This note attempts to study one of the problems that lead to the kind of error distributions that are illustrated by Figs. 4 and 6 of TN-88. The I data runs into an error floor at the brighter magnitudes. This is really not a bright magnitude problem, it is just that other errors dominate for the dimmer stars. All the examples below are done with I camera data. The V camera has similar problems but they are less in magnitude than those of the I data.
First we drag out the light box and take some flat data.
After we run it through the pipeline and subtract darks we take some representative points. Sorry for the small number of points. This was done by moving a cursor by hand and reading off the values. The files are just too big to pass around. The error is the difference between the 1000,1000 center and the indicated position. Here is data for the I lens: (file boxi.fts)
Position Value Error 50,50 12839 -0.13 corner 50,1000 14172 -0.05 side center 50,1950 13185 -0.11 corner 1000,50 14255 -0.04 side center 1000,1000 14876 -0.00 center 1000,1950 13981 -0.06 side center 1950,50 13489 -0.09 corner 1950,1000 14016 -0.06 side center 1950,1950 13245 -0.11 cornerNotice that the flat is pretty symmetrical. There is some structure that looks like off centered rings that I attribute to the slightly tilted camera. Any one who has made a similar measurement with a camera lens will appreciate Elliot Burke's design. About a factor of 5 less droop in the corners compared to an expensive camera lens. We shall have to wait to see if the stiffeners that I have on order will fix the tilt. In any case, I am fixing it in the Mark V design.
Next we look at the flat made from the I data of DS24, run3: (file 7x7n3.fts)
Position Value Error 50,50 2575 -0.09 50,1000 2818 -0.01 50,1950 2516 -0.12 1000,50 2757 -0.03 1000,1000 2852 -0.00 1000,1950 2665 -0.07 1950,50 2507 -0.12 1950,1000 2609 -0.09 1950,1950 2480 -0.13
We can now observe a gradient in the data. The x = 2000 side of the chip is the north side. The north side is darker.
We can emphasize the tilt by dividing the DS24 run flat by the box flat:
Position Value Error 50,50 2791 +0.04 50,1000 2767 +0.04 50,1950 2761 +0.04 1000,50 2673 +0.00 1000,1000 2660 +0.00 1000,1950 2652 -0.00 1950,50 2586 -0.03 1950,1000 2590 -0.03 1950,1950 2606 -0.02
The division removes the bright center and we can much better see the gradient. OK, the North side is the dark side of the flat. The appearance of the flat is just dark on the North side tapering to light on the South side. There is very little East-West structure.
Next we take some recent data where the telescope is moved in Declination. Here the telescope is indexed to even 4 degree centers. The field centers are on even 4 degrees in RA, and this run covered -4 to +12 in declination. We took the data and (by hand) sorted out the runs which were taken at -4 and +12. For the particular data set there were 16 of each. Enough to get a fair master sky flat. Looking at the -4 degree data, we see that the North side is again the dark side. Looking at the +12 degree data, this is not so obvious.
We now take the flat made from the -4 degree data and divide it by the flat made from the +12 degree data. This removes the bright center of a simple master sky flat. The result shows that the North side is still the dark side. This means that the -4 degree flat has a larger gradient than the +12 degree flat. Not a big surprise if one looks out the window. (files m4i.fts, p12i.fits)
Position Value Error 50,50 8881 +0.02 50,1000 8915 +0.03 50,1950 8759 +0.01 1000,50 8810 +0.01 1000,1000 8697 +0.00 1000,1950 8589 -0.01 1950,50 8570 -0.01 1950,1000 8729 +0.00 1950,1950 8739 +0.00
Here the specific data points don't quite tell the story. A bright corner from the output amplifier masks the true value of the 1950,1950 point. The North side is still the dark side. The North side becomes less dark as the declination increases.
If someone wants to study this further, I can supply the flats on a CD.
There is a gradient in a master sky flat made from a median of the I images. The North side of such a flat tends to be dark. The North side gradient is larger at lower declinations. This is not really a surprise. All you have to do to verify this is look out the window.
We have been making a master flat from the evening's run. Just using all the data. This is not advisable since the N-S gradient changes with declination. Most of the data taken through November of 2002 was taken at a single declination, so the variations of the measurement with RA position in the frame cannot be explained by the declination gradient. Sigh! This is just one problem.
I propose to sort the raw data by declination and then run the pipeline separately for each declination. Possibly we can generate a set of flats for each declination (as the present program does for dark exposure time) and have the program select the proper flat.