The Westerbork Northern Sky Survey (WENSS) is being carried out at 325 and 609 MHz (92 and 49 cm) with the Westerbork Synthesis Radio Telescope (WSRT). At 325 MHz, WENSS covers the complete sky north of $30^\circ$ to a limiting flux density of approximately 18 mJy ( $5 \sigma$ ). At 609 MHz, about a quarter of this area, concentrated at high galactic latitudes, has been surveyed to a limiting flux density of approximately 15 mJy ( $5 \sigma$ ). The systematic errors in flux density in WENSS were found to be $\sim 5\%$ (Rengelink et al. 1997). The survey was conducted in mosaicing mode which is very efficient in terms of observing time. In this mode, the telescope cycles through 80 evenly spaced field centres, during each of a number of $12^{\rm h}$ syntheses with different spacings of array elements. The visibilities are sufficiently well sampled for all 80 fields that it is possible to reconstruct the brightness distribution in an area of the sky, $\sim$ 100 square degrees, which is many times larger than the primary beam of the WSRT. Individual fields are referred to as mosaics, and have a resolution (FWHM of the restoring beam) of $54'' \times 54''$ cosec $\delta$ at 325 MHz and $28''\times 28''$ cosec $\delta$ at 609 MHz. From the combined mosaics, maps are made with a uniform sensitivity and regular size, which are called frames. The 325 MHz frames are $6^\circ \times 6^\circ$ in size and positioned on a regular $5^\circ \times 5^\circ$ grid over the sky, which coincides with the position grid of the new Palomar Observatory Sky Survey (POSS II, Reid et al. 1991) plates. A detailed description of WENSS is given by Rengelink et al. (1997)

2.2 Selection of a sample of candidate GPS sources

A deep low frequency radio survey such as WENSS is crucial for selecting a sample of faint GPS sources. It is the inverted spectrum at low frequencies which distinguishes them from other types of radio sources. Figure 1 shows the major large-sky radio surveys in the northern sky with theoretical spectra of homogeneous synchrotron self absorbed radio sources (e.g. Moffet 1975) superimposed, which have spectral peak frequencies of 1 GHz. Samples of GPS sources can be constructed using WENSS flux density measurements in the optically thick part of their spectra which are ten times fainter than samples selected using the Texas Survey (Douglas 1996).

$\begin{figure} \centerline{ \psfig {figure=figure1.ps,width=8cm} }\end{figure}$

Figure 1: Overview of the major radio surveys in the northern sky: the Greenbank Surveys (Condon & Broderick 1985; Gregory & Condon 1991), the Texas Survey (Douglas et al. 1996), and the Cambridge 3C, 4C, and 6C surveys. The curves represent the spectra of a homogeneous synchrotron self absorbed radio source, with a peak frequency of 1 GHz and peak flux density of 300 mJy (lower curve) and 3000 mJy (upper curve). Samples of GPS sources can be constructed using WENSS flux density measurements in the optically thick part of their spectra which are more than an order of magnitude fainter than samples selected using the Texas Survey

When we selected our sample, only a small part of the WENSS region had been observed and the data reduced to the point of providing source lists. The 325 MHz WENSS data used to select GPS sources are from two regions of the sky; one at $15^{\rm h} < \alpha < 20^{\rm h}$ and $58^\circ< \delta < 75^\circ$ , which is called the mini-survey region (Rengelink et al. 1997), and the other at $4^{\rm h}00^{\rm m} < \alpha < 8^{\rm h}30^{\rm m}$ and $58^\circ< \delta < 75^\circ$ ,where $\alpha$ is right ascension and $\delta$ is declination. These were the first two regions observed, reduced and analysed for WENSS. The mini-survey region, which is roughly centered on the North Ecliptic Pole, was chosen as the first area for analysis because it coincides with the NEP-VLA survey at 1.5 GHz (Kollgaard et al. 1994), the deep 7C North Ecliptic Cap survey (Lacy et al. 1995; Visser et al. 1995), the deepest part of the ROSAT All Sky survey (Bower et al. 1996) and the IRAS survey (Hacking & Houck 1987). The high declination of the two regions is very convenient for VLBI experiments, because their locations are circumpolar for almost all the major EVN and VLBA radio telescopes. At the time of selection WENSS 609 MHz data was available for only about one third of the mini-survey region. The regions for which both 325 and 609 MHz source lists were available cover 119 square degrees of the sky. The regions for which only 325 MHz data were available cover 216 square degrees in the mini-survey region and 306 square degrees in the other region.

These source lists were correlated with those from the Greenbank 5 GHz (6 cm) survey (Gregory & Condon 1991; Gregory et al. 1996), which has a limiting flux density of 25 mJy ( $5 \sigma$ ). For the faintest sources the new Greenbank source list (Gregory et al. 1996) was used, which is based on more data. Candidate GPS sources were selected on the basis of a positive spectral index $\alpha$ between 325 MHz and 5 GHz, where the spectral index is defined by $S \sim \nu^{\alpha}$ . If 609 MHz data was also available, an "inverted'' spectrum between 325 MHz and 609 MHz was used as the selection criterion. This in fact increased the sensitivity of the selection process to GPS sources with low peak frequencies (< 1 GHz). Note that in general for a GPS source, the 325 - 609 MHz spectral index will be more positive than the 325 - 5000 MHz spectral index for a spectral peak in the 1 GHz range. Hence, using the 325 - 609 MHz selection criterion will not miss any GPS sources which would have been found using the 325 - 5000 MHz selection criterion, it will only add extra sources with lower peak frequencies.

In total, 117 inverted spectrum sources were selected; 37 using the 325 - 609 MHz selection and 82 using the 325 - 5000 MHz selection. They are listed in Table 1. Columns 1, and 2 give the name, right ascension and declination (B1950) (obtained from the VLA observations), Cols. 3, 4 and 5 the 325 MHz, 609 MHz and 5 GHz flux densities, and Cols. 6 and 7 give the 325 - 609 MHz and 325 - 5000 MHz spectral indices. The uncertainties in the 325 - 5000 MHz spectral indices range from 0.03 to 0.05 (for the faintest objects), and the uncertainties in the 325 - 609 MHz spectral index range from 0.10 to 0.40.

2.3 Additional radio observations

An apparently inverted or peaked spectrum could be caused by variability at any or all of the selection frequencies, due to the fact that the 325, 609 and 5000 MHz surveys were observed at different epochs. The 5 GHz Greenbank survey was made in 1987, while the 325 MHz and 609 MHz data were taken in 1993. To select the genuine GPS sources, additional quasi-simultaneous observations at other frequencies are required to eliminate flat spectrum, variable radio sources. Furthermore, high frequency data is needed to confirm their turnover, and measure the (steep) spectrum in the optically thin part of their spectra. Therefore VLA observations were taken at 8.4 and 15 GHz, and WSRT observations at 1.4 and 5 GHz. Later, after the selection process, data at 1.4 GHz from the NRAO VLA Sky Survey (NVSS, Condon et al. 1996) became available and were used to supplement our spectra.

2.3.1 WSRT observations at 1.4 and 5 GHz

The WSRT was used to observe the candidate GPS sources at 1.4 and 5 GHz. The 1.4 GHz observations were performed on 20 February and 10 March 1994 using 8 bands of 5 MHz between 1377.5 and 1423.5 MHz, providing a total bandwidth of 40 MHz. The sources were all observed for about 100 seconds at two to three different hour angles. This resulted in a noise level of typically 1 mJy/beam and a resolution of $15''\times 15''{\rm cosec} \ \delta$ . The results are shown in Col. 8 of Table 1.

In order to improve the 5 GHz Greenbank flux density measurements, observations were carried out with the WSRT at 4.87 GHz on May 15 1994 using a bandwidth of 80 MHz, at a time when the WSRT was participating a VLBI session. Unfortunately only three telescopes were equipped with 5 GHz receivers. Only sources between 4 and 8 hours right ascension were observed, and the uncertainty in the measured flux densities is large ( $\sim 15$ %). The resulting flux densities are listed in Col. 13 in Table 1.

**Table 1:** The sample of candidate GPS sources. Column 1 gives the B1950 source name, Col. 2 the VLA position, Cols. 3, 4 and 5 the flux densities from WENSS at 325 and 609 MHz and of the Greenbank Survey at 5 GHz. Columns 6 and 7 give the 325 - 609 MHz and the 325 - 5000 MHz spectral indices, Cols. 8, 9 and 10 the flux densities from the WSRT at 1.4 GHz, and from the VLA at 8.4 and 15 GHz. A cross in Col. 11 indicates whether the source was selected in the final sample. Columns 12, 13 and 14 give the NVSS 1.4 GHz, the WSRT 5 GHz and the MERLIN 5 GHz flux densities

2.3.2 VLA Observations at 8.4 and 15 GHz

The candidate GPS sources were observed with the VLA in B-configuration at 8.4 and 15 GHz on 23 July 1994. At both frequencies, the objects were observed in a standard way using a bandwidth of $2\times 25$ MHz. The phases were calibrated using standard nearby VLA phase calibrators. Total integration times were typically 100 seconds at both frequencies, resulting in noise levels of 0.2 and 1.0 mJy/beam respectively. Systematic errors in flux density of VLA observations at these frequencies are typically about $3\%$ (e.g. Carilli et al. 1991). The data were reduced using AIPS in a standard manner, including several iterations of phase self-calibration. The synthesized beams have half widths of $1.5''\times 0.8''$ and $0.8''\times 0.5 ''$ at 8.4 and 15 GHz respectively. Several candidate GPS sources had already been observed at 8.4 GHz on February 26 1994 and April 3 1994 during the Cosmic Lens All Sky Survey (CLASS) program (e.g. Myers et al. 1995); these were not re-observed by us at 8.4 GHz. The CLASS 8.4 GHz observations were made using the VLA in A configuration in a standard way, also with a bandwidth of $2\times 25$ MHz and an average integration time of 30 seconds. The resolution of the CLASS observations was $\sim 0.2''$ , and the noise level $\sim 0.4$ mJy/beam.

The results of the VLA observations are listed in Cols. 9 and 10 in Table 1. All of the sources were unresolved, except for B1608+6540, which was found to be a quadruple gravitational lens (Snellen et al. 1995b; Myers et al. 1995; Fassnacht et al. 1996)

2.3.3 The NRAO VLA Sky Survey at 1.4 GHz

Observations for the 1.4 GHz NRAO VLA Sky Survey (NVSS, Condon et al. 1996) began in September 1993 and are planned to cover the sky north of declination $-40^{\circ}$ (82% of the celestial sphere). Data in our regions of interest were taken on 1 November 1993 for the region $4^{\rm h}00^{\rm m}<{\rm R.A.}<8^{\rm h}00^{\rm m}$ , and on 2 April 1995 for the region between $15^{\rm h}00^{\rm m}<{\rm R.A.}<20^{\rm h}00^{\rm m}$ . The noise level in an image is typically 0.5 mJy/beam and the resolution is 45''.

2.3.4 MERLIN Observations at 5 GHz

The final sample of genuine GPS sources, as selected in Sect. 2.4, was observed with MERLIN at 5 GHz on 15 and 16 May 1995 during our global VLBI measurements. The sources were observed in three to four "snapshots'' of 13 minutes each, resulting in a noise level of typically 0.3 mJy/beam, and a resolution of 0.04''. All the sources were unresolved. The results are listed in Col. 14 of Table 1. Note that these observations were obtained after, and therefore not used for, the final selection.

2.4 Selection of the Genuine GPS sources

The genuine GPS sources were selected using the 325 MHz and, if available, the 609 MHz WENSS data, 1.4 GHz WSRT data, 5 GHz Greenbank data and 8.4 and 15 GHz VLA data. The WSRT 5 GHz data were not used for selection because they were only available for a part of the sample. The NVSS data was not used for selection, being not available at the time of source selection. However, both the 5 GHz WSRT and 1.4 GHz NVSS data were used for variability studies (see Sect. 3).

The resulting sample of 47 sources is listed in Table 2. One of the sources, B1807+5959, did not obey the criterion of decreasing flux density above the peak frequency, because the 5 GHz Greenbank flux density flux point is too low. However it was kept in the sample because the fall off in flux density at both low and high frequencies suggests that the low flux density point at 5 GHz is due to the different epoch of the Greenbank observations. Additional observations showed this to be true.

The spectra of the selected GPS sources were fitted with the following function

$\begin{displaymath} S(\nu ) = S_{\rm max} / (1-{\rm e}^{-1}) \times \left( \frac... ...{\rm e}^{- \left( \frac{\nu }{\nu _{\rm max}}\right) ^{l - k}})\end{displaymath}$

(1)

where k is the optically thick spectral index, l the optically thin spectral index, and $S_{\rm max}$ and $\nu _{\rm max}$ respectively the peak flux density and peak frequency. This equation, which represents a homogeneous synchrotron self absorbed radio source for k = 2.5 (e.g. Moffet 1975), fits the spectral peak well in most cases, however it did not always fit the flux density points at the lowest and highest frequency frequency adequately, and therefore was only used to determine the peak flux density, peak frequency and Full Width Half Maximum of the spectra. The optically thick and thin spectral indices have been determined from the two lowest and the two highest frequency data points respectively. The fitted spectra are shown in Fig. 2. Table 2 gives the characteristics of the GPS sources: Col. 1 gives the source name, Cols. 2 and 3 the peak frequency and peak flux density, Cols. 4 and 5 the optically thick and optically thin spectral indices, and Col. 6 the FWHM of the spectrum in logarithmic units.

Figure 2 shows that three of the 47 sources initially selected probably do not have genuine GPS spectra, namely B0531+6121, B0748+6343 and B0755+6354. In these cases differences between MERLIN, Greenbank and WSRT 5 GHz flux density measurements suggest that the measured spectra are contaminated by flux density variability and it is not clear whether the spectra do indeed exhibit a peak. Although we have included them in the sample, we omit them from the analysis below. Note that no sign of a turnover is seen in B1945+6024, and that there are some sources in the sample which do not have a "clean'' peaked spectrum, like B1954+6146 and B0535+6743. To obtain a better determination of the spectral peak of B1954+6146, the 325 MHz flux density data point is not used to fit the spectrum.

2 Selection of GPS sources