1 Introduction

Radio galaxies have now been found out to redshifts of z=5.19 (van Breugel et al. 1999b) and radio-loud quasars out to z=4.72 (Hook & McMahon 1998). Although new optical selection techniques such as color-dropouts, deep spectroscopy of blank fields, and narrow-band Ly$\alpha$ imaging have now found galaxies at similar (Steidel et al. 1999) and even higher redshifts (up to $z \sim 5.75$; Dey et al. 1998; Weymann et al. 1998; Spinrad et al. 1998, Hu et al. 1999), radio sources are still the only objects that can be selected uniformly over all redshift ranges, and in a way that does not suffer from optical biases such as dust extinction, which is known to be important at these high redshifts (e.g. Hughes et al. 1997; Ivison et al. 1998; Dickinson 1998).

At low to moderate redshift ( $z \lesssim 1$), powerful radio sources are uniquely identified with massive ellipticals (Lilly & Longair 1984; Owen & Laing 1989; Best et al. 1998; McLure & Dunlop 2000). The strongest indications that this is also true at higher redshifts comes from the near-IR Hubble K-z diagram of radio galaxies which shows a remarkably close correlation from the present out to z=5.19 (Lilly 1989; Eales et al. 1997; van Breugel et al. 1998; van Breugel et al. 1999b). This suggests that we can use radio galaxies to study the formation and evolution of the most massive galaxies, which, by their implied star-formation history, can put important constraints on galaxy formation models, and even on cosmological parameters (e.g. Dunlop et al. 1996; Spinrad et al. 1997). Although the unification model for radio galaxies and quasars (e.g. Barthel 1989) suggests we could also use quasars as tracers, a detailed stellar population study of quasar host galaxies is almost impossible due to the extreme luminosity of the AGN. Furthermore, samples of radio sources designed to find large quantities of quasars require additional optical selections (e.g. Gregg et al. 1996; Hook & McMahon 1998; White et al. 2000).

Figure 1: $\alpha _{1400}^{325}$ against z for 2 samples without spectral index selection (3CR, Spinrad et al. 1985 and MRC, McCarthy et al. 1996), and 2 USS samples (4C, Chambers et al. 1996a, and our new WN/TN samples, as defined in this paper). Note that the correlation is present in the spectrally unbiased 3CR and MRC, and that the 4C and our new USS samples are finding three to five times more z>2 radio galaxies than the MRC. The horizontal dotted line indicates the $\alpha _{1400}^{325} < -1.3$ cutoff used in our USS sample

Considerable effort has been spent over the last decade to find these high redshift radio galaxies (HzRGs), which has lead to the discovery of more than 140 radio galaxies at redshifts z>2 (see e.g. De Breuck et al. 1998a for a recent summary). However by z>3, their numbers become increasingly sparse, and using flux limited radio surveys such as the 3CR ( S₁₇₈ > 10 Jy; Laing et al. 1983), or the MRC strip ( S₄₀₈ > 0.95 Jy; McCarthy et al. 1996), the highest redshift radio galaxy found so far is at $z\sim 3.2$ (Fig. 1; Rawlings et al. 1990; McCarthy et al. 1996). This redshift limit arises because radio power is correlated with redshift in bright flux limited samples, and an upper limit exists in the radio luminosity. Lowering the flux limit would not only substantially increase the number of sources in these samples, but at the same time the fraction of luminous very high redshift radio galaxies would decrease (Blundell et al. 1998; Jarvis et al. 1999). This fractional decrease would arise even if there is no decrease in co-moving space density at $z \gtrsim 2.5$. Such a redshift cutoff has been suggested by Bremer et al. (1998), but recently Jarvis et al. (1999) rule out a break at $z\lesssim 2.5$. To efficiently find large numbers of HzRGs in acceptable observing times, it is therefore necessary to apply additional selection criteria, at the expense of completeness.

By far the most successful selection criterion has been the ultra steep spectrum criterion (e.g. Röttgering et al. 1994; Chambers et al. 1996a; Blundell et al. 1998). Selecting sources with very steep radio spectra increases dramatically the chance of finding z>2 radio galaxies (Fig. 1). This technique is based on the results of Tielens et al. (1979) and Blumenthal & Miley (1979), who found that the identification fraction on the POSS ( $R \lesssim 20$) decreases with steepening spectral index, consistent with the steeper sources being at higher redshifts. It is now getting clear that this correlation can be explained by a combination of a K-correction of a concave radio spectrum and an increasing spectral curvature with redshift (Krolik & Chen 1991; Carilli et al. 1998; van Breugel et al. 1999a). To further investigate the $z - \alpha$ correlation, we have calculated spectral indices using the flux densities from the WENSS (Rengelink et al. 1997) and NVSS (Condon et al. 1998) catalogs for four different samples: the flux density limited 3CR (Spinrad et al. 1985) and MRC (McCarthy et al. 1996) surveys, and the USS samples from the 4C (Chambers et al. 1996a) and the one presented in this paper. The results (Fig. 1) show a trend for steeper spectral index sources to have higher redshifts in flux limited, spectrally unbiased samples, confirming the empirical relation out to the highest redshifts. The efficiency of the USS criterion is clearly illustrated by the fact that the 4C USS sample (Chambers et al. 1996a) contains 50% z>2 sources, and by the early spectroscopic results on the USS samples presented in this paper, which indicate that $\sim$2/3 of our sources have z>2. It is even more impressive to note that 13 of the 14 radio galaxies at z>3.5 we know of have been found from samples with a steep spectral index selection! The limitation of this technique is that the steepest spectrum sources are rare, comprising typically only 0.5% (at $\alpha < -1.30$) of a complete low frequency sample; therefore, large and deep all sky surveys are needed to obtain a significant sample of USS sources.

With the advent of several new deep all-sky surveys (Sect. 2), it is now possible for the first time to construct a well defined all-sky USS sample with optimized selection criteria to find large numbers of z>3 radio galaxies. In this paper, we describe the construction of such a sample, and present high resolution radio observations needed to determine accurate positions and morphologies. This information is essential for the optical and near-IR identifications, and subsequent optical spectroscopy of a significant sub-set of our sample, which will be described in future papers. The organization of the paper is as follows: we describe the radio surveys we used in Sect. 2 and define our samples in Sect. 3. We present and discuss our radio observations in Sect. 4, and present our conclusions in Sect. 5.

 

 

Table 1: Radio surveys

	WENSS	TEXAS	MRC
Frequency (MHz)	325	365	408
Sky region (J2000)	$\delta >$ +29 $\hbox{$^\circ$ }$	-35 $.\!\!^\circ$7 $< \delta <$ +71 $.\!\!^\circ$5	-85 $\hbox{$^\circ$ }$ $< \delta <$ +18 $.\!\!^\circ$5
# of sources	229576	67551	12141
Resolution	$54\hbox{$^{\prime\prime}$ }\times$ 54 $^{\prime \prime }$ cosec $\delta$	10 $^{\prime \prime }$a	$2\hbox{$.\mkern-4mu^\prime$ }62 \times 2\hbox{$.\mkern-4mu^\prime$ }86\ \sec\ (\delta - 35\hbox{$.\!\!^\circ$ }5)$
Position uncertainty	1 $.\!\!^{\prime\prime}$5	0 $.\!\!^{\prime\prime}$5-1 $^{\prime \prime }$	8 $^{\prime \prime }$
(strong sources)
RMS noise	$\sim$4 mJy	20 mJy	70 mJy
Flux density limit	18 mJy	150 mJy	670 mJy
Reference	Rengelink et al. 1997	Douglas et al. 1996	Large et al. 1981
	NVSS	FIRST	PMN
Frequency (MHz)	1400	1400	4850
Sky region (J2000)	$\delta > -$40 $\hbox{$^\circ$ }^b$	$7^{\rm h}20^{\rm m} < \alpha < 17^{\rm h}20^{\rm m}$; +22 $.\!\!^\circ$2 $< \delta <$ +57 $.\!\!^\circ$5	-87 $.\!\!^\circ$5 $< \delta <$ +10 $\hbox{$^\circ$ }$
		$21^{\rm h}20^{\rm m} < \alpha < 3^{\rm h}20^{\rm m}$; -2 $.\!\!^\circ$5 $< \delta <$ +1 $.\!\!^\circ$6
# of sources	1689515	437429	50814
Resolution	45 $^{\prime \prime }$$\times$ 45 $^{\prime \prime }$	5 $^{\prime \prime }$$\times$ 5 $^{\prime \prime }$	4 $^{\prime}\mskip-4.7mu.\mskip0.8mu$2
Position uncertainty	1 $^{\prime \prime }$	0 $.\!\!^{\prime\prime}$1	$\sim$45 $^{\prime \prime }$
(strong sources)
RMS noise	0.5 mJy	0.15 mJy	$\sim$8 mJy
Flux density limit	2.5 mJy	1 mJy	20 mJy
Reference	Condon et al. 1998	Becker et al. 1995	Griffith & Wright 1993

^a The Texas interferometer has a complicated beam. However, sources with separations between 10 $^{\prime \prime }$ and 2$^\prime$ can be successfully modeled as doubles, and will have a single entry in the catalog. See Douglas et al. (1996) for details.
^b Some small gaps are not covered. They are listed on the NVSS homepage (1998 January 19 version).