This thesis focuses on the rest-frame optical emission line properties of a small sample of galaxies selected through a large set of methods. By selecting galaxies with a broad range of properties, the diverse set of physical processes is investigated that are likely to influence the evolution of galaxies such as merging and feedback from an active galactic nucleus (AGN). The galaxies were chosen to have redshifts where the important optical diagnostic lines fall into one of the near-infrared windows of the J, H and/or K bands – namely they must lie between z∼1.5−3.5 (see Fig. 3 of Chapter 1). It is both important to study galaxies with a wide range of selection methods and through the measurement and analysis of their optical emission lines. Given the large number of physical processes that might influence or control galaxy evolution, it is important to study classes of objects exhibiting a range of phenomenologies. Astrophysically, this means that the targets were selected with a variety of techniques, such as by their submillimeter flux, emission lines, or UV continuum emission, or by their radio luminosity, to identify galaxies hosting a powerful AGN. In addition, selecting high-redshift galaxies with redshifts such that the optical diagnostic lines are in the near-infrared atmospheric windows means they are studied with the same techniques and analyses that are used to investigate local galaxies. This allows a direct comparison of galaxies at low and high redshift. Moreover, studying galaxies in the rest-frame optical as opposed to the rest-frame UV means that even heavily obscured galaxies can be included with some confidence, and that it is possible to investigate their intrinsic properties instead of being overwhelmed by the high and variable extinction.
Besides some very significant astrophysical advantages to studying galaxies in the near-infrared (NIR), there are also some practical advantages. The smearing of the image, due to turbulence in the Earth’s atmosphere (“seeing”), is strongly wavelength-dependent (with coherence lengths
∝ λ6/5; Lena et al. 1998), so that the loss of spatial resolution due to the turbulent blurring is comparably small in the NIR compared to optical wavebands. Given that typical sizes of even the most extended high-redshift galaxies do not exceed a few seconds of arc, this is clearly a significant advantage. Spiffi’s coarsest pixel scale is 0.2500, well below the typical FWHM of the seeing disk
(&0.500). Hence, apart from observations using adaptive optics, the spatial resolution is limited by
the seeing.
Unfortunately, there are several difficulties in observing high redshift galaxies generally and in the infrared in particular, especially given the goal to obtain spatially extended information. This implies reaching the lowest possible surface brightnesses (O(10−16) W m−2µm−1 ¤00.). Due
to their large distances their integrated fluxes will be low. High redshift galaxies are difficult to resolve. This is both because there may be intrinsic size evolution (e.g., Bouwens et al. 2004) and cosmological surface-brightness dimming (Jz/Jz=0∝(1 +z)4). As stated previously, the important diagnostic optical emission lines fall at wavelengths between 1 and 2.5µm. Near-infrared detectors have historically had some significant limitations compared to optical detectors. Detector cosmetic quality is sometimes not good, with significant numbers of hot and dark pixels which can make data reduction especially difficult for faint objects. They also have had high read noise and dark current which limited their sensitivities (i.e., they are not background limited) between the OH lines and at moderate and high spectral resolutions and small projected pixel scales. The noise of the arrays is compounded by the atmosphere, which at near-infrared wavelengths is not very benevolent, the strong OH night sky lines can vary on timescales of a few minutes. This limits the integration times to 5−10 minutes and thus it is sometimes not possible to over-come the readnoise of the array with long integration times and limiting the performance of the instrument. Besides the emission lines, the transmission is variable, significant over large fractions of the near-infrared, and limits the observations to 3 near-infrared wavebands (J-, H-, and K-bands due to wavelength- dependent absorption of CO2 and H2O). In addition, for wavelengths beyond ∼2.2 to 2.4 µm, thermal background emission becomes a significant source of noise. How significant depends on the altitude of the observatory (the emissivity of the atmosphere), the emissivity of the telescope structure and mirror, and the temperature, thermal homogeneity, and baffling of the instrument.
There are techniques which can mitigate against the brightness and variability of the night sky emission in the near-infrared. Rest-frame optical emission line fluxes of high-redshift galaxies are
O(10−16) W m−2µm−1, a factor∼100 fainter than the night-sky emission. As a consequence, ob- servations are differential, requiring a robust (and nearly instantaneous, due to the time variability) estimate of the night sky emission, to subtract the telluric signal (“night sky subtraction”). This can either be carried out by taking separate “sky frames” or (if the instrument has a sufficiently uniform response all across the field-of-view) by using the regions within the field of view that are devoid of object emission. Both approaches have been used for this work.
Observing faint sources, such as high-redshift galaxies, requires total integration times of sev- eral hours, even with 8-m-class telescopes and the most sensitive near-infrared detectors available today. This is even more a constraint, when one aims at spatially-resolved observations, as is the case here. Hence, a sequence of individual exposures (typically 4−8) will be taken and reduced separately, and in a final step, all of the individual sequences will be combined to yield the final, high-sensitivity, data set. For most of the high-redshift sources signal-to-noise ratios in the single frames are near S/N = 1, so that the source cannot be discerned. As a result, good pointing stability of the telescope, and excellent mechanical stability of the instrument (within fractions of an arcsec, depending on the pixel scale), are mandatory to allow to “blindly” combine these frames. Because the absolute pointing precision is not sufficient, a nearby (within 12000 of the target), com-
parably bright star (K = 15 mag at least) is used for the absolute telescope positioning. After the star’s position within the field of view is known, the telescope can be offset accurately to the target of interest.
Ideally, the telescope pointing will not be identical for all individual frames. Spatial “dithering” (i.e., offsets by a few pixels between individual exposures) allows for the mitigation of static pixel-to- pixel variations across the detector or illumination differences within the field-of-view. In addition, combining total exposures from individual frames with shorter exposure times will provide a good means to identify cosmic rays or temporally variable pixels. For this work, sequences of 4−8 individual exposures (“templates”) were typically taken with predefined dither patterns. If longer exposures were necessary (as was usually the case) then these templates were executed repeatedly.
To optimize the alignment inbetween different templates, it is necessary to blindly co-add all frames taken within the same template, and then derive a template-to-template offset by cross-correlating the images (either the image plane of the collapsed cubes, the continuum image or the line image, depending on the characteristics of the source). This is because templates were not necessarily executed sequentially and may have been executed on different nights. Thus the same pointing could not be guaranteed between templates.
In the GI data, the overall pointing stability was not sufficient to allow blind registration of subsequent on-off pairs at different dither positions. Luckily the dither pattern was such that two subsequent on-source frames were obtained without offsetting the telescope. In all sources, the line emission was bright enough to allow cross-correlation of the combined line images of these two frames, or, in some cases, to orient on bright foreground sources. Observing dates and total exposure times for the objects studied in this work are summarized in Table 1.
Most of the data included in this thesis were taken by theSpiffiteam during the commissioning and an early “guest instrument” run in spring 2003, using the UT1 of the ESO-Very Large Telescope on Paranal. As described in Chapter 2, Spiffi was then brought back to Garching, optimized, furnished with a new (but still equipped with an engineering-grade) detector, and permanently installed on UT-4 of the VLT in summer 2004. In early 2005, this second engineering grade array was replaced by a new, science grade detector.
In spite of these changes, the overall data reduction strategy remained mostly the same, al- though with each subsequent array, reduction became somewhat simplified by the improving cos- metic quality. In many steps, reduction of Spiffidata does not greatly differ from the established reduction schemes used for near-infrared longslit data. It therefore follows the widely used IRAF longslit reduction package (Tody 1993), and was extended by R. Davies in 2003 to meet the special requirements of an integral-field device. However, this initial approach was mainly meant to provide a quick reduction scheme for initial inspection of the data, and not so much for a detailed analysis, so that a few adaptations were necessary afterwards. These improvements included enhancing the robustness of the wavelength calibration, and avoiding a few artifacts introduced by the reduction algorithm. Those changes did improve the quality of the reduction, as will be shown explicitly on the example of an artificial ”toy galaxy” (with a priori known spectral and spatial properties) in Section 5. To illustrate the differences between the old and the new approach, the initial reduction scheme will be quickly described in that context.
A C-based package has been finalized in the meantime, and is, e.g., the basis of the ESO calibration pipeline (Abuter et al. 2005). However, it was not yet available at the time when most of the data for this thesis were reduced, and cannot be used to reduce GI data. For consistency, the GTO data included into this thesis were reduced with the same algorithms as the GI data, with some modifications where necessary. No detailed comparison was tried between the data reduction method used here and the package of Abuter et al. (2005), because the GTO data do not form the backbone of this thesis. However, a quick (eyeballed) comparison of data ofz∼2 BX galaxies reduced with either method in another context than this thesis, does not indicate strongly different performances or systematic differences between the two approaches.
As a first step, the data reduction algorithms will be described that were used for this work. Section 2 gives an overview of the scheme developed to reduce the GI data, and Section 3 will describe the changes that were necessary for the GTO instrument.
2. Guest instrument data
Spifficommissioning and guest instrument (GI) runs were carried out between February and April 2003 on UT1 of the ESO VLT. Due to the overlapping with the instrument’s commissioning, observations were done by theSpiffiinstrument commissioning team.
AlthoughSpiffiuses a technique which is still novel in combination with the 10-m-telescope class – image slicing – to obtain spatially-resolved spectra of a contiguous field of view, data reduction schemes can be developed that are very similar to those commonly used for longslit spectroscopy. Any use of already available packages will have to be extended, however, to account for the uniqueness of the instrument. The basic algorithm, that was used to reduce the GI data and hence the main part of the data used for this work, is illustrated in the middle column of Fig. 1. Dark subtraction
The initial step is to remove the dark current from the individual exposures. As part of the day-time calibration plan, exposures are taken with the same duration as used during the night and with one of the filter wheels in the “closed position”. Since the filter wheels are in thermal equilibrium with the cryostat, which itself is at a low enough temperature to not emit significantly in the near-infrared, the detector is not exposed to any light. Matching integration times with the object exposures is obviously important to provide an accurate estimate of the dark current but also because the number of “bad pixels” depends on integration time. Subsequently, in the data reduction process, these bad pixels need to be accounted for as to not add noise or to be mistaken for signal in the finally reduced frames.
Flat fielding
After the dark current is removed, it is necessary to correct for large-scale variations across the field-of-view. In this step, the data are “flat fielded”. Flat field calibration frames are taken during daylight with the instrument set in exactly the same configuration as for the object frames and are made by observing a uniformly illuminated surface. This surface can either be the lamp-illuminated dome of the VLT or an integrating sphere observed within the AO module. During the period when most of the data presented in this thesis were taken, only the illuminated dome was available for taking the flat field calibration frames. Residuals in the illumination pattern of this “flat” image are used to scale the response in each region of the detector accordingly for all individual frames. Bad pixel correction
To account for static “bad pixels”, i.e., pixels with anomalous response (identified in either the dark frames or in the flat-fields), a bad pixel correction is applied, replacing the “bad” values by the averaged values of the 26 neighboring pixels in the three-dimensional data cube (which for this use is temporarily reconstructed at this step and then re-transformed into the original raw-frame). Wavelength calibration
The individual frames are then wavelength-calibrated. To account for some spectral flexure between the frames, the frames are rectified and wavelength-calibrated using both arc-lamp spectra (taken during day-time calibrations) and the measured OH lines in each frame1. Two columns are clipped on either end of each slitlet to avoid overlapping flux between slitlets that lie adjacent and thus improve the accuracy of the calibration, which is generally ∼1 ˚A in the GI data or within ∼20% of one spectral pixel. In later data, they amount to 0.1-0.2 ˚A. To rectify the frames, the algorithm collapses the spectra within each slitlet, assuming they are exactly 32 pixels long. Although this is not strictly correct, it is of sufficient precision for the GI data. (With the final design and
Fig. 1.— Flow-chart of the various IRAF-based Spiffi data reduction algorithms. Gray-shaded boxes indicate steps common to both approaches. Ovals in the left column name steps that are only in the original scheme, ovals in the right column those that were introduced with the updated method developed for this thesis.
the larger detectors, however, an additional step is required to determine slitlet-lengths, which will be described in section 3.) Arc-lamp spectra are used to define the absolute wavelengths of the night sky lines measured in the frames. The algorithm then calculates individual dispersion solutions in each slitlet by fitting Legendre polynomials. Shifts between slitlets are determined by cross-correlating the night-sky lines.
Night sky subtraction
To mitigate against variations in the night sky emission, which are strongly wavelength-dependent, each sky frame was normalized to the average of the object frame separately for each wavelength. If bright sources were within the field-of-view, they were explicitly masked before calculating the average, applying a sigma-clipping algorithm. In the following, the empty “sky”-frames were sub- tracted from the according “on-source” frames. In spite of the normalization of the background frame, some OH lines show strong residuals in the sky-subtracted frames, leading to spikes that ex- ceed the signal strength by far. They were corrected by averaging over neighboring spectral pixels, when a pixel differs strongly (>20σ) from the average of the surrounding spectral pixels, and has a wavelength close to that of a strong OH line. Thus the strong OH emission lines are clipped. (The better quality of the science grade array installed in the winter of 2005 made better sky-subtraction possible, so that for the data taken with the new array this step was not necessary.)
Cube reconstruction
The three dimensional data cubes were then constructed from the raw frames, assuming that each spatial row of the field was exactly 32 pixels long, and allowing only integer pixel shifts. Although this is not strictly correct, spatial resolution was slightly reduced. However, this method avoided loss of signal–to–noise due to resampling the data.
The cubes were then spatially aligned by cross-correlating the collapsed spectra. Again, maxi- mum signal–to–noise was retained using only integer pixel shifts between the individual data cubes used in constructing the final data set with the total integration time. Depending on the individual source, either line- or continuum-images were used for cross-correlation, or the full data cubes. When combining the cubes deviant pixels were clipped.
3. Guaranteed time observations
Additional data were taken during early guaranteed time observations (GTO) in November and December 2004 and March and April 2005. In the 2004/05 runs, data were only taken in the K band. Individual and total integral times are summarized in Table 1.
Typical source diameters do not exceed a few seconds of arc, so that they occupy only a small area within the field of view. By adopting a dither pattern such that the source falls into distinct quadrants of the field of view, pairs of subsequent frames can be used for background subtraction. Spiffi’s 2nd engineering- and science-grade detectors have a very uniform sensitivity across the entire field of view, which coupled with SINFONIs uniformity of field, variations in the illumination can be neglected relative to the total sky background. Both properties are crucial for this method, because the background subtraction is done using different areas within the field of view (and thus different regions of the detector). As a benefit, overheads are obviously greatly reduced. Instead of requiring ∼2.5 hours of telescope time for an hour on-source as in the GI run, in the GTO observations an hour on-source data were taken in ∼1.25 hours (neglecting the time required to observe spectroscopic standard stars which can be done during twilight).
reduction scheme, although the overall strategy did not change. The most important change was in the size of the science grade array compared to the two engineering grade arrays. The science grade