al.suggestthatthismaybeduetothecoresofFanaroffandRiley above 100 mJy. This survey is complementary to the 9C survey typeI(FRI;Fanaroff&Riley1974)sourcesbecomingdominantat asitcoversalargerareaatashallowerfluxdensity.Morerecently, 15.7GHz. theAT20G-deeppilotsurvey(Franzenetal.2014)surveyed5deg2 Therehavebeenseveralattemptstomodelthehigh-frequency toacompletenesslevelof2.5mJy. radiosky,oftenextrapolatingfromlowerfrequencies.Earlyevolu- The Tenth Cambridge (10C) survey was observed with the tionarymodelsofradiosources(Dunlop&Peacock1990;Jackson Cambridge Arcminute Microkelvin Imager (AMI; Zwart et al. &Wall1999;Toffolattietal.1998)successfullyfittedtheavailable dataatfrequencies(cid:46)10GHzdowntofluxdensitiesofafewmJy. More recently,de Zotti etal. (2005)produced a modelof the ra- (cid:63) email:[email protected] dio source counts at frequencies (cid:38) 5 GHz (up to 30 GHz) which (cid:13)c 2015TheAuthors 2 I.H.Whittametal. successfullyfittedthedataavailableatthetime.ThedeZottietal. modelsplitsthesourcesintoflatandsteep-spectrumpopulations, withtheflat-spectrumpopulationfurtherdividedintoflat-spectrum radioquasarsandBLLacs,anddeterminestheepoch-dependentlu- minosityfunctionforeachpopulation.Starforminggalaxies,GHz- peaked spectrum (GPS) sources and objects in the late stages of AGNevolutionarealsoincludedinthemodel. A more recent model by Tucci et al. (2011) used physically groundedmodelstoextrapolatethe5-GHzsourcecount,whichis wellknownobservationally,tohigherfrequencies.Theyfocuson thespectralbehaviourofblazarsandcomparethreedifferentmod- els which treat flat-spectrum sources differently. This scheme is successfulathighfluxdensitiesbutdoesnotaccuratelyreproduce theobserved15GHzsourcecountbelow≈10mJy.Thenumberof sourcesissignificantlyunderestimated(byafactorof∼ 2),indic- atingthatthepropertiesofthesesourcesarenotwellunderstood, Figure 1. The noise in the 15.7-GHz AMI map of the AMI001 largelyduetothecomplexityanddiversityofthehigh-frequency (J0024+3152)field. spectraofindividualsources. Wilman et al. (2008, 2010) have produced a semi-empirical (J1052+5730), details of which are given in Table 1. From these simulation of the extragalactic radio continuum sky (the SKADS Simulated Sky; S3). The simulation splits the radio sources into two fields we selected sub-fields that we observed further, using arasteringtechniqueinthesamewayasthe10Csurveyobserva- separatepopulations:FRIIsources,FRIsources,radio-quietAGNs tions,withhexagonalrastersof37pointingseachseparatedby4 andstarforminggalaxies,whicharesplitintoquiescentstarforming arcmin.IntheAMI001field,450hoursofnewobservationshave and starbursting galaxies. The observed (and extrapolated) radio beenmadeinseveralhexagons(eachconsistingof37pointings), continuum luminosity functions are used to generate a catalogue of ≈ 320 million simulated sources. This simulation covers 20 × and the rms noise in the centre of the two deepest hexagons is 20 deg2 out to a cosmological redshift of z = 20 and down to a ∼ 16 µJy beam−1. The noise map for the AMI001 field is shown inFig.1.IntheLockmanHolefield,300hoursofadditionalobser- flux density of 10 nJy at 151, 610 MHz, 1.4, 4.86 and 18 GHz. vationswerecarriedoutinone37-pointinghexagon.Thermsnoise Whittametal.(2013)showedthatthesimulationfailstoreproduce inthishexagonis∼21µJybeam−1,asshownbythenoisemapin thespectralindexdistributionof10Csources,dramaticallyunder- Fig.2. predictingthenumberofflatspectrumsources. Allrawdatafileswerereducedwiththeup-to-datereduction Thus,althoughgoodprogresshasbeenmadeinrecentyears pipelineintheAMIin-housesoftwarepackagereduce(PaperF). inmodellingaspectsoftheextragalacticradiosourcepopulation, Thisnewpipelinewasdevelopedfromthatusedforthe10Csur- the high-frequency radio source population at low flux densities veywithenhancedmethodsforflagginginterferenceandproduces is poorly described. To understand the nature of the faint, high- consistentresults.uv-fitsfilesarewrittenoutfromreduceandcon- frequencypopulationandconstrainthemodelsbetterobservations catenatedtoproduceasingleuv-fitsfileforeachfield.Mappingis of the faint, high-frequency sky are required. In this paper we carriedoutinaipsusingthesameimagingpipelineasthatforthe present new, deep observations at 15.7 GHz in two 10C fields, 10C survey, with automated clean procedures and the imean task which when combined with existing 10C data enable the source usedtoestimatethenoise.Thepixelsizeintheimageis5×5arc- countstobeconstraineddownto0.1mJy,afactoroffivedeeper sec. Details of the data reduction and imaging procedure can be thanthe10Csourcecount. foundinPaperF. Thispaperislaidoutasfollows.Section2describestheobser- vationsanddatareduction,andSection3detailsthemethodsused toproducethesourcecatalogue.Section4discussestheeffectsof fluxdensityvariability.Thecatalogueisusedtoderivethesource 3 THESOURCECATALOGUE countsinSection5andtheseresultsarediscussedinSection6be- foresomebriefconclusionsarepresentedinSection7. 3.1 Sourcefitting The source fitting was performed using the Cambridge in-house softwaresource_findinthesamewayasforthe10Csurveyinor- dertomakethenewcatalogueascomparableaspossibletothe10C 2 OBSERVATIONSANDDATAREDUCTION catalogue.Theprocedureimplementedinsource_findisdescribed The15.7-GHzobservationswereconductedbetweenAugust2008 brieflyhere,fulldetailsbeinggiveninPaperF.Theimageswere andJuly2014usingtheArcminuteMicrokelvinImagerLargeAr- usedinconjunctionwiththenoisemapsshowninFigs.1and2to ray(AMILA),locatednearCambridge,UK.TheAMILAconsists identifycomponent‘peaks’aboveagivensignal-to-noiseratioγ(in ofeight13-mantennaswithbaselinesof18to110m,givingita thiscaseγ =4.621).Initially,allpixelsgreaterthan0.6γσ(where primarybeamof5.5arcminandaresolutionof30arcsec. σisthelocalnoise)wereidentifiedtoensureallpeaksgreaterthan For the 10C survey the LA observations were made of 10 γσafterinterpolationarefound.Thepositionandfluxdensityof fieldsofdifferentsizes,coveringatotalof27deg2.Eachfieldcon- eachpeak(RA ,Dec andS )wasthenfoundbyinterpolation pk pk pk sists of an outer region complete to 1 mJy beam−1, and an inner deeperregioncompleteto0.5mJybeam−1. Here we describe further observations of two of the 10C 1 Thisvalueisused,insteadof5σ,totakeaccountofthe1.082phaseerror fields:theAMI001(J0024+3152)fieldandtheLockmanHolefield correctionfactorappliedtothefluxdensities(seeSection3.3fordetails). MNRASinpress,1–11(2015) 10Ccontinued:adeeperradiosurveyat15.7GHz 3 Table1.Summaryofthedifferentregionsinthetwofields. Region Comments Area Observingtime approx.rms /deg2 /hours /µJy AMI001(J0024+3152) 10Cshallow Completeto1mJy 3.56 150 1000 10Cdeep Completeto0.5mJy 1.69 45 10Cultra-deep Thiswork–deepestobs.intwohexagonscentredon 0.38 additional450 16 00h22m19s,+31◦46(cid:48)09.7(cid:48)(cid:48)and00h24m00s,+31◦59(cid:48)29.5(cid:48)(cid:48) LockmanHole(J1052+5730) 10Cshallow Completeto1mJy 1.78 120 450 10Cdeep Completeto0.5mJy 0.54 45 10Cultra-deep Thiswork–hexagoncentredon0h52m22s,+57◦24(cid:48)55.0(cid:48)(cid:48) 0.18 additional300 21 Table2.Thepositionsofthecentresoftheexclusionzonesaroundbright sourcesandtheirradii. RA Dec Radius/arcmin 00:29:33.7 +32:44:52 4.19 00:23:09.8 +31:14:00 3.80 00:21:29.8 +32:26:58 3.41 00:20:50.4 +31:52:28 3.30 00:28:10.5 +31:03:46 3.20 00:29:20.4 +32:16:54 3.06 10:50:07.1 +56:53:37 3.35 10:52:25.1 +57:55:07 3.24 10:54:26.9 +57:36:48 3.69 by: (cid:32) S (cid:33)1/2 r =12 pk arcmin, (1) e 250mJy Figure 2. The noise in the 15.7-GHz AMI map of the Lockman Hole whereS isthepeakfluxdensityofthebrightsource.Theposi- (J1052+5730)field. pk tionsandradiioftheseexclusionzonesareshowninTable2. betweenthepixels,andanypeakslessthanγσwerediscarded.The erroronthepeakfluxdensity(∆S )istakentobethethermalnoise 3.3 Finalfluxdensityvalues pk errorσn comb(cid:112)inedinquadraturewiththefivepercentcalibration ThedeepAMIimagesdonothaveasignal-to-noisehighenough error,∆Spk = σ2n+(0.05Spk)2.Theintegrationarea,consistingof tobeself-calibrated,soacorrectionfactorwasappliedtoallflux contiguouspixelsdowntoalowestcontourvalueof2.5σ,wasthen densityvaluestoaccountfortheresidualphaseerrorsbeforeinclu- calculatedforeachcomponent.Ifmorethanonepeaklayinsidethe sionintothefinalcatalogue(PaperD).Thecorrectionfactorused sameintegrationareathesourceswereclassifiedasa‘group’. hereisthesameastheoneusedforthe10Csurvey,whichwases- Theaipstaskjmfitwasthenusedtofita2DellipticalGaus- timatedbycomparingthepeakfluxdensitiesofbright,unresolved sian to each component. This was used to estimate the integ- sources in the 10C raster maps with their peak flux densities in ratedfluxdensity,positionandangularsize(RAin,Decin,Sin and self-calibrated, pointed maps. All peak flux densities were there- emaj) for each component. The error on the integrated flux dens- fore multiplied by 1.082. The use of this correction factor is the ity (∆Sin) was estimated as the error due to thermal noise (σn) reasonforcarryingoutthesource-fittingdownto4.62σ–asource combine(cid:112)d in quadrature with the five per cent calibration error, nearthesurveylimitwhichshouldbedetectedat5σwillonlybe ∆Sin= σ2n+(0.05Sin)2. detectedat5σ/1.082=4.62σ.Intotal,358sourcesweredetected intheAMI001fieldand134intheLockmanHolefield(including theoriginalshallowerareasfromPaperDaswellasthenewdeep 3.2 Exclusionzones areas). Themapsdisplayanincreaseinnoiseclosetobright(> 15mJy) Asourcewasconsideredtobeextendedifthemajoraxisof sourcesduetoamplitude,phaseanddeconvolutionerrors.Because thedeconvolvedGaussian(emaj)islargerthanacriticalvalueecrit thisnoiseisnon-gaussianitisoftennotincludedinthenoisemaps, (seePaperF),where leading to spurious detections close to bright sources. For this (cid:40) 3.0b ρ−1/2 if3.0b ρ−1/2>25.0arcsec, reason,an‘exclusionzone’wasdefinedaroundanysourcewitha e = maj maj (2) crit 25.0arcsec otherwise, peakfluxdensitygreaterthan15mJyduringthesourcefittingpro- cess,andanysourcedetectedwithinthiszonewasexcludedfrom whereb isthemajoraxisoftherestoringbeamandρ=S /σ maj pk n the catalogue. As derived empirically in Paper D, each exclusion (i.e.thesignal-to-noiseratio).Sourceswithe > e wereclas- maj crit zoneisacirclecentredonthebrightsource,witharadiusdefined sified as extended (flag E), otherwise the source was considered MNRASinpress,1–11(2015) 4 I.H.Whittametal. accountbyaveragingtheprobabilitiesofdetectingasourceateach Fraction of simulated sources detected pixel position in the noise map. As the source fitting was in fact 1 Visibility area Predicted completeness carriedoutto4.62σandthefluxdensitiesweremultipliedby1.082 aftersourceextraction,thisequationbecomes ess0.8 Pi(S)=(cid:90)4.∞62σi (cid:113)21πσ2 exp−(cid:16)X−2σ1.2i0S82(cid:17)2dX, (4) n0.6 i e et pl whereσ isthevalueofthenoisemapattheithpixel.Theouter50 m i Co pixelsofthenoisemapareexcludedfromthiscalculationsothat 0.4 theresultsaredirectlycomparabletothecompletenessestimated usingsimulatedsources. 0.2 The results of both methods used to estimate completeness areshowninFig.3.AsimilaranalysiswasperformedinPaperD toestimatethecompletenessofthe10Csurvey.PaperDfoundthat 0 atthelowestfluxdensitiesprobedbytheirsurvey(S ∼ 0.5mJy) 0 0.2 0.4 0.6 0.8 1 Peak flux density / mJy the fraction of detected sources from the simulation was slightly higher than predicted by the completeness curve, while at higher Figure3.DifferentmethodsusedtoestimatecompletenessintheAMI001 flux densities the fraction of detected sources was slightly lower field.Thefractionsofsimulatedsourcesdetectedatagivenfluxdensityare thanpredicted.Thesameeffectisvisiblehere,withthefractionof shownbyredcircles(errorbarsrepresentPoissonerrors).Thecompleteness detected sources slightly higher than predicted for S < 0.8 mJy curveestimatedfromthenoisemapusingequation4isshownbythegreen andlowerthanpredictedforS >0.8mJy.PaperDsuggestsseveral line.Thebluecrossesshowthevisibilityareausedtocalculatethesource factors which may contribute to this effect. One is source confu- countsinSection5. sion,whichwillincreasethecompletenessatlowfluxdensities,as twosourcesbelowthecompletenesslimitmayliesufficientlyclose point-like(flagP).Intotal,thereare24sourcesclassifiedasexten- together to be detected as one source; in contrast, at higher flux dedintheAMI001fieldandsixintheLockmanHolefield. densitiesconfusionwillreducethecompletenessasasourcemay Forsourcesclassifiedasextended,theintegratedfluxdensity notbedetectedifitliestooclosetoabrightersource.Confusion isused,andforthoseclassifiedaspoint-like,thepeakfluxdensityis thereforepreventsthecompletenessfromreaching100percentas used.Thesevaluesarelistedasthe‘bestflux’inthecatalogue,and quicklyaspredicted. arethevaluesusedfordeterminingthesourcecountinSection5. The‘visibilityarea’(Katgertetal.1973)isalsoplottedinFig. 3.Thisisthefractionofthetotalareaoverwhichasourceofflux density S should be detectable, i.e. the fraction of the area with 3.4 Completeness i 5σ < S .Thisisusedtocalculatethesourcecountandisdis- local i The completeness of the survey was estimated by inserting 250 cussedinmoredetailinSection5. sourcesofequalfluxdensityintotheAMI001mapinrandompo- sitions. The simulated sources were ideal point sources with flux densityS andwereaddedtothemapintheimageplaneusingthe 3.5 Reliability aips task immod; no sources were placed in the border 50 pixels AssumingthatthenoiseinthemapisGaussian,theexpectednum- wide at the edge of the map, as many of these pixels are blank beroffalsepositivessimplyduetothenoisecanbecalculated(the (the blank pixels were excluded in area and source count calcu- noiseisunlikelytobeGaussianclosetobrightsources,butasex- lations). Sources which lay within 2 arcmin of another simulated clusionzonesareplacedaroundsourceswithS >15mJy(seeSec- sourcewereremovedtoavoidthesimulatedsourcescontaminating tion3.2)thisassumptionisvalid).Thetotalareaofthetwofields thefluxmeasurementoftheadjacentsource,leaving224simulated containingthedeeperobservationsis5.3deg2 andthesynthesised sources.Thesourcefittingwasthenperformedinexactlythesame beamareais≈ 700arcsec2,sothenumberofbeamscoveringthe wayasdescribedinSection3.1.Asimulatedsourcewasconsidered twofieldsis≈ 100,000.Theprobabilitythatavaluedrawnfrom tobedetectediftherewasasourceintheoutputcataloguewithin aGaussiandistributionismorethan4.62standarddeviationsaway 30arcsecofthesimulatedsourceposition.Thiswasrepeatedsev- from the mean is ≈ 1.9×10−6, so the expected number of false eraltimesusingarangeoffluxdensities0.1 < S/mJy < 1(keep- positivesis≈ 0.2.Wethereforedonotexpectreliabilitytobean ingthesamepositionseachtime)toestimatethecompletenessasa issueforthissurvey. functionoffluxdensity.Thefractionofsimulatedsourcesdetected as a function of flux density is shown in Fig. 3. The flux densit- iesofthesimulatedsourcesaremultipliedby1.082(thecorrection 3.6 Multiplesources factorappliedtothefinalcataloguetoaccountforphaseerrors,see Section3.3)beforeinclusiononthisplot. Thereare38sourcesintheAMI001fieldwhichareclassifiedasbe- Thecompletenesscanalsobeestimatedfromthenoisemap ingpartofa‘group’,indicatingthattheirfittedGaussiansoverlap. assumingaGaussiannoisedistribution.Theprobabilityofdetect- Twelveofthesesourcesformfourtriples,theremaining26sources ingasourceofpeakfluxdensityS above5σisgivenby form13pairs,meaningthatthereare17groupsintotal.Thesepar- (cid:90) ∞ 1 (cid:32) (X−S)2(cid:33) ationbetweenthesourcesinthese17groupsrangesfrom27to126 P(S)= √ exp − dX. (3) arcsec,withanaverageseparationof49arcsec.Thesourcecounts 5σ 2πσ2 2σ2 canbeusedtoestimatethenumberofsourcesthatwouldbeexpec- In reality the noise varies across the map; this can be taken into tedtobeoverlappingduetoconfusionalone.Integratingthe10C MNRASinpress,1–11(2015) 10Ccontinued:adeeperradiosurveyat15.7GHz 5 4 THEEFFECTSOFVARIABILITY Table3.Theparameterswhichappearinthesourcecataloguesavailable online.FulldetailsofallparametersaregiveninPaperF. Astheseobservationsweremadeoverasixyearperiod,theycan beusedtoinvestigatetheeffectsofvariabilityonthefluxdensities. Source 10CsourcedesignationJ2000 Thisalsoprovidesausefulcheckonthemap-makingprocessand Group 10CgroupdesignationJ2000 thecatalogueitself. αpk RA(peak),inh,m,s(J2000) δpk Dec.(peak),in◦,(cid:48),(cid:48)(cid:48)(J2000) Spk Peakfluxdensity,inmJybeam−1 4.1 Comparisonwiththeoriginal10Ccatalogue δSpk Erroronthepeakfluxdensity,inmJybeam−1 αin RA(fittedpeak),inh,m,s The source catalogue produced in the AMI001 field was com- δin Dec.(fittedpeak),in◦,(cid:48),(cid:48)(cid:48) paredwiththeoriginal10Ccatalogue.Thereare358sourcesinthe Sin Integratedfluxdensity,inmJy newcatalogueand290intheoriginal10Ccatalogue;thereare88 δSin Errorontheintegratedfluxdensity,inmJy sourcesfoundinthenewcataloguewhicharenotfoundinthe10C ecrit Criticalcomponentsize,inarcsec catalogueand20sourcesinthe10Ccataloguewhicharenotinthe emaj Major-axisafterdeconvolution,inarcsec newcatalogue.Ineach10Cfieldtwocompleteareasweredefined; emin Minor-axisafterdeconvolution,inarcsec oneareacompleteto1mJywhichencompassesmostofthefield eθ Positionangleafterdeconvolution,in◦, (the‘shallow’area),andoneareacompleteto0.5mJy(the‘deep’ measuredfromnorthtoeast t Sourcetype(P=point-like,E=extended) area)whichiscontainedwithintheshallowarea.The20sources Flag Astarindicatesthattheapproximationerrorfor notfoundinthenewmapareallattheedgeofthemap,outsidethe thepoint-sourceresponseissignificant completearea. Allofthesourcesinthenew,deepareaswhicharenotpresent inthe10Ccataloguehavefluxdensitieslessthan0.5mJy(thecompleteness limit of the 10C survey), as expected. There are three sourcesinthefieldwhichareincludedinthedeepcomplete10C sourcecountsbetween0.1 < S/mJy < 25(whichinvolvesextra- catalogue, and therefore have flux densities greater than 0.5 mJy, polationfrom0.5to0.1mJy),≈7.2×105sourcespersteradianare whichhavefluxdensitieslessthan0.5mJyinthenewcatalogue. expectedinthisfluxdensityrange.Thereareatotalof358sources However, due to the relatively large errors at low signal-to-noise intheAMI001field,sothetotalareawithin49arcsecofasourceis levels,thesenewfluxdensityvaluesareallconsistentwith0.5mJy 358×A ≈6.4×10−5steradians(whereA istheareaofacircle circ circ withintheerrors. witharadiusof49arcsec).Thus≈ 46sourcesareexpectedtolie Fig. 5 shows a comparison of the peak and integrated flux within49arcsecofanothersourceandwouldconsequentlybeclas- densities of sources which appear in both the original 10C cata- sifiedasoverlapping.Thisishigherthanthenumberofoverlapping logueandthenew,deepercatalogueintheAMI001field.Theflux sourcesobserved,whichisprobablyduetothefactthatsourcesas densitiesfromthetwocataloguesaregenerallyingoodagreement, faint as 0.1 mJy (the lower limit on flux density used in this cal- asexpectedgiventhattwo-thirdsofthedatacamefrom10C. culation) can only be detected in part of the image. This number canthereforebeviewedasanupperlimitontheexpectednumber ofoverlappingsources.Repeatingthiscalculationwithadetection 4.2 Splittingthedatainhalf limitof0.5 mJyshowsthatwewouldexpect≈ 10sourcestobe overlapping; we therefore expect between 10 and 46 overlapping Along-termstabilitytestwasperformedona∼ 1deg2 regionin sourcesinthefieldduetoconfusionalone.Thisestimateassumes the deepest part of the AMI001 field. This involves splitting the nosourceclustering,whichwouldincreasethenumberofoverlap- datainhalfaccordingtowhenitwasobserved,soalltheolderob- pingsources. servationsareinthefirsthalfandallthenewerobservationsarein This analysis suggests that many of the overlapping sources thesecondhalf.Thiswasdoneinsuchawayastoensurethatthe detected in these observations are due to confusion, rather than noisewasthesameinthetwohalves,sothesplitdoesnotnecessar- being genuine multiple sources. No attempt is made therefore to ilyoccuratthesamepointintimeforeverypointing.Themedian combinethefluxdensitiesoftheoverlappingsourcesintoasingle point in time for the split to occur is August 2009, but the exact source;theyarelistedasseparateentriesinthesourcecatalogue, date varies from pointing to pointing, because more observations butflaggedasbeingpartofagroup. weremadeatsomepointingsearlierorlaterduringtheobservation period, or because some of the data were particularly noisy. The twohalvesarethenimagedseparatelyandsource_findwasrunon thetwohalvestocreatetwocatalogues.The‘firsthalf’catalogue contains 63 sources and the ‘second half’ catalogue contains 57 sources;thepositionsofthesesourcesareshowninFig.6.Thema- 3.7 Thefinalcatalogue jorityofsourcesarefoundinbothcatalogues,but16sourcesare The final source catalogue contains 358 sources in the AMI001 onlyfoundinoneorotherofthecatalogues.Inordertocompare field with flux densities 0.088 < S/mJy < 34 and 134 sources thefluxdensitiesinthetwohalves,upperlimitsoffivetimesthe intheLockmanHolefieldwithfluxdensities0.15<S/mJy<22. localnoiseareplacedonthefluxdensitiesofthesesourcesinthe Thefluxdensitydistributionsofthesourcesinthesetwocatalogues imagetheywerenotdetectedin;infourcasesasourcewasvisible areshowninFig.4.Twenty-fourofthesourcesintheAMI001field in the image which had fallen below the 5σ cut-off and in these (7percent)andsixintheLockmanHolefield(4percent)areclas- casesthepeakfluxdensityofthevisiblesourcewasusedinstead. sifiedasextended.Thesourcecataloguesareavailableonline,the Forfiveofthesourceswhicharedetectedinbothimagesthereare parameterswhichappearinthesecataloguesaredescribedinTable smalldifferencesinthepositionsfromthetwohalves;thesesources 3. allhavelowsignaltonoisevaluesandthediscrepancyisduetoa MNRASinpress,1–11(2015) 6 I.H.Whittametal. 70 35 60 30 50 25 s s e e c c our40 our20 s s of of ber 30 ber 15 m m u u N N 20 10 10 5 0 0 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2 log(15.7−GHz flux density / mJy) log(15.7−GHz flux density / mJy) Figure4.FluxdensitydistributionsofsourcesintheAMI001fieldcatalogue(left)andtheLockmanHolefieldcatalogue(right). 102 102 y mJ y mJ C / C / 101 10 101 nsity - 10 density - e x ux d 100 d flu 100 eak fl grate P e nt 10-1 I 10-1 10-1 100 101 102 10-1 100 101 102 Peak flux density - deeper obs. / mJy Integrated flux density - deeper obs. / mJy Figure5.Comparisonofpeak(leftpanel)andintegrated(rightpanel)fluxdensitiesinthe10CanddeepercataloguesintheAMI001field.Theblackline indicateswherethetwofluxdensitiesareequal. differentpixelbeingidentifiedasthepeakinthetwohalves.The sourcesclassifiedasextendedinthefittingprocess,otherwisethe fluxdensitiesofthesourcesinthetwohalvesarecomparedinFig. peakfluxdensityisused.Thenoisevariessignificantlyacrossboth 7.Onesource,whichisnotdetectedinthefirsthalf,isasignificant fields,asisdemonstratedinFigs.1and2,andthisneedstobetaken outlier;thissourceislocatedveryclosetotheedgeofthemapso intoaccountwhencalculatingthesourcecounts.Todothisthevis- could be spurious. There is some scatter in this plot but the ma- ibilityareawascalculated(Katgertetal.1973);thisisthefraction jorityofthevaluesagreewithintheerrors,sotheredonotappear ofthetotalareaoverwhichasourceofafluxdensityS couldbe i tobeanysourceswhicharevaryingsignificantlyonthetimescales detected(i.e.thefractionofthetotalareawith5σ <S ),assuming n i of several years in this sample. This suggests that there are few thenoisemap.Thevisibilityareasofthetwofieldsareshownin beamedobjectsamongsttheflat-spectrumsourcesinthissample. Fig.8. Thereisalsonoevidenceforanysystematicdifferencesbetween To calculate the source count, each source was weighted by thedataobservedatthebeginningandattheendoftheobservation thereciprocalofitsvisibilityarea.Thesourcecountineachflux run. densitybinisthereforegivenbythefollowingexpression: 1 (cid:88)N 1 (5) A x(S ) 5 SOURCECOUNTS i=1 i whereN isthenumberofsourcesinthebin,Aisthetotalareaof 5.1 Calculatingthesourcecounts thefieldandx(S )isthevisibilityareaforasourcewithfluxdensity i The catalogue of sources used to calculate the source counts is S . i described in Section 3.7. The integrated flux density is used for Thedifferentialsourcecountsderivedfromthetwofieldsare MNRASinpress,1–11(2015) 10Ccontinued:adeeperradiosurveyat15.7GHz 7 32.6 Table6.Sourcecountsforthetwofieldscombined. 32.4 s Binfluxdensity dN/dS e e 32.2 range/mJy /Jy−1sr−1 r eg 2.900–5.500 (5.7±1.1)×106 d 32.0 2.050–2.900 (2.1±0.4)×107 n / 1.500–2.050 (3.2±0.6)×107 tio 31.8 1.250–1.500 (5.6±1.1)×107 na 1.000–1.250 (8.8±1.4)×107 Decli 31.6 00..970705––10..090000 ((11..14±±00..33))××110088 0.680–0.775 (1.8±0.4)×108 31.4 0.600–0.680 (2.1±0.5)×108 0.540–0.600 (3.7±0.8)×108 31.2 0.500–0.540 (3.7±1.0)×108 6.6 6.4 6.2 6.0 5.8 5.6 5.4 5.2 0.400–0.500 (3.3±0.6)×108 Right ascension / degrees 0.300–0.400 (6.2±0.9)×108 0.250–0.300 (9.9±1.7)×108 Figure6.Positionsofsourcesinthe‘firsthalf’(red‘×’)and‘secondhalf’ (blue‘+’)cataloguesintheAMI001fieldwhenthetwohalvesareimaged 0.200–0.250 (1.2±0.2)×109 0.100–0.200 (2.2±0.3)×109 separately. showninTables4and5,alongwiththecombinedcountfromthe Table4.SourcecountsintheAMI001field. two fields in Table 6. They are plotted in Fig. 9, along with the 10Ccountforcomparison.Thebrokenpower-lawfittedtothe10C count in Paper D, given by equation 6, is also shown. The com- Binfluxdensity dN/dS bined9C,10Cand10Cultra-deep(thiswork)countsareshownin range/mJy /Jy−1sr−1 Table7. 5221....590500500000––––9522....059000050000 ((((9623....3706±±±±4100....7548))))××××111100005677 n(S)≡ ddNS ≈ 23476(cid:16)J(cid:16)SyJS(cid:17)y−(cid:17)2−.127.80JyJ−y1−s1rs−r1−1 ffoorr02..58≤≤SS <≤22.58mmJJyy,. 1.250–1.500 (6.7±1.5)×107 (6) 1.000–1.250 (8.3±1.7)×107 Higherfluxdensitybinsarenotincludedasthefieldswerechosen 0.900–1.000 (1.3±0.3)×108 tolieawayfrombrightsources.Thepointsareplottedatthe‘centre 0.775–0.900 (1.2±0.3)×108 ofgravity’ofeachfluxdensitybin(theaverageofthedifference 0.680–0.775 (1.8±0.5)×108 betweeneac√hfluxdensityandoneedgeofthebin).Theerrorbars 0.600–0.680 (2.3±0.6)×108 plottedare N Poissonerrors.Thenewcountsareingoodagree- 0.540–0.600 (2.7±0.8)×108 mentwiththe10Ccountwheretheyoverlap,andextendthesource 0.500–0.540 (4.3±1.3)×108 count by a factor of five fainter in flux density. The counts from 0.400–0.500 (4.2±0.8)×108 the two fields agree within the Poisson errors. The brightest flux 0.300–0.400 (5.9±1.0)×108 densitybinsinthetwofieldsappeartounder-predictthe10Cfit, 0.250–0.300 (9.1±1.9)×108 0.200–0.250 (1.7±0.3)×109 thisisprobablyduetothesmallnumberofsourcesinthesebins, 0.100–0.200 (2.5±0.4)×109 withfoursourcesintheAMI001fieldbinandsixintheLockman Hole field bin. The full 10C survey, which covers a much larger area,providesamoreaccuratemeasureofthesourcecountsatflux Table5.SourcecountsintheLockmanHolefield. densities greater than 0.5 mJy. In the faintest bin plotted for the AMI001field(0.08<S /mJy<0.10)thecompletenesscor- 15.7GHz rectionisnotwelldefined,andduetothepoorstatisticsinthisbin Binfluxdensity dN/dS itisomittedfromTable4andlaterdiscussion. range/mJy /Jy−1sr−1 2.900–5.500 (3.7±1.5)×106 2.050–2.900 (2.3±0.6)×107 5.2 Samplevariance 1.500–2.050 (2.4±0.8)×107 1.250–1.500 (3.4±1.5)×107 Heywood et al. (2013) investigated the effects on deep source 1.000–1.250 (9.7±2.6)×107 countsat1.4GHzofsamplevarianceinducedbysourceclustering 0.900–1.000 (7.1±3.5)×107 byextractingaseriesofindependentsamplesfromtheS3catalogue 0.775–0.900 (1.9±0.5)×108 andcomparingtoobservations.Theyusedthistopresentamethod 0.680–0.775 (1.8±0.6)×108 forestimatingtheuncertaintyinthesourcecountcausedbysample 0.500–0.680 (3.2±0.7)×108 0.300–0.500 (4.3±1.1)×108 varianceforanarbitraryradiosurvey.IntheAMI001fieldthereis 0.200–0.300 (6.4±2.0)×108 0.23deg2 withrmsnoise< 20µJyand0.42deg2 withrmsnoise 0.120–0.200 (1.9±0.7)×109 <25µJy,asshowninthecontourplotinFig.10.Itshouldtherefore bepossibletodetectasourcewithS > 0.1mJyinanarea 15.7GHz of0.23deg2;readingoffFig.2inHeywoodetal.(2013)forthis surveylimitandareagivesanuncertaintyduetocosmicvariance MNRASinpress,1–11(2015) 8 I.H.Whittametal. y y J J m m alf / 101 alf / 101 h h d d n n o o c c e e s s − − y y sit sit n n e e d d k flux 100 k flux 100 a a e e P P 100 101 100 101 Peak flux density − first half / mJy Peak flux density − first half / mJy Figure7.Peakfluxdensitiesofsourcesinthe‘firsthalf’and‘secondhalf’cataloguesintheAMI001field,shownwitherrorbarsintheleftpanel,andwithout intherightpanelforclarity.Intherightpanelupperlimitsareincludedforthe16sourceswhicharenotfoundinoneofthehalves–red‘(cid:79)’indicateupper limitsforsourcesfoundinthefirsthalfandnotinthesecondhalf,green‘(cid:47)’indicateupperlimitsforsourcesfoundinthesecondhalfandnotthefirsthalf. Red‘×’indicatethepeakfluxdensityofasourcewhichwasvisibleinthesecondhalfobservationsbutwhichfellbelowthe5σcut-offsowasnotincludedin theoriginalcatalogue. 1 1 0.9 0.9 0.8 0.8 0.7 0.7 a a e0.6 e0.6 ar ar of of n 0.5 n 0.5 o o cti cti Fra0.4 Fra0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 Peak flux density / mJy Peak flux density / mJy Figure8.VisibilityareaoftheAMI001field(left)andLockmanHolefield(right)i.e.fractionofthetotalareaoverwhichasourcewithagivenfluxdensity couldbedetected. of≈14percent.Athigherfluxdensities,thisuncertaintydecreases This means that a resolved source of a given total flux density is astheareaoverwhichasourcecouldbedetected(theeffectivesur- morelikelytofallbelowthepeakfluxdensitydetectionthreshold veyarea)increases.Forexample,forasurveylimitof0.125mJy thanapointsourcewiththesametotalfluxdensity.Duetothere- (equivalenttoanrmsnoiseof25µJy)andanareaof0.42deg2the lativelylargebeamsizeofAMI,only7percentofsourcesinthe uncertaintyis12percent.Wethereforeexpectcosmicvarianceto AMI001fieldand4percentintheLockmanHolefieldareexten- affectthefaintestfluxdensitybinby≈14percent. ded, so resolution is not expected to have a significant effect on thesesourcecounts. (ii) Eddington bias (Eddington 1913). Statistical fluctuations 5.3 Possiblebiases duetothermalnoisecanalterthefluxdensityofsourcesandcan Severaleffectswhichcouldbiasthesourcecountsareconsidered thereforecausesomesourcestobeputintothewrongbins.Given theshapeofthesourcecount,ifthetruenumberofsourcesinone here;howeverforthereasonsdiscussednocorrectionsaremadefor binislargerthaninadjacentbins,moresourceswillbescattered thesebiases. into that bin than out of it. The observed number of sources will (i) Resolution bias. To calculate the source counts a sample thereforebebiasedhigh.InPaperDitisestimatedthatthiseffect which is complete in terms of integrated flux density is required, willincreasethenumberofsourcesinthefaintest10Cfluxdens- butthesourcesaredetectedaccordingtotheirpeakfluxdensities. itybinby≈ 7percent.Asincompletenessisexpectedtoreduce MNRASinpress,1–11(2015) 10Ccontinued:adeeperradiosurveyat15.7GHz 9 1010 New AMI001 count New LH count 10C count 109 10C fit Extrapolated 10C fit −1sr108 −1y J n(S) / 107 106 105 10−4 10−3 10−2 S / Jy Figure 9. New source counts from the AMI001 (red ‘+’) and Lockman Hole(green‘◦’)fields.The10Csourcecountsarealsoshown(blue‘×’)for Figure10.Contoursat20and25µJy,showningtheregionsintheAMI001 comparison,asisthefittothe10Csourcecounts(blackline).Thefaintest fieldwithrmsnoise<25and20µJy. binplottedfortheAMI001fieldcountisbasedononlythreesourcesand thecompletenesscorrectionisnotwelldefinedatthisfluxdensitylevel,so thispointisnotincludedinthediscussionorsubsequentplots. thenumberofsourcesinthisbinbyasimilaramounttheydonot correct for this effect. No correction is applied here for the same reason. Table 7. Source counts for the combined 9C, 10C and 10C ultra-deep (iii) Variability. Variability in flux densities can cause sources counts. neartheedgeoffluxdensitybinstomovebetweenbins.Theshape ofthesourcecountsmeansthatatthebottomofthebin,thenumber ofsourcesinthepositivephaseofvariabilitywhichareincludedin Binfluxdensity dN/dS the bin will be marginally higher than the number of sources in range/mJy /Jy−1sr−1 the negative phase of variability which are excluded. The oppos- 500.0–1000.0 (1.01±0.36)×102 iteeffectoccursattheotherendofthebinbutwillnotbeenough 200.0–500.0 (5.68±1.09)×102 tooffsetthiseffect.Therefore,variabilitywillboostthenumberof 100.0–200.0 (2.97±0.43)×103 sources in a bin, causing a constant shift in the observed source 60.0–100.0 (1.45±0.15)×104 40.0–60.0 (3.06±0.31)×104 count.Thelongtimescaleoverwhichtheobservationsweremade 30.0–40.0 (6.25±0.63)×104 meansthattheeffectsofshort-termvariabilityaregenerallyaver- 25.0–30.0 (1.00±0.11)×105 agedout,andasdiscussedinSection4,wedonotfindanyevidence 16.0–25.0 (1.81±0.23)×105 forsignificantvariabilityonlongertimescales. 12.0–16.0 (4.22±0.53)×105 10.0–12.0 (6.32±0.92)×105 9.00–10.0 (1.03±0.27)×106 6.40–9.00 (1.28±0.19)×106 6 DISCUSSION 5.50–6.40 (3.40±0.51)×106 Figs. 9 and 11 show that the new deeper counts are consistent 2.90–5.50 (6.68±0.56)×106 with the extrapolated 10C fit. There is no sign of the upturn ob- 2.05–2.90 (1.94±0.17)×107 servedinthe1.4-GHzsourcecountsatS ∼ 1mJy(e.g.de 1.50–2.05 (3.29±0.27)×107 1.4GHz 1.20–1.50 (5.70±0.48)×107 Zotti et al. 2010); there is thus no evidence for a new population 1.00–1.20 (8.13±0.70)×107 (e.g.ofstarforminggalaxies)contributingtothe15.7-GHzsource 0.900–1.00 (9.88±1.65)×107 populationabove0.1mJy.Thisisnotsurprising,asasourcewith 0.775–0.900 (1.23±0.16)×108 α=0.8,typicalforastarforminggalaxy(Franzenetal.2014),with 0.680–0.775 (1.47±0.20)×108 S =1mJywillhaveS ∼0.14mJy,andwouldthere- 1.4GHz 15.7GHz 0.600–0.680 (2.19±0.27)×108 foreonlyhaveappearedatthebottomofthefaintestfluxdensity 0.540–0.600 (2.79±0.36)×108 binhere. 0.500–0.540 (3.16±0.47)×108 A model of the high-frequency (ν > 5 GHz) source counts 0.400–0.500 (3.30±0.61)×108 wasproducedbydeZottietal.(2005),asdescribedinSection1. 0.300–0.400 (6.22±0.98)×108 The new 15.7-GHz source counts presented here are compared 0.250–0.300 (9.89±1.75)×108 0.200–0.250 (1.15±0.19)×109 to the latest version of the de Zotti et al. model, extracted from 0.100–0.200 (2.23±0.33)×109 theirwebsite2,inFig.11.PaperDshowedthatthedeZottietal. modelunder-predictsthenumberofsourcesinthe10Csurveybe- low ≈ 5 mJy; it is clear in Fig. 11 that the model continues to 2 http://web.oapd.inaf.it/rstools/srccnt_tables MNRASinpress,1–11(2015) 10 I.H.Whittametal. 1 10 1 − r s 5 1. y J ) / S ( n 0 5 10 2. S New AMI count De Zotti model 3 S Combined 9C and 10C −1 10 −4 −3 −2 10 10 10 S / Jy Figure11.Euclideannormaliseddifferentialsourcecountsfromthenewobservationsinbothfieldscombined(red‘+’)andthecombined9Cand10Csource countsfromPaperD(blue‘×’).ThedeZottietal.modelat15GHz(dottedline)andthe18GHzcountfromtheS3catalogue(dashedline)arealsoshown (noattemptismadetocorrectthelatterto15GHz).Poissonerrorsareplottedfortheobservedcounts. under-predictthenumberofsourcesobservedbyafactoroftwoas used multi-wavelength data to show that these flat-spectrum 10C flux density decreases. Whittam et al. (2013) studied the spectral sources are probably the result of emission from the cores of ra- indicesofasampleof10Csourcesandshowedthattheproportion dio galaxies, which therefore have a far greater contribution than offlat-spectrumsourcesinparticularistoolowinthedeZottietal. predictedbyeitherofthemodelsdiscussedhere. modelbelow≈ 1mJy;themodelpredictsthatatS = 1mJy 15GHz steep-spectrumsourcesoutnumberflat-spectrumsourcebyafactor Below≈0.3mJythereisabetteragreementbetweentheob- ofthree,whiletheobservationsshowthattherearetwiceasmany served and simulated counts due to a slight flattening in the S3 flat-spectrumsourcesassteep-spectrumsources.Itislikelythatthis counts. This flattening is due to the greater contribution of star- under-predictionofthenumberofflat-spectrumsourcesinthesub- formingsources(bothquiescentandstarbursting)tothesimulated mJy population is responsible for the discrepancies between the cataloguebelow≈ 0.3mJy;starformingsourcescomprise21per modelandtheobservedcountseenhereatS <5mJy.Above5mJy centofthesimulatedsourceswith0.09<S /mJy<0.3,com- 18GHz themodelover-predictsthenumberofsourcesobserved,thisisdis- paredtoonly7percentofsourceswithS > 0.5mJy.How- 18GHz cussedfurtherinPaperD.Thehigher-fluxdensityend(0.5mJyto ever,giventhatthereisnoflatteninginthenewAMIcount,itisnot severalJy)ofthe15-GHzsourcecountsisstudiedbycombining clearwhatcontribution,ifany,apopulationofstarformingsources the 9C and 10C counts with the AT20G survey by Franzen et al. ismakingtotheobservedcountsatS >0.1mJy. 15.7GHz (2014), who investigated the source counts of the steep and flat- spectrumpopulationsseparately.Thesecountsarecomparedtothe Astudyofthemulti-wavelengthpropertiesof10Csourcesin de Zotti et al. and Jackson & Wall (1999) models, and they find theLockmanHole(Whittametal.2015)usedradio-to-opticalra- thatbothunderestimatethenumberofflat-spectrumsourcesbelow tiostoshowthatatleast94percentofthe10Csample(S > 15GHz 5mJy. 0.5 mJy) are radio loud, demonstrating that there is no signific- The new 15.7-GHz source count is also compared to the S3 antpopulationofstarforminggalaxiespresentatthesefluxdensity catalogue in Fig. 11. All sources with S > 0.09 mJy were levels.Giventhatthereisnochangeintheslopeofthesourcecount 18GHz selectedfromthesimulation,andthesourcecountwascalculated for0.1<S/mJy<0.5,itseemsunlikelythatstarforminggalaxies inthesamebinsasfortheobservedcount.Thesimulationunder- are making a significantly greater contribution in this flux dens- predicts the observed number of sources in a similar way to the ityrangethanathigherfluxdensities.Therefore,thepopulationof deZottietal.model.Itislikelythatthisunder-predictionisagain starforminggalaxiespredictedtobemakingasignificantcontribu- duetoalackofflatspectrumsourcesinthemodel,asWhittamet tion(∼20percent)inthisfluxdensityrangebytheS3simulation al.showedthattheS3and10Cspectralindexdistributionsaresig- donotappeartobepresent.Thisisconsistentwiththeresultsof nificantlydifferent,withthesimulationmissingalmostalltheflat severalrecentstudiesofthefaint(S <0.1mJy)sourcepop- 1.4GHz spectrumsources.(NotethattheS3 anddeZottietal.modelsare ulation at lower frequencies (Simpson et al. 2012; Lindsay et al. notentirelyindependentastheyarebothextrapolationsfrommod- 2014; Luchsinger et al. 2015) which have also found fewer star- els constructed using low-frequency data.) Whittam et al. (2015) forminggalaxiesthanpredictedbythesimulation. MNRASinpress,1–11(2015)