Identifications with COSMOS

The COSMOS/UKST Southern Sky Object Catalogue is a database containing parameters of optical objects, formed by scanning the blue IIIaJ plates from the UK Schmidt Southern Sky Survey; see Collins et al. (1989), and references therein, for a description. The catalogue is available on-line at the Anglo-Australian Observatory and a description of the software used to extract data and finding charts is given by Drinkwater et al. (1995). An improvement to the astrometric accuracy of the catalogue (by correcting an error in the coordinate system transformation) has been implemented in the on-line software following the discovery of systematic offsets by Unewisse et al. (1993).

The great advantage in using such a catalogue for identification work is that it is a fast process, and gives many useful parameters for the objects. However, it does contain errors, the most common being misclassification of a small fraction of galaxies as stars and vice versa, and the occasional omission of objects altogether. The software that produces the catalogue can also have trouble deblending partly merged images. For a discussion of these and other issues see Unewisse et al. (1993). By combining COSMOS identifications with inspection of the sky survey images, discussed in section 4, we can expect to obtain a more complete and reliable list of identifications than one made with COSMOS alone.

Automated Identifications

A preliminary list of optical identifications for all of the sources found in the MOST fields was compiled by extracting object lists and finding charts centred on each radio position from the COSMOS catalogue. The uncertainties derived from equation (1) were adopted for the radio positions. The uncertainty in the COSMOS positions is quoted at 1-2 (Unewisse et al. 1993) but has been improved somewhat by Drinkwater et al. (1995). A conservative value of 15 was adopted for both and . Each object within of the radio position was assessed for plausibility as a possible identification according to the following criteria. Following Allington-Smith et al. (1982), we calculated the radio-optical position difference, R, normalised by the combined errors, as follows: \

\

and are the separations in RA and Dec respectively, and are the uncertainties in the optical positions (15), and and are the uncertainties in the radio positions (equation 1). For galaxies, an identification was deemed to be genuine if . For unresolved and slightly resolved sources, to which this method can validly be applied, Allington-Smith et al. (1982) show that this gives a formal completeness of 99%. For stellar objects and so-called `faint' objects (objects too faint for reliable classification), the identification was deemed to be genuine if . This gives a lower completeness for these objects, but the high surface density of stellar objects would result in an unacceptably high number of chance coincidences if the criterion was applied, as discussed in section 3.2. Although the set of galaxy identifications of unresolved sources obtained purely from the method described here is highly complete, it is not fully reliable. Section 3.2 describes measures taken to ensure reliability of the identification list.

Chance Coincidences

To assist in deciding which of the preliminary identifications to accept, estimates were made of the probability that any coincidence of radio and optical objects is real or due to chance. The online COSMOS software at AAO was used to find the closest optical counterparts to all the fitted radio positions in the subsample of cluster fields being studied. The same was done for positions offset from the true positions by 2-10, with the data from 40 different offset vectors combined to improve the statistics. Figure 2 shows the normalised detection frequency of stellar objects and galaxies as a function of separation from the true positions, i.e., the probability that an association has occurred by chance. The main conclusions from Figure 2 and additional plots (not shown) were:

1. The probability of finding a genuine stellar identification is very small at radio-optical separations greater than 5 (Fig 2(a)). The same result is found for `faint' objects.
2. Galaxies have a much higher probability of being genuine identifications than stellar objects (compare Figures 2(a) and 2(b)). This is consistent with past work which has shown that radio galaxies are more numerous than quasars in a low frequency survey. No radio stars are expected to be detected to our sensitivity limit.
3. At a given radius vector offset, the likelihood that a galaxy association is real increases as brighter magnitude cutoffs are applied.
4. Because of the broad distribution of quasar optical luminosities, magnitude is generally not a good indicator of the reliability of a candidate stellar identification.

Although these are useful conclusions, they are based on the average density of objects in all the fields when in fact there is significant variation from field to field. To establish a definite reliability limit, the density of galaxies, stars and faint objects was calculated in each individual field. This better accounts for the variation of galaxy density with magnitude in clusters of different richness and redshift, and the variation of stellar density with Galactic latitude.

In re-running the position comparison routine described in section 3.1, the appropriate object density and radio-optical separation were used to calculate the chance probability for each potential identification. Note that this calculation is different from Figure 2 as it only takes account of the object density. A reliability limit was imposed by discarding associations for unresolved sources with a chance probability of more than 2%. Extended sources, for which real radio-optical position offsets can occur, were inspected individually as described in the next section.

© Copyright Astronomical Society of Australia 1997