We selected a sample of 10 red giant branch (RGB) stars in 3 GCs,
representative of the typical range of metal abundance of these objects, i.e.
47 Tuc (high metallicity, [Fe/H]
dex, 3 stars), NGC 6752
(intermediate metallicity, [Fe/H]
, 4 stars) and NGC 6397 (low
metallicity, [Fe/H]
, 3 stars).
Due to obvious flux limitations, we restricted our observations to the
brightest globular cluster stars to obtain observational material good enough
(i.e. adequately high resolution and S/N ratios) for fine abundance analysis
with reasonable exposure times. All observed stars were
brighter than V=12.5; the faintest giant observed, star 5422 in 47 Tuc, has 
(adopting a true distance modulus of 13.31 and 
from
Djorgovski 1993). Moreover, stars were selected to have infrared photometry
available, in particular in the K band, for an accurate determination of the
effective temperature 
.
The observational material was acquired in two runs: October 1990 and June
1991. In both runs, echelle spectra of the program stars were obtained with the
CASPEC spectrograph (in the Long Camera configuration) at the
3.6 m ESO
telescope at La Silla, using a 31.6 lines/mm echelle grating. The slit
width was adjusted in order to give resolution 
.
We tried to obtain a spectral coverage as large as possible, to observe spectral features of different atomic and molecular species and to compare abundances of the same element as derived from different indicators. In Table 1 (click here) we list the main features (abundance indicators) of the spectral regions observed. Table 2 (click here) lists the literature photometric measurements of the program stars, and Table 3 the observed intervals for each star.
Exposure times ranged from 15 to 70 minutes to reduce cosmic ray contamination; we usually tried to obtain more than one spectrum for each object to eliminate the spurious events. A quartz lamp (for flat fielding) and a Thorium-Argon lamp (for wavelength calibration) were acquired after each program star exposure, with the telescope at the same position of the program star exposures. Besides, fast-rotating early-type stars were observed each night to remove telluric lines (see below). Bias frames have been taken at the beginning of each night to account for readout noise.
Table 3: Spectral intervals observed for the program stars
The first stEps in CCD reduction (bias subtraction, echelle order
identification, scattered light subtraction, order extraction and wavelength
calibration) were performed using standard packages implemented in
IRAF
environment. Off-order scattered light was eliminated
through
bi-dimensional fitting along the dispersion and in the orthogonal
direction. The spectra were then wavelength calibrated using a dispersion
solution in two dimensions, derived from the Th-Ar lamps taken after each
spectrum, and one-dimensional spectra were extracted using an optimal
extraction algorithm implemented in the package.
The next stEps of the analysis were then performed using the ISA package (Gratton 1988), purposely developed to deal with high-resolution spectra.
The blaze function was taken into account by dividing the spectra by dome flat
fields; the continuum was then traced on each individual spectrum for every object.
Whenever we had multiple exposures of the same star, spurious events and spikes
due to cosmic rays were eliminated comparing different spectra.
We then used the spectra of featureless, rapidly rotating, early-type stars for
accurate removal of the telluric O
features, affecting in particular the
6300 Å region. We first identified atmospheric features in the solar
spectrum then we measured each line in the spectra of early-type stars
acquired at different airmasses z. From these measurements we derived a mean
relationship between the EWs and z. For each star a synthetic spectrum of
the atmospheric lines was then computed and convolved with the instrumental
profile; O
features were finally cancelled out by dividing the spectrum
of each program star by the appropriate synthetic spectrum. This procedure
allows
to correct each star for the appropriate airmass; moreover it does not
introduce additional noise in the object spectra.
The resulting cleaned spectra were then added to improve the S/N, after correcting for the change in the (geocentric) radial velocity of the star. Finally, a new continuum was traced. Figure 1 (click here) shows the tracing of a portion of our co-added and normalized CASPEC spectra for two program stars.
Figure 1: Tracing of a portion of the co-added, normalized spectra of 2
stars in the 6300 Å region
Final available spectra are listed in Table 3 (click here). We note that the following abundance analysis uses only EWs measured in the spectral regions centered at 5100 and 6300 Å. The main reason is that they are less affected by telluric bands and richer of stellar features (as compared to the 8000 Å region), and there is less concern related to line crowding and continuum level identification (as compared to the 4200 Å region); this last feature is of particular importance for the the spectra of stars in the metal-rich cluster 47 Tuc.
Table 4: Equivalent widths for giants in 47 Tuc, NGC 6397 and NGC 6752
Equivalent widhts of various element were then measured in the two mentioned
regions: in the following
we will refer only to iron lines. Gaussian profiles were fitted
to the observed profiles; when the number of clean lines was
very low (e.g., for
Fe II lines), we first derived a relationship between EW and central depth
from unblended lines and then we used it to add some new
EW's for measured 
's. The number of measured lines depends on
S/N and on the star
metallicity; generally, some 25-50 Fe lines were measured for NGC 6397 stars,
some 50-70 for NGC 6752 stars and from 50 to 100 for 47 Tuc stars.
In the following analysis, only lines with EW>10 mÅ were
used. Line parameters (see Sect. 5) and EWs are listed in
Table 4 (click here), both for Fe I and Fe II lines.
We have 3 stars (namely, 47 Tuc-5529, NGC 6397-C211 and NGC 6752-A45) in common
with another recent, high dispersion analysis by Norris and Da Costa (1995;
hereinafter, NDC), at about the same resolution we used: this allows a
comparison between the two sets of EWs, which is shown in Fig. 2 (click here).
The average difference is: 
mÅ (
=9.2 mÅ,
76 lines).
Note that here we regard this comparison as an external check of the accuracy
of our EW's measurements, but since we used data from NDC to enlarge
the sample of analyzed stars (see below), the above comparison has to be
regarded also as a self-consistency test on our total set of EWs.
Figure 2: Comparison of our EWs with those of NDC for the 3 stars in common