3 Observations and data reduction

3.1 Instrumentation

Spectroscopic observations were performed at the ESO 3.6 m telescope in La Silla. The main body of data was obtained in six observing runs with the OPTOPUS multifiber spectrograph at the Cassegrain focus (Lund 1986; Avila et al. 1989). Further observations were obtained in 1994 with the multifiber spectrograph MEFOS (Avila et al. 1995; Felenbok et al. 1997) at the prime focus of the same telescope. A log of the observations is given in Table 1.

 

Table 1: OPTOPUS-MEFOS observing log

OPTOPUS has a bundle of 50 optical fibers (Polymicro FHP with 320 $\mu$m core and enhanced blue sensitivity), which are manually plugged into aluminium plates with holes and fiberholders at the galaxy positions. These positions are corrected for differential refraction between the wavelength centre of the autoguider sensor and the centre of the spectral range, at the expected zenith angle of observation. The plates are mounted at the Cassegrain focal plane and have a diameter of 32 arcmin on the sky. The core of the fibers corresponds to a sky aperture of 2.3 arcsec in diameter. At the opposite end the fibers are coupled to an ${\it f}/8$ collimator of the Boller & Chivens spectrograph, forming the entrance slit.

Once mounted, the plates are centred using four relatively bright stars at the periphery of the plate, observed through fibers going to an intensified camera. Exposure guiding is accomplished by means of two fiber bundles, which are used to image two relatively bright stars in the field through holes in the plate. The mechanical constraints are such that the minimum object-to-object separation is 24.6 arcsec and minimum object-to-guide star separation is 64.3 arcsec.

For the present program we used the ESO grating # 15 (300 lines mm^-1 and blaze angle of $4^{\circ}~18^{\prime}$), allowing a dispersion of 174 Å mm^-1 with an ${\it f}/1.9$ blue camera in the wavelength range 3750-6150 Å. Initially the detector was a Tektronix 512 $\times$ 512 CCD with a pixel size of 27 microns, corresponding to 4.5 Å/pixel, a velocity bin of $\simeq$ 270 km/s at 5000 Å. After the October 1992 observing run, the detector was replaced with a new Tektronix thinned, back-illuminated CCD (ESO #32), which also had a pixel size of 27 microns, giving a similar spectral coverage but a better blue sensitivity.

The observing time for each field was one hour, split into two 30 mins exposures to ease the removal of cosmic rays. Keeping the telescope in the same position, exposures with a quartz-halogen lamp as well as of a Helium comparison lamp were taken immediately before and after the science exposures. Depending on the length of the night we observed from 4 to 6 different fields per night choosing the sequence of the fields to be observed in such a way so as to minimize the zenith distance and hence the differential refraction.

Further observations were accomplished in October 1994 with the multifiber spectrograph MEFOS, at the prime focus of the 3.6 m telescope.

The MEFOS Spectrograph has 30 robotic arms in a "fishermen-around-the-pond'' configuration over a field of one degree diameter. One arm is used for guiding purposes, while the other 29 arms carry two spectroscopic fibers each, one fiber for the object and one for the sky. Because of the prime focus scale, the optical fibers (Polymicro FBP) have a 135 $\mu$m core, corresponding to a sky aperture of 2.6 arcsec. The object-sky fiber separation is 60 arcsec, and the minimum object-to-object distance is 28 arcsec. Each arm also carries an imaging fiber bundle which is used to accurately centre the fibers on the targets. The output ends of the spectroscopic fibers are arranged in a line so as to form the entrance slit of the same Boller & Chivens spectrograph used for OPTOPUS. With a new optimized ${\it f}/3.03$ collimator the spectrograph-CCD configuration was the same as used for the previous OPTOPUS observations.

3.2 Data reduction

The data reduction was performed using the APEXTRACT package as implemented in IRAF. For each exposure first we identified and followed the spectra on the white lamp frame. Then the solutions obtained for the white lamps were applied to extract the one-dimensional spectra both in the calibration and science frames. Using APEXTRACT we performed an "optimal extraction'' (Horne 1986), also detecting and replacing the highly deviant bad pixels and cosmic ray hits.

The individual wavelength calibration for each fiber is derived from the corresponding arc spectrum. We used a fourth-order polynomial fit with typical rms errors of 0.2-0.3 Å. The accuracy of the calibration was estimated from the measured wavelengths of two [OI] sky lines ($\lambda\lambda$ 5577, 6300), which were always within $\pm~1$ Å (less than a quarter of pixel) from their expected positions.

A critical point in fiber spectroscopy is the sky subtraction because it is impossible to measure the background locally and through the same aperture (fiber) of the object (see Parry & Carrasco 1990 or Wyse & Gilmore 1992 for a detailed discussion). We adopted the following strategy. For OPTOPUS observations we dedicated at least four apertures to the sky in each field. We defined predetermined positions on each plate which were verified not to have any object at the limit of SERC J plates. For MEFOS the sky fibers were as many as the target objects. With both instruments we performed a "mean-sky'' subtraction method (Cuby & Mignoli 1994).

The main problem is the fiber throughput determination, because of the different transmission of each waveguide which could also vary from one exposure to the other. The most intense and isolated sky line in our spectral range is the [OI]5577 emission line, and we used its flux as an estimator of the fiber trasmittance. We directly measured the line counts on the calibrated spectra by mean of a Gaussian profile fit, after subtracting the underlying continuum (due to both sky and object, if any, contribution). Under the assumption that the intensity of the night spectrum does not vary appreciably within the telescope field, the flux of this sky line is the best estimator of the fiber throughput. In fact it is easily measurable both in the sky and object spectra and it is subject to the same temporal transmittance variations as the object flux (e.g. due to the differential fiber flexures during the exposure). In order to reveal possible contaminations (cosmic ray hits or object emission/absorption lines) that could affect the fiber throughput estimate, we also computed the mean and the dispersion of the ratio between the peak of the continuum-subtracted counts of the [OI]5577 and the same quantity for the [OI]6300 sky line: spectra with unusual value of this ratio were re-examined to check the reason for the discrepancy. In all the relatively few cases in which this has occurred, the problem was overcome by interactively fitting the [OI]5577 line and eliminating the cosmic ray contaminations, or was due to an underestimate of the sky [OI]6300 emission because of the coincident absorption feature in stellar objects.

After the normalization to account for the relative fiber transmission, a "mean sky'' was determined using all the sky spectra (four or more for OPTOPUS, up to 29 for MEFOS) and subtracted from all the "object plus sky'' spectra of the same exposure. This procedure was repeated separately for each of the two 30 mins exposures. The comparison of the two exposures of the same field revealed the remaining cosmic rays spikes, which were eliminated by interpolating the adjacent pixels.

The sky-subtraction accuracy is difficult to gauge. The residual counts in the sky-subtracted sky spectra provide a good estimator of this, but caution is necessary: indeed, the simple rms value of the sky-subtracted sky spectrum could overestimate the quality of the sky subtraction because it does not take account for "bias residual'' if the average background has been over/underestimated (see Wyse & Gilmore 1992). Therefore we adopted, as estimator of the sky-subtraction accuracy, the "average absolute sky-residuals'' quantity

$$Q_i = \biggl\langle\Bigl\vert{{\rm sky}_i - \langle {\rm sky... ...langle {\rm sky} \rangle_i }} \Bigr\vert\biggr\rangle_\lambda.$$ (1)

This quantity has been measured, resulting in the range of $2.8\div 7.5$% with a typical value of 4%.