3 Observations and data reduction

We acquired intermediate- to low-resolution optical spectra using the following telescopes: the 3.5-m telescope at Calar Alto (CAHA) Observatory in Almería (Spain), the 2.5-m Isaac Newton telescope (INT) at the Roque de los Muchachos Observatory (ORM) on the island of La Palma (Spain), the 10-m Keck II telescope at Mauna Kea Observatory on the island of Hawaii (U.S.A.), and the Otto Struve 2.1-m telescope at McDonald Observatory in west Texas (U.S.A.). Observing campaigns, spectrographs attached to the Cassegrain focus of each telescope, detectors, gratings and slit widths used for collecting data are summarized in Table 2. The red arm of the TWIN instrument (CAHA) and the 235mm camera of the IDS spectrograph (ORM) were chosen. Table 1 shows the journal of the observations, which includes the nominal dispersions of the instrumental setups. No binning of the pixels along the spectral direction and the projection of the slits onto the detectors yielded spectral resolutions of 1.54Å (R$\sim$4400, CAHA, first run), 4.82Å (R$\sim$1600, CAHA, second run), 1.68Å (R$\sim$3800, ORM), 2.89Å (R$\sim$2500, Keck), 1.4Å (R$\sim$4600, McDonald), and spatial resolutions as listed in Table 2. A binning of 8 pixels along the spatial direction was applied to the CCD at McDonald. Filters blocking the light blueward of 5000 Å were used at the CAHA and ORM telescopes. No order-blocking filter was used at the Keck II telescope; nevertheless, the two targets observed are very red and the contribution of their blue light to the far-red optical spectrum is negligible (Martín et al. 1999). Weather conditions during the four runs (CAHA, ORM, Keck and McDonald) were spectroscopic. The seeing in the visible was stable at around 1 $^{\prime\prime}$ at Keck, and 1.5 $^{\prime\prime}$-2 $^{\prime\prime}$ at CAHA and ORM. The spatial resolution at McDonald was 2.72 $^{\prime\prime}$/pix due to some technical problems related to the instrumentation.

Raw images were reduced with standard procedures including bias subtraction and flat-fielding within NOAO IRAF. We extracted object and sky spectra using the optimal extraction algorithm available in the APEXTRACT package. A full wavelength solution from calibration lamps taken immediately after each target was applied to the spectra. The rms of the fourth-order polynomial fit to the wavelength calibration is typically 5-10% the nominal dispersion. To complete the data reduction, we corrected the extracted spectra for instrumental response using data of spectrophotometric standard stars (HD19445, Feige34, G191B2B, BD+262606) obtained on the same nights and with the same instrumental configurations. These stars have fluxes available in the IRAF environment (Massey et al. 1988).

The resulting spectra are depicted in Figs. 2-5. They are ordered by increasingly late spectral type and shifted by a constant for clarity. The region around the Li I $\lambda$6708Å line is amplified in Figs. 6-9. In Fig. 7 we have included the spectrum of the field M6-type spectroscopic standard star Gl406 for a better comparison.

Figure 6: Region around the Li I $\lambda$6708Å line (CAHA high-resolution spectra). The star 4771-1097 is a fast rotator.

Figure 7: Region around the Li I $\lambda$6708Å line (CAHA and Keck spectra). The spectrum of the M6-type field star Gl406 is included for comparison. Data have been shifted by different constants for clarity.

Figure 8: Region around the Li I $\lambda$6708Å line (ORM spectra). Data have been shifted by 0.6 for clarity.

Figure 9: Region around the Li I $\lambda$6708Å line (McDonald spectra).