Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.atnf.csiro.au/people/Enno.Middelberg/pubs/boyle2007a/boyle2007a.pdf Äàòà èçìåíåíèÿ: Tue Jan 23 03:15:56 2007 Äàòà èíäåêñèðîâàíèÿ: Wed Dec 26 11:05:44 2007 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: m 31

Page 1 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Mon. Not. R. Astron. Soc. 000, 1­8 (2004)

Printed 13 January 2007

A (MN L TEX style file v2.2)

Extending the infrared radio correlation
B.J.Boyle1 , T.J. Cornwell1 , E.Middelberg1 , R.P.Norris1 , P.N.Appleton,2 , Ian Smail 1
2 3

3

Australia Telescope National Facility, PO Box 76, Epping, NSW 1710, Australia Spitzer Science Center, California Institute of Technology, 1200 E.California Blvd., Pasadena, CA 91125, USA Institute for Computational Cosmology, University of Durham, South Road, Durham DH1 3LE, UK

13 January 2007

ABSTRACT

Co-addition of deep (rms 30µJy) 20cm data obtained with the Australia Telescope 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 images with rms < 1µJy. We confirm that the infrared-radio correlation extends to f24µm = 100µJy but with a significantly lower coefficient, f20cm = 0.039f24µm (q24 = log(f24µm /f20cm ) = 1.39 ± 0.02) than hitherto reported. We postulate that this may be due to a change in the mean q24 value ratio for ob jects with f24µm < 1mJy. Key words: infrared: galaxies ­ radio continuum: general ­ galaxies: starburst

1

INTRODUCTION

The tight correlation between the global far-infrared and radio emission from galaxies has been established for over three decades. Early ground-based studies (van der Kruit 1973; Condon et al. 1982) and observations with the infrared Astronomical Satellite (IRAS, Dickey & Salpeter 1984; de Jong et al. 1985) conclusively established a correlation over a broad range of Hubble types and luminosities at low redshifts (z < 0.1). The origin of this correlation is thought to lie in the link between massive stars, which generate farinfrared emission by reheating dust, and supernovae, which accelerate cosmic rays that then generate radio synchrotron radiation (Harwit & Pacini 1975; Condon, 1992; Xu, Lisenfield & Volk 1994). However, alternative mechanisms have also been proposed, including secondary electron production (radio synchrotron) from molecular clouds (far infrared). More recently, by extrapolating observations made with the Infrared Space Observatory (ISO) in the mid-infrared (Cohen et al. 2000, Garrett 2002) and in the sub-millimetre with the SCUBA instrument, (Carilli & Yun 1999, Ivison et al. 2002), it has been argued that the far-infrared/radio correlation extends to z > 1. If confirmed, this suggests that the galaxies at high redshift may share many of the same properties as their low redshift counterparts; whatever the origin of this relation. With the greatly enhanced mid-infrared sensitivity provided by the Spitzer Satellite, it is now possible to investigate directly the far-infrared/radio correlation at redshifts of order unity. An initial attempt has already been made by Appleton et al. (2004), using the Spitzer First Look Survey (FLS) and the Very Large Array (VLA). The flux limits for the FLS and VLA samples were f24µm = 0.3mJy and
c 2004 RAS

f20cm = 115mJy respectively. Although limited by the availability of spectroscopic redshifts, Appleton et al. were able to demonstrate a good correlation between the 24µm and 20cm flux for ob jects with a median z = 0.3 extending out to z 1. Their analysis was limited by the relative brightness of the radio flux limit, which was a factor of three brighter than that corresponding to the 24µm limit, based on the observed far-infrared/radio flux ratio. In this analysis, we extend the flux limits to which the farinfrared/radio correlation may be studied by a further factor of three at 24µm and by a factor of 20 at 20 cm, using the Spitzer Wide Field Survey (SWIRE) and the technique of stacking (see e.g. Wals et al. 2005) applied to deep Australia Telescope Compact Array (ATCA) observations.

2 2.1

DATA ATCA observations

Two regions around the Chandra Deep Field South (CDFS, RA(2000)=3h 31m Dec(2000)=-28 06m ) and the European Large Area ISO Survey S1 (ELAIS, RA(2000)=0h34, DEC(2000)=-43 45m , have been observed with a series of mosaiced pointings at 20cm with the ATCA as part of the Australia Telescope Large Area Survey (ATLAS, Norris et al. 2006; Middelberg et al. 2007, in prep). The goal of the ATLAS program is to provide large area (6 deg2 ), deep surveys of radio sources in fields which are also the sub ject of intensive study at other wavebands. In particular, the areas covered are chosen to be matched to those covered by the SWIRE survey (Lonsdale et al. 2003) as well as enabling optical spectroscopy with instruments such as the Anglo Australian Telescope 2-degree field. A total of 360 hours inte-


Page 2 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

2

B.J.Boyle et al.

Figure 1. ATCA 20cm images of the CDFS field (left) and the ELAIS field (right). The areas used in these analyses is denoted by the circles.

gration has been obtained on the CDFS, and of 234 hours in the ELAIS. In each field, 28 and 20 pointings (each with a mean integration of 13 hours) have been mosaiced together to provide an area of approximately 1.6 deg â 2.2 deg in the CDFS, and of 1.6 deg â 1.6 deg in the ELAIS. The data were calibrated and imaged using Miriad (Sault et al. 1995) following the procedures in the Miriad cookbook for high dynamic range observations. In particular, multi-frequency clean had to be used to account for the high fractional bandwidth of 15 per cent. The mean rms noise in the images is 30µJy, and is limited not by integration time but by sidelobes arising from strong sources in (CDFS) or just outside (ELAIS) the imaged areas (shown in Fig 1). The CDFS image also contains data from an earlier observation by Koekemoer et al. (in preparation), who have covered a region of 1 deg diameter centred on RA(2000)=3h 33, Dec(2000)=-27 with a mosaic of seven pointings, and reach an rms noise level of 15µJy. In order to avoid regions around the edge of the mosaic with increased noise levels, we restricted the analysis to a 48arcmin radius centred on the ATLAS field centres, denoted with circles in Fig 1. As a further check of the calibration, we confirmed that there was no significant offset between the published NVSS fluxes and the flux estimates obtained from the ATCA data for bright sources (f20cm > 2 mJy) in the CDFS field (see also Norris et al. 2006)

square degrees coincident with the ELAIS field. The analysis of those data are described by Surace et al. (2006). Most of the SWIRE data are now in the public domain (SWIRE data Release 31 ). Here we use only the 24µm MIPS data. The fluxes used here are, in the case of unresolved or noisy sources, aperture-corrected fluxes, as described by Surace et al. (2006). In the case of extended sources, Kron fluxes are used (Kron 1980).

3 3.1

STACKING ANALYSIS Metho d

2.2

Spitzer observations

The Spitzer Space Telescope was launched in August 2003. It is equipped with several instruments, and here we use data taken with the Multiband Imaging Photometer for Spitzer (MIPS) in its 24µm band. Before launch, proposals were invited for Legacy Science Programs, which were large coherent programs whose data would be of lasting importance to the broad astronomical community. One of the six Legacy pro jects chosen was the SWIRE (Spitzer Wide Extragalactic) Survey (Lonsdale et al. 2003), which has observed a region of 6 square degrees surrounding the CDFS, and of 5

We present here in detail the analysis of the radio and infrared data obtained in the CDFS and ELAIS field. Both fields were analysed separately. A more detailed analysis was conducted on the CDFS field because of the more extensive data set (deeper SWIRE data, optical CCD data), but the ELAIS field provided a useful independent check of the results. We first removed galactic stars from the SWIRE catalogue in both fields. For the CDFS we used the optical-infrared colour (r - i/r - 3.6µm) selection method employed by Rowan-Robinson et al. (2004). Optical magnitudes for all SWIRE sources were obtained from CTIO CCD observations (Siana et al. in preparation). At bright optical magnitudes (r < 17), image saturation renders the optical magnitudes unreliable and so the UKST R-band sky survey images of bright SWIRE (f24µm > 1000µJy) sources were inspected visually using the SuperCOSMOS sky server to remove any further stellar contamination. For the ELAIS field we did not have access to reliable broadband optical magnitudes and so relied on the stellar classification provided by the SuperCOSMOS catalogue (confirmed by visual inspection for the brighter images). Reliable star/galaxy classification is only possible at r < 20mag from these measurements,

1

http://swire.ipac.caltech.edu/swire/astronomers/data access.html c 2004 RAS, MNRAS 000, 1­8


Page 3 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Extending the infrared radio correlation
Table 1. Flux measurements of stacked images in the CDFS field quiet-source stack Numb er f20cm (20cm) (µJy) (µJy) 200 568 278 972 198 619 193 791 479 296 194 120 73 30 20 5.91 4.88 5.43 8.71 11.73 14.36 18.18 21.45 25.64 33.70 29.92 38.01 50.37 56.59 46.54 2.57 1.60 1.01 0.91 0.87 0.98 1.21 1.62 1.90 2.57 2.80 3.82 5.67 7.30 10.05 al l-source stack Number f20cm (20cm) (µJy) (µJy) 205 581 1317 2039 2295 1685 1267 841 527 325 241 151 97 54 51 6.37 5.21 6.90 10.03 13.60 15.77 21.42 25.59 30.65 37.48 39.30 54.13 65.46 91.15 128.19 2.55 1.61 1.01 0.90 0.89 0.98 1.17 1.57 1.97 2.47 2.55 3.44 4.95 5.31 6.71

3

f24µm (µJy) 115.2 145.1 180.8 225.3 281.0 351.5 442.2 556.6 699.4 870.3 1105.3 1383.7 1722.9 2250.8 2858.9

1 1 2 1 1

and so it is possible that some residual stellar classification remains amongst the fainter optical counterparts. Using AIPS++ we then extracted 43 â 43 arcsec2 subimages from the 20 cm images, centred on each remaining SWIRE extragalactic source above the 5 24µm flux limit in the CDFS (f24µm > 100µJy) and ELAIS (f24µm > 400µJy) fields contained within the 48-arcmin radius region. The 20cm images were then sorted into logarithmic (0.1 dex) 24µm flux bins. We computed a stacked radio image for each bin by forming the median of corresponding pixels in all radio images in that bin. For each bin, two stacks were created: one which included all sources (the al l-source stack) and one which excluded those images whose central pixels had f20cm > 100µJy (the quiet-source stack). The radio flux for each bin was then estimated from the measured peak flux in the stacked image. We also measured fluxes by fitting a point source to the image. This yielded identical results, indicating that the stacked images were not resolved. We also confirmed that stacking similar number of random positions within CDFS field produced no significant detections. Images of the quiet-source stacks are shown in Fig 2. To estimate the accuracy of the flux determinations, we created simulated radio images in AIPS++ with constant noise levels (rms=30µJy) across the CDFS field. To this we added sources at the positions of all SWIRE sources with f24µm > 100µJy with a scaled, f20cm = 0.04f24µm , radio flux. We repeated the analysis above, computing both al lsource and quiet-source stacks. The simulated median images are shown in Fig 3. As a further check on our analysis processes, we repeated the simulation by adding the artificial sources to simulated u­v data, which we then processed in an identical way to the real u­v data. This yielded an essentially identical result to the simulations that had been performed in the image plane.

bins, we have been able to stack over 1000 images. However, we note that the rms in the resultant stacked image (estimated from the image by excluding the central quarter) has not decreased in line with Poisson statistics, probably because of confusion from sub-µJy radio sources. Assuming the infrared/radio correlation identified below, sources with f24µm = 10µJy will have f20cm 0.5µJy, and extrapolating the f24µm number counts (see Table 1) predicts a source density of over 105 sources deg-2 at f24µm = 10µJy, approaching 10 sources per stacked image. A confusion noise threshold of 0.5µJy when added in quadrature to an assumed Poissonian reduction in the noise afforded by the stacking pro cedure, i.e. 30µJy/ n, is consistent with the observed rms noise measurements. The images from the quiet-source stacks in the CDFS field are shown in Fig 2, and the median values of f24µm and f20cm are plotted in Fig. 4. Corresponding images of simulated data are shown in Fig 3, and their f20cm vs f24µm plots are shown in Fig. 5. The qualitative agreement between the corresponding real and simulated images is good but not excellent. The simulations do not reproduce the fine scale noise in the real data, which we believe is partially due to the residual effects of the bright source to the southwest. In support of this conclusion, we note that by moving the field centre closer to that source the level of the fine scale noise increases slightly in the stacked images. Nevertheless, we note that the flux estimates from the simulated data (both the al l-source and quiet-source stacks) are fitted by a linear relation

f20cm = 0.044f24

µm

+ 4.74µJy

3.2

CDFS field

The resultant fluxes for the stacked images in the CDFS are listed in Table 1, together with the median infrared source flux f24µm in each bin. In many of the fainter f24µm
c 2004 RAS, MNRAS 000, 1­8

which is essentially identical to the input simulation. We conclude that the stacking introduces no systematic bias into our flux measurements. We note that the f20cm v f24µm correlation flattens off at the brightest flux bins for the quiet-source stack. This reveals the systematic under-estimation of the radio flux in the quiet-source stack caused by the exclusion of the increasingly large number of radio sources (> 20 per cent) at the


Page 4 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

4

B.J.Boyle et al.

Figure 2. Images from the quiet-source stacks. Upp er left bin corresp onds to faintest bin; 2 < log f24µm < 2.1. Bins increase log f24µm = 0.1 from left to right and top to b ottom.

Figure 3. Images from the simulated quiet-source stacks. Upp er left bin corresp onds to faintest bin; 2 < log f24µm < 2.1. Bins increase log f24µm = 0.1 from left to right and top to b ottom.

f20cm > 100µJy level that are bona-fide counterparts to the SWIRE sources at high f24µm fluxes. Note that this bias is also seen for the simulated quiet-source stacks, although only significant for simulated stacks with f24µm > 2000 µJy. We obtain a fit to both the al l-source stack (all bins) of the form: f20cm = 0.041(±0.002)f24 + 1.35(±0.8)µJy

field (see Table 2). A fit to the al l-source stack gives the relation: f20cm = 0.039(±0.004)f24 + 7.1(±3.3)µJy

µm

µm

or an infrared-to-radio flux ratio, q24 = log(f24µm /f20cm ) = 1.39 ± 0.02. The data points derived for quiet-source stack have a slightly lower normalisation throughout the infrared flux range. This is because ob jects are detected above our radio limit in all f24µm bins used in this analysis. The fraction with radio detections rises from 2 per cent for the lowest f24µm values to 40 per cent at the highest (see Table 1).

for the al l-source stack. The slope is consistent with the result obtained from the CDFS data, giving a q24 = 1.41 ± 0.04. The intercept is different but this is largely dominated by fitting to the data points at the lowest flux levels where the errors are high and the fit poorly constrained. Forcing the correlation to go through the origin of the f24µm / f20cm relationship gives an consistent fit in both cases with q24 = 1.40 ± 0.02. The correlation is shown in Fig. 6 and the fluxes for the stacked images are given in Table 2.

4 3.3 ELAIS field we obtained a value for the the ATCA and SWIRE obThe ELAIS region does not limit as the CDFS SWIRE

DISCUSSION

Following the same procedures infrared-radio correlation using servations of the ELAIS region. extend to as deep a 24µm flux

The most significant result to emerge from this analysis is that infrared-selected sources broadly follow the radio/farinfrared (FIR) correlation down to microJy flux densities, implying that these sources are primarily driven by starformation rather than by AGN activity. This result rules out models such as that of Wall et al. (1986), who propose that
c 2004 RAS, MNRAS 000, 1­8


Page 5 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Extending the infrared radio correlation
Table 2. Flux measurements of stacked images in the ELAIS field quiet-source stack Numb er f20cm (20cm) (µJy) (µJy) 232 607 450 276 122 70 41 27 13 19.75 27.10 31.33 35.02 46.06 52.71 51.89 63.69 53.32 2.27 1.47 1.71 2.16 3.20 4.05 5.53 6.35 8.61 al l-source stack Number f20cm (20cm) (µJy) (µJy) 241 638 487 320 156 98 53 40 28 21.51 28.59 35.07 43.86 59.80 64.61 65.55 76.22 121.77 2.23 1.45 1.69 2.08 2.96 3.43 4.77 5.40 6.28

5

f24µm (µJy) 466.7 568.3 697.5 887.7 1100.4 1416.5 1703.8 2147.9 2811.5

Figure 4. The correlation between median radio flux, f20cm , and infrared flux, f24µm , for the stacked CDFS sources. Flux measurements are included for stacks with (filled circles) and without (op en circles) the inclusion of detected (> 100µJy) radio flux at the SWIRE sources p ositions. The numbers denote the numb ers of SWIRE source p ositions included in each stack). The dashed line is the b est-fit line to the data obtained from the 'al l-source' stack.

Figure 5. The correlation between median radio flux, f20cm , and infrared flux, f24µm , for the simulated data with the scaling 0.04.

the low-flux radio sources are low-luminosity radio galaxies. Moreover, it confirms predictions that the far-infrared emitting starbursts should dominate the radio population at f20cm < 50µJy (Gruppioni et al. 2003) based on extrapolations of models which fit the mid-infrared/radio correlation at brighter fluxes. The results presented here show that, in broad terms, this same correlation holds for sources down to microJy levels, thus extending the range over which the correlation is known to hold by some two orders of magnitude in flux density. However, the data do not immediately tell us whether these galaxies are high-luminosity high-redshift star-forming galaxies or low-luminosity low-redshift star-forming galaxies. Most previous studies of the radio-FIR correlation have measured qFIR = log(SFIR /S20cm ), where SFIR is an integrated far infrared flux typically obtained from IRAS fluxes
c 2004 RAS, MNRAS 000, 1­8

using a weighting algorithm e.g. SFIR (Wm-2 ) = 1.26 â 104 (2.58S60µm + S100µm )Jy. Such studies, whether on radioselected (Condon et al. 1991), infrared-selected (Condon & Broderick 1991), or distance-limited (Helou et al. 1985) samples, all broadly give the same value of qF I R = 2.3 ± 0.2. Here we measure q24 = log(f24µm /f20cm ), rather than qF I R , so to compare our results with previous authors requires a conversion factor between q24 and qF I R , which will depend sensitively on the SED of the parent population and the MIPS filter response, and is beyond the scope of this paper. However, q24 has been measured by Appleton et al. (2005) who obtain a (non-k-corrected) value of q24 = 0.84, which differs significantly from our measured value of q24 = 1.39. Although our value of q is derived from stacked data, whilst the Appleton value is derived from detected sources, this disrepancy is nevertheless a surprising result and merits further scrutiny. First, we note that the simulations discussed in Section 3 have eliminated the possibility that it might be caused by an error in our imaging, stacking, or source-fitting algorithms or software. Another possibility is that the discrepancy might be caused by a difference in k-correction between our analysis and that


Page 6 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

6

B.J.Boyle et al.

Figure 6. The correlation between median radio flux, f20cm , and infrared flux, f24µm , for the stacked ELAIS sources. Flux measurements are included for stacks with (filled circles) and without (op en circles) the inclusion of detected (> 100µJy) radio flux at the SWIRE sources p ositions. The numbers denote the numb ers of SWIRE source p ositions included in each stack). The dashed line is the b est-fit line to the data obtained from the 'al l-source' stack.

Figure 8. Upp er: Radio flux, f20cm , and infrared flux, f24µm , for the individual CDFS sources. Fits using Lower: As for upp er plot, with simulated sources: half with f20cm = 0; half with f20cm = 0.04f24µm , based on CDFS f24µm distribution. See text for more details.

Figure 7. A comp osite plot of the CDFS field shown f24µm and f20cm fluxes for individual sources (small triangles) and the al lsource (filled circles) and radio-quiet (open circles) stacked images.

of Appleton et al. (2004). We have made no k-correction to this flux ratio because we have neither redshifts nor detailed spectral information on these ob jects in either the infrared or radio bands. Appleton et al. (2004) explicitly demonstrate that individual k-corrections have little systematic effect on the mean trend of q24 as a function of redshift at z < 1.

Based on spectral energy distribution modelling in the infrared, and an assumed synchrotron spectral index of -0.7 in the radio, Appleton et al. found that this increased the uncorrected value of q24 = 0.84 by only 0.1­0.2, depending on the SED template used. However, Fadda et al. (2006) demonstrate that the kcorrection arising from the PAH emission features entering the 24µm band at higher redshifts (1.1 < z < 1.8 or 2.2 < z < 2.5) could be much greater. Fadda et al. (2006) found that emission in the 24µm band could be enhanced by as much as a factor 2 ( q = 0.3) for an M82 starburst galaxy template. Thus it is possible that k-corection provides a partial explanation for the differences seen with Appleton et al. (2004), if the bulk of the population used in the stacking analysis are high redshift starbursts. For k-correction to fully explain the difference our sources would need to be dominated by high redshift (z > 1) starburst galaxies with
c 2004 RAS, MNRAS 000, 1­8


Page 7 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Extending the infrared radio correlation

7

stellar 24µm sources are detected at 20cm in the CDFS field (f20cm > 0.4mJy). As a further consistency check of our result we derived the infrared-radio correlation based on radio and infrared fluxes measured at the positions of individual sources used in our stacking sample. Most of the radio sources will lie below the formal 5 detection limit and consequently their fluxes will be sub ject to large errors. Nevertheless the ensemble properties of the distribution will be robust. In each infrared bin used above we computed the median value of the fluxes in each bin and made the fit to those points. For the CDFS and ELAIS the resulting fits were respectively: f20cm = 0.037(±0.001)f24 f20cm = 0.041(±0.002)f24 + 2.6(±0.8)µJy + 6.6(±3.5)µJy Median Median

µm

µm

Figure 10. MIPS 24µm images for four CDFS source with 0.8 < f24µm < 0.83mJy and no detected 20cm emission. Note the complex/extended morphology in the ma jority of cases, broadly consistent with the p opulation at this flux level and q24 value.

PAH emission features approximately double that seen in some of the most intense star-forming galaxies locally. A more mundane source for the discrepancy may arise from the different sample selection effects. We note that q24 derived by Appleton et al. (2004) is based on radio detections of infrared sources. Our values derived here are for all infrared sources, irrespective of radio flux. Thus a lower coefficient could be explained by a population of sources with low radio-infrared flux ratios which have previously not been included in the study of radio-selected samples because they fall below the radio flux limit for the sample. Indeed, we note that Gruppioni et al. (2003) found that the mid-infrared (15µm)/radio (20 cm) luminosity correlation for mid-infrared sources with radio detections was a factor of two higher than the correlation for all sources with upper limits for radio luminosities included in the analysis. However, if this were to be the case we would still have to explain the difference between these lower q24 (passband issues notwithstanding) and those FIR/radio correltions derived from IRAS sources with complete radio detections (Yun, Reddy & Gruppioni 2003). To investigate this further, we plot in Fig. 7 a composite of the radio/infrared fluxes derived for both the individual radio-detected sources in the CDFS (Norris et al. 2006) and the stacked images derived in this paper. The sample 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 q24 . However there is a significant absence of sources below f20cm = 0.1f24µm , (the radio-FIR relation found by Appleton et al.), which appears to mark a lower bound to the distribution. In particular, there is a significant absence of sources along the line extrapolated from our stacked data. Indeed, at the higher (f24µm > 10mJy) fluxes nearly all nonc 2004 RAS, MNRAS 000, 1­8

The median estimator to the individual points in both fields gave a consistent slope to that obtained from the stacked images. We also modelled the effects of a population with a bimodal radio-infrared flux ratio; half of the population having f20cm = 0 and half having f20cm = 0.08f24µm (based on the CDFS MIPS flux distributions). Although the stacked images of this simulated population give similar results to that obtained with the real data, the distribution of the individual simulated population flux values is visibly bimodal in a scatter plot of the individual source fluxes (Fig. 8 upper panel) and clearly different from a similar plot using the real data (Fig. 8 lower panel) Nevertheless, we are faced with the conundrum that, whatever the distribution, the mean observed q24 value is greater than that of any modelled SED. Some of the opticallybrightest CDFS sources with the most extreme q24 values are relatively low redshift galaxies (Fig. 9). However, most of these radio-weak sources have no direct optical counterpart at photographic plate limits (R 21 mag). The ma jority of the ob jects at high f24µm flux end of the correlation f24µm 0.83mJy and show extended or complex morphologies in their 24µm images (Fig. 10). This would be consistent with a population dominated by actively star-forming galaxies at moderate redshifts. To date, optical spectroscopy has been obtained for 130 SWIRE sources in the CDFS (Norris et al. 2007). The median redshift is z 0.4 with a tail to higher redshift z 2, exclusively composed of broad emission line active galactic nuclei at z > 1.2. However, this redshift distribution is strongly influenced by optical selection effects ­ the vast ma jority of the ob jects with measured redshifts having relatively bright optical magnitudes (R < 22.5 mag).

5

CONCLUSIONS

We find that the population of ob jects at µJy flux densities obey a radio-24µm correlation which indicates that they are 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 ob jects, or ob jects with an intrinsically different value of q24 , from ob jects seen in the local Universe. Nevertheless, we acknowledge that the


B.J.Boyle et al.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 9

8

B.J.Boyle et al.

0329110-283237

033235.1-275533

033137.3-282415

z=0.037 0334029-280034 0333571-280358

z=0.215 0328161-280152

z=0.015 032819.6-281941 032930.7-273348 033112.5-284854

z=0.060

z=0.108

midt telescop e images anomalously elowwithio-infrared sly loratios.io-infraredgeux e 9. ne gand UKwith ofdt telescop images anomalou flux wwith Each alouslyi of niR-ba laxies Schmi nine galaxies rad of nine galaxies rad anom ima fl .hFve change µminedgalaxies 13.5, of q24oµ,m when xesREFE13.5,, Sf24µ,m3.98,e3.97 13.5,,3.86mJy, m-righ24 obta fluxesoofmeana-right, ccurs flu ,laxies RENCE 6.99 , 6.06x 5.97, and 6.99,, 6.0 a or f t, the to for the h value 6.99the ga 5.97, have 5.29 flu , s of 5.75 5.29 3. top-left here b ttom ve f 6.06 of 5.75 y) w tw(e 20cm < ioe flux limit ((f20cm < 100µJy)P.N., et al., 2004, ApJS, 154, 147 lo th f o3e rad us of µscrete radio > 5) detections Appleton limi e m v from the 100 diJy)
24

Figure 9. R-band UK Schmidt telescop e images of nine galaxies with anomalously low radio-infrared flux ratios. Each image is 30â30 arcsec. For top-left to b ottom-right, the galaxies have f24µm fluxes of 13.5, 6.99, 6.06, 5.97, 5.75, 5.29, 3.98, 3.97 and 3.86mJy, and all are b elow the 3 radio flux limit (f20cm < 100µJy). As indicated, four galaxies have spectroscopic redshifts from the 2dF Galaxy Redshift Survey (Colless et al. 2001).

e 9. telescop UKwith of nitelescop low with of nine gaflux ratios. Eacfrared slyis of ni R galaxies Schmidt ne galaxies radio-infrared laxies rad anom ima fl midt ne-band e images anomalously e images anomalously lowwithio-inh alouge ux . ave f p-left ga esoof have f t, , 6.06 xes of, have 6.99 3.98, 5.97 and 3.86m 6.0 hFor to to b ttom-righ ga 5.97 5.75 5.29 flu s of 5.75 5.29 3. om-right, thefluxlaxies 13.5, 6.99the flu,laxies 13.5,, f , 6.06xe3.97, 13.5,, 6.99,,Jy, y) w t (e 3 radio flux limit (f lo th f < 100µJy) limi < 100µJy)

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 galaxies 241 mz < 1.8 or 2.2 < z < 2.5. Another m anaat µ< 24µ expl tion for at least part of the discrepancy may arise from the radio-selection biases inherent in some previous surveys. 20cm 20cm Data from high-sensitivity surveys such as ATLAS will help to resolve such issues.

Ca Co Co Co

midt. telescopeUKwith anomalously e images anomalously loratios.io-infraredgeux e 9 ne-band images ofdt telescop lowwithio-innine gaflux wwith Each alouslyi of ni R galaxies Schmi nine galaxies rad of frared laxies rad anom ima fl .m-righ24µm LEDGMENTS 13.5, 6.99thefluxes of, 13.5,, 6.99,, 6.06,, e3.97 13.5,,3.86mJy, have Af t,NOW fluxesoof ha-righ24µ,m galaxies have f For top-left galaxies ve f t, 6.06, 5.97 5.75 5.29 3.98 5.97, and 6.99,, 6.0 flux s of 5.75 5.29 3. CK the to b ttom y) w tRadio20cmas< io fluxhelimit a(Telescope Com- 100µJy) limi thf 3 w obtained µJy) lo (e data rad 100by t Australi f20cm <
pact Array, operated by CSIRO Australia Telescope National Facility. IRS acknowledges support from the Royal Society.

rilli C.L., Yun M.S., 1999, ApJ, 513, 13 hen J., et al. 2000, ApJ, 528, 29 lless M.M. et al., 2001, MNRAS, 328, 1039 ndon J.J., Condon M.A., Gisler G., Puschell J.J., 1982, ApJ, 252, 102 Condon J.J., Anderson M.L., Helou G., 1991, ApJ, 376, 95 24µm Condon J.J., Broderick J.J. 1991, AJ, 102, 1663 Condon J.J., 1992, ARAA, 30, 575 Dickey J.M., Salp eter E.E., 1984, ApJ, 284, 461 de Jong T., Klein U., Wielebinski R., Wunderlich E.,1985, A&A, 147, 6 Fadda D. et al., 2006, AJ, 131, 2859 Garrett M.A. 2002, A&A, 384, 19 Gruppioni C., et al., 2003, MNRAS, 341, 1 24µm Harwit M., Pacini F. 1975, ApJ, 200, 127 Helou G., Soifer B.T., Rowan-Robinson M., 1985, ApJ, 298, 7 Iveson R.J., et al., 2002, MNRAS, 337, 1 Kron R.G., 1980, ApJS, 43, 305 Lonsdale C.J., et al., 2003, PASP, 115, 897 c 2004 RAS, MNRAS 000, 1­8


Page 9 of 9

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Extending the infrared radio correlation
Middelb erg E., 2006, PASA, 2264 Norris R.P. et al., 2006, AJ, 132, 2409 Norris R.P. et al., 2007, MNRAS, in press Rowan-Robinson M. et al., 2004, MNRAS, 351, 1290 Sault R.S. et al., 1995, ASPC, 77, 433S Surace J.A. et al., 2006, AJ, submitted van der Kruit P.C.,1973, A&A, 29, 263 Xu C., Lisenfield U., Volk H.J., 1994, A&A, 285, 19 Wall J.V., Benn C.R., Grueff G., Vignotti M., 1986, in Highlights in Astronomy, Proc. 19th IAU Gen. Assembly, 7, 345 Wals M., Boyle B.J., Cro om S.M., Miller L., Smith R.J., Shanks T., Outram P., 2005, MNRAS, 360, 453 Yun M.S., Reddy N.A., Condon J.J., 2001, ApJ., 554, 803

9

c 2004 RAS, MNRAS 000, 1­8