4 Radio observations

Of all the major radio surveys described in Sect. 2, only FIRST has sufficient positional accuracy and resolution for the optical identification of R > 20 objects. We present FIRST maps of 139 WN and 8 TN sources in Appendices B.2 and B.4.

Outside the area covered by FIRST, we have observed all the remaining WN sources, 30% of the TN sample, and 71% of the MP sources at 0 $.\!\!^{\prime\prime}$3 to 5 $^{\prime \prime }$ resolution using the Very Large Array (VLA; Napier et al. 1983) and Australia Telescope Compact Array (ATCA; Frater et al. 1992) telescopes. A log of the radio observations is given in Table 4. We observed targets for our VLA runs on the basis of declination (A-array for $\delta >$ 0 $\hbox{$^\circ$ }$ and BnA-array for $\delta <$ 0 $\hbox{$^\circ$ }$) and sky coverage of the WN and TN samples, which were still incomplete at the time of the 1996 observations. We observed all WN, and most TN sources with the VLA, and all MP sources with the ATCA. We observed TN sources between $-31\hbox{$^\circ$ }< \delta < -10\hbox{$^\circ$ }$ with either VLA or ATCA, depending on the progress of the NVSS at the time of the observations.

4.1 VLA observations and data reduction

We observed all sources in the standard 4.86 GHz C-band with a 50 MHz bandwidth, resulting in a resolution of $\sim$0 $.\!\!^{\prime\prime}$3 in the A-array and $\sim$1 $^{\prime \prime }$ in the BnA-array. We spent 5 minutes on each source, implying a theoretical rms level of 75 $\mu$Jy, or a ratio of total integrated signal over map noise of 110 for the weakest sources, assuming no spectral curvature beyond 1.4 GHz. We performed calibration and data editing in , the Astronomical Image Processing System from NRAO. We used 3C 286 as the primary flux calibrator in all runs. Comparison of the flux density of 3C 48 with the predicted values indicated the absolute flux density scale was accurate up to 2%. We observed nearby (within 15 $\hbox{$^\circ$ }$) secondary flux calibrators every 15 to 20 minutes to calibrate the phases. After flagging of bad data, we spilt the uv-data up into separate data sets for imaging and self-calibration in DIFMAP, the Caltech difference mapping program (Sheppard et al. 1997). We used field sizes of 164 $^{\prime \prime }$ (A-array) or 256 $^{\prime \prime }$ (BnA-array) with pixel scales of 0 $.\!\!^{\prime\prime}$08/pixel (A-array) or 0 $.\!\!^{\prime\prime}$25/pixel (BnA-array). Even the smallest field of view is still four times larger than the resolution of the NVSS, so all components of an unresolved NVSS source will be covered.

We cleaned each source brighter than the 5$\sigma$ level, followed by a phase-only self-calibration. We repeated the latter for all sources in the field of a source. Next, we made a new model from the (self-calibrated) uv-data, and subsequently cleaned to the level reached before. The last stage in the mapping routine was a deep clean with a 1% gain factor over the entire field. Most of the resulting maps have noise levels in the range 75 to 100 $\mu$Jy, as expected.

4.2 ATCA observations and data reduction

We used the ATCA in the 6C configuration, which has a largest baseline of 6 km. We observed at a central frequency of 1.384 GHz, which was selected to avoid local interference. We used 21 of the 30 frequency channels that had high enough signal, which resulted in an effective central frequency of 1.420 GHz, with a 84 MHz bandwidth. In order to obtain a good uv-coverage, we observed each source eight to ten times for three minutes, spread in hour angle. The primary flux calibrator was the source 1934-648; we used secondary flux calibrators within 20 $\hbox{$^\circ$ }$ of the sources to calibrate the phases. We performed editing and calibration in , following standard procedures. We made maps using the automated mapping/self-calibration procedure MAPIT in . The resulting 1.420 GHz maps (Fig. B.6.) have noise levels of $\sim5$ mJy.

4.3 Results

Of all 343 WN sources, 139 have FIRST maps (Appendix B.2). All remaining 204 sources were observed, and 141 were detected. The remaining 30% were too faint at 4.86 GHz to be detected in 5 min snapshots, because their high frequency spectral index steepens more than expected, or they were over-resolved. Because they are significantly brighter, all the observed 89 TN and 41 MP sources were detected. We present contour maps of all the detections in Appendices B.1, B.3, B.5, and B.6 and list the source parameters in Tables A.1 to A.3.

We have subdivided our sources into 5 morphological classes, using a classification similar to that used by Röttgering et al. (1994). Note that this classification is inevitably a strong function of the resolution, which varies by a factor of 20 between the VLA A-array and the ATCA observations.

We have determined the source parameters by fitting two-dimensional Gaussian profile to all the components of a source. The results are listed in Tables A.1 to A.3 which contain:

Column 1:

Name of the source in IAU J2000 format. The 2-letter prefix indicates the sample: WN: WENSS-NVSS, TN: Texas-NVSS, MP: MRC-PMN.

Column 2:

The integrated flux density from the low-frequency catalog.

Column 3:

The integrated flux density at the intermediate frequency, determined from the NVSS for WN and TN, or from the 1.420 GHz ATCA observations for the MP sample.

Column 4:

The integrated flux density at 4.86 GHz, determined from the VLA observations for WN and TN, and from the PMN survey for the MP sample.

Column 5:

The lower frequency two-point spectral index. This is the spectral index used to define the WN and TN samples.

Column 6:

The higher frequency two-point spectral index. This is the spectral index used to define the MP sample.

Column 7:

Morphological classification code: single (S), double (D), triple (T) and multiple (M) component sources, and irregularly shaped diffuse (DF) sources.

Column 8:

Largest angular size. For single component sources, this is the de-convolved major axis of the elliptical Gaussian, or, for unresolved sources (preceded with <), an upper limit is given by the resolution. For double, triple and multiple component sources, this is the largest separation between their components. For diffuse sources this is the maximum distance between the source boundaries defined by three times the map rms noise.

Column 9:

De-convolved position angle of the radio structure, measured North through East.

Columns 10-11:

J2000 coordinates, determined from the map with position code listed in Col. 12. The positions in the VLA and ATCA maps have been fitted with a single two-dimensional elliptical Gaussian. For double (D) sources, the geometric midpoint is given; for triples (T) and multiples (M), the core position is listed. For diffuse (DF) sources we list the center as determined by eye.

Column 12:

Position code, indicating the origin of the morphological and positional data in Cols. 7 to 11: A = ATCA, F = FIRST, M = MRC, N = NVSS, and V = VLA.

4.4 Notes on individual sources

WN J0043+4719: The source 18 $^{\prime \prime }$ north of the NVSS position is not detected in the NVSS. This is therefore not a real USS source because the NVSS flux density was underestimated.

WN J0048+4137: Our VLA map probably doesn't go deep enough to detect all the flux of this source.

WN J0727+3020: The higher resolution FIRST map shows that both components of this object are indeed identified on the POSS, even though the NVSS position is too far off to satisfy our identification criterion.

WN J0717+4611: Optical and near-IR spectroscopy revealed this object as a red quasar at z=1.462 (De Breuck et al. 1998b).

WN J0725+4123: The extended POSS identification suggest this source is located in a galaxy cluster.

WN J0829+3834: The NVSS position of this unresolved source is 7 $^{\prime \prime }$ ($3\sigma$) from the FIRST position, which itself is only at 2 $^{\prime \prime }$ from the WENSS position.

WN J0850+4830: The difference with the NVSS position indicates that our VLA observations are not deep enough to detect a probable north-eastern component.

WN J0901+6547: This 38 $^{\prime \prime }$ large source is over-resolved in our VLA observations, and probably even misses flux in the NVSS, and is therefore not a real USS source.

WN J1012+3334: The bend morphology and bright optical sources to the east indicate this object is probably located in a galaxy cluster.

WN J1101+3520: The faint FIRST component 20 $^{\prime \prime }$ north of the brighter Southern component is not listed in the FIRST catalog, but is within 1 $^{\prime \prime }$ of a faint optical object. This might be the core of a 70 $^{\prime \prime }$ triple source.

WN J1152+3732: The distorted radio morphology and bright, extended POSS identification suggest this source is located in a galaxy cluster.

WN J1232+4621: This optically identified and diffuse radio source suggest this source is located in a galaxy cluster.

WN J1314+3515: The diffuse radio source appears marginally detected on the POSS.

WN J1329+3046A,B, WN J1330+3037, WN J1332+3009 & WN J1333+3037: The noise in the FIRST image is almost ten times higher than average due to the proximity of the S₁₄₀₀=15 Jy source 3C 286.

WN J1330+5344: The difference with the NVSS position indicates that our VLA observations are not deep enough to detect a probable south-eastern component.

WN J1335+3222: Although the source appears much like the hotspot of a larger source with the core 90 $^{\prime \prime }$ to the east, no other hotspot is detected in the FIRST within 5$^\prime$.

WN J1359+7446: The extended POSS identification suggests this source is located in a galaxy cluster.

WN J1440+3707: The equally bright galaxy 30 $^{\prime \prime }$ south of the POSS identification suggests that this source is located in a galaxy cluster.

WN J1509+5905: The difference with the NVSS position indicates that our VLA observations are not deep enough to detect a probable western component.

WN J1628+3932: This is the well studied galaxy NGC 6166 in the galaxy cluster Abell 2199 (e.g. Zabludoff et al. 1993).

WN J1509+5905: The difference with the NVSS position indicates that our VLA observations are not deep enough to detect a probable west-south-western component.

WN J1821+3601: The source 35 $^{\prime \prime }$ south-west of the NVSS position is not detected in the NVSS. This is therefore not a real USS source because the NVSS flux density was underestimated.

WN J1832+5354: The source 19 $^{\prime \prime }$ north-east of the NVSS position is not detected in the NVSS. This is therefore not a real USS source because the NVSS flux density was underestimated.

WN J1852+5711: The extended POSS identification suggests this source is located in a galaxy cluster.

WN J2313+3842: The extended POSS identification suggests this source is located in a galaxy cluster.

TN J0233+2349: This is probably the north-western hotspot of a 35 $^{\prime \prime }$ source, with the south-eastern component barely detected in our VLA map.

TN J0309-2425: We have classified this source as a 13 $^{\prime \prime }$ double, but the western component might also be the core of a 45 $^{\prime \prime }$ source, with the other hotspot around $\alpha = 3^{\rm h}9^{\rm m}10^{\rm s}, \delta=-24\hbox{$^\circ$ }25\hbox{$^\prime$ }50\hbox{$^{\prime\prime}$ }$.

TN J0349-1207: The core-dominated structure is reminiscent of the red quasar WN J0717+4611.

TN J0352-0355: This is probably the south-western hotspot of a 30 $^{\prime \prime }$ source.

TN J0837-1053: Given the 10 $^{\prime \prime }$ difference between the positions of the NVSS and diffuse VLA source, this is probably the northern component of a larger source.

TN J0408-2418: This is the z=2.44 source MRC 0406-244 (McCarthy et al. 1996). The bright object on the POSS is a foreground star to the north-east of the R=22.7galaxy.

TN J0443-1212: Using the higher resolution VLA image, we can identify this radio source with a faint object on the POSS.

TN J2106-2405: This is the z=2.491 source MRC 2104-242 (McCarthy et al. 1996). The identification is an R=22.7 object, not the star to the north-north-west of the NVSS position.

Figure 9: Radio "color-color'' plots for the WN sample. The abscissa is the $\alpha _{325}^{1400}$ spectral index used to construct the sample. The ordinates are the low-frequency spectral indicesdetermined from the 8C (38 MHz, Rees 1990) or 6C (151 MHz, Hales et al. 1993) and the 325 MHz WENSS (left panel), and the high frequency spectral index determined between the 1.4 GHz NVSS and our 4.86 GHz VLA observations (right panel). The line in each panel indicates a straight power law spectrum. Note the unequal number of points on either side of these lines, indicating substantial spectral curvature

4.5 Radio spectra and spectral curvature

We have used the CATS database at the Special Astronomical Observatory (Verkhodanov et al. 1997) to search for all published radio measurements of the sources in our samples. In Appendix C, we show the radio spectra for all sources with flux density information for more than two frequencies (the S₄₈₆₀ points from our VLA observations are also included). These figures show that most radio spectra have curved spectra, with flatter spectral indices below our selection frequencies, as has been seen in previous USS studies (see e.g. Röttgering et al. 1994; Blundell et al. 1998).

This low frequency flattening and high frequency steepening is obvious in the radio "color-color diagrams'' of the WN sample (Fig. 9). The median spectral index at low frequencies ($\nu < 325$ MHz) is -1.16, while the median $\bar{\alpha}_{325}^{1400} = -1.38$. At higher frequencies ( $\nu > 1400$ MHz), the steepening continues to a median $\bar{\alpha}_{1400}^{4850} = -1.44$. Note that the real value of the latter is probably even steeper, as 30% of the WN sources were not detected in our 4.86 GHz VLA observations, and may therefore have even more steepened high-frequency spectral indices.

4.6 Radio source properties

4.6.1 Radio source structure and angular size

In Table 5, we give the distribution of the radio structures of the 410 USS sources for which we have good radio-maps. At first sight, all three our samples have basically the same percentage of resolved sources, but the similar value for the MP sample is misleading, as it was observed at much lower resolution.

 

 

Table 5: Radio structure distribution

	USS Samples
Morphology	WN	TN	MP	Combined
Single	157 (56 $\pm$ 4%)	43 (48 $\pm$ 7%)	23 (56 $\pm$ 12%)	223 (54 $\pm$ 4%)
Double	81 (29 $\pm$ 3%)	28 (31 $\pm$ 6%)	16 (39 $\pm$ 10%)	125 (31 $\pm$ 3%)
Triple	22 (8 $\pm$ 2%)	9 (10 $\pm$ 3%)	0 (0 $\pm$ 0%)	31 (8 $\pm$ 1%)
Multiple	2 (1 $\pm$ 1%)	4 (5 $\pm$ 2%)	0 (0 $\pm$ 0%)	6 (1 $\pm$ 1%)
Diffuse	18 (6 $\pm$ 2%)	5 (6 $\pm$ 3%)	2 (5 $\pm$ 3%)	25 (6 $\pm$ 1%)
# Observed	280	89	41	410

Figure 10: Median angular size for the flux density limited, spectrally unbiased WSRT samples of Oort 1988 (open triangles), and for our combined USS samples (filled squares). The sources have been binned in equal number bins, and errors represent the 35% and 65% levels of the distribution. Note that our USS selection does not affect the value of the median, and that our USS samples also exclude sources that fall below the break at $S_{1400} \lesssim 10$ mJy

Our results are different from the USS sample of Röttgering et al. (1994), which contains only 18% unresolved sources at comparable resolution (1 $.\!\!^{\prime\prime}$5). To check if this effect is due to the fainter sources in our sample, we compared our sample with the deep high resolution VLA observations of spectrally unbiased sources (Oort 1988; Coleman & Condon 1985). The resolution of our observations is significantly better than the median angular size for S₁₄₀₀ > 1 mJy sources, allowing us to accurately determine the median angular sizes in our samples. We find that our USS sources have a constant median angular size of $\sim6$ $^{\prime \prime }$ between 10 mJy and 1 Jy (Fig. 10). This is indistinguishable from the results from samples without spectral index selection. It indicates that our USS selection of sources with $\alpha < -1.3$ and $\Theta < 1\hbox{$^\prime$ }$ does not bias the angular size distribution in the resulting sample. The "downturn'' in angular sizes that occurs at $\sim 1$ mJy is probably due to a different radio source population, which consist of lower redshift sources in spiral galaxies (see e.g., Coleman & Condon 1985; Oort et al. 1987; Benn et al. 1993). By selecting only sources with S₁₄₀₀ > 10 mJy, we have avoided "contamination'' of our sample by these foreground sources.

We have searched for further correlations between spectral index or spectral curvatures and angular size or flux density, but found no significant results, except for a trend for more extended sources to have lower than expected 4.86 GHz flux densities, but this effect can be explained by missing flux at large scales in our VLA observations.

4.7 Identifications

4.7.1 POSS

We have searched for optical identifications of our USS sources on the digitized POSS-I. We used the likelihood ratio identification criterion as described by e.g. de Ruiter et al. (1977). In short, this criterion compares the probability that a radio and optical source with a certain positional difference are really associated with the probability that this positional difference is due to confusion with a field object (mostly a foreground star), thereby incorporating positional uncertainties in both radio and optical positions. The ratio of these probabilities is expressed as the likelihood ratio LR. In the calculation, we have assumed a density of POSS objects $\rho = 4$ 10^-4''-2, independent of galactic latitude b. We have adopted a likelihood ratio cutoff = 1.0, slightly lower than the values used by de Ruiter et al. (1977) and Röttgering et al. (1994). We list sources with LR> 1.0 for our USS samples in Tables A.4 to A.6. We have included four WN sources (WN J0704+6318, WN J1259+3121, WN J1628+3932 and WN J2313+3842), two TN sources (TN J0510-1838 and TN J1521+0741) and four MP sources (MP J0003-3556, MP J1921-5431, MP J1943-4030 and MP J2357-3445) as identifications because both their optical and radio morphologies are diffuse and overlapping, making it impossible to measure a common radio and optical component, while they are very likely to be associated.

Figure 11: Identification fraction on the POSS as a function of spectral index for the combined WN and TN sample. Note the absence of a further decrease in the identification fraction with steepening spectral index

Figure 11 shows the identification fraction of USS sources on the POSS ( $R \lesssim 20$). Because the distributions for the WN and TN are very similar, we have combined both samples to calculate the identification fraction. Unlike the results for 4C USS (Tielens et al. 1979; Blumenthal & Miley 1979), we do not detect a decrease of the identification fraction with steepening spectral index. We interpret the constant $\sim$15% identification fraction from our sample as a combined population of foreground objects, primarily consisting of clusters (see next section). Our extremely steep spectral index criterion would then select only radio galaxies too distant to be detected on the POSS ( $R \gtrsim 20.0$).

4.7.2 Literature

Using the NASA Extragalactic Database (NED), the SIMBAD database and the W³Browse at the High Energy Astrophysics Science Archive Research Center, we have searched for known optical and X-ray identifications of sources in our samples (see Appendices A.7 to A.9). Of the bright optical ( $R \lesssim 20$) identifications, only one source is a known as a K0-star, three (TN J0055+2624, TN J0102-2152, and TN J1521+0742) are "Relic radio galaxies'' (Komissarov & Gubanov 1994; Giovannini et al. 1999), while all others are known galaxy clusters.

All optical cluster identifications, except MP J1943-4030, are also detected in the ROSAT All-Sky survey Bright Source Catalogue (RASS-BSC; Voges et al. 1999). Conversely, of the 23 X-ray sources, seven are known galaxy clusters, and three known galaxies. The remaining 13 sources are good galaxy cluster candidates because they either show a clear over-density of galaxies on the POSS (eight sources), or they have low X-ray count rates (< 0.02 counts s^-1), suggesting that these might be more distant galaxy clusters too faint to be detected on the POSS. We conclude that probably >3% of our USS sources are associated with galaxy clusters, and that the combined USS + X-ray selection is an efficient (up to 85%) selection technique to find galaxy clusters.

Three of our USS sources (WN J2313+4253, TN J0630-2834 and TN J1136+1551) are previously known pulsars (Kaplan et al. 1998). It is worth noting that two out of nine sources in our USS samples with $\alpha < -2$ are known pulsars. Because Lorimer et al. (1995) found the median spectral index of pulsars to be $\sim -1.6$, we examined the distribution of spectral indices as a function of Galactic latitude. In Fig. 12, we plot the percentage of $\alpha _{325}^{1400} < -1.60$ pulsar candidates as a function of Galactic latitude. The four times higher density near the Galactic plane strongly suggests that the majority of these $\alpha _{325}^{1400} < -1.60$sources are indeed pulsars, which are confined to our Galaxy. A sample of such $\alpha _{325}^{1400} < -1.60$ sources at $\vert b\vert<15\hbox{$^\circ$ }$ would be an efficient pulsar search method.

Figure 12: Percentage of $\alpha _{325}^{1400} < -1.60$ radio sources from a WENSS-NVSS correlation as a function of Galactic latitude. Note the clear peak near the Galactic plane, indicating that these $\alpha _{325}^{1400} < -1.60$ objects might well be Galactic pulsars

We also note that no known quasars are present in our sample. Preliminary results from our optical spectroscopy campaign (De Breuck et al. 1998b, 2000) indicate that $\sim$10% of our sample are quasars. We interpret this lack of previously known quasars as a selection bias in quasar samples against USS sources.

At $R \gtrsim 20$, all five USS sources with known redshift are HzRGs, indicating a selection of sources without detections on the POSS strongly increases our chances of finding HzRGs.