2 Selection criteria for QSO candidates

The objective prism plates are scanned with a PDS1010G microdensitometer in a low-resolution mode (see Paper I). After on-line background reduction and object recognition, the low-resolution density spectra are stored on magneto-optical disc. At a given brightness of an object, its density spectrum has basically a triangular shape, which peaks close to the steep emulsion cut-off at $\lambda$5400Å and declines gradually in density towards the ultraviolet border at $\lambda$3400Å. This characteristic shape is a result of the convolution of the compressed wavelength scale close to the cut-off, which is caused by the non-linear dispersion, and the (usually) rising spectral energy distribution of the objects. In Fig. 1 a few examples are given.

 

Figure 1:   Examples of lrs-spectra with continuum fit. a) QSO HS 0948+4735, displaying a CIV emission line at $\approx$4000Å. b) QSO SBS 0946+50 at z = 1.22 without prominent emission line, c) and d) random stellar spectra

The density spectra are characterized by several parameters as are their spectral length, amplitude, integral density, center of gravity, slope of the spectral density etc. In principle, a multi-parameter space can be constructed in which most of the spectra will occupy a particular volume, the main locus. As quasar spectra often show an ultraviolet excess and possess emission lines, their appearance will differ from those of most of the stars, and they are expected to be found in the multi-parameter space outside the main locus. To ease the handling of the data a principal component analysis can be performed to reduce the dimensions of the parameter space (Francis et al. 1992). In practice, however, objective prism QSO surveys used two-parameter spaces, with the most efficient selection parameter(s) determined by experiment (Hewett et al. 1995; Wisotzki et al. 1996). This approach is also followed by the HQS, with the slope of the spectral density $\Sigma$ used as the selection parameter. For its definition the spectra are fitted with a polynomial of 2nd degree with $\Sigma$ being the slope of this fit at 4400 Å. The fit procedure is iterative omitting density values of individual pixels deviating significantly from the density predicted by the previous fit. The influence of strong emission lines and crippled pixels on the determination of the continuum slope is therefore greatly diminished. Examples for individual fits are given in Fig. 1.

 

Figure 2:   Location of spectra in a plot of the slope of the continuum $\Sigma$at 4400Å against integral density. The density is given in internal machine units (pds-counts). All spectra above the plotted dividing line (see text) are selected

Blue spectra are selected by the determination of a dividing line in the two-parameter space defined by $\Sigma$ and the integral density. The distribution of spectra in this space is non-linear due to spectral variations of the characteristic curve. A typical distribution is shown in Fig. 2, where the distribution of $\Sigma$ as a function of the integral density for all spectra of the plate H1558 is plotted. Spectra are flat for weak and very great densities, and show the steepest slopes in the linear part of the characteristic curve. The reason for this behaviour are changes in the contrast in sensitivity of the plate across the wavelength region $\lambda\, 3400 - 5400$ Å. The contrast first increases as a function of the absolute density, which means that the density spectra of faint objects all have a similar flat slope while the mean slope steepens with increasing absolute density. With further increase of the absolute density, the spectra saturate starting at the red cut-off resulting in a decrease in the contrast and hence in a re-flattening of the spectra. The result is a curved distribution as shown in Fig. 2.

The dividing line to select blue spectra has to be determined in the curved distribution of the chosen two-parameter space, and the determination has to be made individually for each plate, as its locus is also a function of the shape of the characteristic curve of the plate. For each plate spectra are collected into intervals of integral density containing 500 spectra. The widths of the intervals are therefore varying, being small at low densities and increasing towards higher densities. In each interval a limiting slope is determined which separates the bluest spectra from the rest. Depending on the width of the interval the fraction of the selected spectra decreases with increasing integral density, starting with $\approx$20% for the lowest density levels and decreasing to $\approx$5% for high densities. This accounts for the varying effectiveness of our selection criterion with integral density. The dividing line is finally obtained by interpolation of the limiting slope values.

The selection criterion fails for the lowest and highest density levels. Thus spectra with integral densities $\le$2000 and are discarded. The lower limit corresponds to our completeness limit for the extraction of spectra by our digitization technique (see Paper I), making selection of blue spectra below this limit less reliable anyway.

The selected blue spectra are rescanned individually with full resolution and sampling, and are classified visually on a graphics display. The digitized direct plates are used to recognize overlaps, to probe for extended images, and to determine coordinates. Objects are discarded as QSO candidates, if one of the following cases applies:

-

Absorption features, such as the G-Band at 4300Å and the Ca H+K lines qualify the spectrum as stellar,

-

Balmer absorption lines qualify the spectrum as stellar with high effective temperature,

-

an extended image on the direct plate and a strong emission line close to the green head of the spectrum ($\lambda$$\approx$5400Å) suggests the presence of a strong $\lambda$5007Å O[III] emission line, qualifying the correspondent object as a narrow-emission line galaxy (Vogel et al. 1993).

The remaining objects are divided into two categories. In spectra of primary candidates broad emission lines must be detectable while spectra of secondary candidates are featureless and not distinguishable from hot star spectra with small absorption lines which cannot be resolved in our density data.