2. Data reduction
2.1. Observations and wavelength calibration
UV spectra of HS 1216+5032 A and B were obtained on 1996 November 6 with the FOS on board the HST . Target acquisition and spectroscopy were done using Grating G270H with the red detector and the aperture. This configuration yields a spectral resolution of FWHM Å and a wavelength coverage from 2222 Å to 3277 Å (Schneider et al. 1993). Total integration times were 1980 and 10400 s for QSO components A and B, respectively. The spectrum of image B resulted from the variance-weighted addition of four exposures. The signal-to-noise ratios at Ly emission are S/N (A) and 25 per Å pixel, falling down to 15 in the blue part of the A spectrum, and to near one at the BAL troughs in B.
The flux-calibrated spectra and their associated 1 errors are shown in Fig. 1. The dotted lines represent the continua (Sect. 2.3) and the tick-marks indicate the position of absorption lines (Sect. 3).
Since we are interested in comparing absorption systems along both LOSs, we must check for possible small misalignments of the zero-point wavelength scale between both spectra. Such differences may arise when the targets have not been properly centered in the aperture of the FOS, and/or when - unlike here - the science and wavelength calibration exposures have not been performed consecutively. Galactic absorption lines common to both spectra were then required to appear at the same wavelength, under the assumption that both LOSs cross the same cloud (which should be true for the small separation between LOSs at ). There are two Galactic absorption lines, FeII and MgII , common to both spectra and apparently not contaminated with intergalactic lines. For these lines (FeII ) km s-1 and (MgII ) km s-1 are obtained, which lie within the fitting procedure errors ( km s-1) and the FOS limiting accuracy ( km s-1). These small differences imply that both spectra are well-calibrated; consequently, we made no corrections to the zero-point wavelength scale.
2.2. Emission redshifts
To derive the emission redshift of HS 1216+5032 A the Ly emission peak was fitted with a Gaussian in the region between 2959 and 3005 Å. Excluding the wavelength regions with absorption features in the blue wing of the line yields , where the error comes from the wavelength uncertainty. For QSO component B, a similar analysis is extremely difficult because the Ly emission line is severely distorted by the Ly and NV BALs. Fitting instead the NV emission peak between 3016 and 3065 Å, and using the mean rest-frame wavelength of the doublet (NV ) Å, yields . In the following we adopt the nominal values (A) and (B).
The redshifts of A and B are not the same within measurement errors, with QSO component B being blueshifted by km s-1 relative to A. Different emission redshifts are expected in the physical pair hypothesis, but the measured difference between HS 1216+5032 A and B cannot rule out the possibility that the QSO may still be gravitationally lensed. This is because two different transitions, Ly and NV , are being compared and blueshifts of a few hundreds km s-1 between high and low-ionization emission lines are common (e.g., Hamann et al. 1997; Tytler & Fan 1992). This point is further discussed in Sect. 6.1.
Other prominent emission lines in the spectrum of A are: OVI (blended with Ly) and probably CIII and SIV . In the spectrum of B, the BAL troughs allow clear identification of only Ly and NV , but the "undulating" shape of the observed flux for Å suggests that SVI , CIII , NIII , OVI , and SIV have broad absorption troughs on the blue side of the expected QSO rest wavelengths.
2.3. Continuum fitting
Due to line blending in the Ly forest, defining a quasar continuum shortward of Ly emission is not trivial at our resolution. For B, this difficulty is increased by the BAL profiles. A power-law shaped continuum , although an acceptable representation in the optical range, is definitively not able to fit even the UV continuum spectrum of quasar image A, leaving residuals far larger than 3 at apparently featureless spectral regions.
We decided to fit the UV QSO continuum for each spectrum with cubic splines, using the following simple and semi-automated fit algorithm. First, cubic splines are fitted through a given number of points thought to represent the continuum, typically 50 in each spectrum. The point ordinates are then corrected by an amount, such that the mean of the differences between the new ordinate and all the flux values within (defined by the first fit) in a 6 Å wide spectral window around equals zero. The corrected are used to perform a new fit. This procedure is iterated until it converges to a reasonable continuum, provided enough featureless regions are available in the data. The fit routine is robust in the sense that different start values always lead to the same continuum, but the spectral windows need to be carefully chosen, especially in the B spectrum. The final continuum is shown as dotted line in Fig. 1. Note that we have not attempted to model the QSO intrinsic continuum of B, which is evidently distorted by the BALs. Division of the flux by this continuum results in the normalized spectra of HS 1216+5032 A and B.
© European Southern Observatory (ESO) 2000
Online publication: May 3, 2000