The VLA-COSMOS Survey. IV. Deep Data and Joint Catalog

Accepted to ApJS: April 19, 2010 Preprint typeset using LATEX style emulateapj v. 11/10/09 THE VLA-COSMOS SURVEY. IV. DEEP DATA AND JOINT CATALOG E. Schinnerer , M.T. Sargent1 , M. Bondi2 , V. Smolc 1 ´3 , A. Datta4 , C.L. Carilli4 , F. Bertoldi3 , A. Blain5 , P. ˇ ic Ciliegi6 , A. Koekemoer7 , N.Z. Scoville5 Accepted to ApJS: April 19, 2010 ABSTRACT arXiv:1005.1641v1 [astro-ph.CO] 10 May 2010 In the context of the VLA-COSMOS Deep project additional VLA A array observations at 1.4 GHz were obtained for the central degree of the COSMOS field and combined with the existing data from the VLA-COSMOS Large project. A newly constructed Deep mosaic with a resolution of 2.5′′ was used to search for sources down to 4σ with 1σ ≈ 12 µJy/beam in the central 50′ ×50′ . This new catalog is combined with the catalog from the Large project (obtained at 1.5′′ ×1.4′′ resolution) to construct a new Joint catalog. All sources listed in the new Joint catalog have peak flux densities of ≥5σ at 1.5′′ and/or 2.5′′ resolution to account for the fact that a significant fraction of sources at these low flux levels are expected to be slighty resolved at 1.5′′ resolution. All properties listed in the Joint catalog such as peak flux density, integrated flux density and source size are determined in the 2.5′′ resolution Deep image. In addition, the Joint catalog contains 43 newly identified multi-component sources. Subject headings: cosmology: observations – radio continuum: galaxies – surveys 1. INTRODUCTION (COSMOS)8 collaboration has conducted panchromatic In recent years, several cosmological deep fields have imaging and spectroscopy of an equatorial field with been imaged at 20 cm (e.g., Richards 2000; Bondi et al. a size of 2 deg2 (for an overview, see Scoville et al. 2003; Condon et al. 2003; Hopkins et al. 2003; 2007b) ranging from X-ray XMM-Newton and Chan- Seymour et al. 2004; Norris et al. 2005; Huynh et al. dra (Hasinger et al. 2007; Elvis et al. 2009), UV GALEX 2005; Fomalont et al. 2006; Simpson et al. 2006; (Zamojski et al. 2007), optical and near-infrared ground- Ivison et al. 2007; Schinnerer et al. 2007; Miller et al. based (Taniguchi et al. 2007; Capak et al. 2007), optical 2008; Owen & Morrison 2008) providing a few thousand HST (Scoville et al. 2007a; Koekemoer et al. 2007), mid- radio sources down to flux limits of a few 10 µJy. to far-infrared Spitzer (Sanders et al. 2007), millimeter These deep radio imaging data are sensitive enough to (Bertoldi et al. 2007; Scott et al. 2008) and radio VLA detect star forming galaxies with star formation rates (Schinnerer et al. 2004, 2007) imaging to extensive op- of several 10 to 100 M⊙ yr−1 out to and beyond a tical spectroscopy using the VLT/VIMOS and Magel- redshift of z ∼ 1. Similarly, radio galaxies can be seen lan/IMACS instruments (Lilly et al. 2007; Trump et al. out to redshifts of z ∼ 5 and the most luminous ones 2007). Most of these datasets are now publicly available even well into the epoch of reionization. Thus, deep from the COSMOS archive at IPAC/IRSA9 . radio images in conjunction with deep imaging data The VLA-COSMOS survey at 20 cm is part of the at X-ray, optical and infrared wavelengths are ideal overall imaging effort and its scientific goals and moti- to investigate the dust-unbiased star formation, the vation have been described in detail by Schinnerer et al. evolution of radio(-loud) AGN, as well as the population (2007). Initial observations from a pilot project test- mix of radio sources in the first place. ing the mosaicking strategy and giving a first source In order to study the cosmological evolution of galax- catalog are presented by Schinnerer et al. (2004). As ies and black holes, it is not only important to over- a large NRAO/VLA program, the VLA was used in A come the effect of cosmic variance (e.g. by studying and C configuration to cover the entire COSMOS field a large enough area) but also to understand the ef- resulting in an image with uniform noise properties in fect of large scale structure on the evolution (e.g. by the central 1×1 deg2 and an average rms of 10.5 µJy. covering a large contiguous area). To address the sec- Schinnerer et al. (2007) provide a detailed description ond effect in particular, the Cosmic Evolution Survey of the survey set-up, the data reduction, as well as the testing and construction of the final VLA-COSMOS 1 Max-Planck-Institut f¨ ur Astronomie, K¨onigstuhl 17, D-69117 Large project catalog (hereafter: Large catalog). Subse- Heidelberg, Germany 2 INAF-Istituto di Radioastronomia, Via Gobetti 101, I-40129, quently, Bondi et al. (2008) derived the completeness of the Large catalog and also analyzed the effect of band- Bologna, Italy 3 Argelander Institut f¨ ur Astronomie, Universit¨ at Bonn, Auf width smearing on the derived source flux densities to dem H¨ ugel 71, D-53121 Bonn, Germany obtain the source counts. Although the VLA-COSMOS 4 National Radio Astronomy Observatory, P.O. Box O, Socorro, Large Project dataset has been used for several scien- NM, 87801, U.S.A. tific results on, e.g. the faint radio population, the 5 California Institute of Technology, MC 105-24, 1200 East Cal- ifornia Boulevard, Pasadena, CA 91125, U.S.A. radio-derived star formation rate density, the radio AGN 6 INAF-Osservatorio Astronomico di Bologna, Via Ranzani 1, I- population and stacking of high-z galaxy populations 40127 Bologna, Italy (Smolˇci´c et al. 2008, 2009a,b; Carilli et al. 2007, 2008), 7 Space Telescope Science Institute, 3700 San Martin Drive, Bal- timore, MD 21218, U.S.A. 8 http://cosmos.astro.caltech.edu 9 http://irsa.ipac.caltech.edu/Missions/cosmos.html 2 Schinnerer et al. the need for deeper radio imaging data became appar- final CLEAN beam. As for the Large project, each polar- ent during the search for radio counterparts to millime- ization for each intermediate frequency (IF) was CLEANed ter sources from the COSMOS MAMBO mapping data separately and combined afterwards in the image plane. (Bertoldi et al. 2007); the Large project only provided The final seven new fields were combined with the exist- counterparts for about half of the mm-sources. Thus, ing 16 fields from the Large project to obtain a mosaic the Deep project was initiated with the aim of doubling covering the full 2 deg2 of the COSMOS field. The aver- the integration time for the central seven pointings which age rms achieved in the central 30′ was 8 µJy/beam at the fully cover the MAMBO 1.2 mm map of the COSMOS resolution of the Large project. In order to mitigate the field. These new observations are described here. effect of bandwidth smearing on the derived peak flux The paper is organized as follows: After a description densities during source extraction (see §6.1), the final of the new observations and the data reduction (§2), the mosaic was convolved to a lower resolution of 2.5′′ (Fig. revision of the Large catalog (leading to a new version of 1) leading to an increase of the rms to ∼12 µJy/beam. v2.0) and the construction of the new Deep catalog are We tested several convolution kernels by comparing the outlined in §3 and §4. In §5 we explain the construction source fluxes and the number of sources extracted for of the final Joint catalog which is described in detail in two representative sub-images of the mosaic. The res- §6 where we also present all the tests and corrections olution of 2.5′′ provided the best compromise between involved. A summary is given in §7. minimizing the flux losses and maximizing the number of extractable sources. We verified that the noise distri- 2. OBSERVATIONS AND DATA REDUCTION bution in the Deep image follows a Gaussian distribution The central seven pointings of the mosaic for the VLA- as expected (Fig. 2). This Deep mosaic was used for COSMOS Large project (pointing numbers: F07, F08, source detection and extraction (see §4). Fig. 3 shows F11, F12, F13, F16, F17; see Schinnerer et al. 2007) were the fractional area covered as a function of rms for the observed using the VLA A configuration in February and Deep and Large mosaics. March 2006. The observations were executed on 11 days, 3. THE REVISED LARGE PROJECT CATALOG typically lasting for a total of six hours starting at 07hr LST (Local Sidereal Time). The exact coordinates of the The original source catalog for the VLA-COSMOS pointing centers are listed in Tab. 1. We used the same Large Project presented by Schinnerer et al. (2007) was spectral set-up and set of calibrators as for the VLA- created by the following procedure: 1) the AIPS task COSMOS Large project to allow for an easy combination SAD extracted candidate components down to a peak flux of the data from the two projects: the VLA standard L- density limit of 3σ, 2) after fits to the peaks of these band continuum frequencies of 1.3649 and 1.4351 GHz candidates using MAXFIT only those with an measured in multi-channel continuum mode, the quasar 0521+166 (rather than the fitted Gaussian) peak flux density of (3C 138) for flux and bandpass calibration at the be- ≥4.5σ were kept, 3) components potentially missed by ginning of each day, and the quasar 1024-008 as phase SAD were recovered with JMFIT which was used to fit calibrator. In order to obtain a good sampling of the a Gaussian to all peaks above our detection threshold uv-plane, we cycled three times through the pointings that were not included in the SAD component list, and each day with a total integration time per pointing of finally 4) components not in the SAD list but with non- ∼45 min. The final resulting integration time per point- zero major axes found in step 3 were added. This com- ing was ∼8.25 hr. In addition, we changed the order of ponent catalog was transformed into a source catalog, the observations of the individual pointings each day to after multi-component sources, e.g. large radio galaxies, sample the uv plane uniformly in all pointings. were identified by visual inspection. Since in the case of During the data reduction process, we excluded all faint sources, the position and peak values of a Gaus- EVLA antennas10 that were present in the array during sian parametric fit might not be a good representation of the observations. We followed the data reduction pro- the real values, the position and value of the peak found cedure adopted for the Large project (Schinnerer et al. by MAXFIT replaced the results from the Gaussian fit. 2007) which uses the Astronomical Imaging Processing Furthermore, all sources classified as unresolved (for our System (AIPS; Greisen 2003). uv data points affected methodology see Section 6.2 in Schinnerer et al. 2007) by RFI (radio frequency interference) were again flagged had their integrated flux density set equal to the peak by hand using the task TVFLAG and by excluding ampli- flux density. The final VLA-COSMOS Large project cat- tudes greater than 0.55 Jy. The flux density of the phase alog (v1.0) has a total of 3601 entries. calibrator 1024-008 was found to be close to its previous Subsequent use of the Large catalog in conjunction value in the season of 2004/5. After final calibration, with the COSMOS optical and (near- to mid-)IR cat- the Deep data were combined with the uv data of the alogs and images showed that the fraction of spurious corresponding pointings from the Large project. sources increases significantly below 5σ. Spurious sources The final imaging was performed using the task IMAGR in our definition are sources that lie in a noisier environ- with the same settings as for the Large project, i.e. ment than expected for a > 4.5σ detection, i.e. three or the same tiling of the individual pointings and the as- more 3σ peaks are present in a 10′′ ×10′′ box centered on sociated CLEAN boxes were used, the flux cut-off was the source. The misidentification is likely due to a mis- kept at 45 µJy, and the restoring beam was again set match in the derived rms value (in a 17.5′′ ×17.5′′ box) to 1.5′′ ×1.4′′ . The robust parameter was changed from 0 and the local rms distribution at the exact position of to 0.25 to obtain a dirty beam as close as possible to the the source. This finding is also confirmed by a compar- ison of sources present in the Large and Deep catalogs 10 A total of 3 EVLA antennas were included in the array at (see §5.2.1). Thus we exclude all sources below 5σ from the time of the observations. the revised Large catalog. In addition, the correction for  VLA-COSMOS Survey IV. 3 bandwidth smearing as derived by Bondi et al. (2008) imize the number of sources found. Therefore, the com- has been applied to the peak flux densities and the clas- bination of the VLA-COSMOS Large and Deep catalogs sification into resolved and unresolved sources has been is ideal to find a maximum number of sources for inclu- modified accordingly (for details see Bondi et al. 2008). sion in a joint catalog. In the following we describe the The revised version of the VLA-COSMOS Large project steps taken to identify such radio sources in the VLA- catalog [hereafter: Large catalog (v2.0)] contains a new COSMOS mosaics. total of 2417 sources. The revised VLA-COSMOS Large In order to construct a list of radio sources from the project catalog (v2.0) has been publicly released to the two separate catalogs, we used the following process that COSMOS archive at IPAC/IRSA11 . is described in detail below: First, sources present in both catalogs were included (§5.2), then selected sources 4. IDENTIFICATION OF DEEP PROJECT detected only in the Large Project mosaic were added SOURCES (§5.3.1), as were – finally – a number of sources present only in the Deep catalog (§5.3.2). In order to minimize As the rms is less uniform in the Deep mosaic than the inclusion of spurious sources at the low significance in the Large mosaic and shows a steep increase towards end, the local rms was re-examined at 1.5′′ and 2.5′′ res- the edges, applying a simple flux density cut for source olution. All sources having a final peak S/N ≥ 5 at at detection is not possible. In order to use the AIPS least one resolution were included. source/component finding task SAD, a S/N map was cre- An overview of the number of sources selected from the ated. First, a sensitivity map (Fig. 4) was constructed Large (v2.0) and Deep catalog for inclusion in the Joint using the AIPS task RMSD with a box size of 105′′ ×105′′ catalog is given in Tab. 2, together with a listing of some to estimate the rms and a sampling step size of 2.45′′ in key properties. both RA and DEC. To obtain the final input S/N map for SAD, the Deep image was divided by the sensitivity 5.1. Estimate of the Local Background Noise map and blanked at the edges where the rms values in the sensitivity map exceeded 34 µJy/beam (the largest Based on the experience with the Large catalog (v1.0) rms value present in the Large mosaic). This last step it is important to obtain an accurate estimate of the lo- allowed us to search for sources in the corners/edges of cal rms to properly measure the S/N of individual radio the mosaic, whereas the area corresponding to the COS- sources. In order to find the optimal box size for the MOS field was searched for the construction of the Large local rms estimate, we compared the value in the rms catalog (see Schinnerer et al. 2007). map used for source detection (see §4) to that obtained SAD was set to search for sources in six iterations with if a smaller box is used. We chose a box size of 50 pixel cut-off levels for the peak of 10, 8, 7, 6, 5, and 4σ. The de- (=17.5′′) to obtain a good estimate of the rms around rived values in units of S/N were converted to mJy/beam compact sources. Fig. 5 shows the different rms values using the corresponding rms values from the sensitivity obtained with a 105′′ and 17.5′′ box for a number of low map. As in the case of the Large project sources with S/N sources from the deep catalog and highlights the im- complex structures – e.g. radio galaxies – were fitted portance of deriving an accurate local rms estimate. The by multiple Gaussian components, such that the derived choice was based on the finding that a box size of 17.5′′ catalog contains components rather than real sources. In used for the Large catalog still resulted in a number of order to identify these multi-component sources, we com- spurious sources at the low S/N end and the requirement pared by eye the location of the multi-component sources that at 2.5′′ resolution the number of pixels outside the from the Large catalog (v2.0) with the component list source should still be sufficient12 . obtained for the Deep image to identify the components We constructed new rms maps at 1.5′′ and 2.5′′ resolu- belonging to the same source. During the comparison tion using a box size of 50 pixels with the task RMSD and with the Large catalog (v2.0), it was noticed that sev- subsequently used them to estimate the local rms. We eral slightly extended sources were fitted by two separate refer to this revised S/N as the ‘local’ S/N (S/N17.5′′ ). components with a typical separation of ∼1′′ (hereafter: All sources with a S/N17.5′′ ≥ 5 will be included in the ‘twin’ sources). In order to determine the nature (simple joint source list. extended Gaussian vs. truly non-Gaussian geometry) of 5.2. Sources Present in Both Catalogs these ’twin’ sources, a single Gauss component was fitted to these sources using JMFIT. If the integrated flux den- For many sources in the Deep catalog it was pos- sity of this single Gaussian fit was consistent within the sible to identify counterparts in the Large catalog formal errors with the combined integrated flux densities 12 At an image resolution of θB (FWHM), a 17.5′′ rms box holds of the two components derived by SAD, we replaced the ‘twin’ sources by a single object. Nbeam = (17.5′′ )2/Ωbeam × ηh = (17.5′′ )2/kp θ 2 B × ηh (1) independent, hexagonally packed beams. Here ηh ≈ 0.9 is the circle 5. CONSTRUCTION OF THE JOINT SOURCE LIST packing density (e.g., Wells 1981) and kp = 1.13 for an aperture with a Gaussian power pattern (Ko 1964; Otoshi 1981). Hence, as Based on the size distribution of faint radio Nbeam ′′ sources (e.g. Muxlow et al. 2005; Fomalont et al. 2006; √ (2.5 ) ≈ 40, the 5 σ rms estimate has an uncertainty of about 5 σ/ 40 ≈ 0.8 σ, which implies that the local S/N could be 4.2 σ or Owen & Morrison 2008; Bondi et al. 2008) a significant less in some regions of the mosaic. To get an upper bound on how fraction of sources should be resolved at our flux limit many of our sources that were only detected with S/N17.5′′ ≥ 5 at a and resolution. Owen & Morrison (2008) showed that resolution of 2.5′′ might be affected by this, we checked which also had S/N ≥ 5 in 105” rms boxes and found that 61% also satisfied source extraction at different resolutions is ideal to max- the latter criterion. These were assigned a flag ‘det’ = 1 in the Joint catalog, while the remaining sources are given ‘det’ = 2, see Section 11 http://irsa.ipac.caltech.edu/data/COSMOS/tables/vla/ 6.4 and Tab. 3. 4 Schinnerer et al. (Schinnerer et al. 2007) down to 5σ or even 4.5σ. The the VLA-COSMOS Deep mosaic. successfully matched sources include all but one of the Before final inclusion in the source list we checked for objects already known from the Large Project to con- each of the successfully matched sources that it met our sist of multiple Gaussian components and the majority selection criterion of peak S/N17.5′′ ≥ 5 at either 1.5′′ or (∼60%) of the single component sources in the Large 2.5′′ resolution in the deep data. If this requirement was catalog (v1.0). not satisfied the corresponding source was not considered any further. 5.2.1. Identification of Single-Component Sources To identify radio sources present in both the Deep and 5.2.2. Identification of New Multi-Component Sources Large catalog, we used the Large catalog down to 4.5σ. As already stated above, all but one of the multi- We matched the two catalogs with a search radius of 1.5′′ component sources listed in the Large catalog (v2.0) were to account for offsets between the peak positions for ex- also present in the Deep catalog. However, during the tended objects. The spatial characteristics of the noise in merging of the two catalogs 43 new multi-component the Deep and Large image differ due the additional cov- sources were identified. Image cutouts of all new multi- erage of the central seven pointings by the Deep project component sources are shown in Fig. 8. and due to the differences in the final resolution (2.5′′ in During the cross-correlation of the Large and Deep the Deep mosaic vs. ∼1.5′′ in the Large mosaic; see §2). catalog several sources were recognized to constitute As mentioned in Section 3 the occurrence of spurious sources made up of several (Deep) rather than just a sources in the Large catalog is significantly reduced by single (Large) Gaussian component during visual inspec- omitting sources in the range S/NLarge ∈ [4.5, 5[. The tion of all sources. In addition to this, several Deep added depth gained with the Deep project observations sources found in the neighborhood of multi-component in the central 40′ (Fig. 6) should increase the reliabil- objects (identified in the Large catalog) were subse- ity of sources with S/NLarge < 5 in the Large Project quently assigned to these sources as additional com- catalog (v1.0) within this area. Fig. 6 shows the cor- ponents of their extended emission. In one instance responding detection thresholds of the two images. The two of the former multiple component sources were peak flux densities of sources in the Large catalog (v1.0) joined to form a new, larger multiple component source are plotted as a function of their radial distance from (a bipolar jet with nucleus assigned the new Joint the field center, their lower envelope roughly tracing the catalog ID COSMOSVLADP J095758.04+015825.2: - average S/N detection threshold of 4.5σ (used as cut for the three merged components have the following IDs the Large catalog (v1.0)). The radial evolution of the in the VLA-COSMOS Large Project catalogs: COS- mean rms (rms) was calculated in concentric rings with MOSVLA J095755.84+015804.2, J095758.04+015825.2 a width of 2′ using the respective sensitivity maps for & J095800.79+015857.1). Besides these newly identified both images (Fig. 4 for the Deep mosaic and Fig. 12 in or augmented multi-component sources there was also a Schinnerer et al. (2007) for the Large one). The ‘bump’ small number of multi-component sources that had ei- in the measured mean noise level at a radius r ≈ 45′ is ther (i) not been previously listed in the Large catalog due to the sensitivity maps’ rectangular geometry. In or- or (ii) are situated outside the area searched during the der to obtain an estimate of the minimal average noise construction of the Large catalog. level at this distance from the field centre we fitted a cubic spline to the measured rms in this region. The re- 5.3. Sources present in only one catalog sulting curve for the Large mosaic, when scaled to 4.5σ, defines a lower envelope to the measurements (ignoring Apart from the sources that are present in both cata- detections in sites with locally lower rms noise). The 4σ logs, objects present only in one of the two catalogs have limit of the Deep catalog corresponds to the 4.5σ limit of been added to the list as well. The exact procedures the Large catalog for the inner radii with r ≤ 15′ while adopted are described in the following. it increases to the 5σ limit of the Large catalog at larger radial distances. 5.3.1. Sources identified in the Large image only The fraction of sources in the VLA-COSMOS Large All sources in the Large catalog (v2.0) with a catalog (v1.0) which are at a given distance from the S/NLarge ≥ 5 but without a match in the Deep cata- field centre and have Speak < 4σDeep is shown in Fig. 6b. log are in principle valid candidates for inclusion in the The increase of this fraction with distance from the field joint source list, provided they are not flagged as detec- center will affect the radial dependence of the number of tions potentially due to side lobes of strong radio sources. successful matches between the two catalogs. We veri- Since SAD tries to fit Gaussians to all peaks above a cer- fied that the success rate of the catalog matching shows tain limit, the fact that certain sources were not found by no distance dependence (Fig. 7). While the fraction SAD during the construction of the Deep catalog suggests of sources in the Large catalog (v1.0) for which a coun- that some of these objects might be mere noise peaks terpart could be identified decreases at larger distances or have very unusual morphologies. Similarly, we expect (solid black line), the effective matching success rate de- that a fraction of the sources with S/N < 4.5 might be fined as ratio of the measured matching success rate and real. the fraction of sources with peak flux densities below 4σ Therefore, we measured for all sources in the Large in the Deep mosaic stays basically constant (thick grey catalog (v1.0) without a match in the Deep catalog the line) within a range of 60 and 80%. peak flux density in the Deep image at 1.5′′ and 2.5′′ Thus we conclude that the direct positional correlation resolution at the according position from the Large cat- of catalog entries has not caused sources to be system- alog following the steps described in §6.2.1 and subse- atically missed as a function of (radial) field distance in quently discarded 959 candidates with S/N17.5′′ < 5 at  VLA-COSMOS Survey IV. 5 both resolutions while it was possible to keep 393 of the effect Bondi et al. (2008) adopted the approach of boost- unmatched sources with S/NLarge ≥ 4.5 from the Large ing the measured values of the peak flux density Speak catalog (v1.0). by different amounts depending on the distance from the center of the VLA-COSMOS field. The boost factors 5.3.2. Sources identified in the Deep image only were derived from tests comparing the peak flux den- 1155 objects – predominantly in the range of 4≤ sities of sources at the centers in individual pointings S/NDeep <8 – were detected in the Deep mosaic which and the derived flux densities in the final Large mosaic. did not have a counterpart in the original Large catalog This approach can be used because the sensitivity map (v1.0). In order to assess the significance of these sources (cf. Fig. 12 in Schinnerer et al. (2007)) for the VLA- which are potentially new detections due to the different COSMOS Large project is fairly uniform over significant source extraction area for the Deep catalog, the increased areas, i.e. all locations suffer similarly from the combined sensitivity in the central square degree and the fact that BWS effect of the contributing pointings. However, the the larger beam is more sensitive to slightly extended additional data for the central seven pointings for the sources (as SAD works with peak fluxes), we adopted the Deep Project changed the uniformity of the sensitivity approach outlined below. While the S/NDeep = 4 cut distribution over the COSMOS field. We thus exam- corresponds at least to the detection threshold used for ine both the method of Bondi et al. (2008) as well as an the original Large catalog, our experience with this cat- alternative based on a modeled sensitivity map for the alog showed that the number of spurious sources rises VLA-COSMOS Deep project. significantly below S/N < 5 due to small mismatches be- The model sensitivity distribution for an individual tween the real local rms and the estimated value in the pointing was constructed using the beam pattern for a constructed rms map. VLA antenna at 1.465 GHz, as specified in the help page Thus, the local S/N was checked at 2.5′′ resolution as for the AIPS task PBCOR: well as at 1.5′′ resolution. All sources with S/N17.5′′ ≥ 5 were included in the joint list. In total, 178 of 283 1−1.343×10−3 X 2 +6.579×10−7 X 4 −1.186×10−10 X 6 , unmatched Deep detections with S/NDeep ≥ 5 were (2) taken over in the Joint catalog. Of 872 sources with where X is the product of the distance from the point- 4 ≤ S/NDeep < 5 some 136 met our selection criteria. ing center in arcminutes with the observing frequency in GHz. The full model sensitivity map (Fig. 11) was 6. JOINT CATALOG then assembled by adding the individual beam patterns In the following we describe how the flux densities and which were weighted according to the observation time all other measurements provided in the catalog were de- dedicated to each pointing (i.e., the central 7 pointings rived and which corrections have been applied to the tab- were weighted more strongly than the others by a factor ulated values. of 1.4). To estimate the magnitude of the required BWS cor- 6.1. Correction for Bandwidth Smearing rection we closely follow the procedure described by Bandwidth smearing (BWS) can occur in radio synthe- Bondi et al. (2008). In brief, a primary beam correction sis imaging when a finite frequency range is observed. It was applied to the images of the 23 individual point- causes a radial smearing that becomes more severe with ings and they were convolved to the same resolution of increasing distance from the phase center (or center of FWHM 2.5′′ as the Deep mosaic. Then we measured a pointing), as the phase calibration is mathematically the peak flux densities of selected sources in these in- speaking only correct for a given frequency at the phase dividual images and compared their values with those center. Thus the effect is similar to chromatic aberration obtained from the same measurement carried out at the in optical imaging. A detailed explanation and discus- corresponding position in the mosaic. sion of the effect of bandwidth smearing for the VLA is At this stage, it is mandatory to use sources with a given, e.g., by Thompson (1999) and Bridle & Schwab peak flux density that has not been significantly dimin- (1999). ished by the BWS effect. The dimensionless parameter θ0 For the VLA-COSMOS project with an observed band- β = ∆νν0 θFWHM (see Bridle & Schwab (1999) and the VLA width of 3.125 MHz for a single channel, a (final) radial observational status summary13 ) makes it possible to in- smearing of 2.25′′ is expected at a radial distance of 15′ fer the amount by which peak flux densities are reduced. from the pointing center, i.e. close to the cut-off radius of In the case of the VLA-COSMOS Large image with an the individual pointings used when creating the mosaic adopted beam width of θFWHM ∼ 1.5′′ the peak flux den- (for equations and details, see Bridle & Schwab 1999). sities of sources within 5′ of the pointing centers are ex- This bandwidth smearing effect also causes a decrease of pected to be reduced by less than 5%. At a resolution of the peak flux density as the emission is now distributed θFWHM ∼ 2.5′′ , as in the case of the Deep image, the same over a larger area. The integrated flux density, however, small decrease still exists as far as 8′ from the pointing will not be affected. As sources in the final mosaic can center. Thus we were able to use a contiguous area cov- be covered by up to 7 individual pointings, the impact of ering a significant fraction of the entire VLA-COSMOS bandwidth smearing strongly varies at each location in mosaic (see area indicated by dashed white circles in Fig. the mosaic as a given source is separated by a different 11). A correspondingly larger number of sources could amount from the centers of the individual pointings. thus be used to check the validity of the adopted method Thus, the effect of bandwidth smearing depends on for BWS correction. In the following we will quantify the the resolution of the image. In correcting the measure- ments presented in the Large catalog (v2.0) for the BWS 13 http://www.vla.nrao.edu/astro/guides/vlas/current/node15.html 6 Schinnerer et al. magnitude of the BWS effect using only sources within following five steps were required to obtain the position, 5′ of the center of either of the 23 VLA-COSMOS point- peak and integrated flux density, as well as the size for ings – this ensures that peak flux densities measured on each source. All measurements described in the following individual pointings should differ by less than 2% from were carried out on the Deep Mosaic. In particular, this the nominal value – and we will use the larger number also applies to those sources which were not part of the of sources out to 8′ to check the quality of the approach. initial Deep catalog (see §5.3.1). Since the aim is to correct for the BWS effect introduced by overlapping pointings, strictly speaking no (valid) cor- 1. The (parametric) integrated flux Stotal was derived rections can be derived for sources located in the edges using the task SAD. In the case of the ‘twin’ sources, (i.e. those covered by only single pointing, see Fig. 12). JMFIT was used instead (for details see §4). At the Fig. 12 shows the comparison of the two different same time the convolved shape parameters were methods to correct for the BWS effect. It should be determined. noted that the maximum decrease observed is only 10%. 2. The peak flux density was obtained using the task The top row displays the uncorrected values of the quan- MAXFIT15. The square search box for MAXFIT was tity S = Speak, mosaic /Speak, pointing , i.e. the ratio of the centered on the position of the Gaussian fit from peak flux densities measured in the mosaic and in an in- the previous step. We used an initial box size of dividual pointing, as a function of normalized sensitivity 3 pixels that was increased in steps of 2 pixels if (left column) and radial distance from the field center MAXFIT did not converge. The measured peak flux (right column). The median ratio hS i of peak flux den- density Speak was corrected for the effects of band- sities in a specific bin is marked by an open circle. The width smearing (see §6.1) resulting in the corrected associated error bars span the interquartile range of S measured peak flux density Speak, corr. . within a bin in distance or model sensitivity. Median and errors are reported in red when only sources within 5′ of 3. The source position was set to the position of the the center of one of the 23 pointings are used (i.e. only MAXFIT peak. the red points) and in black when all sources within 8′ are considered (i.e. the red and grey points). 4. As a significant fraction of sources is resolved at It is obvious from the upper row of Fig. 12 that the the faint flux levels probed here, it is necessary to effect of bandwidth smearing disappears14 (i.e. the value classify the sources into resolved and unresolved. of the ratio S approaches unity) towards the edge of We followed the approach used by Schinnerer et al. the field where there are no overlapping pointings or, (2007) and Bondi et al. (2008). The method relies conversely, at lower sensitivity values which stem from on the fact that the ratio between integrated and regions in the field covered by only one pointing. The peak flux density is a measure of the spatial extent best-fitting linear trends define which BWS correction of a radio source. A detailed description of the should be applied to sources lying at a given distance or method is given in Appendix A together with a within a region of a specific sensitivity value, depending description of a suite of Monte Carlo simulations on the method considered. In the middle row the ratios which we use to calibrate our classification scheme. Ssensitivity correction are corrected according to the second Fig. 13 is the diagnostic diagram used to identify method using the model sensitivity map (Fig. 11), i.e. the resolved sources based on our simulations. The the the slope of the line as a function of sensitivity (up- line defining the lower boundary to the locus of per left panel). Thus the distribution of S is necessarily resolved sources in Fig. 13 is given by the following flat when plotted as a function of model sensitivity. This equation16 : flatness is also preserved when the corrected peak flux densities are plotted against distance from the center of −11/(S/N )1.45 the field. The BWS correction (third row) based on the Stotal /Speak,corr. = 0.35 , (3) comparison of the peak flux density ratios as a function of distance from the center of the VLA-COSMOS field A discussion on possible systematic biases inherent following Bondi et al. (2008) also fares well in straighten- to this method is presented in the Appendix. We ing the distribution of corrected ratios Sdistance correction performed an independent check of this classifica- with respect to both distance (again, a requirement) and tion method by adopting the JMFIT approach used model sensitivity. However, a weak correlation of peak by, e.g., Miller et al. (2008) and Owen & Morrison flux density ratios with sensitivity is still present after (2008). As JMFIT also derives upper and lower lim- the application of this method. Therefore we conclude its on the deconvolved size of the major and minor that the sensitivity-based approach is slightly better for 15 Note that MAXFIT derives the maximum by fitting a quadratic BWS correction and will use it to compute the corrected function to a 3×3 pixel map. Given that our Large (Deep) CLEAN peak flux density, Speak, corr. , provided in the Joint cata- beam is well sampled with at least 4 (7) pixels per axis the map log (see also Tab. 3). value and the MAXFIT value agree within ∼0.2%. 16 Note that in previously published catalogs (see references 6.2. Catalog Entries above) using this approach, the line described by eq. (3) follows from logarithmically mirroring a lower envelope to the points below 6.2.1. Single-component sources the line Stotal /Speak,corr. = 1 above this very line. In Fig. 13 we 95% of the sources in the VLA-COSMOS Joint cat- show that the negative mirror image of eq. (3) is a very conserva- tive lower envelope of the kind just described. As a consequence, alog are well fit by a single Gaussian component. The the classification into resolved/unresolved sources adopted in this paper based on the simulations in the Appendix, leads to a signif- 14 Note that at these radial distance the BWS effect can no icantly larger fraction of sources being classified as ‘unresolved’ in longer be determined due to the lack of overlapping fields the Joint catalog.  VLA-COSMOS Survey IV. 7 axis of a source, these limits can also be used to for the source shape parameters as these are meant as a divide sources into resolved and unresolved. If the rough measure of the angular size of the source. For the upper limit of the major axis was equal to zero, a same reason we also do not provide values for the er- source is considered unresolved. Comparison be- ror on the major and minor axes of the Gaussian single tween both methods shows that <1% of our re- component sources. solved sources would be considered unresolved us- The error on the peak flux density of single component ing the JMFIT criterion while 66% of the unresolved sources is defined as the local rms noise at the position sources would be resolved. Thus the total fraction of the source. For unresolved sources the integrated flux of resolved sources would increase to ∼70% similar density and peak flux density are identical and hence the to the fractions found by Miller et al. (2008) and error on the integrated flux density also corresponds to Owen & Morrison (2008). Using equation (3) we the local rms noise. have found 405 resolved sources for which the inte- In calculating the error σStotal for resolved sources grated flux is given by the total flux of the Gaus- in the VLA-COSMOS Joint catalog we follow the ap- sian SAD/JMFIT fit, and 2329 unresolved sources for proach of Hopkins et al. (2003) (see also Bondi et al. which the integrated flux is `a posteriori set equal 2003; Schinnerer et al. 2004) based on the assumption to the corrected peak flux density. These num- that the relative uncertainty σStotal /Stotal in the inte- bers correspond to 14% and 81% of the sources in grated flux density is due to uncertainties µdata in the the Joint catalog, respectively, while the remaining data and uncertainties µfit in the Gaussian fit: 5% represent radio sources with multiple Gaussian s components (see following section). σStotal µdata 2  µfit 2 = + . (4) 5. The deconvolved source size parameters (major Stotal Stotal Stotal axis θM, dec , minor axis θm, dec and position angle PAdec ) were calculated for the resolved sources ac- The two factors beneath the square root in the equation cording to the algorithms of the AIPS task JMFIT. are (see Windhorst et al. (1984) and Condon (1997), as For unresolved sources all three are set to zero. well as the explanations in Schinnerer et al. (2004)): s 6.2.2. Multi-Component Sources µdata  −2 S For sources consisting of multiple Gaussian compo- = + 0.012 (5) Stotal N nents the integrated flux density was manually measured on the VLA-COSMOS Deep image with the AIPS task (where S/N = Speak /rms) and TVSTAT in order to encompass all the emission from these irregularly shaped sources. As the individual compo- s    µfit 2 θB θb 2 2 nents all have flux density peaks of their own, no peak = + + . (6) value is specified for multi-component sources in Tab. 3. Stotal ρ2S θM θm ρ2M ρ2m Positions were adopted from the VLA-COSMOS Large project catalogs whenever possible. In all other cases the Here θB and θb are the major and minor axis of the beam position was chosen on an individual basis based on the (in the present case of a circular beam θB = θb = 2.5′′ ) radio morphology of the sources. In practice this usu- and, analogously, θM and θm the major and minor axis ally corresponds to a luminosity weighted mean of the of the measured (i.e. convolved) flux distribution. The positions of the subcomponents (as already done for the S/N values of the fit – ρS , ρM , and ρm – are parameter- multi-component sources in the Large Project catalog, dependent: see §6 of Schinnerer et al. 2007). Measurements of the major and minor axes along the θM θm  θB α  θb β  2 S direction of maximal extension of the source and at right ρ2X = 1+ 1+ , (7) angles to the latter, respectively, are provided for multi- 4θB θb θM θm N component sources. However, no estimate of the position with α = β = 1.5 for ρS , α = 2.5 and β = 0.53 for ρM , angle is provided as it is usually impossible to define for and α = 0.5 and β = 2.5 for ρm . the complex geometry of these sources. Note that while The errors on the position of single-component sources the coordinates were taken over, fluxes and dimensions of are multi-component objects adopted from the Large catalog (v2.0) were re-measured in the Deep project mosaic at s 2.5′′ resolution. 2 θM 2 2 θm 2 ∆α = ǫ2α + 2 sin(PA) + cos(PA) 6.3. Errors on the Catalog Entries 4 ln(2)ρS 4 ln(2)ρ2S s Below we summarize how the errors on the source prop- 2 θM 2 2 θm 2 erties listed in the VLA-COSMOS Joint catalog were ∆δ = ǫ2δ + 2 cos(PA) + sin(PA) , 4 ln(2)ρS 4 ln(2)ρ2S computed. In the case of the single-component ob- jects analytic formulae from the literature are adopted. where ǫα = ǫδ = θB /10 ≈ 0.25′′ are the positional For the multi-component objects which have a manually calibration errors on right ascension and declination and measured integrated flux density Stotal , the accuracy is PA is the measured position angle of the Gaussian fit. about 10% based on the comparison of repeated flux mea- surements on the same objects by different people. For 6.4. Description of the VLA-COSMOS Joint Catalog multi-component objects no error estimates are provided 8 Schinnerer et al. The information on the procedures and equations used 0 – single component source to calculated each entry in the VLA-COSMOS Joint 1 – multi-component source identified in VLA-COS- catalog has been presented above. For each source we MOS Large image list the source name, the ID of the source in the VLA- 2 – multi-component source identified in VLA-COS- COSMOS Large Project catalogs (where available), as MOS Deep image well as the derived source properties and the associated Column(18): Flag for catalog membership errors. Furthermore, the source properties and flags associated with each object are explained. All 2865 -1 – only detected in the VLA-COSMOS Large image radio sources in the VLA-COSMOS Joint catalog are 0 – detected in both the VLA-COSMOS Large & listed by increasing right ascension in Tab. 3 with the Deep image following columns: 1 – only detected in the VLA-COSMOS Deep image Column(19): Flag specifying at which resolu- Column(1): Source name tion the source was detected with S/N17.5′′ ≥ 5 Column(2): Source name in VLA-COSMOS Large -1 – detected with S/N17.5′′ ≥ 5 only at a resolution Project catalog (set to J999999.99+999999.9 if not of 1.5′′ present in the VLA-COSMOS Large catalog) 0 – detected with S/N17.5′′ ≥ 5 at both 1.5′′ and 2.5′′ Column(3): Right ascension (J2000.0) in degrees resolution 1 – detected with S/N ≥ 5 only at a resolution of Column(4): Declination (J2000.0) in degrees 2.5′′ , but in both the large and Column(5): Right ascension (J2000.0) in hexagesimal small scale (105′′ and 17.5′′ box size, respectively) format rms map 2 – detected only with S/N17.5′′ ≥ 5 (but not in the Column(6): Declination (J2000.0) in hexagesimal format large scale rms map) at a Column(7): rms uncertainty in right ascension in arc- resolution of 2.5′′ seconds Column(8): rms uncertainty in declination in arcseconds 6.5. Comparison between Large (v2.0) and Joint Column(9): Peak flux density and its rms uncertainty Catalog in mJy/beam We compared the source properties in the flux-limited Column(10): Peak flux density corrected for bandwidth Large catalog (v2.0) and in the Joint catalog for those smearing in mJy/beam 2256 catalog entries which are common to both catalogs. The different resolution in the VLA-COSMOS Deep and Column(11): Integrated flux density and its rms uncer- Large mosaics leads to a different sensitivity to extended tainty in mJy emission. Hence, source shape parameters cannot be ex- Column(12): rms measured in the RMSD sensitivity map pected to show exact agreement between the two cata- in mJy/beam logs. Therefore the comparison is limited to the peak Column(13): Deconvolved major axis size θM, dec in and integrated flux densities. The results of the compar- arcseconds ison are illustrated in Fig. 15 and Fig. 16 for the peak and integrated flux density, respectively. In particular, Column(14): Deconvolved minor axis size θm, dec in it is instructive to compare the peak values of sources arcseconds that are classified as unresolved in both catalogs. These Column(15): Deconvolved position angle of source sources are highlighted in light grey in Fig. 15 and are PAdec (counterclockwise from North) in degrees seen to scatter well around the diagonal line of 1:1 cor- respondence between the measurements from the Joint Column(16): Flag for resolved and unresolved sources17 and Large mosaics. The comparison of the integrated -2 – unresolved only in VLA-COSMOS Deep image flux densities (Fig. 16) shows that there is a small popu- -1 – resolved only in VLA-COSMOS Large image lation of objects (light grey points) lying above the diago- 0 – unresolved in both VLA-COSMOS Large & nal (i.e. their integrated flux density is higher in the Joint Deep images catalog). These objects are sources that were previously 1 – resolved in both VLA-COSMOS Large & Deep classified as unresolved but have acquired the status of re- images solved in the Joint catalog, implying that their integrated 2 – resolved only in VLA-COSMOS Deep image flux densities have necessarily increased when measured in the Deep mosaic. This class of objects accounts for Column(17): Flag for distinction of multi-component nearly all of the outliers in the plot. Multi-component and single-component sources sources (dark grey crosses) in general lie also above the 17 Note that the flag value of ‘-2’ is only assigned to sources diagonal bisector, reflecting the enhanced sensitivity to that were not in the VLA-COSMOS Large Project catalog. The extended emission in the Deep image. For most other flag ‘-1’, on the other hand, is used for sources that were classified objects the new and previous estimate of integrated flux as resolved in the Large Project catalog but are listed as unresolved density is in good agreement and hence also with flux in the Joint catalog. density estimates of the NVSS and FIRST surveys as shown in Schinnerer et al. (2007). 7. SUMMARY  VLA-COSMOS Survey IV. 9 Continued analysis of the VLA-COSMOS catalog pre- smearing the Deep mosaic has a resolution of 2.5′′ (com- sented in Schinnerer et al. (2007) and the completion of pared to 1.5′′ for the Large mosaic) and it was used to the VLA-COSMOS Deep project motivated the compi- create a corresponding source catalog using the task SAD. lation of a new radio catalog for the COSMOS field. An input list for the new Joint catalog was compiled The VLA-COSMOS Joint catalog was generated by com- by combining the revised Large catalog (v2.0) and the bining the catalogs of the VLA-COSMOS Large Project new Deep catalog. The criteria were set such that no with a newly created source catalog (Deep catalog) from particular bias against slightly extended radio sources the 2.5′′ resolution Deep mosaic. This catalog is already was present when selecting the sources. All properties available for download by the public at the COSMOS of the radio sources listed in the Joint catalog have been archive at IPAC/IRSA18 . A comparison of the depth and derived in the 2.5′′ resolution Deep mosaic. areal coverage for a representative sample of deep field The construction of the Joint catalog was motivated radio surveys at 1.4 GHz shows that the VLA-COSMOS by the desire to provide a catalog of bona-fide radio covers the largest area at its depth and angular resolu- sources in the COSMOS field for distinct science appli- tion (Fig. 17). Thus it should be well suited to also cations that are interested in the radio properties of cer- study effects of the Large Scale Structure on the pres- tain populations of galaxies. On the other hand the re- ence/absence of radio emission. vised Large catalog (v2.0) is flux-limited (in radio), has a The reduction and analysis of the deeper 20 cm obser- fairly uniform sensitivity coverage and its completeness vations of the central 7 pointings of the VLA-COSMOS is well characterized (see Bondi et al. 2008), thus it is projects using the VLA in A configuration have been de- well suited for, e.g., studies of the faint radio population scribed in detail (also referred to as VLA-COSMOS Deep (such as, e.g., Smolˇci´c et al. 2008, 2009a,b). project). In order to minimize the effect of bandwidth APPENDIX BIAS ON THE ESTIMATION OF INTEGRATED FLUXES AND SOURCE SIZES At low signal-to-noise, source fitting algorithms (e.g. JMFIT) are increasingly liable to return false results, an effect which is further amplified by the fact that there are multiple free parameters (e.g. total flux, two angular size components) which cannot be legitimately fixed to make the fit more robust. This Appendix discusses biases in the total flux measurements in our catalog which may arise as a consequence of this. We will not study the uncertainties, systematic or not, which might affect the angular source sizes we quote in the catalog, as we consider these less important for most users of the Joint catalog. Since the unresolved sources in the catalog have a differently determined (MAXFIT) total flux, any biases in the Gaussian source fitting will only be of importance for sources classified as resolved. How many sources are affected is determined by the choice of classification scheme (see Section 6.2.1). Part of this section hence deals with our choice of a criterion that both limits flux biases in the final sample and also performs satisfactorily when it comes to separating resolved from unresolved objects. We first describe our simulations. We performed a set of Monte Carlo simulations following the approach used for the Large project (Bondi et al. 2008). Circular mock sources with random flux densities down to 30 µJy and sizes following the measured radio source counts and an angular size distribution hθi ∼ S m with an exponent m=0.5 (see Bondi et al. 2008, for details) were inserted into the Deep mosaic. They are not subject to bandwidth smearing effects. We searched at the positions of these 16,000 sources using MAXFIT followed by the application of JMFIT (single component fit). We verified that this approach gives the same results as using SAD. Roughly 6,500 sources could be recovered with a S/N ≥ 5. We also ran simulations with different source size distributions which all returned qualitatively similar results. We would like to remind the reader that the final numbers and errors quoted below are sensitive to the adopted intrinsic angular size distribution and therefore also to the resolution of the mosaic used. As a significant fraction of sources is resolved at the faint flux levels probed by the VLA-COSMOS project, it is necessary to classify the sources into resolved and unresolved. We adopt the approach used by Schinnerer et al. (2007) and Bondi et al. (2008) (for applications to other radio surveys, see Prandoni et al. 2000; Bondi et al. 2003). The method relies on the fact that the ratio between integrated and peak flux density is a measure of the spatial extent of a radio source (given by the major and minor axis θM and θm ) in comparison to the size of the synthesized beam (with major and minor axis θB and θb ): Stotal /Speak = (θM θm )/(θB θb ) . (A1) On the other hand we can estimate the limiting (intrinsic) size θlimit at a given S/N above which a source could be classified as resolved. The threshold θlimit is estimated using eq. (16) of Condon (1997): according to this expression the error in size (∆θsrc. ) is proportional to (S/N)−1 (see Fig. 18; dashed line) and a point source would thus on average be liable to havepa convolved size θobs. = θB ± ∆θsrc. . All sources with intrinsic sizes below the lower (red) line in Fig. 2 18, i.e. θlimit = θobs. − θB 2 – the inferred size of a point source subject to the S/N -dependent error ∆θ src. , cannot be expected to be resolvable. The ratio Stotal /Speak of our ‘unresolvable’ (black; θ < θlimit ) and ‘resolvable’ sources (grey; θ > θlimit ) is plotted as a function of the S/N with which the simulated sources are detected in Fig. 19. After calibration with the simulated sources, this diagnostic diagram will be used to identify the unresolved sources in the real catalog. By necessity, measured values with Stotal /Speak < 1 are due to the influence of image noise on the determination of source flux density and/or size. In general, the noise can both lead to an artificial reduction or increase of the true flux density ratio (as peak and integrated flux density are determined independently) implying that not all sources with 10 Schinnerer et al. Stotal /Speak > 1 are genuinely resolved either. As can be seen in Fig. 19 there is a fairly well- defined locus above c which only few unresolvable sources occur. We approximate it by a line with functional form Stotal /Speak = a−b/(S/N ) . As our main goal is to minimize total flux density biases, we define a rather conservative separating line: −11/(S/N )1.45 Stotal /Speak = 0.35 . (A2) This choice ensures that a minimal number of unresolvable mock sources is classified as resolved (in the following a ‘resolved’ source is understood to lie above the line given by eq. (A2) in Fig. 19), especially at low S/N where the errors on total flux measurements from Gaussian fits are largest. A curve rising more slowly toward low S/N would have raised the number of resolvable sources actually classified as resolved. It would also, however, have increased the number of misclassified unresolvable sources. In the upper window of Fig. 19 we illustrate which fraction of resolvable (unresolvable) mock sources is classified as unresolved (resolved) in a given range of S/N . In total, ∼38% of the resolvable and ∼3% of the unresolvable sources are ‘misclassified’, leading to an overall success rate of nearly 87%. Given the classification . of our mock sources . into resolved and unresolved following eq. (A2), their total fluxes can now be set to Stotal = SJMFIT and Stotal = Speak , respectively (see Section 6.2.1), and any systematic flux biases at low S/N quantified. In Fig. 20, we show the difference between the input (Stin ) and output (Stout ) integrated flux densities as a function of S/N (left-hand column) and as histograms (right). As the mock sources were randomly injected into the mosaic and no attempt has been made to avoid confusion with other sources, the outliers in the distributions can be explained due to confusion with real sources. However, the median derived from the distribution should be unaffected by these outliers. The median relative offset between flux values in different ranges of S/N is listed in Tab. 5. In our lowest S/N bin (5 ≤ S/N ≤ 6) we detect a small bias of ∼5% (in the sense that recovered fluxes overestimate the true value) for all sources which becomes even more negligible for higher significance sources. Although this bias is signficantly larger for resolved sources, its effect on statistical tools, such as source counts, is minimal, as these use all sources and the fraction of resolved source is at low S/N is very small. As the scatter in the derived properties (at fixed S/N ) is much larger, we conclude that a systematic correction of the derived integrated flux on a source by source basis is neither straightforward nor indispensable. The National Radio Astronomy Observatory (NRAO) is operated by Associated Universities, Inc., under cooperative agreement with the National Science Foundation. We would like to thank the NRAO for their support during this project. We thank our referee Jim Condon for his insightful comments for improving the paper. CC thanks the Max- Planck-Gesellschaft and the Humboldt-Stiftung for support through the Max-Planck-Forschungspreis. CC and AD acknowledge support through NASA grant HST-GO-09822.33-A. MTS acknowledges support by the German DFG under grant SCHI 536/3-2 and SCHI 536/3-3. VS acknowledges support by the German DFG under grant SCHI 536/3-1. This project has been (partially) funded by the DFG Priority Programme 1177 ‘Galaxy Evolution’. REFERENCES Bertoldi, F., Carilli, C., Aravena, M., Schinnerer, E., Voss, H., Carilli, C. L., Lee, N., Capak, P., Schinnerer, E., Lee, K.-S., Smolˇ ci´ c, V., Jahnke, K., Scoville, N., Blain, A., Menten, K. M., McCraken, H., Yun, M. S., Scoville, N., Smolˇ ci´c, V., Giavalisco, Lutz, D., Brusa, M., Taniguchi, Y., Capak, P., Mobasher, B., M., Datta, A., Taniguchi, Y., & Urry, C. M. 2008, ApJ, 689, Lilly, S., Thompson, D., Aussel, H., Kreysa, E., Hasinger, G., 883 Aguirre, J., Schlaerth, J., & Koekemoer, A. 2007, ApJS, 172, Carilli, C. L., Murayama, T., Wang, R., Schinnerer, E., 132 Taniguchi, Y., Smolˇ ci´ c, V., Bertoldi, F., Ajiki, M., Nagao, T., Bondi, M., Ciliegi, P., Schinnerer, E., Smolˇ ci´ c, V., Jahnke, K., Sasaki, S. S., Shioya, Y., Aguirre, J. E., Blain, A. W., Scoville, Carilli, C., & Zamorani, G. 2008, ApJ, 681, 1129 N., & Sanders, D. B. 2007, ApJS, 172, 518 Bondi, M., Ciliegi, P., Zamorani, G., Gregorini, L., Vettolani, G., Condon, J. J. 1997, PASP, 109, 166 Parma, P., de Ruiter, H., Le Fevre, O., Arnaboldi, M., Guzzo, Condon, J. J., Cotton, W. D., Yin, Q. F., Shupe, D. L., L., Maccagni, D., Scaramella, R., Adami, C., Bardelli, S., Storrie-Lombardi, L. J., Helou, G., Soifer, B. T., & Werner, Bolzonella, M., Bottini, D., Cappi, A., Foucaud, S., Franzetti, M. W. 2003, AJ, 125, 2411 P., Garilli, B., Gwyn, S., Ilbert, O., Iovino, A., Le Brun, V., Elvis, M., Civano, F., Vignali, C., Puccetti, S., Fiore, F., Marano, B., Marinoni, C., McCracken, H. J., Meneux, B., Cappelluti, N., Aldcroft, T. L., Fruscione, A., Zamorani, G., Pollo, A., Pozzetti, L., Radovich, M., Ripepi, V., Rizzo, D., Comastri, A., Brusa, M., Gilli, R., Miyaji, T., Damiani Anton Scodeggio, M., Tresse, L., Zanichelli, A., & Zucca, E. 2003, Koekemoer, F., Finoguenov, A., Brunner, H., Urry, C. M., A&A, 403, 857 Silverman, J., Mainieri, V., Hasinger, G., Griffiths, R., Carollo, Bridle, A. H., & Schwab, F. R. 1999, in Astronomical Society of M., Hao, H., Guzzo, L., Blain, A., Calzetti, D., Carilli, C., the Pacific Conference Series, Vol. 180, Synthesis Imaging in Capak, P., Ettori, S., Fabbiano, G., Impey, C., Lilly, S., Radio Astronomy II, ed. G. B. Taylor, C. L. Carilli, & R. A. Mobasher, B., Rich, M., Salvato, M., Sanders, D. B., Perley, 371–381 Schinnerer, E., Scoville, N., Shopbell, P., Taylor, J. E., Capak, P., Aussel, H., Ajiki, M., McCracken, H. J., Mobasher, B., Taniguchi, Y., & Volonteri, M. 2009, ArXiv e-prints Scoville, N., Shopbell, P., Taniguchi, Y., Thompson, D., Fomalont, E. B., Kellermann, K. I., Cowie, L. L., Capak, P., Tribiano, S., Sasaki, S., Blain, A. W., Brusa, M., Carilli, C., Barger, A. J., Partridge, R. B., Windhorst, R. A., & Richards, Comastri, A., Carollo, C. M., Cassata, P., Colbert, J., Ellis, E. A. 2006, ApJS, 167, 103 R. S., Elvis, M., Giavalisco, M., Green, W., Guzzo, L., Greisen, E. W. 2003, in Astrophysics and Space Science Library, Hasinger, G., Ilbert, O., Impey, C., Jahnke, K., Kartaltepe, J., Vol. 285, Astrophysics and Space Science Library, ed. A. Heck, Kneib, J.-P., Koda, J., Koekemoer, A., Komiyama, Y., 109–+ Leauthaud, A., Lefevre, O., Lilly, S., Liu, C., Massey, R., Hasinger, G., Cappelluti, N., Brunner, H., Brusa, M., Comastri, Miyazaki, S., Murayama, T., Nagao, T., Peacock, J. A., Pickles, A., Elvis, M., Finoguenov, A., Fiore, F., Franceschini, A., Gilli, A., Porciani, C., Renzini, A., Rhodes, J., Rich, M., Salvato, M., R., Griffiths, R. E., Lehmann, I., Mainieri, V., Matt, G., Sanders, D. B., Scarlata, C., Schiminovich, D., Schinnerer, E., Matute, I., Miyaji, T., Molendi, S., Paltani, S., Sanders, D. B., Scodeggio, M., Sheth, K., Shioya, Y., Tasca, L. A. M., Taylor, Scoville, N., Tresse, L., Urry, C. M., Vettolani, P., & Zamorani, J. E., Yan, L., & Zamorani, G. 2007, ApJS, 172, 99 G. 2007, ApJS, 172, 29  VLA-COSMOS Survey IV. 11 Hopkins, A. M., Afonso, J., Chan, B., Cram, L. E., Georgakakis, Scoville, N., Abraham, R. G., Aussel, H., Barnes, J. E., Benson, A., & Mobasher, B. 2003, AJ, 125, 465 A., Blain, A. W., Calzetti, D., Comastri, A., Capak, P., Carilli, Huynh, M. T., Jackson, C. A., Norris, R. P., & Prandoni, I. 2005, C., Carlstrom, J. E., Carollo, C. M., Colbert, J., Daddi, E., AJ, 130, 1373 Ellis, R. S., Elvis, M., Ewald, S. P., Fall, M., Franceschini, A., Ivison, R. J., Chapman, S. C., Faber, S. M., Smail, I., Biggs, Giavalisco, M., Green, W., Griffiths, R. E., Guzzo, L., A. D., Conselice, C. J., Wilson, G., Salim, S., Huang, J.-S., & Hasinger, G., Impey, C., Kneib, J.-P., Koda, J., Koekemoer, Willner, S. P. 2007, ApJ, 660, L77 A., Lefevre, O., Lilly, S., Liu, C. T., McCracken, H. J., Massey, Ko, H. C. 1964, IEEE Transactions on Military Electronics, 8, 225 R., Mellier, Y., Miyazaki, S., Mobasher, B., Mould, J., Norman, Koekemoer, A. M., Aussel, H., Calzetti, D., Capak, P., Giavalisco, C., Refregier, A., Renzini, A., Rhodes, J., Rich, M., Sanders, M., Kneib, J.-P., Leauthaud, A., Le F` evre, O., McCracken, D. B., Schiminovich, D., Schinnerer, E., Scodeggio, M., Sheth, H. J., Massey, R., Mobasher, B., Rhodes, J., Scoville, N., & K., Shopbell, P. L., Taniguchi, Y., Tyson, N. D., Urry, C. M., Shopbell, P. L. 2007, ApJS, 172, 196 Van Waerbeke, L., Vettolani, P., White, S. D. M., & Yan, L. Lilly, S. J., Le F` evre, O., Renzini, A., Zamorani, G., Scodeggio, 2007a, ApJS, 172, 38 M., Contini, T., Carollo, C. M., Hasinger, G., Kneib, J.-P., Scoville, N., Aussel, H., Brusa, M., Capak, P., Carollo, C. M., Iovino, A., Le Brun, V., Maier, C., Mainieri, V., Mignoli, M., Elvis, M., Giavalisco, M., Guzzo, L., Hasinger, G., Impey, C., Silverman, J., Tasca, L. A. M., Bolzonella, M., Bongiorno, A., Kneib, J.-P., LeFevre, O., Lilly, S. J., Mobasher, B., Renzini, Bottini, D., Capak, P., Caputi, K., Cimatti, A., Cucciati, O., A., Rich, R. M., Sanders, D. B., Schinnerer, E., Schminovich, Daddi, E., Feldmann, R., Franzetti, P., Garilli, B., Guzzo, L., D., Shopbell, P., Taniguchi, Y., & Tyson, N. D. 2007b, ApJS, Ilbert, O., Kampczyk, P., Kovac, K., Lamareille, F., 172, 1 Leauthaud, A., Borgne, J.-F. L., McCracken, H. J., Marinoni, Seymour, N., McHardy, I. M., & Gunn, K. F. 2004, MNRAS, 352, C., Pello, R., Ricciardelli, E., Scarlata, C., Vergani, D., 131 Sanders, D. B., Schinnerer, E., Scoville, N., Taniguchi, Y., Simpson, C., Mart´ınez-Sansigre, A., Rawlings, S., Ivison, R., Arnouts, S., Aussel, H., Bardelli, S., Brusa, M., Cappi, A., Akiyama, M., Sekiguchi, K., Takata, T., Ueda, Y., & Watson, Ciliegi, P., Finoguenov, A., Foucaud, S., Franceschini, R., M. 2006, MNRAS, 372, 741 Halliday, C., Impey, C., Knobel, C., Koekemoer, A., Kurk, J., Smolˇci´ c, V., Schinnerer, E., Scodeggio, M., Franzetti, P., Aussel, Maccagni, D., Maddox, S., Marano, B., Marconi, G., Meneux, H., Bondi, M., Brusa, M., Carilli, C. L., Capak, P., Charlot, S., B., Mobasher, B., Moreau, C., Peacock, J. A., Porciani, C., ˇ Jahnke, K., McCracken, H. J., Pozzetti, L., Scaramella, R., Schiminovich, D., Shopbell, P., Ciliegi, P., Ilbert, O., Ivezi´ c, Z., Smail, I., Thompson, D., Tresse, L., Vettolani, G., Zanichelli, Obri´c, M., Salvato, M., Sanders, D. B., Scoville, N., Trump, A., & Zucca, E. 2007, ApJS, 172, 70 J. R., Tremonti, C., Tasca, L., Walcher, C. J., & Zamorani, G. Miller, N. A., Fomalont, E. B., Kellermann, K. I., Mainieri, V., 2008, ApJS, 177, 14 Norman, C., Padovani, P., Rosati, P., & Tozzi, P. 2008, ApJS, Smolˇci´ c, V., Schinnerer, E., Zamorani, G., Bell, E. F., Bondi, M., 179, 114 Carilli, C. L., Ciliegi, P., Mobasher, B., Paglione, T., Muxlow, T. W. B., Richards, A. M. S., Garrington, S. T., Scodeggio, M., & Scoville, N. 2009a, ApJ, 690, 610 Wilkinson, P. N., Anderson, B., Richards, E. A., Axon, D. J., Smolˇci´ c, V., Zamorani, G., Schinnerer, E., Bardelli, S., Bondi, M., Fomalont, E. B., Kellermann, K. I., Partridge, R. B., & Birzan, L., Carilli, C. L., Ciliegi, P., Ilbert, O., Koekemoer, Windhorst, R. A. 2005, MNRAS, 358, 1159 A. M., Merloni, A., Paglione, T., Salvato, M., Scodeggio, M., & Norris, R. P., Huynh, M. T., Jackson, C. A., Boyle, B. J., Ekers, Scoville, N. 2009b, ArXiv e-prints R. D., Mitchell, D. A., Sault, R. J., Wieringa, M. H., Williams, Taniguchi, Y., Scoville, N., Murayama, T., Sanders, D. B., R. E., Hopkins, A. M., & Higdon, J. 2005, AJ, 130, 1358 Mobasher, B., Aussel, H., Capak, P., Ajiki, M., Miyazaki, S., Otoshi, T. Y. 1981, Telecommunications and Data Acquisition Komiyama, Y., Shioya, Y., Nagao, T., Sasaki, S. S., Koda, J., Progress Report, 64, 168 Carilli, C., Giavalisco, M., Guzzo, L., Hasinger, G., Impey, C., Owen, F. N., & Morrison, G. E. 2008, AJ, 136, 1889 LeFevre, O., Lilly, S., Renzini, A., Rich, M., Schinnerer, E., Prandoni, I., Gregorini, L., Parma, P., de Ruiter, H. R., Vettolani, Shopbell, P., Kaifu, N., Karoji, H., Arimoto, N., Okamura, S., G., Wieringa, M. H., & Ekers, R. D. 2000, A&AS, 146, 41 & Ohta, K. 2007, ApJS, 172, 9 Richards, E. A. 2000, ApJ, 533, 611 Thompson, A. R. 1999, in Astronomical Society of the Pacific Sanders, D. B., Salvato, M., Aussel, H., Ilbert, O., Scoville, N., Conference Series, Vol. 180, Synthesis Imaging in Radio Surace, J. A., Frayer, D. T., Sheth, K., Helou, G., Brooke, T., Astronomy II, ed. G. B. Taylor, C. L. Carilli, & R. A. Perley, Bhattacharya, B., Yan, L., Kartaltepe, J. S., Barnes, J. E., 11–36 Blain, A. W., Calzetti, D., Capak, P., Carilli, C., Carollo, Trump, J. R., Impey, C. D., McCarthy, P. J., Elvis, M., Huchra, C. M., Comastri, A., Daddi, E., Ellis, R. S., Elvis, M., Fall, J. P., Brusa, M., Hasinger, G., Schinnerer, E., Capak, P., Lilly, S. M., Franceschini, A., Giavalisco, M., Hasinger, G., Impey, S. J., & Scoville, N. Z. 2007, ApJS, 172, 383 C., Koekemoer, A., Le F` evre, O., Lilly, S., Liu, M. C., Wells, D. The Penguin Dictionary of Curious and Interesting McCracken, H. J., Mobasher, B., Renzini, A., Rich, M., Geometry, London: Penguin Books, pp. 30-31, 1991 Schinnerer, E., Shopbell, P. L., Taniguchi, Y., Thompson, Windhorst, R. A., van Heerde, G. M., & Katgert, P. 1984, A&AS, D. J., Urry, C. M., & Williams, J. P. 2007, ApJS, 172, 86 58, 1 Schinnerer, E., Carilli, C. L., Scoville, N. Z., Bondi, M., Ciliegi, Zamojski, M. A., Schiminovich, D., Rich, R. M., Mobasher, B., P., Vettolani, P., Le F` evre, O., Koekemoer, A. M., Bertoldi, F., Koekemoer, A. M., Capak, P., Taniguchi, Y., Sasaki, S. S., & Impey, C. D. 2004, AJ, 128, 1974 McCracken, H. J., Mellier, Y., Bertin, E., Aussel, H., Sanders, Schinnerer, E., Smolˇ ci´ c, V., Carilli, C. L., Bondi, M., Ciliegi, P., D. B., Le F` evre, O., Ilbert, O., Salvato, M., Thompson, D. J., Jahnke, K., Scoville, N. Z., Aussel, H., Bertoldi, F., Blain, Kartaltepe, J. S., Scoville, N., Barlow, T. A., Forster, K., A. W., Impey, C. D., Koekemoer, A. M., Le Fevre, O., & Urry, Friedman, P. G., Martin, D. C., Morrissey, P., Neff, S. G., C. M. 2007, ApJS, 172, 46 Seibert, M., Small, T., Wyder, T. K., Bianchi, L., Donas, J., Scott, K. S., Austermann, J. E., Perera, T. A., Wilson, G. W., Heckman, T. M., Lee, Y.-W., Madore, B. F., Milliard, B., Aretxaga, I., Bock, J. J., Hughes, D. H., Kang, Y., Kim, S., Szalay, A. S., Welsh, B. Y., & Yi, S. K. 2007, ApJS, 172, 468 Mauskopf, P. D., Sanders, D. B., Scoville, N., & Yun, M. S. 2008, MNRAS, 385, 2225 12 Schinnerer et al. Fig. 1.— The VLA-COSMOS Deep image at 2.5′′ resolution.  VLA-COSMOS Survey IV. 13 Fig. 2.— Distribution of the noise in the VLA-COSMOS Deep mosaic. Pixel values extracted from a 52′ × 52′ box centered on the COSMOS field center show a Gaussian distribution in agreement with our assumption of Gaussian noise. The fitted Gaussian (dashed line) has a rms of 12.09 µJy/beam (σ). The tail at high flux densities is consistent with the presence of sources in the image which were not excluded during the extraction process. Fig. 3.— The rms level vs. the cumulative (left axis) and relative area (right axis) covered by the VLA-COSMOS Large (at a resolution of 1.5′′ ×1.4′′ ) and Deep (with a resolution of 2.5′′ ×2.5′′ ) project mosaics. 14 Schinnerer et al. Fig. 4.— Sensitivity map of the area covered by the VLA Deep Project, derived using the AIPS task RMSD (see text for details). The rms is fairly uniform except for areas around strong radio sources. Lighter shades indicate lower rms noise values. The contours correspond to rms levels of 10, 12, 15, 20, 25, 30, 34 and 40 µJy/beam. The search for radio sources was limited to the area enclosed by the (light grey) 34 µJy/beam contour. For comparison, the dashed box indicates the area used to construct the catalog of the Large Project.  VLA-COSMOS Survey IV. 15 Fig. 5.— Comparison of the noise estimate in the rms map used to generate the S/N map for source detection in the 2.5′′ resolution Deep image (original rms) and in the local rms map generated with a 50 pixel mesh size (corresponding to 17.5′′ ). Contours illustrate the distribution of points in the most densely populated part of the scatter plot. 16 Schinnerer et al. Fig. 6.— Correspondence of the detection thresholds used for the Deep and Large catalog. a) Peak flux densities from the Large catalog (v1.0) are compared to the mean rms noise level (rms, averaged in concentric rings of width 2′ ), and the detection thresholds applied in the VLA-COSMOS Large (light grey curves) and Deep mosaics (dark grey curve). A cubic spline fit to the measured rms around the bump at r ≈ 45′ gives an estimate of the minimal average noise level at a given radial distance from the center (solid lines at the bottom of the panel). b) Fraction of sources in the Large catalog (v1.0) at a given radial distance r that have a S/N smaller than 4 σDeep in the Deep mosaic. Fig. 7.— Fraction of sources in the VLA-COSMOS Large catalog (v1.0) which could (solid black line) or could not (dashed black line) be successfully assigned a counterpart within 1.5′′ in the VLA-COSMOS Deep catalog. The effective success rate (grey line) is determined by accounting for the fraction of sources that lie below 4σ in the Deep image (cf. panel (b) of Fig. 6.) and thus cannot be included in the Deep catalog. The vertical axis is normalized to the total number of sources in the catalog that lie in a given concentric ring of width 2′ .  VLA-COSMOS Survey IV. 17 Fig. 8.— See last panel of figure for explanations. 18 Schinnerer et al. Fig. 8.— cont.  VLA-COSMOS Survey IV. 19 Fig. 8.— cont. 20 Schinnerer et al. Fig. 8.— cont. Radio sources newly fitted with multiple Gaussian components in the VLA-COSMOS Deep image (cf. Fig. 16 of Schinnerer et al. (2007) for images of all previously identified multi-component sources). The source name is specified at the top of the individual panels. The grey-scale ranges from -2σ to 7σ of the local rms (Tab. 3). The contours start at 3σ and increase to a maximum of 11σ in steps of 2σ. The circular beam with FWHM of 2.5′′ is shown for reference in the lower left corner of each panel.  VLA-COSMOS Survey IV. 21 Fig. 9.— Distribution of the ‘local’ S/N measure, S/N17.5′′ , for sources present only in the Large (top) or Deep (bottom) catalog. The highest measured S/N17.5′′ was adopted for each source (i.e. the larger of the two S/N values determined at 1.5′′ or 2.5′′ resolution, S/N17.5′′ [1.5′′ ] and S/N17.5′′ [2.5′′ ], respectively). All sources with S/N17.5′′ ≥ 5 have been included in the Joint catalog. Grey histograms show the distribution of objects which are only detected at 2.5′′ resolution and for which the detection threshold is only reached in the rms box of 17.5′′ (flag ‘det’ equals 2 in the Joint catalog; see Table 3 and end of Section 5.2.1 for additional explanations), but not in the 105′′ rms box used for the initial source extraction in the Deep mosaic. Fig. 10.— Spatial distribution of sources in the VLA-COSMOS Joint catalog that were either not listed in the Deep (light grey) or Large catalog (v1.0) (dark grey). The size of the sign scales with the significance of the detection (see scheme at right). Note that the newly identified bright sources along the eastern and western edge of the field did not figure in the VLA-COSMOS Large project catalog as the search area was previously restricted to the nominal size of the COSMOS field (see Fig. 4). This geometric restriction was dropped in the detection of sources in the mosaic of the Deep Project. 22 Schinnerer et al. Fig. 11.— Model map used for the correction of the bandwidth smearing effect. The model map is based on the sensitivity drop due to the shape of the primary beam response. The intensity scale is normalized to the sensitivity at the center of the field. The contour levels are at 0.1, 0.3, 0.5, 0.7 and 0.9. The area used to extract sources (r ≤ 8′ ) in each individual pointing for the analysis of Section 6.1 is indicated with dashed white circles.  VLA-COSMOS Survey IV. 23 Fig. 12.— Comparison of two different approaches to correct for the bandwidth smearing (BWS) effect. (See text for details) 24 Schinnerer et al. Fig. 13.— Ratio of total Stotal and peak flux density Speak, corr. (BWS corrected) as a function of the detection S/N , defined as the ratio of the (uncorrected) peak flux density and the local (17.5′′ box) rms. The solid lines show the upper and lower envelope of the region defined to contain unresolved sources (small dots), based on the simulations presented in the Appendix. Sources lying above the upper envelope are considered resolved (large dots). Fig. 14.— Number distribution of unresolved and resolved sources in the VLA-COSMOS Joint catalog as a function of corrected peak (top) and integrated flux density (bottom). Sources with Speak, corr. or Stotal > 1 mJy are added to the right-most bin of the histograms. −1 In panel (b) the dashed line (normalized to the counts at 0.1 mJy) illustrates the decline of the integral sources counts ∼Stotal which is expected from the flat part of the Euclidan source counts present at low flux levels (e.g. Bondi et al. 2008).  VLA-COSMOS Survey IV. 25 Fig. 15.— Comparison of the peak flux densities Speak (both BWS corrected) in the VLA-COSMOS Large (v2.0) and Joint catalog. The inset covers the entire range of peak flux densities while the main part of the figure is limited to peak flux densities less than 4 mJy/beam. The measured values of sources classified as unresolved in both the Large (v2.0) as well as the Joint catalog are in good agreement (grey dots). Sources classified as resolved in the Joint catalog tend to have larger values, as the Deep image has a higher sensitivity to extended emission. 26 Schinnerer et al. Fig. 16.— Comparison of the integrated flux Stotal in the VLA-COSMOS Large (v2.0) and Joint catalog (the layout of the figure matches that of 15). The majority of the multi-component sources (dark grey crosses) have larger measured integrated fluxes in the Joint catalog as the Deep mosaic is more sensitive to extended emission. Most other sources lie along the diagonal bisector of the plot (indicating a good correspondence between the catalogs), except for sources that were previously classified as unresolved in the Large catalog (v2.0) and are now classified as resolved in the VLA-COSMOS Joint catalog (light grey points).  VLA-COSMOS Survey IV. 27 Fig. 17.— Comparison of representative radio surveys conducted at 20 cm with the VLA and ATCA. Filled squares denote surveys that were conducted in the VLA A configuration reaching resolutions of ≈2′′ , while surveys having resolutions of 5-10′′ (from the VLA B array and ATCA) are shown as filled circles. The VLA-COSMOS surveys are marked by open squares (Deep: top, Large: bottom), showing that they cover the largest area at their resolution and sensitivity. In order to reasonably compare the different surveys, we used the sensitivity that is achieved for at least 80% of the area covered. Thus in the case of surveys that have a non-uniform rms pattern the sensitivity value used here can be significantly higher than the best value present in the deepest part of the images (for detailed numbers see Tab. 4). 28 Schinnerer et al. Fig. 18.— Intrinsic source sizes of simulated Gaussian sources vs. their detection S/N . The dashed line marks the upper expectation value θobs. of the observed size of a point source of varying S/N in the Deep mosaic. At low S/N , point sources may be significantly larger than the beam size, θB . The intrinsic (deconvolved) source size, θlimit , inferred from the dashed line is shown in red. It defines an upper limit to the intrinsic source size below which a conclusive classification as resolved or unresolved is no longer possible. Potentially resolvable simulated sources are plotted in grey for easier identification in Fig. 19.  VLA-COSMOS Survey IV. 29 . . . . Fig. 19.— Diagnostic diagrams for the separation of simulated Gaussian sources into resolved and unresolved objects based on their location in Fig. 18 (see text for details). The color of the dots matches that used in Fig. 18. In the upper panel we show the fraction of sources in a given bin of S/N that (1) have an intrinsic source size θ < θlimit but which nevertheless come to lie above the curve – given by eq. (A2; see lower panel) – separating resolved (above curve) from unresolved (below curve) sources, or (2) have θ > θlimit but are found in the area attributed to unresolved objects. (The curves are discontinued at S/N = 100 because small number fluctuations dominate beyond this value.) 30 Schinnerer et al. 1 1500 1000 0 500 -1 0 5 6 7 8 9 10 20 -1 0 1 200 1 150 0 100 50 -1 0 5 5.2 5.4 5.6 5.8 6 -1 0 1 Fig. 20.— Comparison of the recovered total flux density Stout of simulated Gaussian sources with the input value, Stin . Top row: distribution of relative errors (right) and plot of the relative errors as a function of signal-to-noise ratio (SNR; left) for all simulated sources after these have been classified as resolved or unresolved using eq. (A2). (See also Fig. 19.) Bottom row: close-up of the simulated sources in the bin with the smallest simulated S/N . Total flux densities tend to be overestimated by an amount that increases towards lower S/N . The according average bias is listed in Tab. 5 for resolved sources as well as for the complete sample of resolved and unresolved simulated sources.  VLA-COSMOS Survey IV. 31 TABLE 1 VLA Pointing Centers for VLA-COSMOS Deep Project Pointing # R.A. (J2000) DEC (J2000) F07 10:00:58.62 +02:25:20.42 F08 09:59:58.58 +02:25:20.42 F11 10:01:28.64 +02:12:21.00 F12a 10:00:28.60 +02:12:21.00 F13 09:59:28.56 +02:12:21.00 F16 10:00:58.62 +01:59:21.58 F17 09:59:58.58 +01:59:21.58 Note. — The naming convention of the VLA-COSMOS Large project (Schinnerer et al. 2007) has been kept for the pointing centers of the VLA-COSMOS Deep project at 1.4 GHz. a COSMOS field center TABLE 2 Source Numbers in the VLA-COSMOS catalogs Catalog Type # of objects Remarks Deep Project componentsa 3744 4σ (∼45 µJy); total Deep Project sourcesb 3441 4σ (∼45 µJy); total Large Project (v1.0) sources 3643 4.5σ; total 1226 < 5σ Large Project (v2.0) sources 2417 5σ (∼60 µJy); total 1611 unresolved sources 806 resolved sources 78 multi-component sources Joint sources 2865 total; combined Large (v2.0) and Deep Project 2159 detected in both Large (v1.0) and Deep image 392 from Large Project (v2.0) catalog 314 from Deep Project source catalog 2329 unresolved sources 405 resolved sources 131 multi-component sources (incl. 43 new ones) Note. — See text for details. a Direct product of running the AIPS task SAD on the Deep image b Cleaned for multi-component and ’twin’ sources  TABLE 3 1.4 GHz Joint Source Catalog of the VLA-COSMOS Project (abridged) Name former Name R.A. Dec. R.A. Dec. σR.A. σDec. (in Large Project catalog, v2) [deg], (J2000.0) [deg], (J2000.0) (J2000.0) (J2000.0) [′′ ] [′′ ] COSMOSVLADP J095821.65+024628.1 COSMOSVLA J095821.65+024628.1 149.5902208 2.7744972 09 58 21.653 +02 46 28.19 0.25 0.25 COSMOSVLADP J095821.78+024820.6 COSMOSVLA J095821.78+024820.6 149.5907833 2.8057333 09 58 21.788 +02 48 20.64 0.35 0.36 COSMOSVLADP J095821.81+014550.7 COSMOSVLA J095821.82+014550.8 149.5908875 1.7640944 09 58 21.813 +01 45 50.74 0.38 0.41 COSMOSVLADP J095821.82+014724.1 COSMOSVLA J095821.81+014724.2 149.5909333 1.7900389 09 58 21.824 +01 47 24.14 0.26 0.26 COSMOSVLADP J095821.94+020707.7 COSMOSVLA J095821.94+020707.7 149.5914333 2.1188167 09 58 21.944 +02 07 07.74 0.27 0.27 COSMOSVLADP J095822.11+014058.9 COSMOSVLA J095822.10+014058.7 149.5921333 1.6830278 09 58 22.112 +01 40 58.90 0.29 0.27 COSMOSVLADP J095822.18+014524.3 COSMOSVLA J095822.18+014524.3 149.5924333 1.7567500 09 58 22.184 +01 45 24.30 0.29 0.30 COSMOSVLADP J095822.25+013512.3 COSMOSVLA J999999.99+999999.9 149.5927083 1.5867500 09 58 22.250 +01 35 12.30 0.44 0.38 COSMOSVLADP J095822.30+024721.3 COSMOSVLA J095822.30+024721.3 149.5929250 2.7892583 09 58 22.302 +02 47 21.33 -99.00 -99.00 COSMOSVLADP J095822.57+020239.1 COSMOSVLA J095822.57+020239.1 149.5940667 2.0441972 09 58 22.576 +02 02 39.11 0.27 0.29 COSMOSVLADP J095822.81+023604.3 COSMOSVLA J095822.81+023604.5 149.5950583 2.6012000 09 58 22.814 +02 36 04.32 0.48 0.43 COSMOSVLADP J095822.93+022619.8 COSMOSVLA J095822.93+022619.8 149.5955667 2.4388333 09 58 22.936 +02 26 19.80 -99.00 -99.00 COSMOSVLADP J095823.25+020859.4 COSMOSVLA J095823.25+020859.4 149.5969083 2.1498417 09 58 23.258 +02 08 59.43 -99.00 -99.00 COSMOSVLADP J095823.27+021455.9 COSMOSVLA J095823.27+021455.5 149.5969917 2.2488639 09 58 23.278 +02 14 55.91 0.39 0.35 COSMOSVLADP J095823.67+021201.4 COSMOSVLA J095823.68+021201.4 149.5986333 2.2004000 09 58 23.672 +02 12 01.44 0.29 0.31 COSMOSVLADP J095824.02+024916.0 COSMOSVLA J095824.02+024916.0 149.6000833 2.8211167 09 58 24.020 +02 49 16.02 -99.00 -99.00 COSMOSVLADP J095824.02+025029.5 COSMOSVLA J095824.02+025029.3 149.6000875 2.8415333 09 58 24.021 +02 50 29.52 0.26 0.26 COSMOSVLADP J095824.13+013836.6 COSMOSVLA J095824.14+013836.6 149.6005542 1.6435250 09 58 24.133 +01 38 36.69 0.29 0.31 Schinnerer et al. Note. — VLA-COSMOS Joint Catalog of radio sources at 1.4 GHz representing all reliable radio sources from the Large catalog and a search for sources in the Deep mosaic (see text for details). All measurements were made on the 2.5′′ ×2′′ Deep image. The complete table is available via the link to a machine-readable version above and/or at the COSMOS archive at IPAC/IRSAa . A portion is shown here as an example of its form and content. a http://www.irsa.ipac.edu/data/COSMOS/tables/vla/ 32  VLA-COSMOS Survey IV. 33 TABLE 3 cont. Speak Speak,corr. a Stotal rms θM, dec θm, dec PAdec Flags [mJy/beam] [mJy/beam] [mJy] [mJy/beam] [′′ ] [′′ ] [◦ ] resb multc membd dete 5.218±0.029 5.218 5.218±0.029 0.029 0.00 0.00 0.00 -1 0 0 0 0.201±0.035 0.201 0.201±0.035 0.035 0.00 0.00 0.00 0 0 -1 0 0.079±0.018 0.081 0.081±0.018 0.018 0.00 0.00 0.00 0 0 0 -1 0.267±0.014 0.274 0.328±0.034 0.014 1.43 0.97 60.80 1 0 0 0 0.179±0.016 0.182 0.182±0.016 0.016 0.00 0.00 0.00 0 0 0 0 0.154±0.016 0.154 0.154±0.016 0.016 0.00 0.00 0.00 0 0 0 0 0.181±0.017 0.185 0.341±0.057 0.017 3.04 1.99 94.80 1 0 0 0 0.135±0.026 0.135 0.135±0.026 0.026 0.00 0.00 0.00 -2 0 1 1 -99.000 -99.000 30.120 0.083 35.97 9.90 -99.00 1 1 0 -99 0.153±0.017 0.157 0.157±0.017 0.017 0.00 0.00 0.00 0 0 0 0 0.093±0.019 0.096 0.096±0.019 0.019 0.00 0.00 0.00 0 0 0 -1 -99.000 -99.000 116.500 0.033 74.99 12.92 -99.00 1 1 0 -99 -99.000 -99.000 5.025 0.026 9.90 3.75 -99.00 1 1 0 -99 0.077±0.016 0.079 0.079±0.016 0.016 0.00 0.00 0.00 0 0 0 -1 0.162±0.021 0.169 0.162±0.021 0.021 0.00 0.00 0.00 -1 0 0 0 -99.000 -99.000 56.620 0.067 56.98 13.93 -99.00 1 1 0 -99 0.461±0.029 0.461 0.461±0.029 0.029 0.00 0.00 0.00 -1 0 0 0 0.141±0.018 0.141 0.141±0.018 0.018 0.00 0.00 0.00 0 0 0 0 a Peak flux value corrected for the bandwidth smearing effect. b Flag for resolved sources (based on Fig. 13): (-2) if unresolved only in VLA-COSMOS Deep image; (-1) if resolved only in VLA-COSMOS Large image; (0) if unresolved in both VLA-COSMOS Large and Deep images; (1) if resolved in both VLA- COSMOS Large and Deep images; (2) if resolved only in VLA-COSMOS Deep image. Note that the flag value of ‘-2’ is only assigned to sources that were not in the VLA-COSMOS Large Project catalog. The flag ‘-1’, on the other hand, is used for sources that were classified as resolved in the Large Project catalog but are listed as unresolved in the Joint catalog. c Flag for distinction between sources consisting of multiple components (1) or a single component (0). d Flag identifying whether the source was only detected in the VLA-COSMOS Large image (-1), in both the VLA-COSMOS Large and Deep image (0), or only in the VLA-COSMOS Deep image (1). e Flag specifying if the source was detected with S/N ′′ 17.5′′ ≥ 5 only at 1.5 resolution (-1); with S/N17.5′′ ≥ 5 at a resolution of both 1.5′′ and 2.5′′ (0); or only at a resolution of 2.5′′ , either in both the large and small scale rms map (1), or only in the small scale rms map (2). TABLE 4 Selected Radio Surveys at 1.4 GHz Field Areaa rmsb Instrument Reference [deg2 ] [µJy/beam] VLA-COSMOS (Deep) 1.7 27 VLA-A this paper VLA-COSMOS (Large) 1.6 20 VLA-A Schinnerer et al. 2007 ECDF-S 0.23 8.5 VLA-A Miller et al. 2008 SSA 13 0.25 16 VLA-A Fomalont et al. 2006 FIRST 10,000 150 VLA-B Becker et al. 1995 FLS 5 23 VLA-B Condon et al. 2003 SXDF 1.1 27 VLA-A Simpson et al. 2006 VVDS 0.8 17 VLA-B Bondi et al. 2003 ATHDFS 0.19 80 ATCA Norris et al. 2005, Huynh et al. 2005 PDS 3.65 100 ATCA Hopkins et al. 2003 Note. — Radio surveys displayed in Fig. 17. The values listed were derived from figures similar to Fig. 3 presented in the corresponding references, except for The FIRST and FLS surveys where the full areas are used. a 80% of the area covered by the survey b Highest rms value occurring for 80% of the area covered 34 Schinnerer et al. TABLE 5 Median of relative error on recovered integrated flux densities for simulated Gaussian sources (see Appendix). S/N range (Stin − Stout )/Stin (all sources) (Stin − Stout )/Stin (resolved sources) 5-6 -0.05±0.25 -0.17±0.42 6-7 -0.02±0.21 -0.28±0.39 7-8 0.01±0.17 -0.16±0.33 8-9 -0.01±0.17 -0.19±0.29 9-10 -0.01±0.15 -0.20±0.23 10-15 0.01±0.12 -0.10±0.18 15-20 0.01±0.10 -0.08±0.15 Note. — Errors span the range (+/-) containing 2/3 of the measurements.