2. Optical data
The data were collected in 4 nights at the 3.6 m CFHT Telescope in May 1993. Two 10 minutes exposures in B band and two 15 minutes exposures in R band were obtained. Exposures of 30 to 55 minutes per spectroscopic mask was obtained for 3 separate masks, each containing about 30 slits (Fig. 3). The focal reducer MOS/SIS together with CCD Lick2 ( pixels of 15 µm) were used during the run. This CCD is a thick device having a quantum efficiency of in the blue. The observing configuration provides a pixel size of over a field of view of about . The overall image quality was good (stellar FWHM ) although some optical distortions were conspicuous near the edges of the images due to the optics of the focal reducer.
2.2. Photometric analysis
The B and R frames were prepared using standard pre-reduction techniques. Since there were only 2 frames per filter, cosmic rays were removed by taking the lower pixel value in cases where a pixel in one frame is significantly higher than the corresponding pixel in the other frame. The photometric analysis was performed by means of the SExtractor package (Bertin & Arnouts 1996) in the same way as Pierre et al. (1997), but adapted to our data. The images were first slightly smoothed to give the same PSF in B and R frames, then the background was estimated using a 64 64 pixel mesh. Source detections were claimed if at least 9 adjacent pixels were above a threshold corresponding to 1.5 times the local noise level. The CCD Sequence in M 92 (Christian et al. 1985) observed during the same run was used for photometric calibration. Stars VCS1, A, B (probably variable) had to be removed because of obvious inconsistencies. Estimates of the photometric errors were taken directly from the SExtractor analysis, and are less than for R and less than for B.
The catalogue is estimated to be complete to R = 22.5 and B = 23.5. On inspection of the detected objects above the completeness limit, we found those objects with a SExtractor classification may be assumed to be galaxies, i.e., 275 objects. Changing the threshold does not affect the outcome significantly because most of the galaxies are well separated from stars (3/4 of the objects fall below 0.05 or above 0.95).
Fig. 4 shows the colour magnitude diagram for all the galaxies detected in the R-frame, and the corresponding magnitudes in the R and B bands were measured within the same apertures. The band of E/S0 sequence galaxies is discernable in Fig. 4 ; the spectroscopically confirmed cluster members are shown to fall mostly on the E/S0 sequence confirming that a large fraction of the galaxies on the E/S0 sequence belongs to the cluster. The mean error in B-R colour is .
Grism O300 was used for the spectroscopy. It has a zero deviation at 5900 Å, covers approximately 4700-7900 Å, and gives a dispersion of 3.59 Åpixel (). The slit has a width of 2", i.e. 6.4 pixels, yielding a resolution of 23 Å FWHM. Since there was only 1 frame per mask, cosmic rays were picked out individually by eye and replaced by the median of the surrounding pixels. The internal Helium and Argon lamps were used for wavelength calibration. The subsequent reduction was performed as described in Pierre et al. (1997). Redshifts were measured by a cross-correlation method implemented in the MIDAS environment following Tonry & Davis (1979). The cross-correlation results for each spectrum were checked independently by eye.
The results from the cross-correlation analysis for all spectra are presented in Table 2. Heliocentric correction has not been applied, but is negligible at this resolution. The absolute error in the velocity calibration is km s-1.
Table 1. Results on spectral fit to ASCA data.
Table 2. Spectral analysis of Abell 2104
As a first guess, galaxies are considered to be cluster members if they lie within 3000 km s-1 of the central cD galaxy, which selects 47 (the main sample) out of the 60 galaxies. This procedure eliminates most of the foreground and background galaxies without affecting the dispersion measurements significantly. If we relax the velocity constraint and apply the usual clipping technique then we have 51 cluster members (the extended sample). The cluster redshift distribution for both samples is displayed in Fig. 5. The histogram includes all galaxies in the redshift range in Table 2 and a Gaussian corresponding to the velocity distribution of the main sample. The bi-weighted mean and scale for the main sample are and km s-1 correspondingly; and and km s-1 for the extended sample. It is difficult to find an objective criterion for deciding which galaxies are cluster members. Even with the sophisticated weighting scheme employed by Carlberg et al. (1997), the determination of the weight for each galaxy is still subjective. In Table 2, we have marked only the galaxies from the main sample as cluster members.
For the main sample we have enough redshifts to test whether or not the galaxy velocities are drawn from a Gaussian distribution applying various statistical tests for normality (e.g. D'Agostino & Stephens 1986; ROSTAT - Beers et al. 1990; Bird & Beers 1993). As a result Anderson-Darling test (A2) accepts the hypothesis for normal distribution at 90% significance, the combined skewness and kurtosis test (B1 & B2 omnibus test) at 97% level and the alternative shape estimators, asymmetry index and tail index based on order statistics, were found to be -0.21 and 0.98 respectively, which also shows that the velocity distribution is drawn from a Gaussian.
We can obtain a conservative estimate of the errors on the velocity dispersion by comparing the dispersion from the extended and main samples. When we take account of the uncertainties in cluster membership, a more conservative estimate of the errors should give the velocity dispersion as km s-1.
We also investigated the presence of substructures in (, , z) space but no obvious signal was detected (see Fig. 6). More redshifts are required for a proper statistical analysis.
© European Southern Observatory (ESO) 2000
Online publication: December 11, 2000