A Statistical overview of the wavelength calibration accuracy
These figures show the errors in the wavelength calibration. The inset in the top left corner shows that at most times the anchor point that was predicted using the lenticular filters taken just before and after the observation, for the first calibration has a systematic offset of -8.0 Å averaged over the calibration observations (blue dots). Ignoring the systematic offset, the anchor is found with a spread of 33Å. The contours give a better picture of where the calibration for the anchor position results in an offset in the wavelengths. In the corners and near the edges, only a few observations were available for the calibration, so the values should be taken as an indication of the size of the error, rather than an accurate value that can be used to improve the wavelength scale knowledge.
Looking at the second inset "accuracy", and disregarding the
systematic wavelength scale offset from the top left inset, wavelengths
from the dispersion relation are found to typically fall within 14 Å
(about 4 pixels) of the
correct value. The "accuracy" is the RMS dispersion of the dispersion
errors over the whole range as shown in the accuracy plots below in
more detail, and not the error themselves.
The variation of the accuracy of the dispersion relation over the
detector can be seen from the contour plot below. The contours are the
RMS values of the wavelength errors of known lines in the calibration
spectra. There is only a small variation over the face of the detector
present, with the largest discrepancies in the top left and bottom
right corners. This can possibly be improved by applying a bilinear
correction as a function of the detector position to the Zemax model
dispersion constants. In the UV nominal grism, a constant was used. In
the other grism modes which were done later, a bilinear correction was
used. 
The overall accuracy is generally sufficient to avoid making misidentifications of the larger spectral features which have a FWHM which is comparable in magnitude, although wavelength scale errors of up to 40 Å were seen for the calibration. For details, the actual fits given in the maps on this page should be consulted.
How to read the wavelength accuracy plots
Wavelength accuracy plots were made for each useful calibration spectrum. The format of the accuracy plots is to have two panels. Since the main variation of the dispersion can be represented by a constant with a linear term, the top panel of the accuracy plot has taken those two terms out. The higher order terms tend to zero near the adopted anchor point, which for the V-clocked grism mode is ~4200 Å. The dispersion from the scaled optical model has been approximated with a polynomial. The observed line positions from lines identified in the calibration spectrum have been plotted as blue dots. The lower panel of these wavelength accuracy plots shows the remainder after subtraction of the predicted position. The observed position was found from the pixel distance to the anchor point and the dispersion relation, while the predicted position was found using the known wavelength. If the model dispersion is good, the difference can be divided into an offset and a random looking spread around that offset. The values for that case are given in the plot. In some points near the edges of the detector, the scaled model dispersion deviates and the points will not evenly be distributed around some mean offset.The upper left corner (low det-X, high det-Y) has no first orders because of the aperture effect from rotating the filterwheel.
A map for wavelength accuracy plots
The map below provides links to the wavelength accuracy plots.
It should be noted that the contours are a bit misleading. Take for example the 57021002 observation in the top right corner. While the contours indicate a wavelength scale offset of +2 A, the actual value is -12.4 A. The reason is that there is an uncertainty in the position of the anchor point as determined here from the aspect solution based on observations in a lenticular filter. That uncertainty is due to a possible drift of the pointing during the observation which can typically be around 4 pixels, which for a dispersion of 3.2A/pixel translates to about 13A.
A Map for count rate spectra
Here the count rate calibration spectra can be found. The top panel as a function of fitted wavelength, the bottom panel as function of pixel coordinate. The stronger lines have been identified. Any shifts between the predicted line position and the actual spectrum can be seen in the top panel as an offset.
The count rate in the spectra can be seen to vary across the detector, which is partly due to variations in sensitivity, and perhaps also to variablility in the source, which was chosen for calibrating the wavelengths using the many spectral lines.
The calibration file used was swwavcal20090406_v1_mssl_ug200.fits. In some browsers the maps do not work. In that case the plots/images can be found here.
