Searching for the earliest massive galaxies
We are conducting a program to find the first massive galaxies to form in the early universe, the "high-redshift radio galaxies" (HzRGs). These are the earliest galaxies in which we can deduce, from their powerful radio emission that a supermassive black hole (109 MSun) is in place. Powerful radio galaxies in the local universe are hosted only by elliptical galaxies, with a close connection between the mass of the black hole and that of the galaxy (Kormendy & Richstone 1995, Gebhardt et al. 2000). The presence of a supermassive black hole establishes a direct link between HzRGs and present day giant ellipticals, making HzRGs particularly important for testing current theories of galaxy formation.
Figure 1: K-z diagram for optically selected (o, +) and radio-selected (squares) galaxies (De Breuck et al. 2002). The radio galaxies trace the bright upper envelope in K-band luminosity, suggesting that they are also the most massive galaxies at a given redshift.
The motivation for selecting high redshift candidates from radio surveys is illustrated in Figure 1 which shows the relationship between redshift and K-band magnitude. It is clear from this diagram that at z > 1, radio-selected galaxies are 2 3 mag brighter than optically-selected galaxies. Their near-IR K-band light is dominated by stellar emission rather than AGN `contamination' (e.g. Jarvis et al. 2001). Radio selection therefore traces the most massive and luminous galaxies out to very high redshift. The furthest radio galaxy found to date is at z = 5.19 (van Breugel et al. 1999). All this makes HzRGs ideal probes for investigating the formation and evolution of massive galaxies, and even protoclusters (e.g. Kurk et al. 2000, Pentericci et al. 2000, Venemans et al. 2002). However, HzRGs are extremely rare, and the only way to find them is by "data mining" observations of very large areas of sky at radio wavelengths using selection techniques that isolate the most likely high redshift candidates.
Steep spectrum selection
In the late 1970s, Blumenthal & Miley (1979) found that radio sources without optical counterparts generally had steeper spectral indices. Therefore, in order to separate the high redshift candidates from the bulk of the nearby radio galaxy population, galaxies are chosen to have an ultra-steep radio spectral index (a < -1.0, where Snµ na). A number of HzRG surveys have adopted this technique, which has proved to be very successful, with thirty new z > 3 galaxies found so far.
The relationship between spectral index and redshift is shown in Figure 2 (De Breuck et al. 2000, 2004) for the MRC and 3CR samples and for samples selected to have ultra-steep spectra (USS). The combined USS samples have ~70% with z > 2 and ~30% with z > 3.
Figure 2: Spectral index vs redshift for radio galaxies from our 843 – 1400 MHz USS sample, previous USS samples and samples with no spectral bias (3CR, MRC). The dotted line marks the nominal cut-off at a = -1.3, while the solid line defines the trend for the 3CR and MRC samples.
The success of steep-spectrum selection in finding high redshift objects has previously been attributed to several effects. The spectral energy distributions of radio galaxies tend to steepen towards high frequencies and this steeper region is then redshifted to lower frequencies (below 1 GHz); the so-called k-correction. Other effects, such as inverse Compton losses, enhance the spectral steepening.
While USS techniques have resulted in discovery of the most distant radio galaxies known, many low-redshift radio galaxies also have a steep spectrum and need to be eliminated from the sample. Spectroscopic redshifts cannot be measured for every steep spectrum target because this would require a prohibitively large observing time. Therefore, other selection criteria are needed to refine the steep spectrum sample to include only the most likely high-redshift candidates for spectroscopic follow-up.
Regions with |b| < 20° are excluded to avoid confusion.
Targets with optical counterparts are removed as they are likely to be at low redshift.
High resolution radio images from the Compact Array at 13 and 20 cm are then essential to pinpoint the exact position of the nucleus and hence find the K-band counterpart.
Using the K-z relation, the faintest K-band magnitudes indicate the best high redshift candidates. These are selected for spectroscopic follow-up.
Once a redshift is measured, the Compact Array images again come into play in looking at the relationship between the morphology and the spectral energy distribution.
The Southern HzRGs sample
Until recently, radio searches for these rare galaxies were based almost exclusively in the northern hemisphere, due to a lack of large, sensitive multi-frequency radio surveys in the south. This imbalance has now been removed with the advent of the 843-MHz Sydney University Molonglo Sky Survey (SUMSS; Bock et al. 1999, Mauch et al. 2003), which covers the sky south of declination 30°. Furthermore, the 408-MHz survey from the original Molonglo Cross telescope in the 1970s has been re-analysed (D. Crawford 2005, priv. comm.), and the flux density limit lowered from 1 Jy to 200 mJy. The highest redshift objects are likely to have 408-MHz fluxes below the limit of the MRC because the maximum flux density of known USS radio sources drops well below 1 Jy at 408 MHz for z > 3.
The new 408-MHz catalogue and SUMSS overlap the NVSS 1.4-GHz catalogue in the range 40°< DEC < 30°. We have therefore limited our project to these declinations so that two pairs of selection frequencies are available, 408 – 843 MHz and 843 – 1400 MHz. We can then compare the effect of selection at different frequencies to test which is more efficient for finding high redshift galaxies. The extra frequency also allows us to characterise the spectral energy distribution. Selection in the narrow frequency ranges of 408 – 843 MHz and 843 – 1400 MHz is a slightly different approach to previous USS searches which have used a wide frequency range (~300 to ~1400 MHz) or high frequencies (> 1400 MHz).
Results so far
Figure 3: Compact Array 20-cm radio contours overlaid on the K-band IRIS2 image of the most distant radio galaxy confirmed in our sample so far, at z = 3.980. Image size is 56², and the beam is 11.4 ²× 6.4², p.a. = 68º.
The 843 – 1400-MHz sample was observed in the infrared using a combination of the AAT and NTT, followed by spectroscopy on the NTT and VLT. The highest redshift from the 843 1400-MHz sample so far is z = 3.980 (Figure 3). Five galaxies have z > 3 and nine have z > 2. A further six are undetected in deep VLT integrations implying z > 7 or heavy dust obscuration (De Breuck et al. 2004, De Breuck et al. 2005) and Gemini time has been awarded for NIR spectra of these. Our results increase the number of z > 3 galaxies in the southern hemisphere to 12.
An interesting member of the 843 1400-MHz sample is shown in Figure 4. A deep I-band image with the VLT showed only one object aligned with the 1.4-GHz Compact Array contours (marked b). A spectrum gave a redshift of only 0.8 despite having a K-band magnitude of K > 20, making us suspicious of the identification because the point lay well off the radio galaxy K-z distribution (see Figure1). A higher resolution 4.8-GHz Compact Array image, when overlaid on a deep NTT/SofI K-band image, revealed the true counterpart for this radio source. It is interesting that there is almost no correspondence between the K-band and I-band images. The I-band image does not have an obvious detection at the location of the K-band source (an I-band dropout), which indicates that the radio source may be at very high redshift.
The 408-SUMSS sample is in the K-band imaging stage, with 42 galaxies already observed by IRIS2. Follow-up spectroscopy of the faintest two of these has been awarded GEMINI time, while the brighter ones are being observed with the ANU MSSSO 2.3-m telescope.
Figure 4: Left: Compact Array 1.4-GHz radio contours on a VLT I-band image showing the I-band identification, marked b, which proved to not be the counterpart to this radio source. Right: Compact Array 4.8-GHz contours overlaid on the NTT K-band image isolating the correct counterpart, marked a. Both images are 30² on a side and cover the same sky area. Note that the two images have very few objects in common.
Automatically constrained CLEANing
In order to identify the faintest K-band targets in confused fields, and to improve flux densities for our spectral energy analysis, we have had to look more carefully at the Compact Array radio reduction. This has led to the adoption of new techniques which have benefits beyond our current project.
Most CLEAN algorithms work effectively for isolated point sources. However, the presence of other sources of significant flux density results in their sidelobes contributing to the measured flux density of the source of interest. Clearly how large an effect these sources have is a function of their brightness and their proximity to the source of interest in the image; the brighter, closer sources cause the most significant problems. Running unconstrained CLEAN on an image containing multiple sources can result in a significant underestimate of the flux density of the source of interest. It can also introduce spurious sources into the image. The standard way of resolving these problems is to manually select CLEAN regions around all of the sources in the field. However, this process is time-consuming when dealing with a large number of observations. In order to measure reliable flux densities for our sources we have developed a technique to define CLEAN regions without manual intervention.
Large radio surveys such as NVSS and SUMSS have catalogued the positions of the majority of strong radio sources in the southern sky. Hence it makes sense to use this information to pre-define CLEAN regions around these known source positions. Our technique involved searching these catalogues for all sources within the image region, above a pre-defined flux density cut-off. Then a box is fitted around each of these sources and passed to the normal Miriad CLEAN routine.
We tested this method on a sample of 147 radio galaxies, comparing the 20-cm Compact Array flux density with values from the NVSS catalogue. As expected, constraining CLEANing to specific regions around known sources improved the flux densities obtained and removed the artefacts. We also compared the results against the same process using manually selected CLEAN regions. Again, our flux densities were closer to the NVSS in all cases. For example, one of our sources has a 1400-MHz NVSS flux density of 43.2 ± 1.7 mJy, and the flux density measured after unconstrained CLEANing was 31.7 ± 1.1 mJy. However, constrained CLEANing using our automatically defined regions (for the same number of iterations) gave 41.5 ± 1.4 mJy, consistent with NVSS.
Future plan
We plan to continue our campaign of NIR and radio imaging with follow-up spectroscopy for our full sample of targets matched among the three catalogues. The aim is to identify 50 z > 3 galaxies, enough for statistical studies of their radio spectrum and space density.
References
Blumenthal, G. & Miley, G. 1979, A&A, 80, 13
Bock, D.C-J., Large, M.I., Sadler, E.M. 1999, AJ, 117, 1578
Mauch, T. et al. 2003, MNRAS 342, 1117
De Breuck, C. et al. 2000, A&A 362, 519
De Breuck, C. et al. 2002, AJ 123, 637
De Breuck, C., Hunstead, R.W., Sadler, E.M., Rocca-Volmerange, B. & Klamer, I. 2004 MNRAS 347, 837
De Breuck, C., Klamer, I., Johnston, H., Hunstead, R.W., Bryant, J., Rocca-Volmerange, B. & Sadler, E.M. 2005, MNRAS, in press.
Kormendy, J. & Richstone, D. 1995 ARA&A, 33, 581
Gebhardt, K. et al. 2000. ApJ 539, L13
van Breugel, W. et al. 1999, ApJ 518, L61
Jarvis, M.J. et al. 2001, MNRAS 326, 1585
Kurk, J.D. et al. 2000, A&A 358, 1
Pentericci, L. et al. 2000, A&A 361, 25
Venemans, B. et al. 2002, ApJ 569, L11
Julia Bryant, Dick Hunstead, Tara Murphy, Jess Broderick, Elaine Sadler, Helen Johnston, Ilana Klamer (University of Sydney) and Carlos de Breuck (ESO)
(jbryant@physics.usyd.edu.au)