The absolute magnitudes of the G to M type MK standards from the Hipparcos parallaxes

We analyse a sample of about 500 MK standards of cool spectral types (G to M) for to compare the visual absolute magnitudes obtained from both Hipparcos data and Schmidt-Kaler calibrations. Our purpose is to validate our spectroscopic work (Ginestet et al. 1997, 1999) on stars with composite spectra with the help of Hipparcos data. Contrary to what is claimed in other papers, the absolute magnitude domain devoted to the giant stars does not overlap the domain of dwarfs. We find that the discrepancies between absolute magnitudes from Hipparcos data and absolute magnitudes deduced from Schmidt-Kaler calibrations increase with the relative error sigma(pi)/pi on the parallaxes. So, for sigma(pi)/pi less than or equal to 0.05 only 3% of the stars present a discrepancy of one luminosity class, while this percentage reaches 54% for 0.25 < sigma(pi)/pi less than or equal to 0.50. Curiously, the luminosity of the giants seems to increase with the distance of the stars, whereas the supergiants of the sample appear underluminous at least for d < 600 pc! We point out a list of 14 MK standards whose luminosity classes may be erroneous and need a new spectral classification, in the near infrared. The case of composite-spectrum binaries is also discussed. Most of these are too distant for accurate parallaxes ever? with Hipparcos: only sixteen stars have sigma(pi)/pi less than or equal to 0.10; for these, we give new spectral classifications in agreement with both our classifications in the near infrared of the cool components and Hipparcos data. Finally, for stars having high-precision parallaxes (sigma(pi)/pi less than or equal to 5%) there is no serious problem for Schmidt-Kaler calibrations with respect to Hipparcos data. The data corresponding to parallaxes of lower precisions should be used with caution and only for statistical analyses.