Selection of ELG candidates from objective prism spectra

The ELG candidates were selected from the digitized objective prism plates of the Hamburg Quasar Survey. The original plates were obtained with the 80 cm Schmidt telescope at Calar Alto, Spain. The telescope is equipped with an 1.7 $^\circ$ objective prism allowing to obtain spectra with a dispersion of $\sim 1390$ Å/mm at H $\gamma$ . The size of plates is 24 $\times$ 24 cm covering a field of $5.5^\circ \times 5.5^\circ$ on the sky. The Kodak IIIa-J emulsion is used, giving a spectral range from 3400 to 5400 Å. The plates were digitized with a PDS 1010G microdensitometer in a low-resolution mode using 100 $\times$ 20 $\mu$ m slit and 10 $\mu$ m step size and the extracted spectra are stored on optical disks. Once spectra of candidates are taken from this database, high-resolution PDS scans (with a slit size of 20 $\times$ 20 $\mu$ m) are made of them for further refinement of the classification. Additional information about the HQS can be obtained from Hagen et al. ([1995]). The low-resolution spectra, which have 10-20 spectral points per object, already allow the detection of strong emission lines, in particular for ELGs with prominent [O III] $\lambda\lambda$ 4959,5007 Å and H ${\beta}$ lines. The faintest ELG candidates have B magnitudes up to 20^m. In general their redshifts are limited to z $\leq$ 0.075 because the [O III] $\lambda\lambda$ 4959,5007 Å lines are shifted for more distant galaxies outside the spectral range of the photoplates. Sometimes prominent H ${\beta}$ may allow detection up to z $\leq$ 0.1, and in rare cases galaxies may be selected due to their strong [O II] $\lambda$ 3727 Å emission line up to a redshift of z = 0.43. The selection is biased for the [O III] lines compared to the [O II] line because of linear dispersion, favoring the detection of narrow lines near the red end of the objective prism spectra. Only in bright galaxies (B < 17.5 the [O II] line was well detected (for example, HS 0132+0610 and HS 0935+4726, see Vogel et al. [1993]). Quasars with strong emission lines near $\lambda$ 5000 Å can also be detected, in particular, if Ly $\alpha$ or Mg II $\lambda$ 2798 Å enter to this wavelength region (corresponding to redshifts of $\sim$ 3.0 and 0.9, respectively).

**Figure:** Distribution of equivalent widths of the [O III] $\lambda$ 5007 Å emission line in Å as function of photographic magnitude for the training sample of known BCGs
$\begin{figure}\centering \psfig{figure=FIGURE1.ps,width=9.0cm,angle=270} \end{figure}$

The adopted search method consists of various selection techniques. It is applied first to the low-resolution and subsequently to the high-resolution density spectra. The low-resolution density spectra are characterized by several parameters, and we chose after some experiments the integral density and the slope at 4000 Å as the main selection parameters. The integral density is the sum of the densities of all pixels contributing to a spectrum and the slope is determined by a 2nd order polynomial fitting of the spectrum. The length of the selected spectra was limited to 5-20 pixels to avoid spectra dominated by noise or affected by overlaps. To determine the range of slopes shown by ELGs for a given brightness (represented by the integral density), we used a training sample of $\sim$ 50 blue compact galaxies with known spectral properties taken from the SBS (Izotov et al. [1993a], [1994]), which were separated on HQS plates in the zone of the SBS. This sample contains galaxies with a wide range of sizes, luminosities and [O IIi] $\lambda$ 5007 Å emission line equivalent widths between 10 and more than 1000 Å. Figure 1 shows the distribution of the training sample in a photographic magnitude - equivalent width plane. For magnitudes fainter than about 18.0, our training sample is running out of objects with moderate or weak [OIII] line strength. This yielded some uncertainties at the fainter end for our first definition of the selection rules (see below).

**Figure:** Positions of known BCGs from the training sample in the unitless coordinates integral density (brightness) (dns) -- slope of spectra (slo) (large symbols) compared to objects of all types as recorded from the HQS plates. The various symbols denote different ranges of the equivalent width of the [O III] $\lambda$ 5007 Å emission line. The lower part of the figure is a magnified part of the upper panel within the rectangle. This rectangle indicates our final selection box (see text for details)
$\begin{figure}\centering \psfig{figure=FIGURE2.ps,width=10.2cm}\vspace{-3.0cm}\end{figure}$

Figure 2 shows the location of the SBS BCGs in an integral density - slope diagram (in the relative units of the digitisation of the HQS plates). According to the distribution of BCGs in this diagram we chose limits $-0.3 \div 0.15$ for the slope and $400\div 6500$ for the densities. The slope range is rather wide due to both the intrinsic properties of the galaxies and variations of the plates quality. Nearly 30 % of the typically 30000-50000 spectra per plate pass this filter. This number of spectra is still too large to proceed with high-resolution scans. Therefore, all preselected low-resolution spectra were visually classified on a vector graphics screen and candidates with density peaks close to the green head of the spectrum were selected as possible [O III] $\lambda$ 5007 Å hits. This procedure is quite efficient to keep good ELG candidates and to reduce the total number of selected candidates to $3000\div5000$ . Tests with several plates of the same field have shown that 5 to 15 % of good candidates are lost, if the selection was not made on the best plate.

All selected low-resolution spectra were then rescanned with high resolution (PDS scanning time $\sim$ 3-4 hours/plate) in order to select visually first and second priority candidates according to the following criteria:

a) Objects showing a clear density peak near 5000 Å and blue continuum in the high resolution spectra are adopted as first priority candidates.

b) Some bright (B $\le$ $18\fm5$ ) BCGs with weaker emission lines (EW([O III] $\lambda$ 5007 Å) $\sim$ 50 Å) in the training sample have density spectra with blue continuum but no clearly detectable emission peak. Therefore, candidates with a blue continuum but without prominent emission features or candidates with indications of emission peaks but with an unusual continuum shape are kept as second priority candidates in the attempt to avoid losses of true ELGs.

Examples of the high resolution spectra of the first and the second priority candidates are shown in Fig. 3 (small boxes include the low resolution spectra). As a result we have got up to 30 first priority candidates per plate and a similar number of second priority objects for follow-up spectroscopy. The resulting surface density of emission-line candidates for each priority varies from 0.3 to 1 candidate per square degree.

It is reasonable to outline here the ELG types which we wish to discriminate with the selection procedure used. First, due to the adopted slope limits of the spectra in the parameter space, we miss ELGs with redder continuum distribution. This is usually the case for galaxies dominated by the emission of an old stellar population. Such objects are found efficiently using wide-field objective prism surveys sensitive to the H $\alpha$ spectral range (Zamorano et al. [1994], [1996]) or by surveys based on narrow-band H $\alpha$ imaging (Boroson et al. [1993]). Second, due to the combined effect of the upper limit on the redshift of [O III] $\lambda\lambda$ 4959,5007 Å and the lower limit on the object's apparent magnitude (corresponding to the upper limit in integral density), we miss all objects with absolute magnitudes brighter than M_B $\sim$ -21. Therefore, many Seyfert galaxies are not selected into our candidate lists.

**Figure:** Examples of candidates selected as first (left panel) and second (right panel) priorities. The high resolution scans are shown, the inlets present the low resolution scans used for the first selection step. For details see text
$\begin{figure}\centering \psfig{figure=FIGURE3.ps,width=10.5cm,angle=0}\vspace{0.5cm}\end{figure}$

We apply the above selection procedure to the first set of 38 fields with boundaries $\alpha = 8^h - 17^h30^m, \delta = +45^\circ - +50^\circ$ and $\alpha = 8^h - 15^h30^m, \delta = +40^\circ - +45^\circ$ , comprising an area of about 900 square degrees. This survey strip partly overlaps with the fields of Popescu et al ([1996], [1997]). The resulting list consists of about 1000 first priority candidates, and a similar number of second priority ones. To check the contents of these categories we observed both first and second priority candidates in two pilot follow-up spectroscopy runs. The candidate lists contain also many ELGs with known redshifts but with somewhat uncertain classification. Due to the lack of time we could observe only 11 of them.

After the analysis of the results of these observations (see Section 5), we improved the efficiency of the overall selection procedure by applying additional criteria, which resulted in significant increase of ELG detection efficiency during follow-up spectroscopy.