Revised the orientations in the FullXnoTPD.sdf model to match Oleg's findings. I had a few errors too. The result (no TP calib or DCOPS calib) had a measurement chisquared of 20690 and an "object shifting" chisquared of 8516, half of which came from CCD shifts internal to the DCOPS (and 30% from non-dowel shifts with pulls greater than 5). That says that the side to side difference in DCOPS position is substantial. Perhaps I should loosen the tilt parameters? But Bob's work suggests that that may not help.
At any rate the transfer plate positions fit by Cocoa still don't correlate with the photogrammetry positions, though the range is now much smaller. If I rerun with the photogrammetry measurements applied as starting positions, the chisquared contributions are 20672 and 8778 respectively: the measurement chisquared getting worse instead of better. A pull of 5 for a DCOPS CCD shift represents a .325mm shift in position--much larger than calibration shifts, so I don't think this is attributable to uncalibrated DCOPS.
Inspection shows some of the large chi2meas contributions are from junk non-profiles, which should have been cut away.
Looking at fit info from the data shows an odd pattern:
In the following plot I only look at events with signal/background greater than 10 but where the A parameter of the gaussian (the height) is less than 200 counts. The sigma is large (greater than 200), and the signal small in height, leading one to wonder if the result is bogus.
I think that it may be useful to save an additional peice of info: the average value for a profile.
For comparison, this is the fit result for one profile:
MINUIT WARNING IN PARAM DEF ============== LIMITS ON PARAM1 TOO FAR APART. MINUIT WARNING IN PARAM DEF ============== LIMITS ON PARAM1 TOO FAR APART. MINUIT WARNING IN PARAM DEF ============== STARTING VALUE OUTSIDE LIMITS. MINUIT WARNING IN PARAMETR ============== VARIABLE3 BROUGHT BACK INSIDE LIMITS. MINUIT WARNING IN PARAMETR ============== VARIABLE3 BROUGHT BACK INSIDE LIMITS. HIST_417_1251259165_CCD1_UP 1395.51 51.9475 257.976 157.494 7.79996 12.9566 5043.85 63797.4 0 106.747 -0.0293545 7.64624e-06 912 1812 name center errcenter sigma errsigma A errA gausarea background area sbits p1 p2 p3 range_low range_high
MINUIT WARNING IN PARAM DEF ============== LIMITS ON PARAM1 TOO FAR APART. MINUIT WARNING IN PARAM DEF ============== LIMITS ON PARAM1 TOO FAR APART. MINUIT WARNING IN PARAMETR ============== VARIABLE3 BROUGHT BACK INSIDE LIMITS. HIST_393_1251259165_CCD1_UP 375.857 42.0848 458.592 57.6974 94.4991 7.86251 108628 7338.24 0 13.2278 -0.0443898 4.47473e-05 0 869 name center errcenter sigma errsigma A errA gausarea background area sbits p1 p2 p3 range_low range_high
As you can see, the A parameter is reasonably large for this: 94 counts; and the gaussian area looks ok: 109000. But the profile is junk (it isn't even a shadow). I don't want to fiddle with this stuff by hand, but I may have to if I can't find some decent signal/noise cuts. I have to allow for fairly large sigmas, but 459 may be too large. An example of a very wide profile follows: the sigma is 220 pixels. Perhaps a cut of 250 pixels will not do any harm.
MINUIT WARNING IN PARAM DEF ============== LIMITS ON PARAM1 TOO FAR APART. MINUIT WARNING IN PARAM DEF ============== LIMITS ON PARAM1 TOO FAR APART. HIST_393_1251259165_CCD3_DOWN 1370.28 3.69442 218.305 34.1089 128.132 48.6567 70114.8 132747 0 -55.528 0.136671 -2.93301e-05 909 1809 name center errcenter sigma errsigma A errA gausarea background area sbits p1 p2 p3 range_low range_high
Modified 21-September-2009 at 14:21
http://hep.physics.wisc.edu/~jnb/cms/21Sep2009
| Previous notes | Next notes | Main slide directory |