1 Introduction

Modern radial-velocity spectrometers permit to measure the absolute wavelength shifts of stellar spectral features to better than 100 m s^-1 (Udry et al. 1999; Nidever et al. 2002). At this accuracy level the interpretation of the observed spectral shifts in terms of stellar radial velocities is non-trivial, due to many factors such as gravitational and convective shifts, template mismatch, and the ambiguity of the classical radial-velocity concept (Lindegren et al. 1999; Lindegren & Dravins 2002). Recognising this difficulty, the IAU has adopted a resolution (Rickman 2002) identifying the "barycentric radial-velocity measure'' ( $cz_{\rm B}$) as the appropriate quantity to be determined by accurate radial-velocity spectrometry. The radial-velocity measure is further explained below; briefly, it is the absolute spectral shift corrected only for the accurately known local effects such as the motion of the observer. It is expressed as an apparent velocity, which for normal stars to first order ($\la$1 km s^-1) coincides with the classical radial velocity.

In order to apply this new concept in accurate radial-velocity work, it is necessary to review many of the established procedures and to modify, or even abandon, some of them. For instance, the practice of using standard stars or minor planets to define a velocity zero point is inconsistent with this aim, except at a superficial accuracy level ($\sim$0.5 km s^-1 for normal stars). In this paper we describe and apply a procedure to derive accurate radial-velocity measures from digital échelle spectra. While there may be many other (and also better) ways to achieve this, we hope that the paper may serve as a practical illustration of the new concept, in addition to providing accurate radial-velocity measures for a number of stars.

The procedure is intended to give results that are reproducible in an absolute sense, i.e. without systematic shifts caused by poorly understood or controlled conditions, such as which spectral lines are being used, which parts of the lines define the central wavelengths, and the definition of wavelength scales at the spectrometer and in the laboratory. The means to achieve these goals are not necessarily consistent with techniques that aim for maximum precision, and our procedure is therefore not optimal e.g. for the search of exoplanets.

The spectra used here were obtained in 1997 with the ELODIE spectrometer at Observatoire de Haute-Provence (Baranne et al. 1996) as part of a larger programme to compare the spectroscopic results with astrometric radial velocities (Dravins et al. 1999b; Lindegren et al. 2000; Madsen et al. 2002) in order to study the line shifts intrinsic to the stars (Dravins et al. 1999a). Radial-velocity measures are derived by means of a special reduction procedure using a synthetic template of Fe I lines, and an absolute zero point defined by the Kurucz et al. (1984) Solar Flux Atlas via observations of the Moon. Results for the Sun and 42 stars are given in Tables 1 and 2.