4 Observations and data reduction

The target stars were observed during two nights, in December 1999, with the ESO New Technology Telescope (NTT) and EMMI in the red mid-resolution spectrographic mode. The selected grating (#4) has a nominal resolution of 5500 (for the 1''-wide slit we used), however the seeing conditions of one night were so good ( 0.5''-0.6'') that for a part of the spectra the actual resolution is actually higher than the nominal value. A total of 23 long slit spectra were obtained for 16 out of the 17 stars in the initial sample, all of them are 640 Å long and are centered on the LiI 6707.8 Å line; the details are given in the last column of Table 1. Some of the stars of NGC 1866 and NGC 2031 were observed several times. The number of spectra available for each target is listed in Col. 7 of Table 1 together with the mean S/N ratio of the individual spectra. Note that star #4 of NGC 2031, though present in the initial target list, was not observed because of its rather low priority on the basis of its blue color and uncertain classification, however to avoid confusion with the data file names we kept the initial numbering of the sample. The data reduction was performed by means of standard techniques (the long slit reduction routines of the IRAF package). The steps were bias subtraction, flat fielding, (partial) cosmic ray removal, normalization, wavelength calibration. The Doppler shifts were derived from H and a few other reference lines (mainly FeI), and the corresponding radial velocities (or a mean value when more spectra were available) are given in the last column of Table 1. This simple derivation provides relatively large errors, but the results are at least consistent with those from more accurate estimations: for instance, the Doppler shifts derived for the whole sample of nine NGC 1866 AGB stars, yield a mean velocity  km s^-1, in good agreement with previous dynamic studies: according to Fischer et al. (1992) the systemic velocity of NGC 1866 is  km s^-1. The small deviations of the individual stars from the mean value confirm as well their cluster membership.

Figures 3 and 4 show the reduced spectra normalized to the maximum flux in the wavelength range displayed.

Figure 3: The spectra in the LiI 6707.8 Å region of the nine AGB stars observed in NGC 1866. The stars are ordered from top to bottom by decreasing near-IR luminosity.

Figure 4: The spectra in the LiI 6707.8 Å region of the seven AGB stars observed in NGC 2031, together with those of the AGB stars observed in the comparison clusters NGC 2214 and NGC 2107.

A LiI line at 6708 Å of decreasing strength is clearly visible in stars #2, #3, #4 of NGC 1866, but is absent in the K-band brightest star #1. The same seems to happen in NGC 2031, where again the spectrum of star #2 shows a strong LiI line, while that of star#1 one does not. Figure 5 shows a comparison between the observed spectrum of NGC 1866 and two synthetic spectra with reasonable $T_{\rm eff}$ and lithium content, where it is evident that the lithium abundance of this star must be large. A detailed analysis of the lithium abundances of each star is presented in the next section.

Figure 5: Comparison among the spectrum of NGC 1866#2 and two synthetic spectra computed for $T_{\rm eff}$ = 3250, 3350 K, $\log N(\element[][]{Li})=1.5$, $\log g=0.0$ and [ Fe/ H] =-0.4.

In addition, three different images of the NGC 1866 field obtained with ISOCAM, the mid-IR camera onboard the Infrared Space Observatory (ISO), were retrieved from the ISO Data Archive (TDT numbers: 59 000 438, 59 000 439 and 59 000 440) in order to study the mid-IR emission of the AGB sources under analysis in this paper and search for other possible heavily obscured AGB stars in the same field not listed in Table 1. The images were taken through the broadband ISOCAM LW1, LW2 and LW10 filters, whose passbands are centered at 4.5, 6.7 and 12.0 $\mu$m respectively (Blommaert et al. 2002). They were originally part of the proposal TTANABE.REDSTAR. They were all taken on 28 June 1997 and cover a region of $192''\times 192''$ with a pixel scale of 3.0'' pix$^{\rm -1}$.

Data reduction was performed using the "CAM Interactive Analysis'' (CIA, version 4.0) starting from the raw data products which were corrected for dark current, glitches, transients and flat-field, following the standard routines available within CIA. An accuracy better than 20% is expected even at these very low flux levels in the absolute flux assigned to the end products.