2000-3000) region and has a comparable quality.
Thus, we had to resort to alternative strategies to evaluate zero-point
errors in long-wavelength camera data. Although it is possible in
principle to cross-correlate echelle orders for SWP and
long-wavelength camera data over their overlapping wavelength range,
practical difficulties intervene when carrying out this operation. One of
these is that a loss of camera sensitivity in these regions causes a steep
gradient in signal-to-noise ratio along the spectrum. In order to compare
spectral lines of comparable quality, this constraint further limits the
already narrow wavelength interval. In addition, spectra of even cool stars do
not clearly exhibit many features. Despite these limitations, we were able to
compute reliable cross-correlation shifts for the sharp-lined stars 10 Lac
and 
Sco for wavelengths (1949-1959) in common with the two
cameras, namely order m = 71 for SWP and m = 118 for LWP/LWR.
For a comparison with the LWR camera, we find apparent radial velocity
differences, RVSWP - RVLWR = -0.1 and -3.0 km s-1 for 10 Lac
and Sco, respectively. The corresponding results with the LWP
camera are RVSWP - RVLWP = +1.1 and +5.4 km s-1, respectively.
As usual, these values refer to means of the ensembles of
many LGAP spectra for each star. We estimate the errors in these
comparisons to be about
± 4.4 km s-1 by adding in quadrature the
typical order-to-order fluctuations (
± 3 km s-1) with the
average velocity difference (
± 3.2 km s-1) between large- and
small-apertures. Since the derived zero-point differences are
comparable with this r.m.s., our comparison suggests that there are no
significant wavelength differences in the
1950-2050
region between the SWP and long-wavelength camera data.
An alternative method of evaluating mean radial velocity errors for the long-wavelength errors is to cross-correlate the fluxes of IUE spectra with those of HST atlases of Procyon and Arcturus. Note that the suitability of this technique relies upon our previously tying the zero-points of these atlases to the 10 Lac atlas via IUE data. This method is reliable in principle, but in practice it has the drawback that rather few IUE observations are available for these particular stars. In the case of LWR camera for Arcturus, we were able to cross-correlate 9 orders in just three spectra and found a difference, RVLWR - RVSTIS, of +2.7 ± 4 km s-1. These quoted errors and others quoted below include both the r.m.s. velocities for individual orders and the r.m.s. arising from the small numbers of observations. We cross-correlated seven available spectra with the Procyon atlas and found a difference of -3.5 ± 6 km s-1. A comparison between the Arcturus atlas and the LWP datasets is not possible because the IUE observations were made with the star image placed in various ``nonstandard" positions in the large aperture. For the Procyon atlas (3 orders; 4 LWP spectra) the difference, RVLWP - RVGHRS, is +7.9 ± 6 km s-1. However, this difference is noticeably affected by a wavelength shift of one spectrum (LWP 13112) with respect to the three. Although we could find no a priori reason to exclude the LWP 13112 spectrum from our analysis, its omission would result in an RVLWP - RVGHRS difference of +2.4 km s-1. This would bring the LWP scale fully into agreement with the essentially null difference found in the preceding paragraph. Additionally, we compared the mean zero-point difference between the LWP and LWR cameras by cross-correlating many spectra in 59 orders of six stars. These results are summarized in Table 4. The entries in this table give a mean difference of only +0.4 ± 3 km s-1, which is shown as the last entry in Table 3. T his agreement suggests again that the zero-points of these two cameras agree to within errors of about ± 5--6 km s-1.
To summarize all these comparisons, the radial velocity differences of the three IUE cameras, inter alia, are zero to within the measurement errors.