Extending the infrared radio correlation

0.35 MB·English

by  B. J. Boyle

Mon.Not.R.Astron.Soc.000,1–8(2004) Printed5February2008 (MNLATEXstylefilev2.2)
Extending the infrared radio correlation
B.J.Boyle1, T.J. Cornwell1, E.Middelberg1, R.P.Norris1, P.N.Appleton,2, Ian Smail3
1 Australia Telescope National Facility, PO Box 76, Epping, NSW 1710, Australia
2 SpitzerScience Center, California Institute of Technology, 1200 E.California Blvd., Pasadena, CA 91125, USA
3 Institute for Computational Cosmology, University of Durham, South Road, Durham DH13LE, UK
7
0
0
2
5February2008

n
a
J ABSTRACT

2 Co-addition of deep (rms ∼ 30µJy) 20cm data obtained with the Australia Tele-
2 scope Compact Array at the location of Spitzer Wide field survey (SWIRE) sources

has yielded statistics of radio source counterparts to faint 24µm sources in stacked
1
images with rms < 1µJy. We confirm that the infrared-radio correlation extends
v
2 to f24µm = 100µJy but with a significantly lower coefficient, f20cm = 0.039f24µm
0 (q24 =log(f24µm/f20cm)=1.39±0.02)thanhitherto reported.We postulatethatthis
6 may be due to a change in the mean q24 value ratio for objects with f24µm <1mJy.
1
Key words: infrared: galaxies – radio continuum: general – galaxies: starburst
0
7
0
/
h
p 1 INTRODUCTION f20cm =115mJyrespectively.Althoughlimitedbytheavail-
- ability of spectroscopic redshifts, Appleton et al. were able
o The tight correlation between the global far-infrared and to demonstrate a good correlation between the 24µm and
tr radio emission from galaxies has been established for over 20cmfluxforobjectswithamedianz=0.3extendingoutto
s three decades. Early ground-based studies (van der Kruit z 1. Their analysis was limited by the relative brightness
:a 1973;Condonetal.1982)andobservationswiththeinfrared of∼the radio flux limit, which was a factor of three brighter
v Astronomical Satellite (IRAS, Dickey & Salpeter 1984; de than that corresponding to the 24µm limit, based on the
Xi Jong et al. 1985) conclusively established a correlation over observed far-infrared/radio flux ratio.
a broad range of Hubbletypes and luminosities at low red- In this analysis, we extend the flux limits to which the far-
r
a shifts (z < 0.1). The origin of this correlation is thought infrared/radiocorrelationmaybestudiedbyafurtherfactor
tolie in thelink between massive stars, which generate far- of three at 24µm and by a factor of 20 at 20cm, using the
infrared emission by reheating dust,and supernovae,which Spitzer Wide Field Survey (SWIRE) and the technique of
acceleratecosmicraysthatthengenerateradiosynchrotron stacking(seee.g.Walsetal.2005)appliedtodeepAustralia
radiation (Harwit & Pacini 1975; Condon, 1992; Xu,Lisen- Telescope Compact Array (ATCA) observations.
field & Volk 1994). However, alternative mechanisms have
alsobeenproposed,includingsecondaryelectronproduction
(radio synchrotron) from molecular clouds (far infrared). 2 DATA
Morerecently,byextrapolatingobservationsmadewiththe
2.1 ATCA observations
Infrared Space Observatory (ISO) in the mid-infrared (Co-
henetal.2000,Garrett2002)andinthesub-millimetrewith TworegionsaroundtheChandraDeepFieldSouth(CDFS,
the SCUBA instrument, (Carilli & Yun 1999, Ivison et al. RA(2000)=3h31m Dec(2000)= 28◦06m) and the European
2002), it has been argued that the far-infrared/radio corre- Large Area ISO Survey S1−(ELAIS, RA(2000)=0h34,
lation extendsto z>1. Ifconfirmed, this suggests that the DEC(2000)= 43◦45m, have been observed with a series of
galaxies at highredshift maysharemany ofthesameprop- mosaiced poin−tings at 20cm with the ATCA as part of the
ertiesastheirlowredshiftcounterparts;whatevertheorigin Australia Telescope Large Area Survey (ATLAS, Norris et
of this relation. al. 2006; Middelberg et al. 2007, in prep). The goal of the
Withthegreatlyenhancedmid-infraredsensitivityprovided ATLASprogramistoprovidelargearea(6deg2),deepsur-
by the Spitzer Satellite, it is now possible to investigate di- veys of radio sources in fields which are also the subject of
rectlythefar-infrared/radiocorrelationatredshiftsoforder intensivestudyat otherwavebands.Inparticular,theareas
unity. An initial attempt has already been made by Ap- covered are chosen to be matched to those covered by the
pleton et al. (2004), using the Spitzer First Look Survey SWIREsurvey(Lonsdaleetal.2003)aswellasenablingop-
(FLS) and the Very Large Array (VLA). The flux limits tical spectroscopy with instrumentssuch astheAngloAus-
for the FLS and VLA samples were f24µm = 0.3mJy and tralian Telescope 2-degree field. A total of 360 hours inte-
2 B.J.Boyle et al.
Figure 1.ATCA20cmimagesoftheCDFSfield(left)andtheELAISfield(right).Theareasusedintheseanalyses isdenotedbythe
circles.
grationhasbeenobtainedontheCDFS,andof234hoursin data Release 31). Here we use only the 24µm MIPS data.
the ELAIS. In each field, 28 and 20 pointings (each with a The fluxes used here are, in the case of unresolved or noisy
mean integration of 13 hours) have been mosaiced together sources,aperture-correctedfluxes,asdescribedbySuraceet
toprovideanareaofapproximately1.6deg 2.2deginthe al. (2006). In the case of extended sources, Kron fluxes are
×
CDFS, and of 1.6deg 1.6degin theELAIS. used (Kron 1980).
×
The data were calibrated and imaged using Miriad (Sault
et al. 1995) following the procedures in the Miriad cook-
book for high dynamic range observations. In particular,
3 STACKING ANALYSIS
multi-frequency clean had to be used to account for the
high fractional bandwidth of 15 per cent. The mean rms 3.1 Method
noise in the images is 30µJy, and is limited not by integra-
We present here in detail the analysis of the radio and in-
tion time but by sidelobes arising from strong sources in
frared data obtained in the CDFS and ELAIS field. Both
(CDFS) or just outside (ELAIS) the imaged areas (shown
fieldswereanalysedseparately.Amoredetailedanalysiswas
in Fig 1). The CDFS image also contains data from an
conductedontheCDFSfieldbecauseofthemoreextensive
earlier observation by Koekemoer et al. (in preparation),
data set (deeper SWIRE data, optical CCD data), but the
who have covered a region of 1 deg diameter centred on
RA(2000)=3h33, Dec(2000)= 27◦ with a mosaic of seven ELAIS field provided a useful independent check of the re-
− sults.
pointings, and reach an rms noise level of 15µJy. In or-
∼ We first removed galactic stars from the SWIRE catalogue
der to avoid regions around the edge of the mosaic with
in both fields. For the CDFS we used the optical-infrared
increased noise levels, we restricted the analysis to a 48-
colour (r i/r 3.6µm) selection method employed by
arcmin radius centred on theATLAS field centres, denoted
− −
Rowan-Robinson et al. (2004). Optical magnitudes for all
with circles in Fig 1. As a further check of the calibration,
SWIRE sources were obtained from CTIO CCD observa-
we confirmed that there was no significant offset between
tions(Sianaet al. in preparation).Atbright optical magni-
the published NVSS fluxes and the flux estimates obtained
tudes(r<17),image saturation renderstheoptical magni-
from the ATCA data for bright sources (f20cm >2mJy) in
tudesunreliableandsotheUKSTR-bandskysurveyimages
theCDFS field (see also Norris et al. 2006)
ofbrightSWIRE(f24µm >1000µJy)sourceswereinspected
visually usingtheSuperCOSMOSskyservertoremoveany
further stellar contamination. For the ELAIS field we did
not have access to reliable broadband optical magnitudes
2.2 Spitzer observations and so relied on the stellar classification provided by the
SuperCOSMOS catalogue (confirmed by visual inspection
TheSpitzerSpaceTelescopewaslaunchedinAugust2003.It for the brighter images). Reliable star/galaxy classification
is equipped with several instruments, and here we use data is only possible at r < 20mag from these measurements,
taken with the Multiband Imaging Photometer for Spitzer and so it is possible that some residual stellar classification
(MIPS) in its 24µm band. Before launch, proposals were remains amongst thefainter optical counterparts.
invited for Legacy Science Programs, which were large co- Using AIPS++ we then extracted 43 43 arcsec2 sub-
×
herentprogramswhosedatawouldbeoflastingimportance images from the 20 cm images, centred on each remaining
tothebroadastronomicalcommunity.OneofthesixLegacy SWIREextragalacticsourceabovethe5σ24µmfluxlimitin
projects chosen was the SWIRE (Spitzer Wide Extragalac- the CDFS (f24µm >100µJy) and ELAIS (f24µm >400µJy)
tic) Survey (Lonsdale et al. 2003), which has observed a fieldscontained within the48-arcmin radius region.
region of 6 square degrees surrounding the CDFS, and of 5 The20cmimageswerethensortedintologarithmic(0.1dex)
square degrees coincident with the ELAIS field. The analy-
sisofthosedataaredescribedbySuraceetal.(2006).Most
of the SWIRE data are now in the public domain (SWIRE 1 http://swire.ipac.caltech.edu/swire/astronomers/data access.html
Extending the infrared radio correlation 3
Table 1.FluxmeasurementsofstackedimagesintheCDFSfield
quiet-sourcestack all-source stack
f24µm Number f20cm σ (20cm) Number f20cm σ (20cm)
(µJy) (µJy) (µJy) (µJy) (µJy)
115.2 200 5.91 2.57 205 6.37 2.55
145.1 568 4.88 1.60 581 5.21 1.61
180.8 1278 5.43 1.01 1317 6.90 1.01
225.3 1972 8.71 0.91 2039 10.03 0.90
281.0 2198 11.73 0.87 2295 13.60 0.89
351.5 1619 14.36 0.98 1685 15.77 0.98
442.2 1193 18.18 1.21 1267 21.42 1.17
556.6 791 21.45 1.62 841 25.59 1.57
699.4 479 25.64 1.90 527 30.65 1.97
870.3 296 33.70 2.57 325 37.48 2.47
1105.3 194 29.92 2.80 241 39.30 2.55
1383.7 120 38.01 3.82 151 54.13 3.44
1722.9 73 50.37 5.67 97 65.46 4.95
2250.8 30 56.59 7.30 54 91.15 5.31
2858.9 20 46.54 10.05 51 128.19 6.71
24µmfluxbins.Wecomputedastackedradioimageforeach sityofover105sourcesdeg−2atf24µm =10µJy,approaching
binbyformingthemedianofcorrespondingpixelsinallra- 10 sources per stacked image. A confusion noise threshold
dio images in that bin. For each bin, two stacks were cre- of 0.5µJy whenaddedinquadraturetoan assumedPois-
∼
ated: one which included all sources (the all-source stack) sonian reduction in the noise afforded by the stacking pro-
and one which excluded those images whose central pixels cedure, i.e. 30µJy/√n, is consistent with the observed rms
hadf20cm >100µJy(thequiet-source stack).Theradioflux noise measurements.
foreachbinwasthenestimatedfromthemeasuredpeakflux The images from the quiet-source stacks in the CDFS field
in the stacked image. We also measured fluxes by fitting a areshowninFig2,andthemedianvaluesoff24µmandf20cm
pointsourcetotheimage. Thisyieldedidenticalresults,in- are plotted in Fig. 4. Corresponding images of simulated
dicatingthatthestackedimageswerenotresolved.Wealso data are shown in Fig 3, and their f20cm vs f24µm plots
confirmedthatstackingsimilarnumberofrandompositions areshowninFig. 5.Thequalitativeagreement between the
within CDFS field produced no significant detections. Im- corresponding real and simulated images is good but not
ages of the quiet-source stacks are shown in Fig 2. excellent. The simulations do not reproduce the fine scale
To estimate the accuracy of the flux determinations, we noise in the real data, which we believe is partially due to
created simulated radio images in AIPS++ with constant theresidualeffectsofthebrightsourcetothesouthwest.In
noise levels(rms=30µJy) across theCDFS field.Tothiswe supportofthisconclusion,wenotethatbymovingthefield
added sources at the positions of all SWIRE sources with centre closer to that source the level of the fine scale noise
f24µm > 100µJy with a scaled, f20cm = 0.04f24µm, radio increases slightly in thestacked images.
flux. We repeated the analysis above, computing both all- Nevertheless, we note that theflux estimates from thesim-
source and quiet-source stacks. The simulated median im- ulateddata(boththeall-sourceandquiet-sourcestacks)are
ages are shown in Fig 3. Asa further check on our analysis fitted bya linear relation
processes, we repeated the simulation by adding the artifi-
cialsourcestosimulatedu–v data,whichwethenprocessed
in an identical way to the real u–v data. This yielded an f20cm =0.044f24µm+4.74µJy
essentially identical result to thesimulations that had been which is essentially identical to the input simulation. We
performed in theimage plane. conclude that the stacking introduces no systematic bias
intoour flux measurements.
We note that the f20cm v f24µm correlation flattens off at
3.2 CDFS field the brightest flux bins for the quiet-source stack. This re-
veals the systematic under-estimation of the radio flux in
The resultant fluxes for the stacked images in the CDFS
thequiet-sourcestackcausedbytheexclusionoftheincreas-
are listed in Table 1, together with the median infrared
ingly large number of radio sources (> 20 per cent) at the
source flux f24µm in each bin. In many of the fainter f24µm f20cm >100µJy levelthatare bona-fidecounterpartstothe
bins, we have been able to stack over 1000 images. How-
SWIRE sources at high f24µm fluxes. Note that this bias
ever, we note that the rms in the resultant stacked image
is also seen for the simulated quiet-source stacks, although
(estimatedfromtheimagebyexcludingthecentralquarter)
onlysignificant forsimulated stackswith f24µm >2000µJy.
has not decreased in line with Poisson statistics, probably
Weobtain afit toboththeall-source stack (allbins) of the
because ofconfusion from sub-µJy radio sources. Assuming
form:
theinfrared/radio correlation identifiedbelow,sourceswith
f24µm =10µJy will have f20cm 0.5µJy, and extrapolating
∼
thef24µmnumbercounts(seeTable1)predictsasourceden- f20cm =0.041( 0.002)f24µm +1.35( 0.8)µJy
± ±
4 B.J.Boyle et al.
Figure 2. Images from the quiet-source stacks. Upper left bin Figure3.Imagesfromthesimulatedquiet-sourcestacks.Upper
corresponds to faintest bin; 2 < logf24µm < 2.1. Bins increase left bin corresponds to faintest bin; 2 < logf24µm < 2.1. Bins
logf24µm =0.1fromlefttorightandtoptobottom. increaselogf24µm=0.1fromlefttorightandtoptobottom.
or an infrared-to-radio flux ratio, q24 =log(f24µm/f20cm)=
1.39 0.02. The data points derived for quiet-source stack
±
havea slightly lower normalisation throughout the infrared
flux range. This is because objects are detected above our
radio limit in all f24µm bins used in this analysis. The frac-
tionwithradiodetectionsrisesfrom2percentforthelowest
f24µm values to 40 per cent at thehighest (see Table 1).
3.3 ELAIS field
Following the same procedures we obtained a value for the
infrared-radio correlation using the ATCA and SWIRE ob-
servationsoftheELAISregion.TheELAISregion doesnot
extend to as deep a 24µm flux limit as the CDFS SWIRE
field (see Table 2). A fit to the all-source stack gives the
relation:
Figure4.Thecorrelationbetweenmedianradioflux,f20cm,and
infrared flux, f24µm, for the stacked CDFS sources. Flux mea-
f20cm =0.039( 0.004)f24µm +7.1( 3.3)µJy surementsareincludedforstackswith(filledcircles)andwithout
± ± (open circles)the inclusion of detected (>100µJy) radio flux at
for the all-source stack. The slope is consistent with the the SWIRE sources positions.The numbers denote the numbers
result obtained from the CDFS data, giving a q24 =1.41 of SWIRE source positions included in each stack). The dashed
0.04.Theinterceptisdifferentbutthisislargelydominate±d lineisthe best-fitlinetothedata obtained fromthe ’all-source’
stack.
by fitting to the data points at the lowest flux levels where
the errors are high and the fit poorly constrained. Forcing
Extending the infrared radio correlation 5
Table 2.FluxmeasurementsofstackedimagesintheELAISfield
quiet-sourcestack all-source stack
f24µm Number f20cm σ (20cm) Number f20cm σ (20cm)
(µJy) (µJy) (µJy) (µJy) (µJy)
466.7 232 19.75 2.27 241 21.51 2.23
568.3 607 27.10 1.47 638 28.59 1.45
697.5 450 31.33 1.71 487 35.07 1.69
887.7 276 35.02 2.16 320 43.86 2.08
1100.4 122 46.06 3.20 156 59.80 2.96
1416.5 70 52.71 4.05 98 64.61 3.43
1703.8 41 51.89 5.53 53 65.55 4.77
2147.9 27 63.69 6.35 40 76.22 5.40
2811.5 13 53.32 8.61 28 121.77 6.28
Figure5.Thecorrelationbetweenmedianradioflux,f20cm,and
infraredflux,f24µm,forthesimulateddatawiththescaling0.04.
Figure7.AcompositeplotoftheCDFSfieldshownf24µm and
f20cm fluxes for individual sources (small triangles) and the all-
source (filled circles) and radio-quiet (open circles) stacked im-
ages.
thecorrelation togothroughtheorigin ofthef24µm/f20cm
relationship gives an consistent fit in both cases with q24 =
1.40 0.02.ThecorrelationisshowninFig.6andthefluxes
±
for thestacked images are given in Table 2.
4 DISCUSSION
Figure6.Thecorrelationbetweenmedianradioflux,f20cm,and The most significant result to emerge from this analysis is
infrared flux, f24µm, for the stacked ELAIS sources. Flux mea- that infrared-selected sources broadly follow the radio/far-
surementsareincludedforstackswith(filledcircles)andwithout
infrared (FIR) correlation down to microJy flux densities,
(open circles)the inclusion of detected (>100µJy) radioflux at
implying that these sources are primarily driven by star-
the SWIRE sources positions.The numbers denote the numbers
formationratherthanbyAGNactivity.Thisresultrulesout
of SWIRE source positions included in each stack). The dashed
lineisthe best-fitlinetothe data obtained fromthe ’all-source’ modelssuch asthatof Wallet al. (1986), whoproposethat
stack. the low-flux radio sources are low-luminosity radio galax-
ies. Moreover, it confirms predictions that the far-infrared
emittingstarburstsshoulddominatetheradiopopulationat
f20cm <50µJy (Gruppioni et al. 2003) based on extrapola-
6 B.J.Boyle et al.
Figure 10. MIPS 24µm images for four CDFS source with
0.8<f24µm <0.83mJyandnodetected20cmemission.Notethe
complex/extended morphology in the majorityof cases, broadly
consistentwiththepopulationatthisfluxlevelandq24 value.
Figure8.Upper:Radioflux,f20cm,andinfraredflux,f24µm,for spectral information on these objects in either the infrared
theindividualCDFSsources.FitsusingLower:Asforupperplot, orradiobands.Appletonetal.(2004)explicitlydemonstrate
with simulated sources: half with f20cm = 0; half with f20cm = thatindividualk-correctionshavelittlesystematiceffecton
0.04f24µm,basedonCDFSf24µm distribution.Seetextformore the mean trend of q24 as a function of redshift at z < 1.
details.
Based on spectral energy distribution modelling in the in-
frared, and an assumed synchrotron spectral index of 0.7
−
in the radio, Appleton et al. found that this increased the
tions of models which fit themid-infrared/radio correlation uncorrected value of q24 = 0.84 by only 0.1–0.2, depending
at brighterfluxes. on the SED template used.
The results presented here show that, in broad terms, this However, Fadda et al. (2006) demonstrate that the k-
same correlation holds for sources down to microJy lev- correction arising from the PAH emission features enter-
els, thus extending the range over which the correlation is ing the 24µm band at higher redshifts (1.1 < z < 1.8 or
knowntoholdbysometwoordersofmagnitudeinfluxden- 2.2 < z < 2.5) could be much greater. Fadda et al. (2006)
sity. However, the data do not immediately tell us whether foundthatemissioninthe24µmbandcouldbeenhancedby
thesegalaxiesarehigh-luminosityhigh-redshiftstar-forming asmuchasafactor2(δq=0.3)foranM82starburstgalaxy
galaxies or low-luminosity low-redshift star-forming galax- template.Thusitispossiblethatk-corectionprovidesapar-
ies. tial explanation for the differences seen with Appleton et
Most previous studies of the radio-FIR correlation have al. (2004), if the bulk of the population used in the stack-
measured qFIR = log(SFIR/S20cm), where SFIR is an inte- inganalysisarehighredshiftstarbursts.Fork-correctionto
gratedfarinfrared fluxtypicallyobtainedfrom IRASfluxes fully explain the difference our sources would need to be
using a weighting algorithm e.g. SFIR(Wm−2) = 1.26 dominated by high redshift (z >1) starburst galaxies with
104(2.58S60µm+S100µm)Jy.Suchstudies,whetheronradio×- PAH emission features approximately double that seen in
selected (Condon et al. 1991), infrared-selected (Condon & some of themost intense star-forming galaxies locally.
Broderick1991),ordistance-limited(Helouetal.1985)sam- Amoremundanesourceforthediscrepancymayarisefrom
ples, all broadly givethe same value of qFIR =2.3 0.2. the different sample selection effects. We note that q24 de-
±
Herewemeasureq24 =log(f24µm/f20cm),ratherthanqFIR, rivedbyAppletonet al. (2004) isbased on radio detections
so to compare our results with previous authors requires a of infrared sources. Our values derived here are for all in-
conversion factor between q24 and qFIR, which will depend frared sources, irrespective of radio flux.Thus a lower coef-
sensitively on the SED of the parent population and the ficient could be explained by a population of sources with
MIPSfilterresponse,andisbeyondthescopeofthispaper. lowradio-infraredfluxratioswhichhavepreviouslynotbeen
However, q24 has been measured by Appleton et al. (2005) includedinthestudyofradio-selectedsamplesbecausethey
who obtain a (non-k-corrected) value of q24 = 0.84, which fall below the radio flux limit for thesample.
differs significantly from our measured valueof q24 =1.39. Indeed,wenotethatGruppionietal.(2003) foundthatthe
Althoughourvalueofq isderivedfromstackeddata,whilst mid-infrared (15µm)/radio (20 cm) luminosity correlation
the Appleton value is derived from detected sources, this for mid-infrared sources with radio detections was a factor
disrepancyisneverthelessasurprisingresultandmeritsfur- oftwohigherthanthecorrelationforallsourceswithupper
ther scrutiny. First, we note that the simulations discussed limits for radio luminosities included in the analysis. How-
inSection3haveeliminatedthepossibilitythatitmightbe ever, if this were to be the case we would still have to ex-
causedbyanerrorinourimaging,stacking,orsource-fitting plainthedifferencebetweentheselowerq24(passbandissues
algorithms or software. notwithstanding) and those FIR/radio correltions derived
Anotherpossibility isthatthediscrepancymightbecaused from IRAS sources with complete radio detections (Yun,
byadifferenceink-correctionbetweenouranalysisandthat Reddy& Gruppioni 2003).
of Appleton et al. (2004). Wehavemade nok-correction to To investigate this further, we plot in Fig. 7 a composite
thisfluxratiobecausewehaveneitherredshiftsnordetailed of the radio/infrared fluxes derived for both the individ-
Extending the infrared radio correlation 7
Figure 9.R-bandUKSchmidttelescope imagesofninegalaxies withanomalouslylowradio-infraredfluxratios.Eachimageis30×30
arcsec.Fortop-lefttobottom-right,thegalaxieshavef24µm fluxesof13.5,6.99,6.06,5.97,5.75,5.29,3.98,3.97and3.86mJy,andallare
below the 3σ radiofluxlimit(f20cm <100µJy). As indicated, fourgalaxies have spectroscopic redshifts fromthe 2dFGalaxy Redshift
Survey(Collessetal.2001).
ual radio-detected sources in theCDFS (Norris et al. 2006) ineach bin andmadethefit tothosepoints.FortheCDFS
and the stacked images derived in this paper. The sample and ELAIS theresulting fits were respectively:
of detected sources includes both AGN and star forming
galaxies, resulting in a significant tail in Fig. 7 extending
to high radio-FIR ratios, and precluding a measurement of f20cm =0.037(±0.001)f24µm +2.6(±0.8)µJy Median
q24. However there is a significant absence of sources below
f20cm = 0.1f24µm, (the radio-FIR relation found by Apple- f20cm =0.041(±0.002)f24µm +6.6(±3.5)µJy Median
ton et al.), which appears to mark a lower bound to the
Themedianestimatortotheindividualpointsinbothfields
distribution. In particular, there is a significant absence of
gave a consistent slope to that obtained from the stacked
sources along the line extrapolated from our stacked data.
images.
Indeed,atthehigher(f24µm >10mJy)fluxesnearlyallnon- We also modelled the effects of a population with a bi-
stellar24µmsourcesaredetectedat20cmintheCDFSfield
modalradio-infraredfluxratio;halfofthepopulationhaving
(f20cm >0.4mJy). f20cm =0 and half having f20cm =0.08f24µm (based on the
As a further consistency check of our result we derived the CDFS MIPS flux distributions). Although the stacked im-
infrared-radiocorrelationbasedonradioandinfraredfluxes agesofthissimulatedpopulationgivesimilarresultstothat
measured at the positions of individual sources used in our obtained with thereal data, thedistribution oftheindivid-
stacking sample. Most of the radio sources will lie below ual simulated population flux values is visibly bimodal in
the formal 5σ detection limit and consequently their fluxes a scatter plot of the individual source fluxes (Fig. 8 upper
will be subject to large errors. Nevertheless the ensemble panel)andclearlydifferentfromasimilarplotusingthereal
propertiesofthedistributionwillberobust.Ineachinfrared data (Fig. 8 lower panel)
binused abovewecomputed themedian valueof thefluxes Nevertheless, weare faced with the conundrumthat,what-
8 B.J.Boyle et al.
everthedistribution,themeanobservedq24 valueisgreater CondonJ.J.,1992,ARAA,30,575
than that of any modelled SED. Some of the optically- DickeyJ.M.,Salpeter E.E.,1984,ApJ,284,461
brightest CDFS sources with the most extreme q24 values deJongT.,KleinU.,WielebinskiR.,WunderlichE.,1985,A&A,
are relatively low redshift galaxies (Fig. 9). However, most 147,6
FaddaD.etal.,2006, AJ,131,2859
of these radio-weak sources have no direct optical counter-
GarrettM.A.2002, A&A,384,19
partatphotographicplatelimits(R 21mag).Themajor-
∼ GruppioniC.,etal.,2003,MNRAS,341,1
ity of the objects at high f24µm flux end of the correlation HarwitM.,PaciniF.1975,ApJ,200,127
f24µm ∼0.83mJy andshowextendedorcomplex morpholo- HelouG.,SoiferB.T.,Rowan-RobinsonM.,1985,ApJ,298,7
giesintheir24µmimages(Fig.10).Thiswouldbeconsistent IvesonR.J.,etal.,2002,MNRAS,337,1
withapopulationdominatedbyactivelystar-forminggalax- KronR.G.,1980,ApJS,43,305
ies at moderate redshifts. Todate, optical spectroscopy has LonsdaleC.J.,etal.,2003,PASP,115,897
beenobtainedfor 130SWIREsourcesintheCDFS(Nor- MiddelbergE.,2006,PASA,2264
≈
risetal.2007).Themedianredshiftisz 0.4withatailto NorrisR.P.etal.,2006,AJ,132,2409
higher redshift z 2, exclusively compo∼sed of broad emis- NorrisR.P.etal.,2007,MNRAS,inpress
∼ Rowan-RobinsonM.etal.,2004,MNRAS,351,1290
sionlineactivegalacticnucleiatz>1.2.However,thisred-
SaultR.S.etal.,1995,ASPC,77,433S
shift distribution is strongly influenced by optical selection
SuraceJ.A.etal.,2006,AJ,submitted
effects–thevastmajorityoftheobjectswithmeasuredred-
vanderKruitP.C.,1973,A&A,29,263
shiftshavingrelativelybrightopticalmagnitudes(R<22.5
XuC.,LisenfieldU.,VolkH.J.,1994,A&A,285,19
mag). WallJ.V.,BennC.R.,GrueffG.,VignottiM.,1986,inHighlights
inAstronomy,Proc.19thIAUGen.Assembly,7,345
WalsM.,BoyleB.J.,CroomS.M.,MillerL.,SmithR.J.,Shanks
5 CONCLUSIONS T.,OutramP.,2005,MNRAS,360,453
YunM.S.,ReddyN.A.,CondonJ.J.,2001,ApJ.,554,803
Wefindthatthepopulationof objectsat µJyfluxdensities
obeyaradio-24µmcorrelationwhichindicatesthattheyare
powered by star formation rather than by AGN, but their
mean value of q24 is inconsistent with previous results. We
therefore propose that at microJy sensitivity, radio surveys
are sampling a different population of objects, or objects
withanintrinsicallydifferentvalueofq24,fromobjectsseen
inthelocalUniverse.Nevertheless,weacknowledgethatthe
changeobtainedhereforthemeanvalueofq24 occurswhen
we move from the use of discrete radio (> 5σ) detections
to fainter fluxes. Despite extensive testing and simulations
we cannot rule out some unsuspected selection effect which
may also contribute to this result. A partial explanation of
the discrepancy may arise from the k-correction, but only
if the population is dominated by extreme starburst galax-
ies at 1 < z < 1.8 or 2.2 < z < 2.5. Another explana-
tion for at least part of the discrepancy may arise from
theradio-selectionbiasesinherentinsomeprevioussurveys.
Datafromhigh-sensitivitysurveyssuchasATLASwillhelp
to resolve such issues.
ACKNOWLEDGMENTS
Radio data was obtained by the Australia Telescope Com-
pact Array, operated by CSIRO Australia Telescope Na-
tional Facility. IRS acknowledges support from the Royal
Society.
REFERENCES
AppletonP.N.,etal.,2004,ApJS,154,147
CarilliC.L.,YunM.S.,1999,ApJ,513,13
CohenJ.,etal.2000,ApJ,528,29
CollessM.M.etal.,2001,MNRAS,328,1039
Condon J.J., Condon M.A., Gisler G., Puschell J.J., 1982, ApJ,
252,102
CondonJ.J.,AndersonM.L.,HelouG.,1991, ApJ,376,95
CondonJ.J.,BroderickJ.J.1991,AJ,102,1663