The observations were performed in different observing runs carried out from 1990 to 1994 at two different observatories: the Observatorio del Roque de los Muchachos (La Palma, Spain) and the McDonald Observatory (Texas, U.S.A.).
Two telescopes were used in the Observatorio del Roque de los
Muchachos: the 1.0 m Jakobus Kapteyn Telescope (hereafter referred
to as JKT) and the 2.5 m Isaac Newton
Telescope (hereafter, INT). In the first case we used the Richardson-Brealey
spectrograph working with a 385 
578 pixels GEC CCD detector
whereas the Intermediate
Dispersion Spectrograph together with an 1280 
1180
pixels EEV5 CCD detector was used for the observations done with the INT. In both cases,
the wavelength range covers 
(
) nearly centered at 
. The
resolving power was 12000 for the JKT and 13000 for the
INT which correspond to a reciprocal spectral dispersion of 0.41 and 0.37
Å/pixel at 
respectively.
Those stars with 
observed in La
Palma were further observed at the McDonald Observatory with the 2.1 m
telescope using the Sandiford echelle spectrograph together with a 
Reticon CCD. The covered spectral range was
and the resolving power
110 000 which corresponds to a reciprocal spectral dispersion of
0.06 Å/pixel at 
.
La Palma spectra were reduced using the MIDAS 1-D image reduction package: Firstly, the cosmic rays were removed by replacing the number of counts in the affected pixel(s) by the mean of the neighboring pixels. The raw spectra were then bias-subtracted, flatfield-corrected, (two or more master flatfield exposures were taken at the beginning of every night) and background-subtracted. The 1-D spectra were then extracted using the optimal extraction algorithm as proposed by Horne (1986). Since we were not interested in deriving radial velocities the wavelength calibration was made using the stellar spectral lines themselves and fitting a first-order polynomial. Wavelengths were taken from Moore et al. (1966). The continuum windows were chosen with the aid of the Atlas of Procyon (Griffin & Griffin 1979) and the continuum level calculated by fitting a third-degree polynomial. The normalized spectra were derived by dividing the extracted spectra by this continuum level. A typical S/N ratio of 150-200 was reached both in the JKT and in the INT observations whereas S/N ratios ranging from 50 to 190 were achieved in the McDonald observations. The S/N ratio was determined by measuring the standard deviation of the intensity of the normalized spectrum in a spectral window free of absorption lines.
The reduction of the McDonald spectra was performed under the context echelle within the MIDAS reduction package. A typical session of echelle reduction comprises the following steps: removal of wrong columns and saturated pixels, bias subtraction, spatial positioning of the different spectral orders, flatfield correction, order extraction and wavelength calibration. The sky contribution was found not to be significant and no background subtraction was done. Although the ripple effect is an interference pattern inside the detector that depends on the angle of the rays entering the detector it was seen that the flatfield correction was enough to remove an important part of this effect permitting a linear fit to the spectral continuum. The wavelength calibration and continuum normalization were similar to those of La Palma spectra.