From: Christopher Tunnell (tunnell@mail.utexas.edu)
Date: Sat Sep 24 2005 - 16:51:41 CDT
(Note: If you just want plots, there are 3 sets of three plots which can
be easily scanned for)
Hey all,
Before I go into what I put into RAT, there are two questions we should
think about. First, how do we want reconstruction data stored and also
do we want to focus on keeping RAT's fit in sync with Fsim's?
Right now, reconstruction data is stored in a class called RAT_PosFit
since that is the only reconstruction class RAT has at the moment. The
class was originally intended to just store position information, yet
the reconstruction code I put in also has energy information. Are there
any objections to just having a single reconstruction class for energy
and also position information?
Secondly, in porting over Dave's code, I fixed a few bugs and also did
the small design changes I thought of, so how important is it to make
sure that non-trivial changes to reconstruction get back into Fsim?
Now onto what I committed:
I added a processor called fitpoisson. To see how to use it, I added 6
new macro files which tests the fitter at various locations and
energies. I also created a quick how-to on adding processor, which if
you are curious on how to add a processor or how I added mine should
answer your questions. It is on the UT BW site.
Here are some plots. First, let's look at the q2-weighted fitter for 1
MeV positrons events at various locations (x,0,0), where x can be
determined by the filename:
http://bogosort.net/~xichimos/q2_0m.gif
http://bogosort.net/~xichimos/q2_1m.gif
http://bogosort.net/~xichimos/q2_2m.gif
Notice that there isn't much of an offset for the q2-weighted fitter,
except for a little bias in the (2m,0,0) case near the acrylic vessel
(2.6m). So the q2-weighted fitter, which is used as a guess for the
Fsim fitters, is near the actual event location. The problem with it,
however, is the sigma resolution, which shows about a 14-20 cm
statistical offset for all cases. This is a good situation to apply a
minimization fitter to, however, since the minimization fitter shouldn't
find false local minimums since it should be near the real one.
http://bogosort.net/~xichimos/poisson_0m.gif
http://bogosort.net/~xichimos/poisson_1m.gif
http://bogosort.net/~xichimos/poisson_2m.gif
The Poisson fitter, which is the fitter that Dave put into ReactorFsim
and which Dave said was the most accurate, reduces the resolution down
to 7 cm, but has a bias towards the center of the detector which is
significant (see above plots, where the 0m, 1m, 2m in the filename is
the vertex displacement from the center in the x-direction).
Like I said in the meeting, the fitter is a copy of ReactorFsim's with a
wrapper around it, and I tested Fsim's version against this one with the
same input and got the same output before I started making fixes. This
means one of the problems with this reconstruction code lies in tweaking
variables like the Escale for our purpose. I also believe that error
is caused by energy guesser, which like Fsim, ignores reflections and
such. There are also some low-level comments on what I think is causing
error, like the way step-size is set, but I will leave this for people
working on reconstruction.
Lastly, the plots people will want to see are the energy versus r^3
plots. The simulation is of 1 MeV positrons. Here is reconstructed
energy versus actual r^3, then versus q2-weighted r^3, then versus
Poisson ^3, respectively:
http://bogosort.net/~xichimos/recoq_realr3.gif
http://bogosort.net/~xichimos/recoq_q2r3.gif
http://bogosort.net/~xichimos/recoq_poissonr3.gif
Before people get too excited, I just started putting time into energy
reconstruction some-time last week, and the day of our meeting, I
discovered during tests that even though our shape is good for the
E-plots, there is not much variation in reconstructed E between 10 MeV,
1 MeV and 0.1 MeV positrons. For example, the mean E for 10 MeVs MC
positrons is about 1.3 MeV. The mean reconstructed energy for 0.1 MeV
MC positrons is about 0.7 MeV. This just shows a scaling issue.
Now that I have committed my code, if people want to help out on
reconstruction, the biggest problem is adjusting parameters. Anybody
who is interested can e-mail me for more information.
Chris
This archive was generated by hypermail 2.1.6 : Tue Sep 27 2005 - 00:01:02 CDT
