The LABOCA survey of the Extended Chandra Deep Field-South – radio and mid-infrared counterparts to submillimetre galaxies Free

Abstract

1 INTRODUCTION

Although rare today, ultraluminous infrared galaxies – galaxies with infrared (IR) luminosities exceeding 10¹² L_⊙– were extremely common in the early Universe, signposting systems undergoing intense, dust-obscured star formation. Moreover, they contribute a significant fraction of the submillimetre (submm) background (Fixsen et al. 1998). This important high-redshift population was first discovered in the form of bright submm sources behind massive, lensing clusters (Smail, Ivison & Blain 1997) and in blank fields (e.g. Barger et al. 1998; Hughes et al. 1998; Eales et al. 1999), using the Submm Common User Bolometer Array (SCUBA; Holland et al. 1999) on the 15-m James Clerk Maxwell Telescope (JCMT); a number of surveys with a variety of instruments have now brought the number of known submm galaxies (SMGs) to several hundred (e.g. Coppin et al. 2006; Bertoldi et al. 2007; Greve et al. 2008; Scott et al. 2008).

Cross-identifying the submm sources with emission at other wavelengths is made difficult by the poor spatial resolution of even the largest submm telescopes. For example, the combination of JCMT and SCUBA resulted in a resolution of 14 arcsec (FWHM) at 850 μm. The best way to overcome this would be with mm/submm interferometric observations – capable of locating the submm emission directly, with arcsec accuracy (e.g. Downes et al. 1999; Gear et al. 2000; Iono et al. 2006; Wang et al. 2007; Younger et al. 2007; Ivison et al. 2008; Cowie et al. 2009). Such observations, however, require a large investment of observing time with the few existing facilities that are capable, although the advent of the Atacama Large Millimeter/Submillimeter Array (ALMA) will make this strategy much easier in the future.

In the meantime, attaining higher resolution is possible using radio interferometric and IR observations, where the empirical correlations between the far-IR and radio wavebands (Condon 1992) or the bolometric IR/mid-IR (Elbaz et al. 2002) make it much easier to identify the submm emitter, particularly given the low source densities in the radio (Ivison et al. 1998, 2000, 2002; Smail et al. 2000; Dannerbauer et al. 2004). This work has typically relied on data from the Very Large Array (VLA) at 1.4 GHz and Spitzer using the 24-μm channel of the MIPS instrument (Rieke et al. 2004; Werner et al. 2004). In addition, high-redshift SMGs can be identified through their IR colours as measured by Spitzer’s Infrared Array Camera (IRAC) camera (e.g. Pope et al. 2006).

Here we present radio, mid-IR (24 μm) and IRAC counterparts to the 126 SMGs that have been detected in the Large APEX Bolometer Camera (LABOCA) Extended Chandra Deep Field-South (ECDF-S) Submm Survey (LESS), a deep blank-field 870-μm survey, down to a 3.7σ limit of 4.4 mJy (Weiß et al. 2009). The ECDF-S is an exceptional area for multiwavelength, wide-field studies of galaxy evolution due to deep X-ray (Giacconi et al. 2001; Lehmer et al. 2005; Luo et al. 2008), optical (Giavalisco et al. 2004; Beckwith et al. 2006), IR (Dickinson et al., in preparation) and radio (Miller et al. 2008; Ivison et al. 2010) data. The CDF-S portion of the field has also been surveyed (Scott et al. 2010) with the AzTEC 1.1-mm bolometric camera (Wilson et al. 2008) on the Atacama Submillimeter Telescope Experiment.

The paper is organized as follows. In Section 2 we describe the submm, radio, 24 μm and IRAC data that have been used to identify counterparts to the submm sources, with particular emphasis on the techniques used to extract source fluxes and positions from the radio map. Section 3 contains details of our counterpart identification strategy, and in Sections 4 and 5 we present lists of the likely counterparts and their properties. Section 6 discusses these results in detail, ascertaining the effectiveness of our strategy. We also derive the redshift distribution of the radio-detected robust counterparts using the radio-submm spectral index relation of Carilli & Yun (1999, 2000) before drawing our conclusions in Section 7. In an appendix, we present detailed notes on some of the sources as well as multiwavelength maps with the counterparts marked.

We assume a flat Λ cold dark matter (ΛCDM) cosmology of Ω_Λ = 0.73, Ω_m = 0.27 and H₀ = 70.5 km s⁻¹ Mpc⁻¹ (Hinshaw et al. 2009).

2 OBSERVATIONS, REDUCTION AND ANALYSIS

2.1 APEX 870-m catalogue

LABOCA (Siringo et al. 2009) is a 295-element bolometer camera operating at the 12-m Atacama Pathfinder Telescope (APEX)1 in the exceptionally dry environment of the Atacama desert in Chile (Güsten et al. 2006). The LESS map comprises 200 h of on-sky integration (excluding overheads) and has extremely uniform noise coverage (average rms = 1.2 mJy beam⁻¹) over the 30 × 30 arcmin² extent of the ECDF-S, with a resolution of 19 arcsec (FWHM). The catalogue of submm sources identified by LESS is described in detail by Weiß et al. (2009). The full catalogue comprises 126 sources above 3.7σ with a false-detection expectation of ≈5. This is based on extensive simulations as described in Weiß et al. (2009).

2.2 VLA 1.4-GHz catalogue

To identify the radio counterparts to the LESS SMGs, we use the VLA 1.4-GHz map of Miller et al. (2008) which we briefly describe here. The map is constructed from six separate pointings arranged in a hexagonal pattern, centred on the coordinates 03^h32^m28^s, −27°48′30″ (J2000). Each pointing consists of approximately eight separate ∼5 h observations. The noise in the final 34 × 34 arcmin² mosaic is ∼6.5 μJy beam⁻¹ at its deepest. All data were taken in ‘A’ configuration, resulting in a synthesized beam with dimensions 2.8 × 1.6 arcsec², aligned north–south. When looking for radio counterparts to the SMGs, we do not use the Miller et al. (2008) catalogue as this is truncated at a signal-to-noise ratio (S/N) of 7; instead, we have created our own catalogue containing sources down to an S/N of 3.

Seven of the SMGs in the LESS catalogue lie outside the 34 × 34 arcmin² area of the radio map. For these, we use our own reduction of the VLA data to search for counterparts. Our map was created in a similar fashion to that of Miller et al. (2008) and achieves an rms just below 7 μJy beam⁻¹. The flux density of the brightest of the SMG counterparts has a flux density in the two maps that differs by less than 1 per cent, and thus we are confident that the two maps are tied to the same flux scale.

2.2.1 Source extraction

The first step in producing a catalogue of radio sources is to create a map of the noise across the field. Sources with a peak-flux-density-to-noise ratio (PNR) greater than 5 are detected and removed using the standard aips source-extraction code, sad. The residual image is then inverted and the source extraction process repeated in order to remove ‘sources’ with negative flux – mainly prominent side lobes caused by Gibbs ringing (associated with high-S/N sources) which become increasingly prominent with distance from the phase centre of each pointing. Once all significant sources have been removed, a noise map is created for each pixel by fitting a Gaussian to the histogram of pixel values contained within a surrounding circle of diameter 50 arcsec (using rmsd with optype =‘hist’). Aided by the accurate noise map, we start the source extraction again, this time restricting the fitting to positive sources with a PNR equal to or greater than 3.

To improve the accuracy of our extracted flux densities, we extract sources in two ways. In the vast majority of cases, we assume that the source is unresolved and fix the size of the fitted Gaussian to that of the restoring beam. For those sources that are significantly resolved, we instead allow the size of the Gaussian to vary. The reason for this approach is that allowing the size of unresolved sources to vary often produces cases where the peak flux density is greater than the total, a consequence of its measured size being smaller than the beam. The result is that the measured flux densities are less accurate than if their sizes had been held fixed at the width of the restoring beam. We have simulated this effect by injecting multiple point sources into our residual map and extracting them, as with the real map, first with the source size unconstrained, then again with the size fixed to that of the beam. The scatter in the ratio of injected and extracted flux densities was significantly reduced in the latter case (see also Ibar et al. 2009). In all the radio source simulations described in this section, we created 50 fake maps, each containing 500 sources, i.e. a total of 25 000 sources.

In order to use this approach, it is obviously necessary to decide which sources are unresolved and which are resolved. We do this in the following way. When the source size is allowed to vary, the uncertainties in the fitting process cause the ratio of peak to total flux density to increase from unity as often (and by as much) as it decreases; this symmetry is illustrated using our simulated data in the inset of Fig. 1. Each point represents a source injected with the same size as the synthesized beam, but which has increased or decreased in size upon being extracted. The envelope of this plot locates sources in the real data which are inherently unresolved and which should be fitted as such, yielding a more accurate flux density. A similar approach was adopted by Bondi et al. (2003), but using the observed data only and not simulations. Fig. 1 also shows the envelope (containing 98 per cent of the simulated sources) plotted over the real data. Those sources lying above the upper envelope are fitted using a variable width; all other sources are constrained to be point sources.

Figure 1

Left: plots of the ratio of the extracted integrated and peak flux densities for simulated sources as a function of PNR. The inset shows the results for simulated point sources and demonstrates how at low S/N the ratio deviates symmetrically from its initial value of unity. The main plot shows the same for the real data along with the upper envelope derived from the simulated data. All sources below the green line were fitted as point sources. Right: ratio of the injected and extracted flux densities for simulated sources as a function of PNR. The points with error bars plot the median gain weighted by the differential source counts (see text) in consecutive bins. The vertical error bars give the range including 68 per cent of the sources in that bin. The dot–dashed line is a polynomial fit to the points and is used to correct the flux densities of the real sources for flux boosting. The lower line shows the median gain without weighting by the source counts – its value is approximately equal to one, independent of PNR.

Open in new tabDownload slide

2.2.2 Bias correction

We have also studied the effects of biases in the model fitting by comparing the fluxes that we recover from our simulations to those that were injected. A plot of this flux density ratio against PNR (calculated based on the recovered peak flux, a measure against which we can correct our data) is also shown in the right-hand panel of Fig. 1 (lower line) – we find that the median value is close to unity, independent of PNR i.e. there is no bias in the measured flux densities. This is in contrast to the findings of Seymour, McHardy & Gunn (2004), who find a significant positive bias. This is because Seymour et al. plot their flux ratios as a function of input flux density, a quantity which is unknown in the real radio data and which is biased towards sources whose flux densities have increased due to the model-fitting uncertainties.

2.2.3 Flux boosting

‘Flux boosting’ is an effect regularly taken into account when estimating the flux densities of SMGs (e.g. Coppin et al. 2006; Austermann et al. 2009; Weiß et al. 2009), but very rarely with radio sources. The apparent flux density of such a source deviates from its true value if it sits on/in a noise peak/trough. Because faint sources are more numerous than bright ones, the measured flux density of a catalogued source (i.e. a source lying above the chosen S/N threshold) is more likely to have been boosted than reduced. The most likely flux density is produced by ‘deboosting’ the measured flux densities by the appropriate factor.

We have measured the magnitude of the flux boosting as a function of recovered peak flux density by using the same simulations that were used to investigate the biases in the model fitting. These were performed using equal numbers of sources per flux density bin and therefore do not show the effects of flux boosting (the dots in the right-hand panel of Fig. 1). The source counts can however be added retrospectively by defining flux density bins and again forming a median, but this time weighting each point (source) in a bin by its differential source count (dN/dS). Using our catalogue, we measured a Euclidean slope for the source counts, based on sources above a flux density of 100 μJy. This should be valid for all sources fainter than this limit as extremely deep radio observations have shown that there is no change in the slope of the counts down to flux densities as low as ∼15 μJy (Owen & Morrison 2008). The bins and the value of the flux boosting correction in each are overplotted on the unweighted data in the right-hand panel of Fig. 1 as red points with error bars; the flux boosting at any value of PNR is calculated by fitting a function to these points (also shown in the figure). For a 3σ source, the flux boosting is equal to 36 per cent.

2.3 Spitzer MIPS catalogues

The 24-μm data are taken from the Far-Infrared Deep Extragalactic Legacy Survey (FIDEL; Dickinson et al., in preparation), a programme to map the ECDF-S (as well as the Extended Groth Strip and GOODS-N) at 24 μm using the MIPS camera on board Spitzer. The FIDEL MIPS data were reduced following the procedures given in Chary et al. (2004), Frayer et al. (2006) and Frayer et al. (2009). The final image depth at 24 μm varies across the field, with typical exposure times ranging from 11 000 to 30 000 s (with a maximum of approximately 36 000 s). The 24-μm image almost completely covers the area mapped by LABOCA and only one submm source (LESS046) falls off its edge.

For the counterpart analysis, we have used a catalogue produced by the daophot package from iraf; the source extraction was not guided by information on positions from other wavelengths. Examination of the differential number counts in the 24-μm data shows that these turn over at ∼30 μJy due to incompleteness; thus we have not considered sources with fluxes lower than this. The flux errors reported by the daophot software are gross underestimates, but simulations have shown that the values of S/N reported by the apex point-source extraction software specifically developed for Spitzer (Makovoz et al. 2002) are accurate. Although we have not used the apex catalogue for our counterpart analysis (it does not go as deep as that produced using daophot), matching sources from the two catalogues to within 1 arcsec shows that the apex S/N and daophot flux/Δflux measurements are linearly related and that the latter need to be multiplied by a factor of 3; the simulations also showed that the flux measurements from each catalogue were consistent.

2.4 Spitzer IRAC catalogues

The Spitzer IRAC (Fazio et al. 2004) images are taken from the Spitzer IRAC and MUSYC Public Legacy in ECDF-S (SIMPLE) survey (Damen et al. 2011). We use SExtractor (Bertin & Arnouts 1996) to extract source positions on a summed image of all four IRAC channels, weighted such that a source of a given magnitude in each image is equally represented. The areas within 15 arcsec of each LESS source were checked visually to ensure the catalogues were complete. We then use apphot in iraf to extract fluxes in 3.8-arcsec diameter apertures on the 3.6- and 5.8-μm images, and apply aperture corrections as derived by the Spitzer Wide-area Infrared Survey (SWIRE) team (Surace et al. 2005) to obtain total source magnitudes.

3 IDENTIFYING COUNTERPARTS TO SMGS

Following several other authors (e.g. Ivison et al. 2002, 2007; Pope et al. 2006; Chapin et al. 2009a), we have identified the most likely radio and 24-μm counterparts to the LESS sources by calculating the corrected Poissonian probability (Browne & Cohen 1978; Downes et al. 1986) of radio and 24-μm sources that lie within a search radius, r_s, of each SMG. Given a potential counterpart at radius, r, with flux density, S, we can calculate the a priori probability, p, of finding at least one object within that radius of at least that flux density from the expected number of events:

1

where n_S is the surface density of sources with fluxes >S. The probability is

2

However, as the search is being conducted over the (generally) larger radius r_s, this is not the probability we require, i.e. searching at random locations will find more sources as extreme as the one found than would be expected given its measured probability, p. Having found a source of probability p, at radius r, we need to know the number of similar events that would be found in our random search out to r_s. Provided that p≪ 1, this is given by

3

where the so-called critical probability is defined as

4

and n_lim is the source surface density at our lowest detectable flux density.2 The final probability of a counterpart being a chance coincidence is calculated by inserting the corrected number of events (μ_cor) into equation (2) in place of μ. As has been typical in the literature (Ivison et al. 2002, 2007; Pope et al. 2006; Chapin et al. 2009a), we take a value of p≤ 0.05 to indicate a secure association.

Offsets between the SMG and radio/24-μm sources will be dominated by the uncertainty in the SMG positions, this being a function of the S/N of the submm detection. Therefore, in contrast to some studies of this type that use a search radius based on some representative S/N, we have chosen a different search radius for each SMG that is some multiple of its 1σ positional uncertainty in right ascension/declination (≈1–3 arcsec); Smail et al. (2000) similarly used an S/N-dependent search radius. Effects such as telescope pointing errors might conspire to produce systematic offsets between the submm source and its counterpart, but as the final map is an average of multiple observations taken at different times, any such systematic offsets are minimal. Greve et al. (2010) come to the same conclusion from a stacking analysis of the LESS data which confirms the absolute astrometry of the submm map.

In choosing a value for r_s, our overriding concern has been to make it large enough to avoid missing significant numbers of counterparts, but small enough to avoid choosing counterparts from unrelated, bright field sources; too large a radius also overestimates the value of p. A reasonable maximum value for r_s is 3σ as this will ensure that we only miss the counterpart for 1 per cent of the SMGs3 i.e. 1.3 sources. The number of SMGs with missed counterparts is plotted in Fig. 2 over a wide function of radius, 0–5σ. Also plotted in Fig. 2 are the numbers of secure counterparts found as a function of radius for both the radio and 24-μm catalogues. Both rise steeply between 0 and 2σ and gently decline above ∼3σ. We have therefore set our search radius to r_s = 3 σ.

Figure 2

The solid lines show the number of secure counterparts (p≤ 0.05) as a function of radius (in units of the SMG 1σ positional uncertainty) for the radio catalogue (thin line) and 24-μm catalogue (thick line). Also shown is the number of SMGs for which the counterpart will not have been found, based on the cumulative distribution function of the Rayleigh distribution (dot–dashed line). The low dashed lines show the results of the Monte Carlo simulations, i.e. the number of secure counterparts found as a function of radius for randomly distributed SMGs. Choosing a radius of 3σ produces close to the maximum number of counterparts and results in only ∼1 per cent of the SMGs not being searched out to a sufficient radius. Note that we do not show the IRAC IDs here as those were only searched for in error circles devoid of radio and MIPS counterparts, using a selection guided by the radio/MIPS IDs.

Open in new tabDownload slide

We have also used Monte Carlo simulations to investigate the effect of varying the search radius, producing 100 realizations of the SMG catalogue at radii between 0 and 5σ in steps of 0.5σ and searching for secure counterparts in the same way as with the real data. Each simulated catalogue has the same distribution of flux densities, and therefore search radii, as the real catalogue, but with randomized positions. The results are again shown in Fig. 2 and illustrate that the number of false detections is approximately constant beyond r_s∼ 1.5σ.

The positional errors for each SMG have been determined using the simulated source extractions of Weiß et al. (2009). This offers advantages over analytical formulae such as equation (B22) of Ivison et al. (2007), in that it includes all sources of uncertainty, including those originating in the data reduction and source extraction processes. The empirical formula for the positional uncertainties is

5

where a = 6.08, b = 0.14, c = 0.56 and S_in is the intrinsic flux of the source, i.e. the observed flux after deboosting.

Integrated source counts were calculated for both the radio and 24-μm data from the respective catalogues; these are used to calculate the value of n_S at both the flux density of the potential counterpart and the flux limit (for the radio catalogue, we formed the counts using the undeboosted fluxes as these correspond to the actual source densities in the radio map). In order to test these and the entire p-statistic procedure, we have again performed Monte Carlo simulations, producing 500 realizations of the 126-source submm catalogue as described above. On average, 5 per cent of the SMGs should have a counterpart with p≤ 0.05. This corresponds to 6.3 sources on average and we indeed find values of 6.326 for the radio and 6.302 for the 24-μm data. We are thus confident that we are measuring the correct probabilities for each counterpart.

4 THE RADIO AND MIPS COUNTERPARTS

In the following, we try and identify counterparts from the radio and MIPS catalogues. We then use the properties of these counterparts to select the parameter space in IRAC colour and flux to identify potential counterparts to the radio and MIPS-undetected sources.

The results of our counterpart search are given in Tables 1 (radio) and 2 (24 μm), where we include the position and flux of the counterpart as well as its radial offset from the submm source and the value of the search radius. The first column also gives the ‘ID’ of each SMG, an integer describing the position of each in a ranked list of decreasing S/N (as they appear in each table). For brevity, we will often refer to an individual source using this integer, e.g. LESS001; the integers also refer to the S/N-ranked list in Weiß et al. (2009). Postage stamp maps with a size of 36 × 36 arcsec² centred on the submm position are shown in Fig. A1. We show radio contours superimposed on IRAC 3.6-μm grey-scales. The IRAC images are taken from the SIMPLE Legacy Program (P.I.: P. van Dokkum) and where the SMGs do not lie fully within the SIMPLE coverage, we replace the images with the ones from SWIRE; LESS046 also has its FIDEL image replaced with one from SWIRE. We also show the LESS contours, overplotted on 24-μm grey-scales, in a separate panel.