1. INTRODUCTION

Martin Harwit. "Yes. I think that if there were less requirement for theoretical justification for building a instrument that is very powerful, you would then have a better chance of making a discovery."
Bernard Burke. "The record of radio astronomy is clear. Quasars were not found by the desire to find black holes or by the need to find long-distance cosmological probes."

Proceedings of Greenbank Workshop on Serendipitous Discovery in Radio Astronomy, 1983

Distant radio galaxies are among the largest, most luminous, most massive and most beautiful objects in the Universe. They are energetic sources of radiation throughout most of the electromagnetic spectrum. The radio sources are believed to be powered by accretion of matter onto supermassive black holes in the nuclei of their host galaxies. Not only are distant radio galaxies fascinating objects in their own right, but they also have several properties that make them unique probes of the early Universe.

1.1. Scope of this article - HzRGs

Our review will be restricted to galaxies that have redshifts, z > 2 and radio luminosities at 500 MHz (rest frame) L_500(rest) > 10^27.5 W Hz^-1. Radio emission from such objects has a steep nonthermal spectrum, is collimated and is usually extended by tens of kiloparsec. There are very few known z > 2 radio galaxies associated with compact flat-spectrum radio sources.

This definition is somewhat arbitrary. Luminous distant radio galaxies have properties that are different from less powerful radio galaxies at low redshifts and the properties of radio galaxies change gradually with luminosity and redshift. Likewise, z > 2 quasars that are associated with steep-spectrum extended radio emission have many similar properties to our distant radio galaxies (see Section 5.1) and are located in similar environments. However, without some restriction, our task would not have been tractable.

For conciseness, we shall frequently refer to such high-redshift radio galaxies as "HzRGs". McCarthy (1993) gave an extensive review of this topic in 1993. Since then there have been several workshops on high-redshift radio galaxies, whose proceedings have been published, including Amsterdam (Röttgering et al. 1999), Leiden (Jarvis and Röttgering 2003) and Granada (Villar-Martín et al. 2006). Thousands of refereed papers dealing with distant radio galaxies and related topics have been published during the last fifteen years. Although our goal is to be comprehensive, some personal bias will inevitably have entered in selecting the topics and literature. Our apologies for this.

The structure of this review is as follows. We shall first set the scene by reviewing the history of the field. Then we shall describe techniques used for finding HzRGs and discuss how HzRGs are distributed in redshift. Following an overview of the diverse constituents of HzRGs, we shall discuss each HzRG component in detail. The nature of HzRGs and their role in the general evolution of galaxies will then be covered. We shall present evidence that HzRGs are the massive progenitors of dominant cluster galaxies in the local Universe and that they are located in forming galaxy clusters. The properties of these radio-selected protoclusters are then reviewed. Finally, we shall discuss some of the exciting prospects for future HzRG research. An appendix includes a list of known HzRGs at the time of writing (October 2007).

1.2. History

The study of distant radio galaxies has progressed in several phases:

(i) Infancy. Mid-1940s to mid-1960s: After the discovery that Cygnus A is associated with a faint distant galaxy (Baade and Minkowski 1954), radio astronomy became one of the most important tools of observational cosmology. Redshifts of up to z = 0.45 were measured for galaxies associated with radio sources (Minkowski 1960). Two fundamental breakthroughs were made towards the end of this period. First, quasi-stellar radio sources or "quasars" were discovered, with much larger redshifts even than radio galaxies. Secondly, it was demonstrated that the space density of radio sources varies with cosmic epoch, an observation that sounded the death knell for the Steady State cosmology.

(ii) Childhood. Mid-1960s to mid-1980s: This can be termed the "Spinrad era". Hy Spinrad devoted a considerable amount of his time, and that of the 120-inch telescope at Lick, to measuring the redshifts of faint radio galaxies by means of long photographic exposures. With considerable effort, he pushed the highest radio galaxy redshift out to z ~ 1 (Spinrad 1976, Spinrad et al. 1977, Smith and Spinrad 1980). At that time, it was not realised that radio sources interact with their optical hosts and the different wavelength regimes were usually studied in isolation from one another. Radio sources were regarded as interesting high-energy exotica that were useful for pinpointing distant elliptical galaxies. They were not believed to play an important role in the general scheme of galaxy evolution. During this period the paradigm that radio sources are powered by accretion of matter onto rotating black holes was developed.

(iii) Teens. Mid-1980s to mid-1990s: The replacement of photographic techniques by CCDs revolutionised optical astronomy. Large numbers of very distant radio galaxies were discovered with redshifts out to z ~ 5 (HzRGs). The surprising discovery that the radio sources and the optical host galaxies are aligned showed that there must be considerable interaction between the radio sources and their host galaxies. The role that orientation could have in determining observed properties was realised."Orientation unification" that explained differences in the observed types of active galaxies as merely due to the viewing angle of the observer became a popular interpretative "do-it-all".

(iv) Maturity. Mid-1990s to the present: New efficient techniques for finding distant galaxies were developed. These included photometric searches for "dropout" objects with redshifted Lyman break features and narrow-band searches for objects with excess redshifted Lyman fluxes. The advent of these new techniques placed the cosmological use of radio galaxies in a new perspective. Although radio astronomy lost its prime position as a technique for finding the most distant galaxies, it became clear that luminous radio galaxies play a highly important role in the evolution of galaxies and the emergence of large scale structure. The discovery of a relation between the masses of elliptical galaxies and the inferred masses of nuclear black holes led to the conclusion that all galaxies may have undergone nuclear activity at some time in their histories. This brought nuclear activity into the mainstream of galaxy evolution studies.

1.3. Hunting for HzRGs

Appendix A contains a compendium of 178 radio galaxies with z > 2 known to the authors. This list is the result of more than two decades of hard work. The relatively small number of known HzRGs illustrates both their rarity and the difficulty of detecting them.

Finding distant radio galaxies involves a multistage process.

• The first step (e.g. Röttgering et al. 1994, Blundell et al. 1998, De Breuck et al. 2000, De Breuck et al. 2004) is to filter out probable HzRGs from the huge number of radio sources contained in low-frequency radio surveys (e.g. Condon et al. 1998). The usual criteria used for filtering likely candidates are (i) radio colours - extremely steep spectra (Section 2.4) and (ii) small angular sizes (Section 2.3).

• In the second step, relatively bright nearby objects are discarded, by comparing the radio positions with existing wide-field shallow surveys at optical and infrared wavelengths.

• The third step is to refine the positions of the remaining radio sources to arc-second accuracy, thereby facilitating their identification with faint galaxies. Until the 1990s, radio identifications were made using optical CCD data (e.g. Röttgering et al. 1995, McCarthy et al. 1996). However, after sensitive near-IR detectors became available, the near-IR K-band was found to be more efficient for identifying distant radio sources (e.g. Jarvis et al. 2001, De Breuck et al. 2002, De Breuck et al. 2004, Jarvis et al. 2004, Cruz et al. 2006). More than 94% of bright radio sources are identified down to K = 22. Furthermore, the K-band magnitude of an HzRG provides a first indication of its redshift, by means of the K-z relationship (see Section 4.1).

• The fourth step is to carry out spectroscopic observations of the HzRG candidates with large optical and infrared telescopes. These are needed to determine redshifts from the wavelengths of HzRG emission lines (e.g. Röttgering et al. 1997, De Breuck et al. 2001, Jarvis et al. 2001, De Breuck et al. 2006, Bornancini et al. 2007) (see Section 3.2).

The main limitation on finding HzRGs has been the scarcity of available observing time on large optical/IR telescopes for the last stage in the process. Because candidates are located too far apart in the sky to permit multi-object spectroscopy, tedious long-slit spectroscopy on many individual fields is required to determine the redshifts.

Although radio selection ensures that there is no a priori selection against dust properties (e.g. extinction), there are other observational selection effects that introduce bias into the redshift distributions of HzRGs and the list given in the appendix. The first obvious source of bias is that the determination of a spectroscopic redshift is dependent on being able to observe bright emission lines with ground-based telescopes. The primary line for such redshift measurement (e.g. McCarthy 1993) is Ly 1216 Å (typical equivalent widths of several hundred Å). Useful additional lines are CIV 1549 Å, HeII 1640 Å and CIII] 1909 Å (equivalent widths in excess of ~ 60 Å). There is a so-called "redshift desert" 1.2 < z < 1.8, (e.g. Cruz et al. 2006), where [OII] 3727 Å is too red and Ly 1216 Å is too blue to be easily observed from ground-based optical telescopes. Objects located within the redshift desert are thus under-represented in samples of HzRGs.

A second source of bias is redshift incompleteness. A small fraction of radio sources, (~ 4%) are not identified to K ~ 22 and about a third of those with K-band identifications do not show any emission or absorption lines, even after long exposures on 8 - 10m class optical/IR telescopes. These objects either (i) have emission lines that are hidden by substantial dust obscuration (e.g. De Breuck et al. 2001, Reuland et al. 2003), (ii) are located at such large redshifts that Ly and other bright emission lines fall outside the easily observable spectral windows, or (iii) are peculiar in that they radiate no strong emission lines.

1.4. Redshift distribution of HzRGs

HzRGs are extremely rare. Luminous steep-spectrum radio sources associated with HzRGs have typical luminosities of L_{2.7 GHz} > 10³³ erg s^-1 Hz^-1 ster^-1. The number density of radio sources with this luminosity in the redshift range 2 < z < 5 is a few times 10^-8 Mpc^-3, with large uncertainties (Dunlop and Peacock 1990, Willott et al. 2001, Venemans et al. 2007).

Although HzRGs are sparsely distributed in the early Universe, such objects are almost nonexistent at low redshifts. The co-moving space density of luminous steep-spectrum radio sources increases dramatically by a factor of 100 - 1000 between 0 z < 2.5 and then appears to flatten out (Willott et al. 2001, Jarvis et al. 2001). Although there are huge uncertainties in the evolution of the luminosity function at higher redshifts, no significant cut-off in space density has yet been observed.

The redshifts at which the space density of radio sources is maximum correspond to a crucial era in the evolution of the Universe. It coincides with the epoch when (i) luminous quasars also appear to have had their maximum space density (e.g. Pei 1995, Fan et al. 2001) (ii) star formation was rampant and more than an order of magnitude larger than the present (Madau et al. 1998, Lilly et al. 1996, Schiminovich et al. 2005) and (iii) galaxy clusters were forming, but were not yet gravitationally bound structures.

The question of why the most luminous radio sources (and the most luminous quasars) became extinct in the local Universe is an intriguing one that is still not understood (Section 5.3).

1.5. Constituents of HzRGs

Radio galaxies have several distinct emitting components which provide diagnostics about various physical constituents of the early Universe. A list of known HzRG building blocks is given in Table 1, together with a summary of techniques used to study them. Also included are a list of the resultant diagnostics, some useful relevant references and our best estimate for the typical mass of the component in HzRGs. We caution the reader to thoroughly examine the assumptions inherent in deriving each of the diagnostics. Although, these assumptions are needed in order to reach some conclusions about the nature of HzRGs, it is important never to forget that they exist and to be sceptical about the "houses of cards" that are frequently involved.

Table 1. Constituents of distant radio galaxies

Constituent	Observable	Typical Diagnostics	Refs. 1	Mass (M)
Relativistic plasma	Radio continuum	Magnetic field, age, energetics, pressure, particle acceleration. Jet collimation and propagation	1,2
	X-ray continuum	Magnetic field, equipartition, pressures	3,4,1
Hot ionized gas Te ~ 107-108K ne ~ 10-1.5cm-3	Radio (de)polarisation	Density, magnetic field,	1	1011 - 12
	X-rays	Temperature, density mass
Warm ionized gas Te ~ 104-105K ne ~ 100.5 - 1.5cm-3	UV-optical emission lines	Temperature, density, kinematics, mass, ionisation, metallicity, filling factor	5,6,7,8	109 - 10.5
	Nebular continuum	SED contamination	9,10
Cool atomic gas Ts ~103K n(HI) ~ 101cm-3	HI absorption	Kinematics, column densities, spin temperature, sizes, mass	11,8	107 - 8
	UV-optical absorption lines	Kinematics, mass, column densities, metallicity	8,12 13,14
Molecular gas T ~ 50 - 500K n(H2) > 102 cm-3	(Sub)millimeter lines	Temperature, density, mass	15	1010-11
Dust T ~ 50 - 500K	UV-optical polarisation	Dust composition, scattering, mass, hidden quasar	16 17	108 - 9
	(Sub)millimeter continuum	Temperature, mass, heating source	18
Old stars t > 1 Gyr	Optical to near IR continuum	Age, mass, formation epoch	19	1011 -12
Young stars t <0.5 Gyr	UV-optical	Star formation rates, ages, history	20,8	109 - 10
	Ly	Star formation rate	20
Quasar (hidden or dormant)	UV-optical polarisation broad lines	Luminosity	21,22
Supermassive black hole(SMBH)	Extended radio, Quasar	Formation, evolution	23,24	~ 109
1 References: 1 = Miley (1980), 2 = Klamer et al. (2006), 3 = Felten and Morrison (1966), 4 = Schwartz (2002), 5 = Osterbrock and Ferland (2006), 6 = Groves et al. (2004), 7 = Groves et al. (2004), 8 = Dopita and Sutherland (2003), 9 = Aller (1984), 10 = Dickson et al. (1995), 11 = Morganti (2006), 12 = van Ojik et al. (1997), 13 = Binette et al. (2000), 14 = Binette et al. (2006), 15 = Downes et al. (1993), 16 = Cimatti et al. (1993), 17 = Vernet et al. (2001), 18 = Reuland et al. (2004), 19 = Seymour et al. (2007), 20 = Madau et al. (1998), 21 = Cimatti et al. (1993), 22 = Vernet et al. (2001), 23 = Blandford (2001), 24 = Peacock (1999)

As can be seen in Table 1, several constituents of HzRGs are inferred to be extremely massive, including old stars (up to ~ 10¹² M), hot gas (up to ~ 10¹² M) and molecular gas (up to ~ 10¹¹ M).

Figure 1 shows the spectral energy distribution (SED) of a typical HzRG from radio to X-ray wavelengths, together with a decomposition into various observable HzRG constituents - relativistic plasma, gas and dust, stars and the active galactic nuclei (AGN). We shall discuss each of these building blocks individually in Sections 2 to 5. Note from Figure 1 that disentangling the various components in the optical and the infrared is difficult and the results are often extremely model dependent and several alternative solutions may be equally consistent with the available multi-wavelength data.

Figure 1. Spectral energy distribution (SED) of the continuum emission from the HzRG 4C 23.56 at z = 2.5, illustrating the contributions from the various constituents. [From De Breuck et al. in preparation]. Coloured lines show the decomposition of the SED into individual components, under many assumptions. Cyan = radio synchrotron (Section 2). Black = Absorbed nonthermal X-ray AGN (Section 2.5). Yellow = nebular continuum (Section 3.2.6). Blue = AGN-heated thermal dust emission (Section 3.5). Red = Starburst-heated dust emission (Section 3.5). Green = Stars (Section 4). Magenta = scattered quasar (Section 5). The addition of the overlapping modeled components fits the SED well.