3.1. Detector Linearity
Because our observations of the ZL have total count levels in the range 20-50 DN pixel-1, and the standard star observations have close to 5000 DN pixel-1, it is crucial to verify that the CCD is linear over this broad range. In these data, 16 rows are read off beyond the physical extent of the chip, averaged together, and recorded as a bias row, which can be used as an accurate diagnostic of the charge transfer efficiency (CTE) of the CCD. Because the slit only illuminates the central third of the chip, any residual charge which is not passed through the parallel gates (due to low CTE) will appear as a jump in the charge level of the bias row at the boundary between the exposed and unexposed regions of the chip. It is evident from this diagnostic that the temperature regulation of the chip became erratic during the third night of the run, causing an increase in the spurious charge and causing the CTE to drop to an unacceptable level (~ 99.995% per transfer, or 95.0% over 1024 rows). The data from this night were excluded from the analysis because charge shared between rows due to incomplete charge transfer will affect the apparent strength of spectral features. On the nights during which the temperature of the CCD remained stable, the mean level in the bias row was not detectably higher in columns illuminated by the slit than in those which were not illuminated. This was true even for dome flats images, in which the mean level of the illuminated columns was ~ 7000 DN pixel-1. In addition, our final results are based on data imaged in rows 1-700 of the CCD. The data which contribute to our final results were obtained with a minimum CTE of 99.9995% per transfer, or 99.7% over 700 rows.
Another possible cause of non-linearity at low count levels is deferred charge. We performed the standard tests for deferred charge as described in Gilliland (1992) and find that only 0.3% of the pixels showed deviations from linearity greater than four times the read-noise. Pixels exhibiting non-linearity were flagged in all images and excluded from analysis. We also performed a standard linearity test by taking dome flat-field exposures with integration times between 0.5 and 200 seconds and looking for variation in the detected count-rate between 20-20,000 DN pixel-1. The influence of lamp instability was minimized by taking several series of exposures and averaging the results. The detector response was linear to the limits of the sampled range which more than brackets the signal level of the standard star observations (peaking at roughly 5,000DN) and the program observations (20-50DN on average). Non-repeating deviations of less than 1% were attributed to the instability of the lamp.