2 Observations and data reduction

2.1 Observational program

The aim of the ESO key-program required large wavelength range to allow to detect possible peculiarities, a 20000 minimum resolution for vsini, $T_{\rm eff}$, $\log\,g$ and metallicity determinations and an accessible instrumentation several times a year. These conditions were met by the Echelec Spectrograph equipped with a CCD camera at the 1.52m ESO telescope at La Silla. Priority was given to radial velocities determinations. The initial program of B8-F2 stars nearer than 100pc changed with time, the lack of transparency of the instrumentation not allowing the observation of stars fainter than V magnitude 7.5. So the number of F-type stars was reduced and priority was given to early A-type stars. Reference stars were observed for the different researched parameters. Some supergiants were added to obtain their vsini. We observed, only once, stars for the obtention of vsini and twice for radial velocity. Meanwhile it appeared interesting to determine radial velocity for the whole sample, even when there was one observation only. Keeping in mind that Hipparcos Catalogue is complete until V magnitude 7.5, it contains 3545 south normal stars from B8 to F2 and V < 7.5, 58% of which being A-type, and 39% having a radial velocity now. As a result, 610 stars were observed during 21 observation runs, from June 1989 to January 1995, 581 being in the range B8 to F2 type, 41% of which having two or more measurements.

2.2 Instrumentation

The ESO Echelec spectrograph was used at the coudé focus of the 1.52 m telescope at La Silla (Chile). The 9 central orders around $\lambda_{\rm c}$=4350Å were used. They covered the spectral range 4210 to 4500Å. The dispersion was 3.1Åmm^-1. The detector was a RCA CCD with 640$\times$1024 pixels of 15 $\times$ 15${\mu}{\rm m}^{2}$, the pixel size corresponding to 0$.\!\!^{\prime\prime}$65 on the sky. The nominal spectrograph resolution was about 70000. The characteristics are described in detail in Gilliotte & Lindgren (1989). Using a slit width of 320$\mu$m (1$.\!\!^{\prime\prime}$5 on the sky), the instrumental resolving power was degraded to 28000 and the ratio S/N, highly variable from the center to the edges of each order covered the range 50 to 200. The reduction from a CCD image to a complete linear spectrum (calibration frames, orders extraction, wavelength calibration, connection of the orders) is described with details by Burnage & Gerbaldi (1990, 1992).

2.3 Correlation method

A cross-correlation method with synthetic spectra was chosen to determine radial velocities. This method, described by Tonry & Davis (1979), to determine galaxy redshifts was since used by several authors for late-type stars, but not often for early-type stars. Main difficulties in A-type stars come from the small number of lines and the fact that the H$\gamma$ line covers most of the spectral range of the spectra at our disposal. It would contribute far too much in the computation of a correlation index with respect to the faint metallic lines more or less washed out by the high projected rotational velocity of most stars. In order to use metallic lines only, a pseudonormalised spectrum has been computed whose continuum follows the profile of H$\gamma$. So the correlation was independent of this line. A programme of automatic normalisation has been elaborated and applied.

Synthetic spectra have several advantages compared to actual stellar spectra for the use as templates: they are perfectly adapted to the linear response of CCD detector; they are noise-free; their radial velocity is zero and they allow to compute an homogeneous and regular grid of reference spectra. They do not incorporate the instrumental profile, whose influence is negligible on the spectra of these mean-high rotating stars. They were computed with Kurucz (1993) models. The range of $T_{\rm eff}$ was 6000 to 15000K, logg=4.0 and metallicity solar (justified by the sample of normal stars) and lines were widened by rotation with a path of 25kms^-1. The resolution of synthetic spectra was adapted to the observed spectra (same step and start). As to avoid the differential dispersion due to the radial velocity, all spectra were rebinned in Naperian logarithms.

All reduction programs were built into the MIDAS environment. Systematically the correlation program identified the synthetic spectrum giving the best correlation index (see the description of the grid of the synthetic spectrum in the Sect. 3.1). This method allowed to test the spectral types and rotation velocities found in the literature. Discrepancies came from binaries generally. The range of correlation index varied from 0.5 to 0.95 in most cases. The maximum of the cross-correlation function which was usually quasi Gaussian was estimated by fitting its peak with a parabola.

2.4 Zero point of radial velocities

The zero-point was determined from IAU standard which were observed every night, when it was possible. To have an homogeneous determination, we adopted the radial velocities obtained with Coravel and given by Latham & Stefanik (1991). The zero-point was determined for each run. The result on IAU standards is summarized in Table1, where $Vr_{\rm cor}$ is the Coravel value, $Vr_{\rm ech}$ the mean observed value obtained with spectra with its rms (root mean square) and n the number of spectra.

 

Table 1: IAU standards used for zero-point

The independence of obtained radial velocities with the spectral type was tested using the homogeneous radial velocities of B to F type stars from Andersen & Nordström (1983) and Nordström & Andersen (1985) (see the comparison in the Sect. 3.3).

2.5 Correlation peaks analysis

Generally the correlation peak is symmetric and roughly Gaussian. Nevertheless some of them were double, asymmetric or not Gaussian and their interpretation was not obvious. To detect an eventual asymmetry, each peak was fitted with a Gaussian profile and a new radial velocity was calculated (the parabola only fits the top of the peak whereas the Gaussian fits the whole profile). The difference $dv_{\rm gp}$ of this new radial velocity with the value obtained with a parabola was computed. Several cases, function of the shape of peaks, are considered. Separated peaks of binaries are listed in Table7 with a and b components. Some radial velocities were considered as not significant: when the correlation coefficient was lower than 0.3 or when the peak was too asymmetric (detected by a Chi-Square test upon $dv_{\rm gp}$). The Hipparcos (HIP) and HD numbers of these stars are listed in Table5. Symmetric but wide peaks could mask a binary system. To analyse the shape of some peaks, simulations of binary and triple systems were made with synthetic spectra.

In the simulations, the input parameters were $T_{\rm eff}$ and rotation broadening of components, their radial velocities difference and the ratio of their intrinsic luminosities. It was not possible to do this simulation for each peak but this has allowed to recognize characterized shapes of binary peaks. With the aim of detecting a possible double or multiple system, hidden by rotation, all spectra were correlated with the minimum rotation available in our grid of synthetic spectra. A number ranging from 0 to 10 refers to the shape of the corresponding correlation curve in Tables8 and 9. The meaning of this flag is summarized in Table4. The agreement between $T_{\rm eff}$ and rotation broadening of synthetic spectra found for the different observed spectra of a same star is considered: in 79% stars these parameters are the same. For the others, 13% show a duplicity criterion and in the remaining 8% the real values are between the two found values, probably.