Final Results

In spite of the overall agreement in Figures 3 and 4 and the numbers above, there is in all panels a significant number of stars that deviate significantly from the identity relations. This is not negligible, because the spectral library is somewhat sparse in some areas of stellar parameter space, where a few stars with badly wrong stellar parameters may have an important weight on the resulting fitting functions.

We tried to improve the quality of our determinations by inspecting significantly deviant stars on a case-by-case basis. In $T_{\rm eff}$ space, dwarf (giant) stars cooler than 7000 K were deemed significantly deviant, and thus worthy of further scrutiny, when our determinations differed from those of JW95 by more than $\sim$ 200 K (300 K). In $\log g$ space, and in the same $T_{\rm eff}$ range, dwarf (giant) stars were considered significantly deviant when differences were higher than $\sim$ 0.5 ($\sim$ 1.0) dex. In [Fe/H] space, we decided to further investigate the cases of all stars cooler than 7000 K, and for which discrepancies were larger than $\sim$ 0.4 dex. For approximately 1/3 of the hotter stars we needed to double-check our determinations, because there the uncertainties are significantly higher, and therefore agreement with JW95 is poorer.

Determining the best values of [Fe/H] was quite laborious, because iron abundance estimates from any method are subject to larger uncertainties than the other parameters. For the same reason, the scatter in the values found in the literature is likewise higher. Following the above criteria, we found 59 stars (almost 10% of the spectral library) for which our [Fe/H] estimates were significantly different from those of JW95, according to the criteria defined above. For each star, we performed a critical, non-exhaustive, revision of the available literature, in order to select those values which we regarded as more robust. Determinations based on classical abundance analyses of high-resolution spectra were given precedence, and amongst the latter, higher weight was given to those involving recent, high S/N CCD observations, and updated model atmospheres.

Deciding for the best values of $T_{\rm eff}$ and $\log g$ was relatively simple, as these determinations tend by themselves to be fairly robust, and besides it is possible to compare our values with estimates made using independent methods. According to the above criteria, 27 of our $T_{\rm eff}$ determinations were found to significantly disagree with those of JW95. In order to decide for the best value, we compared the two sets of $T_{\rm eff}$ with those inferred from photometry from the literature. Consistency with the observed spectra, and in particular with the measured EWs of key absorption features was also required in order to help choosing that which seemed the most reliable $T_{\rm eff}$ determination. For stars hotter than $\sim$ 8000 K, the lack of robust calibrations of $T_{\rm eff}$ against photometric indices other than the ones employed in our own estimates made us resort to spectroscopic determinations from the literature, based on the analysis of intermediate-to-high resolution spectra on the basis of model atmospheres.

Deciding for the best choice of $\log g$ is very important, as this is the parameter that, for a given $T_{\rm eff}$, discriminates between the giant or dwarf nature of a given star, thus deciding for its allocation as input for different sets of fitting functions (see Section 3). Luckily, our $\log g$ estimates were found to disagree strongly with those of JW95 for only 16 stars. In order to decide for the best value, we looked in the literature for spectroscopic $\log g$ determinations.

In most cases, our stellar parameter determinations, being based on recent, more robust calibrations and high S/N spectra, were found to be in better agreement with those from the literature and/or other methods, than those of JW95. In some cases we gave preference to the latter values, and for very few stars we chose to adopt values from the literature.