The purpose of this paper is to provide input data for Th-Ar
wavelength calibration lamps that allow to approach the random noise limited
centering accuracy on individual calibration lines. The rich Th
spectrum produces many blended features at moderate resolutions.
The majority of the lines is unblended only at pixel-scales 
.
Even when a blending line is weak
compared to the intensity of the principal component, its influence on
centering algorithms is shown to exceed often by far the uncertainty due to
random noise. We produce a table of calibration lines for different moderate
resolutions, from which the user can easily extract a line list with
resolution-dependent blend wavelengths. In this way, the user keeps the
freedom to apply the selection at a level of rigidity compatible with his goals
and the specific format of the observations. The tables encompass the range in
pixel-scale from a lower limit 
below which so many Th
lines are badly blended that another lamp should be used for accurate
calibration, up to an upper limit 
above which blending is a
minor problem (selecting all completely unblended lines at
already gives ample calibrator lines).
The application that profits most from the use of a list of line positions with
high individual accuracy is the derivation of an adequate analytical
calibration relation 
where x1 and x2 refer to the position on the detector. Usually
x1 is along the detector rows or columns, whichever is approximately
parallel to the spectral orders, and x2 varies with the number of the
spectral order or along the slit. Accurate input data allow to derive a more
robust relation (Hensberge & Verschueren 1989), rather than using
(bivariate) polynomial approximations, that, especially in the case of
multi-order (echelle) registrations, induce many uncoupled parameters without
physical meaning.
The availability of few-parameter accurate fit relations will further gain importance with the use of arrays of CCD's in spectroscopy. Indeed, one echelle order may then be projected in parts on different detectors which will never be perfectly aligned. With this orientation problem entering the calibration procedure, an inclusion of unnecessary degrees of freedom would induce still more instability.
In Sect. 2, general arguments on the attainable precision and on line blending effects are summarized. The method used to determine the corrected blend wavelengths is presented in Sect. 3. A short statistical discussion of the results is given in Sect. 4. The availability and the applicability of the selection tables are described in Sect. 5. In Sect. 6 we briefly discuss to what extent the line selection requires the use of robust, few-parameter calibration relations, and comment on the risks involved in some common alternative approaches.