6. STELLAR POPULATIONS AT HIGH REDSHIFT

6.1 The golden age

Finally, for a line of argument that turns out to lead in completely the opposite direction - i.e. to galaxy formation at rather high redshift - we turn to the stellar populations in high-redshift radio galaxies. Most of the 1980s constitute a vanished age of innocence for the radio cosmologists: at this time, they were the only ones able to find galaxies at z 1 in any sort of numbers. A series of investigations established several interesting properties for these objects, in particular

1. The well-defined K-z relation for 3CR radio galaxies, consistent with purely passive evolution of their stellar populations, and producing 1 mag. of brightening by z 1 (Lilly & Longair 1984).

2. The large scatter in the optical-IR (Lilly & Longair 1984) and optical (e.g. Spinrad & Djorgovski 1987) colors of 3CR galaxies, which was interpreted as reflecting the occurrence of bursts of star formation in otherwise passively evolving objects.

3. Lilly (1989) argued for a two-component model in which a bluer component was superimposed onto a rather red underlying galaxy. To reproduce the spectral energy distribution of the red component (and thus of the reddest radio galaxies) required ages greater than 1 Gyr, pointing to high formation redshifts (Lilly 1989, Dunlop et al. 1989b, Windhorst, Koo & Spinrad 1986), although there was some controversy over the model dependence of the exact ages (Chambers & Charlot 1990).

4. Perhaps the high-water mark of this period was the discovery by Lilly (1988) of 0902+34 at z = 3.4 (at a time when the galaxy redshift record was 1.8). The apparently red colors of this object argued for a large enough age that the bulk of the stars must have formed at z 6 - an inference of enormous importance for models of galaxy formation.

However, over the last few years a revisionist tendency has appeared - leading to all the above achievements being questioned. Even at the time, there was some doubt whether we could be sure that the above behavior was representative of all galaxies. Fears of a radio-induced bias appeared well founded with the discovery of what has become known as the `alignment effect': the realization that at large redshifts (z 0.8) the optical and radio axes of many of the most powerful radio galaxies are aligned (McCarthy et al. 1987; Chambers, Miley & van Breugel 1987). Near-IR images of 3CR galaxies appeared to confirm that the infrared morphologies of these objects were in general just as peculiar as their optical morphologies (Chambers, Miley & Joyce 1988; Eisenhardt & Chokshi 1990; Eales & Rawlings 1990). These discoveries provide direct evidence of radio-induced `pollution' of the UV-optical light of radio galaxies, and this led some authors to suggest that these sources are thus useless as probes of galaxy evolution in general (e.g. Eisenhardt & Chokshi 1990).

Furthermore, it has become apparent that Lilly's galaxy 0902+34 does not have the properties initially claimed. The K flux is rather lower than Lilly's measurement, and a large fraction of this smaller total is contributed by the [OII] 3727Å line, which is redshifted into the K window. The result is that the galaxy in fact looks very young: nearly flat-spectrum with no evidence for the presence of an old component. On this basis, and considering other similar objects at extreme redshifts, Eales et al. (1993) have argued that radio galaxies at z 2 are in effect protogalaxies observed in the process of formation.

Before accepting this remarkable reversal of conventional wisdom, however, it is worth bearing in mind that the galaxies under discussion are among the most luminous few dozen radio AGN in the entire universe (inevitably: they are the high-redshift members of bright samples with S ~ 1-10 Jy). In order to draw any general conclusions about galaxy formation, it is necessary to understand the effect the AGN has on the optical/IR properties of the galaxy within which it is embedded.

6.2 Alignments as a function of power

What is required is to be able to study the properties of galaxies with a wide range of radio powers, and this is what James Dunlop & I have attempted in some recent work (Dunlop & Peacock 1993). In order to eliminate possible confusion with any epoch dependence, we worked with a redshift band around z 1. At this redshift, it is relatively easy to select samples unbiased by optical selection, and the objects are bright enough that high-quality data can be obtained. We considered galaxies from two catalogues: 19 high-power 3CR galaxies; 14 low-power comparison galaxies with S_{2.7 GHz} > 0.1 Jy from the Parkes Selected Regions (PSR) (Downes et al. 1986; Dunlop et al. 1989a). The PSR galaxies are a factor 20 less radio luminous than their 3CR counterparts. Radio luminosity is the only significant difference between the radio properties of the two samples.

Our principal dataset on these galaxies is deep infrared images, taken with the 62 x 58 pixel InSb array camera IRCAM, on the 3.9m United Kingdom Infrared Telescope (UKIRT), with the camera operating in the 0.62-arcsec/pixel mode. From these images, we investigated the extent of the alignment effect at z 1. To avoid subjective factors, the infrared position angles were determined automatically by using the moments of the sky-subtracted flux within some circular aperture. We decided to vary the diameter of the aperture to adapt to the size of the radio source, because there are virtually no examples of optical or IR emission extending beyond the radio lobes. If the diameter of the radio source lay between 5 and 8 arcsec, an aperture equal in diameter to the radio source was used. If the radio source was greater than 8 arcsec in diameter, an 8 arcsec diameter was used (larger apertures generally contain foreground objects). If the radio source was smaller than 5 arcsec in diameter, a 5-arcsec diameter was used.

Figure 5 shows the resulting IR-radio alignment histogram for the 3CR and PSR subsamples. The infrared alignment effect is extremely obvious in our data for the 3CR galaxies, which appears to contrast with the conclusions of Rigler et al. (1992). Much of the apparent discrepancy arises from the fact that we have a larger sample. Position angles for objects in common generally agree well, but with some exceptions which are due to different methods of analysis; Rigler et al. (1992) sometimes use a large aperture where their position angle is affected by companion objects. In contrast to the 3CR sub-sample, there is no evidence of any significant alignment between the infrared and radio morphologies of the PSR galaxies. This result is very robust and quite obvious given the images: the PSR galaxies are rounder, with generally little sign of the disturbance evident in many of the 3CR images.

Figure 5. Histograms of (IR-Radio) position angle differences for the 3CR and PSR samples. The clear difference seen here is completely robust to different methods for determining position angles. It is related to the fact that the PSR galaxies are also rounder, and generally lacking in an extended aligned component of blue light.

This argues in favor of the two-component model advanced by Lilly (1989) and Rigler et al. (1992). In this, the underlying galaxy is round, but there is a component of variable amplitude which is elongated along the radio axis, and it is this which leads to the alignment. Our data demonstrate that the strength of this component correlates well with radio power, as is perhaps not so surprising in retrospect. Certainly, several models for the production of this light exist that predict a correlation with radio power (scattering, induced star formation, inverse Compton emission - see e.g. Daly 1992 for a review). We shall not be concerned here with having to plump for a specific model, but it is worth noting that evidence is starting to mount in favor of the explanation in terms of scattering from a hidden blazar. The main argument in this direction is the measurement of polarization with B-vector perpendicular to the radio axis. The first measurements of this effect gave very low percentage polarizations, implying that this could not be the dominant mechanism. However, with better resolution, imaging polarimetry is now producing polarized fractions of 20% in the outer parts of strongly aligned galaxies (Jannuzi & Elston 1991; Tadhunter et al. 1992; Cimatti et al. 1993). Given geometrical dilution, it now seems plausible that the aligned component results from scattering in at least some objects.

6.3 Colors and ages of radio galaxies

Having seen that the extent of the aligned component scales so dramatically with radio power, we now look for other optical/infrared properties which correlate with power. Given that the aligned component is often bluer than the nucleus of the galaxy, we should certainly expect to see some correlation between color and power. A useful way of quantifying the degree of UV activity was introduced by Lilly (1989). He assumes that the observed spectrum of a radio galaxy arises from a combination of two distinct components - an `old' population with a well-developed 4000Å break, and a `young' flat-spectrum component. This simple model can be fitted to the observed colors by varying one parameter. This is f₅₀₀₀: the fractional contribution of the flat-spectrum component to the galaxy light at a rest wavelength of 5000Å. This method can also be used with some success to estimate the redshift for objects which lack spectroscopy (see Lilly 1989; Dunlop & Peacock 1993). Some of the PSR objects had their redshifts estimated in exactly this way: the redder objects with low f₅₀₀₀ also have low levels of emission-line activity and so are of course the hardest spectroscopic targets.

This procedure is illustrated in Figure 6. For the `old' or `red' component we chose to adopt a spectrum capable of producing the reddest colors seen in radio galaxies at z 1 (e.g. 3C65); in practice this was achieved using the spectrum produced by a stellar population of age 10 Gyr in an updated version of the models of Guiderdoni & Rocca-Volmerange (1987). For the `young' or `blue' component, we decided to adopt a power-law spectrum (f ^-) with a spectral index = 0.2. This choice of spectrum can be justified at two different levels. First, the exact value of was chosen in the spirit of scattered quasar light; Barvainis (1990) concluded that the mean value for the optical spectral index in high luminosity quasars (i.e. those whose optical spectra are essentially uncontaminated by a host galaxy contribution) is = 0.2. Second, empirically, this form of spectrum is an excellent representation of the approximately flat f optical-UV continuum actually observed in high-redshift radio galaxies.

Figure 6. Two examples of the spectral fitting used to determine estimated redshifts and f5000, the relative contribution of the flat-spectrum component. The `red' component is the spectrum produced by a 1-Gyr `Burst' model of galaxy evolution at an age of 10 Gyr. The blue component is a power-law with spectral index = 0.2 (f -). the mean optical spectral index found for quasars by Barvainis (1990). 2355-010 is a red radio galaxy with only a very small value of f5000, while 0059+027 is one of the bluer galaxies in the PSR sample.

In Figure 7 we show the quantitative relation between this definition of UV activity and radio power. Radio power and f₅₀₀₀ appear to be strongly correlated (no PSR galaxy has f₅₀₀₀ > 0.19 whereas more than half the 3CR galaxies have f₅₀₀₀ > 0.20). This result contrasts sharply with that of Lilly (1989), who reported that in his combined 3CR and 1-Jy sample there was no significant correlation between f₅₀₀₀ and P_{408 MHz}. The origin of the difference appears to be an error in Lilly's calculation of radio luminosity. An interesting aspect of the relation with power is that all sub-samples appear to possess a range of f₅₀₀₀ values, but with power apparently setting the upper limit in f₅₀₀₀. This suggests the existence of a second parameter which determines the actual level of UV light - see Dunlop & Peacock (1993) for further discussion.

Figure 7. Radio power, P2.7 GHz, versus f5000 for the combined 46-source 3CR / 1-Jy / PSR sample. 3CR sources are shown as squares, 1-Jy sources as triangles, and the PSR sources as circles. Notice that the correlation is mainly in the sense of setting an upper limit to f5000 at given power.

In Figure 8 we compare the R - K colors of the PSR and 3CR galaxies. Several other objects which are not part of our PSR and 3CR subsamples have been included here for comparison purposes. These are (i) the very red 3CR galaxy 3C65, (ii) the five 1-Jy galaxies with measured redshifts for which r - K colors are given by Lilly (1989), and (iii) all spectroscopically confirmed quasars with 0.5 < z < 2.0 in the Parkes Selected Regions sample for which R - K colors exist (Dunlop et al. 1989a).

Figure 8. Comparison of the R - K colors of the PSR galaxies (solid squares and triangles) and the 3CR galaxies in the subsample (open circles and diamonds). PSR galaxies with measured redshifts are denoted by solid triangles, those with estimated redshifts by solid squares. 3CR galaxies whose K-band morphologies are aligned with 15° of the radio axis are denoted by diamonds, and the remainder by open circles. Also shown are five 1-Jy galaxies (from Lilly 1989) (asterisks), and all spectroscopically confirmed quasars with 0.5 < z < 2.0 in the PSR sample (stars). The dashed line shows the effect of simply k-correcting the spectrum. The solid line shows a very old (zf = 50, 0 = 0, H0 = 50 kms-1 Mpc-1) UV-hot model of elliptical galaxy evolution (Rocca-Volmerange 1989).

This diagram displays a number of important features. First, with the obvious exception of 3C65, the PSR galaxies are consistently redder than the 3CR galaxies; moreover, the PSR galaxies display remarkably little dispersion in their optical-infrared colors. This is well consistent with the findings of Rixon, Wall & Benn (1991) at lower redshift: they found the rest-frame colors of radio ellipticals at z < 0.3 to be constant to within a few hundredths of a magnitude. In contrast, the 3CR galaxies scatter downwards from the well-defined PSR locus towards the region of color space occupied by the PSR quasars (the very red galaxies 3C65 and 1129+37 appear to be exceptional). Of the six 3CR galaxies with R - K 4.0, all but one (3C252) have K-band morphologies clearly aligned with their radio axes.

The homogeneity of the PSR galaxies, along with the lack of any dramatic alignment effect in the redder galaxies, suggests instead that the true optical-infrared color of a radio-quiet elliptical at z 1 is actually R - K 4.8. Values of f₅₀₀₀ 0.05 might be a feature of most elliptical galaxies at z 1. This is certainly consistent with the results of Aragón-Salamanca et al. (1993). From optical/IR photometry of clusters of galaxies up to z = 0.8, they conclude that ellipticals (mainly radio-quiet) in the highest-redshift clusters are slightly bluer than present-day ellipticals. On the assumption that these galaxies formed in a single burst, their data allow the epoch of formation to be as low as z = 2. However, the radio-selected samples extend the range still further. Although the above discussion has concentrated on the situation at z 1, the PSR sample contains a number of galaxies inferred from color-estimated redshifts and from the K-z relation to have z 2. These also are apparently old and red, with R - K 4 - 5. If this is taken to imply a minimum age of 1h^-1 Gyr, the formation redshift is pushed out to between 3.3 and 7.2, depending on . Note that this is the epoch at which the whole galaxy must be assembled: ellipticals cannot have been assembled from many small clumps after star formation had ceased (Bower, Lucey & Ellis 1992). It will be fascinating to pursue this line of argument in mJy samples, where we may hope to find `normal' radio galaxies at z > 3. If these are still red, the consequences for galaxy formation models will be radical indeed.