All objects from our sample are found to have a FIR loudness . In contrast, the highest value of R among the sources that were removed from the sample is 76. Therefore our approach of selecting those sources which do not have reliable counterparts above the COSMOS plate limit, or for which the counterpart is so faint that misidentification is no longer unlikely, proves to be very effective in selecting sources with extreme values of R. What is the nature of these objects? Of our sample of 6 sources, one is a HyLIG and five are non-hyperluminous ULIGs. The five non-hyperluminous ULIGs are all detected on the COSMOS plates and have and . Their mean redshift is higher than the highest known redshift of any non-hyperluminous ULIG prior to this study, indicating that our procedure is also a powerful method for selecting distant ULIGs. The HyLIG in our sample is the only object not detected on the COSMOS plates and this object has and . This result confirms that HyLIGs can be found by selecting objects with extreme values of R. The main difficulty in applying this method is the large size of the IRAS position error ellipses, which precludes a direct optical identification at the faint magnitude levels expected for distant HyLIGs. However, future surveys, such as the ongoing European Large Area ISO Survey (ELAIS; Oliver 1996), and surveys with SIRTF and FIRST, and with SCUBA on the James Clerk Maxwell Telescope (JCMT) will provide substantially better positional accuracy and not suffer from this identification ambiguity. The method used here for selecting the most luminous and distant objects can be adapted directly to those surveys.
The small size of our sample, which contains only one HyLIG, precludes any detailed statistical inferences, which must await more extensive programmes using this selection and identification method, based on IRAS data or on the surveys mentioned previously. However, a number of trends in our data merit further discussion. In the first place, the detection of  emission in the only HyLIG in our sample shows that this object contains an AGN. Thus all three IRAS-selected HyLIGs discovered so far (, (Cutri et al 1994; Hines et al. 1995) and (this work)) contain AGNs. While the statistics for HyLIGs is still based on small numbers, the result is significant, since the  line was not detected in any of the non-hyperluminous ULIGs in our sample, while our spectra did cover the wavelength where this line would be expected. Thus the HyLIGs form a remarkable contrast with the non-hyperluminous ULIGs, where the presence or absence of AGNs is a strongly debated issue, and direct evidence for the presence for an AGN is very scarce.
Our procedure brings about incompleteness in our sample of objects in two ways: identification incompleteness and selection incompleteness. The former effect arises if objects with fail to be selected by our criterion, which occurs if a bright galaxy lies close to the line-of-sight to a distant FSC source, giving rise to erroneous identification with the bright galaxy. As noted in Sect. 2, the probability of misidentification in this situation is only about 2% for galaxies with . Since the large majority of our identifications have counterparts significantly brighter than (for 85% of the objects with , the counterpart has ), the probability of chance superpositions is much less than 2%, and the identification incompleteness can thus be neglected.
However, the sample of 313 objects used for our identification programme does suffer from selection incompleteness. Our selection method was aimed at rejecting spurious sources; however, as shown below, it must have removed a significant number of real sources from the sample as well. The relevant selection criteria are the requirement to have a high-quality 60 µm detection, no cirrus confusion, and a detection at 100 µm. While these criteria were effective at rejecting spurious detections, they also introduce a selection incompleteness, and may have rejected some distant objects. In order to assess the magnitude of this effect, we compare our sample to the FSS-z I sample described by Oliver et al. (1996). This sample has been constructed using low-cirrus regions with good IRAS 60 µm coverage and is estimated to be 99% complete for , which is the same flux limit as the sample described in the present paper. It contains 1931 IRAS FSC galaxies over an area of , giving a source density of 2.30 per deg2. Adopting this source density as characteristic for the present survey shows that a total of 2483 expected IRAS FSC galaxies over the entire survey area should be expected, a plausible number given that, including spurious sources, our initial extragalactic sample in this area contained 2719 objects (see Sect. 2). In contrast, only 313 objects were retained in our sample of candidate objects after the strict selection criteria described in Sect. 2 had been applied. However, since none of these criteria introduces a bias in luminosity or distance, our sample is unbiased and our survey thus constitutes a sparse (approximately 1 in 8) survey of infrared galaxies with over the 1079deg2 area. Hence we can use our results to estimate a number density for HyLIGs at of approximately 710-3deg-2, with considerable uncertainty due to the small numbers involved. We note that, adopting the local 60 µm luminosity function of Saunders et al. (1990), this estimate implies significant evolution in the infrared galaxy population to . Only in the unlikely case that the HyLIG detected in our sparse survey was the only HyLIG in the entire 1079deg2 survey area, no evolution would be needed.
We finally note that since we are using to select luminous objects, our selection method is robust against the presence of gravitational lensing, provided the corresponding magnification factors are similar at 60 µm and B. As a result, once a redshift and hence an infrared luminosity is available, R and may be combined to address the possibility of gravitational lensing. We illustrate the method using the lensed HyLIG and the HyLIG , identified in the present work. As noted in Sect. 2, has . Using the bivariate B-60 µm luminosity function of Saunders et al. (1990), we find a most likely intrinsic infrared luminosity of about . The apparent luminosity following from the redshift of 2.28 on the other hand, is . The large discrepancy between and suggests gravitational amplification by a factor of about 60. Using the same reasoning, for we find and . Because of the similarity of the two values, there is in this case no indication for gravitational lensing. Caution is required when applying this method, since the underlying assumption of similar magnification factors at optical and infrared wavelengths may easily be violated, as is the case in , where an optical magnification by approximately a factor of 100 is found (Eisenhardt et al. 1996), whereas the infrared magnification is only approximately a factor of 10 (Downes et al. 1995; Green & Rowan-Robinson 1996; Serjeant et al. 1998). Therefore the actual presence or absence of gravitational amplification must always be established by additional observations. However this method may be useful for selecting candidate gravitationally lensed sources for further study.
© European Southern Observatory (ESO) 1999
Online publication: February 23, 1999