Determining an accurate position for a submm galaxy (SMG) is the crucial step that enables us to move from the basic properties of an SMG sample - source counts and 2-D clustering - to an assessment of their detailed, multi-wavelength properties, their con

10Ivisonet

al.

Figure4.Deboostedsubmm uxdensityversustheangularseparationofthecounterpartsforSMGswithmorethanonerobustradioidenti cation(P 0.05)within8arcsecofthe850-µmposition.Pointsdenoteradiodoubles(D)andtriples(T).Theaverageofthesingle,unresolvedorbarelyresolvedradiocounterpartsisrepresentedbyadashedline.Thepaucityofdataatverylowandhighseparationsisduetoour nitespatialresolutionontheonehandandtoouruseofa nitesearchareaandthePstatisticontheother.Thehistogramshowsthedistributionofangularseparations,scaledarbitrarily,formultipleidenti cationsfoundintheMonte-Carlosimulationsdescribedin§5.

Wereturnnowtoourinitialmotivationforthisstudyofposi-tionaluncertainty,namelythefeasibilityofaspectroscopicredshiftdistributionforSMGsbasedonKMOSnear-IRspectroscopyofanunbiasedsample.SuchaprogrammecouldaffordtoloseoneSMGduetopositionalerrorduringasingledeploymentofthe24KMOSIFUs.EachIFUcovers2.8×2.8-arcsec2so,leavingroomforsee-ingeffects,werequire2σ～2.5arcsectoensurethat95.6percentofSMGsfallwithinthecentral5arcsec2ofeachIFU.Ourpa-rameterisationsuggeststhatthislevelofaccuracyrequiresanSMGsamplecutatSNR 20.AdoptingthesourcecountsofCoppinetal.(2006),asourcedensityof～2200deg 2–suf cienttoemployall24KMOSIFUs–wouldrequirethatweprobethe3-mJySMGpopulation;this,inturn,wouldrequirethatwedelvewellbelowthe850-µmconfusionlimittoensureSNR 20,orthatweutilisepositionsdeterminedusingthe450-µmdatathatareacquiredsi-multaneouslybySCUBA-2.OptimalexploitationofKMOSmayrequiresharingtheIFUswithotherprogrammesinallbutthedeep-estSCUBA-2survey elds.

5MULTIPLERADIOCOUNTERPARTS

AnumberofSMGswithmore

thanonerobust(P 0.05)radiocounterpartareapparentinTables1–2andFigsA1–A2:sevenintheLHand veintheSXDF.Thistendencyfor～10percentofSMGstohavemultipleradioidenti cationswasnotedpreviouslybyIvisonetal.(2002)andPopeetal.(2006).TheprobabilityofanSMGpossessingtwostatisticallysigni cantradiocounterpartswasquanti edbyplacing106fakesourcesintotherealLHandSXDFradio eldsandcountingthenumberofP<0.05radiocoun-terparts–asimpleMonte-Carloapproach.ThisrevealedthatthecalibrationofthePstatisticissecure,withP=0.05yielding5.05spuriousassociationsforevery100fakeSMGs.Multiplerobustcounterpartsarefarrarer,however.Forevery100fakeSMGsthesimulationssuggestthatonly0.22willhavemorethanonesecure

Figure5.Histogramofdeboostedsubmm uxdensityforthefullSHADESSourceCatalogue.Cross-hatchedareasrepresentthe12SMGswithtwoormoreradiocomponentswithin8arcsecofthe850-µmposition,associatedrobustlywiththeSMG(P 0.05);single-hatchedareasrepresentthesevenSMGswithmultiple,signi cant24-µmidenti cations.FiveSMGshavemultiple,signi cantradioand24-µmidenti cations.

radioidenti cationbychance,a guredominatedbydoubles,soat rstsighttheobservedtendencyformultiplerobustradiocoun-terpartsishighlysigni cant.However,weknowthataroundhalf(65)oftheSHADESSMGshavearealassociationwitharadioemitter,or59afteraccountingforthesixspuriousidenti cationsweexpect(0.05×120),soshouldwebesurprisedto ndadozenSMGswithmultipleradioidenti cations?Oftheradio-identi edSHADESSMGs,5percentwillbespuriouslyassociatedwithan-otherradiosource.Wethusexpectthreemultipleidenti cationswhereasweseeadozen:asigni cantdifference.

Lookingatthisanotherway,thefractionofradio-identi edSMGswithmultipleradiocounterpartsis18.5±5.3percent(12/65),15.4±4.9percent(10)withseparationsbelow6arcsec.Howfrequentaresuchcasesamongstthegeneralradiopopu-lation?TheproportionofradiosourcesintheSHADES eldswithradiocompanionswithin4,6,8and10arcsecare(cumula-tively)1.2±0.3,3.9±0.5,7.1±0.6and10.3±0.7percent(Pois-sonuncertainties).ThenumberofSMGswithseparationsbelow10arcsec,andparticularlybelow6arcsec,isthussigni cant.Inter-estingly,brightSMGsmakeuponeinsevenofallradiomultipleswithseparationsbelow6arcsec.

Whatcausesthismultiplicity?Atleastthreemechanismscouldberesponsible:AGN-drivenjets;physicalinteractions;andconfusion.

Discriminatingbetweenthesemechanismsisextremelydif -cult.The rst–jets–couldberevealedviatheirmorphologyortheirradiospectralindex,buttodateneitherpropertyhasbeenprobedforasigni cantsample.Thespectroscopicevidencere-quiredtorevealthesecondpossibility–aphysicalassociation–isavailableonlyrarelyintheSHADES elds,althoughanumberoflinked,multiplesystemswithfew-arcsecseparationsandnear-identicalredshiftshavebeendocumentedelsewhere(Ivisonetal.1998,2000;Ledlowetal.2002;Nerietal.2003;Smailetal.2003a;Chapmanetal.2005;Tacconietal.2006)whichleaveslittledoubtthatmanySMGswithmultipleradioidenti cationsareinteraction-drivenstarburstswithseparationsoften(orafewtens)ofkpc.

Fig.4showsaplotofsubmm uxdensityversusangularsep-

c0000RAS,MNRAS000,000–000