To increase our sample and to obtain a more statistically significant basis for our discussion, we also analyzed in a homogeneous way the globular cluster giants studied in two recent surveys: G8689 (41 red giants in 17 globular clusters) and SKPL (82 giants in 7 clusters). To these sets, we also added a full re-analysis of 3 giants in NGC 2298 (from McWilliam et al. 1992; McW92), 18 stars studied by Minniti et al. (1993; M93) in 8 clusters, and 8 stars in 2 clusters from NDC. Table 5 (click here) lists the whole sample of globular clusters red giants analyzed in the present paper, with the source for each subset of data.
Table 5: Complete list of globular clusters analyzed
Figure 3: a) Comparison between EWs of G8689 sample after the correction and
the EWs of the present analysis (sample CG96) for NGC 6752-C9 and NGC
6397-C428; b) another test of self-consistency of our EWs, done by comparing
the corrected EWs from G8689 sample with those from SKPL sample for 3 stars
in common between the two sets
We simply adopted published EWs for all samples, since the observational
material consists in high-resolution (
), high S/N spectra,
fairly comparable to our own spectra. However, for the G8689 sample we had to
make some corrections to the published EWs, before using them for a new
analysis. In fact, in G87 the authors themselves noted that their EWs were
systematically higher than Mäckle et al. (1975) EWs for 
Boo
(Arcturus). This is due to the lower resolution (
) used by
G8689: unnoticed lines blend with the measured
ones, resulting in an overall overestimate of the EWs. We derived the
following relation between G8689 EWs (
) and Mäckle et al.
(1975) EWs (
) for 
Boo :
Strictly, this correction applies only to EWs measured on spectra having
metallicities similar to that of Arcturus ([Fe/H]
); we expect
smaller corrections for more metal-poor stars. To verify this point we
considered the stars in common between G8689 and the present data (NGC
6397-C428 and NGC 6752-C9); in both cases there are 17 lines in common between
the old and the new analyses. The mean difference between the EWs of our
new analysis 
and those of G8689 for these two stars are:
(
mÅ, 34 lines). Had we applied first Eq. (1 (click here)) to
correct the G8689's EWs, this difference would have been:
(
mÅ, 34 lines), where 
are G8689 EWs after
transformation using Eq. (1 (click here)). The overall correction is 
mÅ. We then estimate that while Eq. (1 (click here)) gives the appropriate
corrections to G8689 EWs for metallicities similar to that of Arcturus
([Fe/H]
), these corrections should reduce to 
of
this amount
for a metallicity in between those of NGC 6752 and NGC 6397 ([Fe/H]
). We then included a metallicity-dependent correction, obtaining:
Equation (4 (click here)) was then applied to the old 
s of G8689 to bring
them to the same system of our higher resolution spectra.
Figure 3 (click here)a
compares 
with 
for the two stars in common between
the G8689 and the CG96 sample; the average difference is 
mÅ (
= 10 mÅ, 34 lines). The best fit line (with
constant term put to zero) to the data in Figure 3 (click here)a is: 
, with a scatter of 11 mÅ. In the following
analisys, we used only lines with 
20 mÅ.
Figure 3 (click here)b shows another internal comparison, using
G8689 (corrected) and SKPL EWs for 3 stars: the agreement is very good, the
average difference being
mÅ (
=14 mÅ, 36
lines). We estimated internal errors in EWs of about 
9 mÅ in G8689
set and 
5 mÅ in CG96, SKPL and the other sets analyzed.
We can then be confident that our results are based on an homogeneous and
self-consistent dataset.