3.3. Dark Current Subtraction
Dark frames were taken immediately before and after observations on each night, and 20 darks were taken at the start and end of the run, which were combined to make a "superdark," the mean level of which drifted by ~ 1.0 DN over the extent of the frame. Unfortunately, the 20-frame superdark is read-noise dominated and cannot provide a pixel-to-pixel correction. The superdark was therefore smoothed using a sliding 3 × 3 boxcar median filter, in order to avoid adding noise and to allow the removal of the mean dark level. The frames taken during the run were used to test the accuracy of the superdark: after bias subtraction (as described above) and dark subtraction, the test-reduced darks taken at the beginning and end of the observations on 1995 November 27 and 29 had a mean level of ± 0.25 DN with no coherent pattern.
3.4. Flat Fielding and Illumination Correction
We used a 1.5 arcsec slit for the program observations in order to preserve resolution. However, we used a 10.8 arcsec slit for the standard star observations in order to collect as much light as possible. Different sets of flat-field and illumination corrections were therefore required for two reasons. First, microscopic roughness on the edges of the slit jaws caused shadowing which changed as a function of slit-width. Second, the slit-jaw mechanism on this spectrograph is such that the jaws are parallel for separations up to ~ 5.4 arcsec (500µm) but are not parallel when the jaws are set to 10.8arcsec in the center. Variation in slit-width is almost 10% from end-to-end when the width at the center is 10.8arcsec. To compensate for the variable slit-width for standards, the illumination corrections for both slit-widths were normalized to the spatial center of the slit. Standard stars were all observed within two pixels of the central column used for normalization, which places them within 99.98% of the nominal value for the wide-slit illumination correction. Flat-field and slit-illumination corrections were created from dome and twilight sky flats, respectively, using the tasks RESPONSE and ILLUMINATION in the IRAF SPECRED package.
3.5. Wavelength Calibration
Wavelength solutions were based on He-Ar comparison spectra taken before and after each 1800 second program exposure and immediately after each standard star observation. At the beginning of the run, great care was taken to align the dispersion axis with the pixel rows: the rms variation in the centroid position of arc lines over the full 3.4 arcmin spatial extent of the images is typically 0.05 - 0.12 pixels. This is true over the full wavelength range. Consequently, it was not necessary to rectify the two-dimensional program spectra of the zodiacal light (night sky). Each program spectrum was simply averaged along rows (with 5 cosmic ray rejection) to obtain a one-dimensional spectrum, and a wavelength solution based on the spatial center of the image was applied afterward. Standard stars were first extracted to produce one-dimensional spectra and then wavelength calibrated in the usual way.
We identified 21-25 features in each comparison spectrum visually and fitted a third order (four term) Legendre polynomial to the pixel-wavelength solution to obtain a dispersion curve. This was done using the IDENTIFY task in IRAF. The rms residuals in the wavelength solution were 0.016-0.04Å. Shifts in the wavelength solution between program observations were less than ± 1.5 pixels from the start to the end of the night. Linear dispersion solutions were applied using the task DISPCOR.