Two papers (González-Riestra et al. 2000, Holberg, Barstow, & Sion 1998) have reported that NEWSIPS-processed, high-dispersion SWP images have negative zero-points errors relative to other standards. Both rely upon measurements of centroids of individual interstellar lines, generally in spectra of OB stars. González-Riestra et al. reported differences of -17.7 ± 3.7 km s-1 for the SWP camera with respect to ground-based measurements of interstellar lines in the optical spectra of the same stars. A smaller discrepancy was found for small-aperture, long-wavelength camera data. As a consequence, a correction of +17.7 km s-1 was applied to all SWP high-dispersion spectra in the IUE Newly Extracted Spectra (``INES") system (González-Riestra et al. 2000).
Upon investigation we are unable to support the conclusion by González-Riestra et al. that a significant difference in radial velocities exists for NEWSIPS SWP spectra. We have attempted without success to reconcile the differences between their results and ours. These authors kindly provided us with program data lists, so we were able to compare cross-correlation shifts from their samples with those from entire sets of obervations of these stars in the archives. A comparison of the results showed no noticeable difference between the two samples of observations, and thus one must look elsewhere to resolve the difference. Next, we cross-correlated the same interstellar lines used in the González-Riestra et al. analysis for in a sample of 145 large-aperture SWP spectra of 10 Lac with the interstellar lines in the Brandt et al. (1998) GHRS atlas. For the three sets of lines they measured (Si II 1259-60, O I 1302, Si II 1304, and C III 1335-6), we found a difference RVSWP - RVGHRS = -0.9 ± 1.5 km s-1. Thus, there is essentially no difference between the systemic velocities of these interstellar lines and the photospheric lines in the IUE spectrum for these two instruments. González-Riestra et al. did not measure SWP observations of 10 Lac, but their measurements of lines in HD 93521 and HD 60753 spectra suggest that the wavelengths in these stars typify the zero-point offsets they found for their total sample of six stars: -18.8 km s-1 for HD 93521 and -12.9 km s-1 for HD 60753. In 4.2.3 we reported that the relative velocities of the photospheric lines for these three stars were in agreement with ground-based studies, so in this important respect all our comparisons are self-consistent (as are those of González-Riestra et al.).
Since we can find no other discrepancy among our results with respect to the atlas results or with respect to photospheric lines of the stars in our common sample, we are left with the possibility that the measurement positions of the interstellar lines by González-Riestra et al. are misleading for some subtle reason. One possibility is that the multi-component nature of the interstellar clouds toward these stars gives different centroid velocities for some groups of lines than for others. In particular, differing degrees of saturation among interstellar lines at low spectral resolutions can give rise to rather large centroid velocities differences among various interstellar line complexes in the spectrum of a given star. For example, differences between the 1304 and 1335-6 complexes are quite large in the spectrum of HD 93521, as was indeed noted by González-Riestra et al. themselves.
In 3.1.1 we noted a very similar time-dependence of apparent velocities between IUESIPS and NEWSIPS spectra with respect to the dependence noted by Holberg, Sion, & Barstow (1998). These authors also found a mean difference of -8.3 km s-1 ± 1.4 between NEWSIPS -derived and ground-based radial velocities for four white dwarfs. This finding appears to contradict the conclusion from our work on 10 Lac and other bright B stars (4.2) that the velocity zero-point for SWP NEWSIPS is correct within -1 km s-1. We have attempted to track down the source of this discrepancy in a number of ways. First, in 3.2 we discussed the possibility that the mode of centering on bright stars and faint stars could lead to different results (see Table 1). Yet, we found no such effect. We may also consult Fig. 4 for clues of an apparent velocity difference between white dwarfs and bright OB stars. The figure implies that the mean of the white dwarf velocity offset should agree within about ± 1 km s-1 of our 10 Lac result. This statement holds true for the velocity average of the four white dwarfs selected by HSB. We have also cross-correlated the groups of interstellar lines considered by HSB in selected echelle orders. These comparisons produce a mean IUESIPS - NEWSIPS velocity shift of -1.3 ± 0.6 km s-1 relative to differences derived from orders containing only photospheric lines. Thus, there is nothing special or ``peculiar" about the spatial mapping of the ISM lines in NEWSIPS that would give results systematically different from an analysis of photospheric features over the whole spectrum.
Unable to resolve a disagreement with HSB's results in this manner, we proceeded with another test. In their discussion of the apparent wavelength error, HSB compared their results with two earlier studies of interstellar lines in their white dwarf sample. The more accurate of these was a study by Lemoine et al. (1996; Small Science Aperture observations) of the ISM line spectrum of the white dwarf G191-B2B. To make a comparison with NEWSIPS wavelengths, we acquired the Lemoine et al. data from the the MAST archives. These observations were originally made by substepping the grating to minimize fixed pattern detector noise, so we made corrections for these shifts by comparing emission line spectra obtained at nearly the same time for wavelength calibration (this comparison was made by measuring centroid line positions of the two groups of spectra interactively). The differences we determined were applied to the raw data in order to form high-quality, co-added spectra. We next cross-correlated 13 SWP LGAP spectra of the star with the mean GHRS spectrum in narrow wavelength regions centered on 12 interstellar features studied by these Lemoine et al. These cross-correlations gave a mean shift, RVNEWSIPS - RVGHRS, of -2.3 ± 3.0 km s-1, i.e. zero within the errors (see Table 3). This finding is at mild variance with the HSB prediction of -8.2 km s-1, but it is consistent with our other results.