Since the first version of the library, the reduction procedure described in 2001A&A...369.1048P has been improved in several aspects but the general philosophy remains the same. The basic steps in the reduction are to

(1) correct the different orders in the spectra for the blaze effect and connect the orders together

(2) mask the telluric lines and spikes due to cosmic rays : in the new version, the mask of the spikes due to cosmic rays has been suppressed because it was found to alter real features. Because of it, in the previous version of the library Lick indices measured on our spectra presented a slope compared to measurements on Jones spectra, the strongest features were found weaker than real (thanks to G. Worthey who stressed our attention on this important problem).

(3) make the flux calibration

(4) build an interpolator which in turn is used to generate the grid that feeds the population synthesis program (PEGASE.HR).

The strengthening of the spectra have been improved: artifacts near Balmer's lines were corrected. The wavelength interval has been extended by adding 5 orders in the blue. It was 410-680 nm, it is now 400-680. The correction for telluric lines has been improved. We corrected some small bugs in the FITS keywords (in particular an unfortunate bad rounding of CRVAL1).

Flux calibration:
The main improvement in the flux calibration is due to the increase of the number of external and internal comparisons. We made 2 observing runs to have more intercomparisons. We have also improved the color equations. Finally, we have improved the determination of the Galactic extinction. It is determined from either Klare and Neckel, Hakkila et al. or Chen et al. after a consistency test between these various 3D model. Correction of the galactic extinction is primordial for population
synthesis and quite critical because there exists a correlation in the library between the spectral type and the extinction (both hot
stars and red giants are often in the disk and have high extinction).

Interpolator:
The main goal of project was to make a stellar library for synthesis of stellar population. This require to provide a spectrum at any desired point of the parameter space, and actually the spectral systhesis tools we are using, PEGASE.HR, uses a grid of spectra with a defined mesh.

The interpolator produces a spectrum at given Teff, log g and [Fe/H] by evaluating polynomials in a way quite similar to the Lick fitting functions (Worthey 1994). The polynomials include with up to 21 terms and up to the power 4 in each of the parameters with a selection of cross terms. Each wavelength element is represented with an independent polynomial but the functional form is the same for the whole spectrum. As well as for the Lick fitting functions the parameter space is divided in a few overlapping regions where different polynomials are fitted. The overlap between these regions is used to give a smooth transition between the solutions obtained in each region. In the overlap the fluxes are a linear interpolation between the polynomials of each region. The terms choosen, and the validity range of the polynomials were chosen empirically by trial and errors.

The quality of the interpolator can be checked by comparing the interpolated fluxes with the observed fluxes for the spectra in the library. So, at each iteration we searched these residuals for discrepancies and systematic effects and we modified the form of the polynomials and the validity regions in order to improve the model. The solution adopted in the present version reflect the intrinsic variation of the lines and the actual distribution of the stars in the space of parameters.

Because the library contains a large number of F-K stars, we have also weighted the spectra in order to re-equilibrate the distributions. The polynomials were fitted by least-square with a kappa-sigma rejection of outliers (un-corrected bad pixels in some spectra). In the new version we have much more hot stars (>10000K) and cool stars (<4500K) than in previous version and the interpolator is therefore considerably improved with respect to the previous version.

Absolute and internal interpolators: The interpolator was first computed using the stars with known atmospheric parameters, giving each stars a weight function of the quality of these determination. This version is called the "absolute interpolator". Inverting the interpolator for a spectrum corresponding to a star with unknown parameters allows to determine these parameters. For the stars used to build the interpolator, the comparison between the parameters resulting for inversion and those from our catalogue allow to detect errors on the input parameters of some stars (or errors on the identification of some stars at the telescope). In some cases these errors could be corrected and the interpolator were re-computed. This comparison also allowed to assess the weaknesses of the interpolator and drived its improvement (see above) and finally the statistics on this comparison reflect the errors on the parameters in our catalogue plus the errors due to the processing and modelling.

We then determined the parameters for all the stars using the absolute interpolator and used them to compute a new interpolator: the internal interpolator. This internal interpolator may suffer from some biases introduced by the absolute interpolator, but it has smaller ramdom errors because the errors on the input atmospheric parameters are decreased. It is also more stable because it is computed on more spectra, which is particularly sensitive in the less populated regions of the space of parameters. We used the internal interpolator to compute the internal determination of the atmospheric parameters listed in table 1.

Physical flux interpolator, correction for Galactic extinction: The interpolator was yet computed using the fluxes normalized to the continuum, because it is better suited to the determination of atmospheric parameters. The reason is that our flux calibration is not absolutely precise and since the information on the atmospheric parameters is essentially contained in the lines, we wanted to avoid to bias the atmospheric parameters with errors in the flux calibration. But to make the population synthesis we need to generate spectra calibrated in fluxes as emitted by the stars (ie. above the earth atmosphere and corrected for interstellar extinction). The next step was therefore to correct galactic extinction. We used for that the Schild (1977) extinction law (the choice of the extinction law is not critical in this region of wavelengths) scaled with the E(B-V) excess listed in our catalogue. We build an interpolator for source flux spectra using the internal determination of the atmospheric parameters. Again it was possible to compare the results of the interpolation with all the input spectra. This allowed to detect spectra with discrepant SED, either due to flux calibration, to erroneous value of the Galactic extinction. The most discrepant case were closely examined, searching the literature for possible explanations or hints for errors on the parameters and checking the consistency of the estimates of the Galactic extinction and the various photometric measurements for the objects. Errors in the catalogues were detected and corrected and in other cases we substituted the catalogued value of the Galactic extinction with a value estimated from the photometry (and atmospheric parameters).

A final test on the quality of the source flux internal interpolator was to compare the colour of the stars in different regions of the HR diagram with the colors measured on the spectra (Table xxx). The result of the comparison is quite accpetable but we still see a systematic effect that we believe is due to the color equations used to convert between the colors measured on Elodie spectra, the Johnson colors and the Tycho colors. We would like to solve this discrepancy in the near future.

Consistency and external comparisons: Measurements of Lick indices and determination of the atmospheric parameters are described in the article and the results are presented in.

Description of FITS keywords specific to these datasets

TEFF 	[K]	Teff from literature

Q_TEFF 		Quality flag for Teff 
		Q_TEFF=1:Poor determination -> Q_Teff=4: Excellent (see Sect. 2) 
		Q_TEFF=0:means that we have estimated Teff from the B-V colour 
                        (Tycho2 catalogue),assuming the empirical colour-temperature relation
                         for a main sequence star and neglecting the interstellar extinction 
		Q_Teff=-1: Internal determination of the effective temperature

LOGG 	[log cm/s**2)] 	log g from literature

Q_LOGG 		Quality flag for LOGG 
		Q_LOGG=1 means that the determination of LOGG is taken from the literature. 
		Q_LOGG=0 means that LOGG was converted from the V absolute magnitude from
                         Hiparcos and Teff, using a bolometric correction valid for a main
                         sequence start and an empirical mass-to-light relation.
		Q_LOGG=-1 Internal determination of the surface gravity


FE_H 	[Sun] 	[Fe/H] from literature

Q_FE_H	Quality flag for [Fe/H] 
		Q_Fe/H=1: Poor determination -> Q_Fe/H=4: Excellent
		Q_Fe/H=-1: Internal determination of [Fe/H]


Mv 	[mag] 	Mv from Hipparcos Q_MV Quality flag for Mv: 4=excellent


S/N 		S/N-per-pixel of original spectrum at 555 nm


VR 	[km/s] 	heliocentric radial velocity


FWHM 	[km/s] 	FWHM line broadening


Q_VR 		comment on Vr determination


SPTYPE 		spectral type from INCA


VMAG 	[mag] 	V-magnitude


E_VMAG 	[mag] 	error on V-magnitude
	

B-V 	[mag] 	B-V Johnson


R_PHOT 		Source of photometry: TYCHO2, INCA