Observations and Data Reduction

We have used the 4.1m SOAR telescope and Goodman Spectrograph (Clemens et al. 2004) to carry out deep longslit or image slicer spectroscopic observations of our RESOLVE S0 sample. Typical major axis exposure times of∼2 hours, carried out primarily in “darkest” sky conditions were necessary to measure stellar kinematics to sufficient galactocentric radii to identify extended secondary disk structures in our targets. Our standard observing pattern involves multiple 1200s major axis exposures with a comparison lamp spectrum between each exposure. To mitigate the effects of atmospheric differential refraction (see Filippenko 1982), we confine our typical observations to airmass<2 and further to airmass<1.2 for the broadest wavelength coverage observations (taken in the 1200 l/mm grating setup; see Table 4.1). Strict airmass requirements combined with long exposure times necessitates spanning most of our observations over multiple nights. On each night, we take bias, flat field, and scattered light “pseudo-dark” calibration frames. These pseudo-dark

calibrations are exposures taken in a dark dome with exposure times ideally matching the maximum exposure times of our science frames and were needed during periods when stray light signatures were observed in Goodman Spectrograph data. Since the level and structure of the stray light were both constant over the course of a given observing night, except on A-semester 2011 nights as noted in Table 4.1, subtracting these pseudo-dark calibration frames from the science frames in our data reduction procedure allowed generally clean removal of stray light from our data.

Our instrumental setup was chosen to target a wavelength range covering 4500-5550 Angstroms at minimum, allowing the examination of absorption line features that trace the stellar kinematics (Hβ, Mgb, Fe5270 and Fe5335). Due to evolving capabilities of the Goodman Spectrograph, we have used a variety of grating and slit combinations to achieve this wavelength coverage and our desired resolution targets. Our typical instrumental setup used the Goodman Spectrograph’s 2100 l/mm VPH grating and 1.03” longslit, with resulting typical FWHM resolution∼1 Angstrom or equivalent velocity resolution ofσ∼25 km/s. A wider 1.68” longslit was chosen for the most massive targets so that their larger velocity widths could be similarly well sampled but observed more efficiently, requiring less observing time per object. Before the 2100 l/mm grating was installed on the Goodman Spectrograph, different longslits were chosen to approximate these same resolution constraints while using a 1200 l/mm grating. Lower-resolution observations taken prior to the B-semester of 2010 will also be used in forthcoming analysis for this project as appropriate (see full setup details in Table 4.1).

At the time of observation, initial binning in the spatial direction to ∼0.3” pixels was applied, which is well within the typical minimum seeing of∼1” on our observing nights. Further spatial bin- ning is applied in post processing using an adaptive binning algorithm that bins together individual spectral rows of our final reduced frames until a specified continuum S/N target is achieved. In this analysis, we use a S/N target of 10 per Angstrom, estimated using the final error frames that result from our data reduction procedure (see description below). Even though absorption line kinematics studies often target S/N&20, lower S/N levels are reasonable for the cross-correlation analysis we perform (e.g., Kregel et al. 2004 and references therein), and several spatial sampling points are averaged together when interpreting our velocity dispersion measurements. Each adaptively binned spectral row we analyze is considered to lie at the galactocentric radius at which half its total flux was reached.

spectroscopic data reduction, which consists of both customized IDL reduction routines and wrap- per codes for standard IRAF routines. This pipeline has been optimized for processing Goodman Spectrograph data taken in stock RESOLVE survey configurations, and minor modifications to the pipeline procedures have been made to allow processing of our deep S0 data, with its variety of configurations that differ from typical RESOLVE survey products. The data reduction procedures include standard methods of overscan/bias subtraction, flat fielding, and wavelength calibration along with custom methods of exposure alignment and stacking, spatial curvature rectification, and sky subtraction originally developed by Kannappan (2001). We also apply cosmic ray removal using the Laplacian cosmic ray identification algorithm of van Dokkum (2001, L.A.Cosmic), which allows cosmic ray rejection even when insufficient numbers of exposures exist for rejection during the frame stacking process. For data taken on observing nights with detectable stray light, we subtract a com- bined pseudo-dark frame from the science frames. When pseudo-dark frames have been taken that match the exposure times of the science frames, we subtract the pseudo-dark levels in these frames directly, but otherwise we scale the pseudo-dark signal by the ratio of the science exposure time to the pseudo-dark exposure time. After pseudo-dark subtraction, we inspect the resulting science frames for under- or over-subtraction. If under- or over-subtraction has occurred based on the sim- ple exposure time scaling of the pseudo-dark frame, we rescale the pseudo-dark frames accordingly and repeat the subtraction procedure. This approach typically removes the stray light signal from our data cleanly, adding only to our noise levels, except in cases noted in Table 4.1 where variable stray light levels and/or structure yield low-level residuals in the science frames after pseudo-dark subtraction.

We carefully track all sources of error associated with each final reduced science frame by com- bining error frames that include the bias variations, Poisson errors on pseudo-dark levels (where applicable), flat-field variations, and Poisson errors on signal in each science frame. The combined error frame associated with each science frame is then transformed in the same manner as the data when a wavelength solution is applied. The resulting error frame is also rectified in the same manner as the data when the spatial curvature correction is applied. Individual error frames are aligned and combined, adding in quadrature, to match the corresponding final stacked science frame. These final error frames are used in our adaptive spatial binning procedure as previously described.