Context and Motivation
In a recent paper, our group published the identification of a z=10.0 galaxy strongly magnified by the lensing cluster A1835 (Pello et al. , 2004a, A&A, 416, L35, hereafter P04). This galaxy is one of our first priority targets in a long term pilot project conducted at ESO/VLT, aiming at identifying and studying sources beyond z>7 using lensing clusters as natural telescopes. Galaxies are selected using photometric criteria based on broad-band near-IR and optical photometry in combination with the traditional Lyman drop-out technique. The magnification in the core of lensing clusters improves the search efficiency and subsequent spectroscopic follow up.
The spectroscopic confirmation of this source (A1835/#1916) was based on the detection of a faint emission-line obtained with ISAAC/VLT between 29 June and 3 July 2003, under excellent seeing conditions, using a 1" slit width. The observed spectral interval covers the range from 1.162 to 1.399 micron. The observations resulted in the detection of one weak emission line at the 4-5 sigma level above the neighbouring background, with an integrated flux of (4.1 +/- 0.5) 10^{-18} erg cm^{-2} s^{-1} at a wavelength of 1.33745 micron (see Fig. 1 below). This line appears on 2 different overlapping wavelength settings (1.315 and 1.365 microns). When identified as Ly alpha, in good agreement with the photometric SED, the observed line gives a redshift of z=10.00, the most likely one given the data set available at this epoch (see a detailed discussion in P04). Further observations by our group and other authors indicate that the nature and precise redshift of this source are still unclear, i.e. we no longer consider this source as a genuine z=10 galaxy. Taken together, all spectroscopic and photometric detections, albeit individually of relatively low significance, indicate that this source is most likely not a spurious, but an intrinsically variable object, as discussed in Richard et al. (2006,astro-ph/0606134) (see also Pello et al. 2004b [astro-ph/0410132], Bremer et al. 2005, Lehnert et al. 2005, Smith et al. 2006).
The original ISAAC/VLT data are public since the end of June 04. In a recent astro-ph posting of a submitted paper, Weatherley et al. (2004, astro-ph/0407150) re-analyse the archival spectroscopic data from our observations of Abell 1835 IR 1916. The conclusion of their paper, submitted to MNRAS, is that the emission line is not detected in their analysis, neither in the 1.315 nor in the 1.365 micron bands. Although their reduction procedure is accurate and robust, there are several important differences between their procedure and ours. We have investigated these differences in order to understand the origin of the discrepancy. The first difference (discussed in Weatherley et al's paper) is the way of dealing with distorsion and wavelength calibration of 2D spectra. According to our tests, this factor could hardly been responsible for our detection (or not-detection) differences. On the contrary, the main difference we found is in the way of stacking the spectral frames. This difference alone seems crucial to understand the contradictory results. This difference is related to the way of recentering the compact target sources (seeing-limited objects, ~3 pixels FWHM in the spectral direction) on the 1" wide slit (~7 pixels). Weatherley et al.'s procedure as given in the paper neglects this effect. When we use (as explained in Pello et al. 04) the [OIII]5007 strong emission line of a reference compact object (located on the central region of the slit) to track and to correct for drift across the slit, the line is detected again, even when the combination scheme adopted is as close as possible to the one proposed by Weatherley et al. (i.e., to correct for distorsion/ wavelength at the very last stage of the reduction process). It is important to correct for drift effects across the slit when dealing with faint seeing limited sources as the ones studied here.
The Weatherley et al's paper was finally published in A&A after revision (2004, A&A 428, L29). According to these authors, the final explanation for P04 detection is spurious positive flux introduced in the sky-subtraction stage as a result of variable hot pixels. On the other hand, we have re-analysed our spectroscopic data, and we found an error in the absolute wavelength calibration of our extracted spectra: the actual position of the signal is 1.33790 microns, i.e. ~4-5 A shift with respect to the position given in P04. This fact does not modify neither the conclusions of P04 nor the puzzling nature of this source but, together with the stacking issues mentioned above, it could explain the discrepancy. Note that in Weatherley et al's procedure the information is only preserved at the original (wrong) P04 wavelength position and smeared elsewhere.
The seeing/slit-width and recentering issue was already mentioned in P04, but it is difficult for archive users to take into consideration all the important points such as the ones discussed here. On the other hand, although the wavelength shift in the line position is small, it is worth to take it into account. The aim of this brief report is to help researchers to re-analyse and to discuss independently the ISAAC spectroscopic observations of this faint source. We summarize below:
Reduction
Procedure used in P04
The reduction of our ISAAC data presented in P04 was performed according to a classical scheme, using IRAF procedures and conforming to the ISAAC Data Reduction Guide 1.5 (http://www.hq.eso.org/instruments/isaac/index.html), using the same procedure described in Richard et al. (2003, A&A 412, L57). Spectra were obtained in beam-switching mode between two positions A and B, following the usual sequence ABBA. In summary, these were the steps included of the reduction process of #1916, for the 2 frame sets corresponding to the 1.315 and the 1.365 micron bands:
Figure 1 : 2D spectra presented in P04 showing the detected emission line of #1916, as well as the nearby field galaxy and the [OIII]5007 line galaxy #2582 (z=1.68) used as a reference to stack the frames. 2D spectra are sky subtracted and show the spectral region around the emission-line at 1.33745 micron (using P04 reference; it is actually around 1.33790 micron), leading to z=10.0 when identified as Lya (1215.67 A). The line is seen on the 2 independent overlapping bands at 1.315 and 1.365 micron. No smoothing was applied to these spectra.
Seeing and slit-width effects
Given the excellent seeing conditions during the run, with
a target much smaller than the slit width, our observations were basically
seeing-limited. This fact has an important implication for positionning
in lambda: a shift in position across the slit translates into
a shift in the spectral direction. For this reason, given that
object #2582 (see Richard et al. 2003A&A 412 L57) is located on
the central region of the slit, and it has strong enough e-lines ( in particular
[OIII]5007, which is close to our target in the spectral direction -see Fig.
1-), we used the position of this emission line to recenter the final stack
on #1916. Indeed, when the frames are recentered using the OH sky lines only,
the SN of the final composite [OIII]5007 line itself deteriorates. The effect
is more dramatic for a very faint emission line.
Figure 2 provides an histogram of the residual differences
between the position of the [OIII]5007 line and the neighbouring OH
sky line, for the 21 frames in the 1.365 micron band, measured on frames
at steps 4 and 5, for the sky and the emission line respectively. The rms
of the distribution is 0.98 pixels, but the distribution is not gaussian.
For comparison, the rms of the distance between 2 OH lines is of the order
of 0.1 pixels. With a conservative gaussian approximation, the dilution of
the energy measured at the maximum of the emission line will range between
20% (0.5" seeing) and 30% (0.4" seeing) assuming ideal conditions
. But the distribution is not gaussian (there is no reason to assume a gaussian
distribution in this case), and the IRAF rejection schemes used for stacking
will worsen the situation.
Figure 2 : Histogram of the residual shifts due to slit-drift between the reference [OIII]5007 line and the neighbouring OH sky line, for the 21 frames in the 1.365 micron band.
Results and Discussion
There are several important differences between the reduction procedure adopted by Weatherley et al. and ours. The main ones are the following:
In order to investigate the effects of these choices on the
final results (and discrepancies), we have adopted a combination scheme
as close as possible to the one proposed by Weatherley et al. (i.e., to
correct for distorsion/ wavelength at the very last stage of the reduction
process). This means stacking the frames after step 6 above, using only
integer number of pixels for recentering both spatially and in the lambda
direction, but using the [OIII]5007 line to correct for a residual
drift. Because of the 2D distorsion at this stage (sky lines are curved
on the image), this stacking procedure is not rigurously correct as it is
when using distorsion corrected images (as in the "classical" method used
in P04). We consider only the spectral region close
to #1916 but, instead of recentering in lambda on the closest OH line alone,
we include the shifts measured on the [OIII]5007 line. If there is a systematic
trend on #1916 in this procedure, it will go in the sens of diluting the
signal. We have checked that he relative distance between the 2
neighbouring OH lines redwards from the [OIII]5007 line remains
the same (within 0.1 pixels) all along the spatial direction.
The first results are shown below (Figure 3), on the
stack of the best 16/21 frames in the 1.365 micron band. These were the
frames with the highest weigth in this band. Even without any particular
optimization, the emission line shows up as in our previous paper (P04),
although with a lower S/N. The S/N measured for the peak of the line on the
smoothed image (with respect to the 10x10 pixels neighbouring region) is
3.7, without any particular optimization at this stage.
|
|
Figure 3 : 2D composite spectra corresponding to our first test using the 16/21 best frames in the 1.365 micron band, showing the re-detected emission line of #1916, as in Figure 1, together with a gaussian-smoothed image (1.5 pixel) around the emission-line (SN~3.7), without any particular optimization.
Absolute wavelength calibration in P04 extracted spectra
We have re-analised the spectra used in P04, and we came to the conclusion that the absolute wavelength calibration of the extracted spectra was slightly off in the original paper: ~4-5A, equivalent to 7-9 pixels in linear wavelength scale, depending on the band, leading to a line actually detected at 1.33790 (+/- 0.0001) microns instead of the 1.33745 microns published by P04 and used by Weatherley et al. (04). The line is still detected in the two independent bands. The final 2D spectra are similar to the ones presented above (P04) and fully consistent with the results published by P04: S/N between 4 and 8 depending on the band and aperture considered. Different rejection schemes were used with similar results (e.g. IRAF sigma-clip or minmax rejection with nhigh=4, i.e. different rejection schemes wich are intended to remove the hot pixels reported by Weatherley et al.).
In the method used by Weatherley et al. (04) to combine their spectra, the information is preserved only around the pixels of interest, and smeared elsewhere. Thus, this shift could be responsible for their non-detection, or at least seriously contribute to their non-detection, together with the other stacking issues mentioned above.
Conclusions
The effect of an accurate recentering in wavelength before stacking seems to be more important than rebinning effects, since we re-detect the emission line on our non-distortion/wavelength corrected stack. On the other hand, correcting for drift effects across the slit is an important issue when dealing with faint seeing-limited sources such as the ones studied here. However, the main problem in the present case seems to be an error in the absolute wavelength calibration of P04 extracted spectrum, discovered afterwards as reported above, combined with the fact that in Weatherley et al.'s stacking the information is only preserved at the original (wrong) P04 wavelength position, and smeared elsewhere . The reduction procedure used by Weatherley et al. in their paper seems to be an excellent approach for the detection and analysis of faint e-lines in 2D spectra, provided that the 2D reconstruction is performed on the full spectral and spatial directions.