Author: Russell Cannon
Email: rdc@aaoepp.aao.gov.au
A progress report presented by Russell Cannon at the Dunk Island Cosmology Conference on 28 August 1999.
The work I am describing here is a joint project by Elaine Sadler and Vince McIntyre (U. Sydney), Carole Jackson (RSAA, Mt Stromlo) and myself at the AAO. We are identifying and studying radio sources that happen to be included in the optically-selected 2dF GRS.
The primary radio source catalogue is the NRAO VLA Sky Survey (NVSS: Condon et al. 1998, AJ 115 1693). This is a 1.4 GHz all-sky survey down to declination -40 deg, containing 2 million sources with flux density greater than 2.5 mJy. The spatial resolution is 45 arcsec, with positional accuracy ranging from < 1 arcsec for the strongest sources to 7 arcsec at the faint limit.
Additional radio sources will be selected from the Sydney University Molonglo Sky Survey (SUMSS: Bock et al. 1999, AJ 117 1578) which covers the southern sky, with declination < -30 deg, to a comparable depth and resolution at 843 MHz; some radio spectral information will be available for sources in the overlap zone. However, all the 2dF fields available to date are in the northern NVSS region only; the distribution of fields has been shown by Matthew Colless and others.
The identification of radio sources with optical objects is based simply on positional coincidence. Early tests showed that objects lying within 10 arcsec have about a 90% chance of being genuine identifications (both the optical and radio surveys have minimum separation of targets greater than 30 arcsec, so there are no multiple identifications). Two statistical checks were used to estimate the 10% level of random coincidences: (i) counting the number of "identifications" when all the radio sources were offset by a few arcminutes, and (ii) simply considering the average densities of optical and radio sources on the sky. The reliability can probably be improved in at least two ways, by eliminating the up to 5% of GRS objects that turn out to be Galactic stars, and by taking account of the variable accuracy of the radio positions as a function of flux density.
About 1.5% of the GRS optical galaxies turn out to be coincident with NVSS radio sources; conversely, about 5% of the NVSS sources are included in the GRS survey. This means that there must be about 4000 known radio sources in the full GRS survey. Already we have some 700 identifications, making this the largest homogeneous sample of optical spectra of radio galaxies. A strength of the program is that the optical and radio surveys were carried out completely independently, so there are no explicit selection effects in the sample.
An initial sample of about 100 radio galaxies was identified in the fields observed by 2dF prior to March 1998. These were used by Elaine Sadler to do a preliminary classification of the optical spectra by eye; the results were presented at the 1998 ASA meeting and are being published in PASA (this paper is available on el-PASA and as astro-ph/9909171).
The optical spectra can be separated into three main classes: star-forming galaxies (either normal late-type galaxies at very low redshift, or starbursts) with relatively strong Balmer emission lines; optical AGNs, with emission lines of [OII] and [OIII] dominant (a small fraction of them with broad Seyfert-type Balmer lines); and radio AGNs, presumably mostly early-type galaxies, showing only composite stellar absorption-line spectra. The approximately equal distribution of the radio sources into these three categories is a consequence of the relatively bright optical limit, so that almost all the GRS galaxies have z < 0.3; most "classical" identification projects have been dominated by powerful radio galaxies and quasars at much larger redshifts.
Similar results from a search for NVSS identifications in the Las Campanas Redshift Survey have recently been published by Machalski and Condon (ApJSupp 123 41 1999) but they base their classification on the radio/IR/optical flux ratios, since they have optical spectra for only a third of their sample.
A more sophisticated spectral classification of the small initial sample, based on emission line ratios, was done by Diana Londish (UNSW) and Carole Jackson (paper in preparation). Eventually it should be possible to relate the classification of the radio galaxies to the overall classification of the entire GRS sample, for example using the PCA-type of analysis of Folkes, Ronen et al. (MNRAS in press; astro-ph/9903456).
Fig.1. A plot of redshift v. blue magnitude shows that the identified radio sources populate the upper optical luminosity range at all redshifts. Crosses represent starburst galaxies; these predominate at low redshifts and bright apparent magnitudes (z < 0.1, B(J) < 16.5). Shaded circles are emission line AGNs, while open circles are radio-only AGNs; these predominate for 0.1 < z < 0.3 and 17 < B(J) < 19.5. (Note that the names of the axes are accidentally transposed in this plot.)
Fig.2. Examples of the three main galaxy types, based on their optical spectra; these can be compared with:
Fig.3. The PCA-derived types of Folkes et al. (from astro-ph/9903456, their Fig.9).
Fig.4, Fig.5, Fig.6. The distributions of the SF and AGN types with B(J) magnitude, redshift and radio power, for the small initial sample (all from astro-ph/9909171).
Fig.7. The local Radio Luminosity Function, based on a sample of ~700 sources. The shaded circles joined by the line are our new data; the crosses, with larger error bars, are from Condon (ApJ 338 13 1989), based on fewer and less homogeneous data. The new data are smoother but still show a double hump, which evidently has a physical origin (see Fig.8 below).
Fig.8. Separate RLFs for the star-forming galaxies and AGNs. This confirms the finding of Condon (1989) but note that the separation of galaxy types is done here purely on the basis of the optical spectra, not their morphology or fluxes.
(i) Check the dodgy 2dF spectra; a few percent do not yield good redshifts and classifications. In some cases this only requires re-analysis of the 2dF data, for some new optical spectra are needed. This careful scrutiny of a small subset of the full GRS sample provides a useful quality control check on the main survey.
(ii) Do a better reliability analysis, taking account of varying radio position errors and the fact that some radio (and some optical) sources are significantly extended. Also, check for IDs of possible double radio sources listed as two separate NVSS sources.
(iii) Relate this large but low-redshift sample to the more powerful classical radio galaxies. Obtain higher resolution radio maps with the ATCA to see if these low-power galaxies can be classified as FR-I, FR-II and nuclear sources, and if their statistics are consistent with unified pictures.
(iv) Relate the optical properties of the radio galaxies to those of the full GRS sample. Look at the numbers and properties of different spectral sub-classes; find the relative numbers of radio-loud and radio-quiet Seyferts.
(v) How can 2dF be used directly to extend this work? It is not easy to obtain large samples of bright enough candidates to utilise all 400 fibres in 2dF. Currently we have only half a dozen radio IDs per 2dF field. For full use of 2dF this would have to be increased by two orders of magnitude, requiring about a factor of ten increase in both the radio and optical flux limits, or an even larger gain in at least one of these. Any deeper radio survey also has to be done to much higher spatial resolution to avoid confusion, while to go a lot fainter optically will require much better sky subtraction than has traditionally been achieved with fibres. However, the "nod and shuffle" technique described here by Karl Glazebrook may provide a crucial breakthrough in this regard; it is already quite possible to generate deeper radio source lists which satisfy the above criteria, for example using Sydney University's MOST, while the optical limits attainable with "nod and shuffle" would be well-matched to the upper limit of 200 fibres available with that technique.
There have been several references here to the desirability of getting accurate spectrophotometry for galaxies in the 2dF, Sloan and other surveys. However, Alex Szalay has warned us about the dangers of complicated hidden systematic errors which are going to make it hard to analyse the huge survey databases, and I believe this is a case in point. Speaking as an observer with both spectroscopic and photometric experience, I have to say that I am not sure we will ever be able to do spectrophotometry with fibres to much better than 10%. At present we are not achieving better than 20%.
The fundamental problem is that we are using very small apertures (2 arcsec diameter for 2dF), and both the atmosphere and the telescope optics introduce complicated wavelength-dependent effects that vary across the 2 degree wide field of view. There are chromatic effects at the level of a few tenths of an arcsecond, even when you use an Atmospheric Dispersion Compensator (ADC), as with 2dF on the AAT. Ian Lewis gave us a graphic demonstration of how large the atmospheric effects become when you are working over a wide wavelength range at large zenith distances; the raw UV and IR images can easily be displaced by more than 1.5 arcsec.
These effects are compounded by small errors in the fibre placement, at the level of say 0.3 arcsec rms. The light that goes down the fibres is a convolution of the seeing, the optics and positional errors (and of course of the source itself, for resolved galaxies). I should stress that these effects appear to dominate over any intrinsic wavelength-dependence of throughput from fibre to fibre, which could in principle be calibrated out.
The resultant errors were demonstrated when we did tests of the 2dF ADC, taking successive spectra of the same stars going down the same fibres. The expected large variations in UV throughput were seen when the ADC settings were varied (although initially the system appeared to work better without the ADC, due to several errors in the control system!). However, in addition to the UV variations, there were large variations in spectral slope and shape at longer wavelengths, at the level of about 20%. I believe that these must be due to small variations in the position of the star image relative to the fibre (in this case due to seeing variations, since the exposure times were short).
My conclusion is that in general, by far the best way to get the correct large-scale continuum shape for 2dF spectra will be to use broad-band photometric colours to correct the fibre spectra. Such photometry can be done to an accuracy of a few percent, with care. It certainly does not seem to be safe to use the continuum slope or overall spectral shape as one of the parameters in a PCA type of analysis.