4. Stars and galaxies
4.1. Star/galaxy separation
SExtractor computes a stellarity index for each detected object (in the interval 0-1, with 1 for stars, and 0 for galaxies). The stellarity index is determined from a non-linear set of equations (Trained Neural Network) (Bertin & Arnouts, 1996). The good seeing of the images () allows a robust classification to and . According to Bertin & Arnouts, the algorithm success rate at these magnitudes is 95% using data with a similar sampling and a seeing . Fig. 7 shows the index of the objects detected both in V and I. For or , all objects with an index are classified as galaxies. This criterion classifies as galaxies only the objects showing a clear evidence of extendedness. For or objects with an index are classified as galaxies. Because at these faint magnitudes most objects have an index (the great majority of objects with a stellarity index higher than 0.95 are spurious detections), the threshold of 0.95 does not remove the remaining stars from the sample. To correct for the fact that most of the stars with have been misclassified as galaxies, we need to apply a correction for the star dilution (see subsection on data-induced errors in Sect. 5). We evaluated the stellar contamination with the Galaxy star-count model of Bahcall (1986), which is compared on Fig. 8 to the number counts of galaxies. The galaxies outnumber the stars by nearly an order of magnitude where the classification algorithm efficiency is less than 95% (vertical dotted line in the lower part of the diagrams).
4.2. Galaxy counts
The galaxy counts shown in Fig. 8 are in good agreement with other measurements. In the I band, we systematically measure more galaxies than Postman et al. (1998) up to . At , we count 5626 galaxies per deg2, and Postman et al. find 4057. Given the errors in the I zero-point calibration (), the possible difference in the magnitude scale, and the intrinsic cosmic variance, we do not consider this difference to be significant.
We model the galaxy counts of Fig. 8 following the method described by Cole et al. (1992). They give the equations of the volume element, the comoving distance and the luminosity distance of the objects for three cosmologies (, Einstein de Sitter; Open; , Flat, a factor is missing in their equation of the volume element for the flat universe). Yoshii (1993) also details a similar method. The luminosity function (LF) is chosen to be similar to the CFRS for which LFs have been measured separately for blue and red galaxies (Lilly et al., 1995). Here we approximate the LF by its red component. The Schechter parameterization is used (Schechter, 1976),
where M is the absolute magnitude in the V or I band, and , and are the Schechter parameters (Here, , , ). K corrections are determined using 13-Gy-old elliptical galaxy template spectra from the PEGASE atlas (Fioc & Rocca-Volmerange, 1997), between redshifts . As already noted by many authors, this simple model does not provide a satisfactory fit to the number counts at faint and bright magnitudes simultaneously for any cosmologies in the V-band. A better fit would include a more realistic luminosity function accounting for both the red and blue galaxy populations, and for either a density or a luminosity evolution (see Sect. 6.1). As our purpose here is not to model the number counts, we limit ourselves to this partial model.
Fig. 8 shows that our UH8K number counts deviate from the predicted counts in non-evolving Einstein-de Sitter and open universes at and at . Postman et al. (1998) observed a clear departure from these two cosmological models in their I counts using the no-evolution model of Ferguson & Babul (1998) (FB). All other surveys displayed in both panels of Fig. 8 (Arnouts et al., 1999; Cowie et al., 1988; Driver et al., 1994; Gardner et al., 1996; Postman et al., 1998; Woods & Fahlman, 1997; Mamon, 1998; McCracken et al., 2000a) have the same behaviour, except for the V number counts of Cowie et al. (1988), which agree well with the Einstein-de Sitter model at and deviate only at fainter magnitudes.
In contrast, Fig. 8 shows that the non-evolving flat universe model provides a marginal agreement with our UH8K galaxy number counts in the I band at our magnitude limit of . In fact, most surveys displayed in the right panel of Fig. 8 agree with this cosmological model at , as expected from the results of the CFRS (Lilly et al., 1996), where giant red galaxies evolve little in the redshift interval . In Fig. 8, the deviations from the non-evolving flat universe occur in dataset which probe the faintest magnitudes, near . Our sample is not deep enough to show a departure from this cosmological model. Lidman et al. also find that their I number counts (not shown in Fig. 8) are compatible with a no-evolution flat universe (Lidman & Peterson, 1996) out to , whereas they deviate from the evolution model of Yoshii (1993) (this model uses a dwarf galaxy blue LF, which yields very different K corrections at faint magnitudes compared to a no evolution model). From the mentioned surveys, there are indications that evolution should be used in the modeling of the I counts at , where significant departures from a non-evolving distribution occur.
There is also marginal agreement of most V number counts displayed in the left panel of Fig. 8, including our UH8K data, with a non-evolving flat universe model. The deviations from the model occur in a wider range of V magnitude, namely at , depending on the data set, and the deviation is greater than for the I counts. This may be partly explained by the fact that the V number counts are more sensitive to evolution of the blue galaxy population than the I counts.
4.3. Galaxy colours
Fig. 9 shows histograms of galaxies and stars versus colours. For galaxies, histograms are given for three limiting magnitudes, , , and . The corresponding median colours are 1.45, 1.57, and 1.38. Fig. 10 shows colours vs I magnitudes for the sample of detected galaxies. A vertical line shows the I-band completion limit and an oblique line shows colour limits accessible due the V-band completion limit of . There is no visible trend towards a strong colour evolution of the galaxies with their magnitudes, but a robust conclusion is not possible because the I sample is depleted in red objects. A natural although arbitrary value to divide red galaxies from blue galaxies is the median of the histogram with in Fig. 9, located near . In the rest of the paper red galaxies will be those having and blue galaxies will be those having .
© European Southern Observatory (ESO) 2000
Online publication: January 29, 2001