Accuracy of radial-velocity measurements for early-type stars - II. Investigations of spectrum mismatch from high-resolution observations

The accuracy with which radial velocities of early-type stars can be measured is limited in practice by the existence of asymmetrical differences between object and template spectrum, constituting "spectrum mismatch". Our studies of the magnitude of spectrum-mismatch errors, commenced in Paper I (Verschueren et al. 1999) on the basis of synthetic spectra having different attributes of effective temperature (T-eff) and log g, are continued here in a complementary approach that employs observed spectra. From over 60 de-archived observations rye derive accurate wavelength scales for the spectra of 16 dwarfs of spectral types B8-F7, and examine the results of cross-correlating the spectra against different (observed) template spectra. We also test the effects of (a) truncating the spectra at different levels below the continuum, (b) adding rotational broadening to enforce a visual match of line-width between object and template, (c) applying rotational broadening to exacerbate a rotational mismatch, and (d) neglecting the presence of faint companion spectra. We also cross-correlate pairs of spectra such that the differences between their T-eff are minimal. We conclude that it will be possible to measure radial velocities to an accuracy considerably better than 1 km s(-1) for slowly-rotating stars in the range of spectral types examined, and a careful discussion of the nature and sources of the random and systematic errors that become significant in work of this nature enables us to specify conditions that are important for achieving such accuracy routinely. We find that both rotational broadening, and the star-to-star variations in line strengths that are so prevelant among A-type spectra, can give rise to more deleterious mismatch shifts (RV errors) than do differences in T-eff alone, even for DeltaT(eff) as great as 300 - 400 K. By intercomparing the results given by wide regions of spectrum (similar to 800 Angstrom) with those obtained by isolating small groups of features in very narrow windows (similar to 30 Angstrom), we have been able to designate a window near lambda 4570 Angstrom, that should be particularly reliable for high-accuracy results; and we propose further studies at very high S/N ratio in that specific window to complement and extend the results of the present paper.