2 Samples, observations, calibrations and classification criteria

$\bullet$ Samples
We have studied the behavior of 57 MK standards and of 137 stars signaled as having composite spectra, on the basis of equivalent widths.

- Composite spectra

The sample of CS stars was selected, as for Paper II, from the list of Hynek (1938); some additional objects were taken from Markowitz (1969); Cowley (1973, 1976), Hoffleit & Jaschek (1982) and Stickland (1988).

We did observe 86 stars designed as or suspected to be CS; we have also reobserved 51 CS already taken with CARELEC, both to ensure a perfect fit and to supplement the material for some stars of Paper II which present problems.

- MK standards

We observed 57 standards stars; 40 of these had already been studied in Paper I and were reobserved to permit the calibration of AURELIE with respect to CARELEC. Seventeen new stars were included to study the lines MgI 8806 and FeI 8824. This complete sample is representative of stars of spectral types G, K and M and luminosity classes I, II, III and V.

The observational limits of our samples are V < 9.5 and $\delta \gt -25^\circ$.

$\bullet$ Observations
The observations were made at the 152 cm telescope with the AURELIE spectrograph (Gillet et al. 1994) at a dispersion of 33 Å/mm. Five observational runs were made in 1995 and 1996 and permitted to secure 194 objects (standards plus CS), usually with two spectra for each star.

At this dispersion one has access to 880 Å, instead of the 400 Å available at the CARELEC spectrograph. We have used a region of 500 Å extending between 8370 and 8870 Å, because for $\lambda < 8370$ Å the spectra are dominated by strong telluric lines. The receiver for the AURELIE spectrograph is a Thomson double bar (TH 7832) constituted by two lines of 2048 photodiodes of $750\times 13\,\mu$m, each line (even and uneven pixels) being read by two lateral CCD's.

The treatment of spectra is here definitely simplified respect to CARELEC, since we have to deal only with unidimensional images. As with CARELEC a tungsten lamp has been used for flat field, and a neon spectrum for calibrations in wavelength. The normalisation of the spectra and the measurement of the equivalent widths has been made in a similar way to those made with CARELEC (see Paper I) with the help of the IHAP software available at the OHP.

$\bullet$ Calibrations CARELEC/AURELIE
To calibrate the spectra obtained with Aurelie with regard to those obtained with Carelec, we have measured the equivalent widths of some blends and lines selected over the length of the spectrum for the 40 MK standards. We have chosen the blend 8468 Å, the CaII triplet (8498, 8542, 8662 Å), FeI 8688 Å and the total absorption between 8390 and 8775 Å (TA).

The measurements carried out with the two spectrographs are given in Figs. 1-4; on these graphs we have also plotted the equivalent width's of the CS measured.

  

Figure 1: Comparison of equivalent widths measured from CARELEC and AURELIE spectrographs for the 8468 Å blend, the Triplet of CaII, the FeI 8688 line and the Total Absorption between 8390 Å and 8775 Å. Filled squares: MK standards; open circles: composite spectra

  

Figure 2: Behaviour of the equivalent widths of the MgI 8806 and FeI 8824 lines versus R (correlated with the spectral type, see text) for MK standards. Open squares: supergiants; asterisks: bright giants; open circles: giants; dots: dwarfs

  

Figure 3: Behaviour of the MgI 8806/FeI 8824 and MgI 8806/FeI 8688 equivalent width ratios versus R (correlated with the spectral type, see text) for MK standards. Open squares: supergiants; asterisks: bright giants; open circles: giants; dots: dwarfs

  

Figure 4: Comparison of the Spectral Types (Sp T) and Luminosity Classes (Lumi.) for 51 composite spectra classified with both the CARELEC and AURELIE spectrographs. Sp T: G...1, K...2, M...3; Lumi.: Ia...0, Ib...1, II...2, III...3; numbers of the plots are weights

We find that for the shorter wavelengths (blend 8468 and CaII triplet) the equivalent widths obtained with Aurelie are, in the mean, weaker than those measured with Carelec, whereas for the longer wavelengths (FeI 8688) and for the total absorption TA there is no significant difference.

A careful comparison of the spectra show effectively that the lines observed with Aurelie are less deep than when observed with Carelec. This translates into a correction of the sum of the three equivalent widths of the CaII lines by an average correction of 0.5 Å, i.e. a 3 to 7% correction of the equivalent width. On the other side, the continuous background of the spectrum is sligthly inferior to the pseudocontinuum at the longer wavelengths (essentially for $\lambda \gt 8662$ Å) because we had to adjust it over a longer wavelength range than for the Carelec spectra and these two effects seem to compensate for TA and FeI 8688.

We still compare in Sect. 3 the classifications made with material from the two spectrographs to see if with all corrections made there remains still a systematic effect.

$\bullet$ Classification criteria
The classification of the 115 objects which present a cool stellar spectrum (G, K or M) in the near infrared has been carried out using the same criteria defined in Paper I and using the same technique as described in Paper II. But since the zone observable with Aurelie is larger than the zone observable with Carelec, we have been able to study with the help of the 57 MK standards the behavior of two supplementary intense lines, MgI 8806 and FeI 8824.

We have found in Paper I that a relation exists between the spectral type and the ratio of the central depths (R) of the neighbouring lines TiI 8683 and FeI 8679.

The mean relation was as follows:

We have therefore used this ratio R instead of the spectral type (assumed unknown for the CS) for the study of the behaviour of the lines MgI 8806 and FeI 8824: both lines present a positive luminosity effect (see Fig. 2). We have also considered the ratio of the equivalent widths of MgI 8806 and FeI 8688, as well as the ratio MgI 8806/FeI 8824. These two relations present a negative luminosity effect and permit a separation of giants and supergiants (Fig. 3). Dwarfs figure also on these graphs, in spite of their small number; they seem to behave like giants.