Home » PANDA » PandaRoot » Bugs, Fixes, Releases » [FIXED] Ideal Tracking Bugs(?) when using FairLinks
[FIXED] Ideal Tracking Bugs(?) when using FairLinks [message #18211] |
Tue, 12 May 2015 16:38 |
André Zambanini
Messages: 17 Registered: February 2012 Location: FZ Jülich
|
occasional visitor |
From: *ext.kfa-juelich.de
|
|
Hello everyone,
for my analysis I had a closer look at some events in the eventDisplay and found some strange behavior of the ideal track finding. Basically, I found two problems and my guess is, they are independent of each other, but I'm not sure.
General Information
First, some information beforehand.
FairSoft: mar15
FairRoot: master (fb738d60 from 26.03.2015)
PandaRoot: r27581
I'm simulating events with EvtGen (momentum 4.07 GeV/c) using the decay chain: pbar p -> Xi+ Xi(1690)- -> Lambdabar pi+ Lambda K- and the lambdas decaying to pi p. The main message from the decay for you is that I have a lot of displaced vertices with distances of several centimeters to the IP.
For the reconstruction I'm using ideal track finding, both with the old track finder (PndSttMvdGemTrackingIdeal) and the new one (PndMCIdealTrackFinderNewLinks). The simulation chain uses FairLinks (fRun->SetUseFairLinks(kTRUE).
Wrongly Assigned GEM Hits
The first thing Tobias and me noticed were wrongly assigned GEM hits. From what I understood from Tobias and Stefano, this is a known issue. To illustrate this a bit, see these screenshots here:
Both, the old and the new track finder seem to assign GEM hits which don't belong to the track. The white dots indicate the MC points associated with the selected track - which works fine for MVD (blue squares) and STT (purple) hits, but seems to be off for GEM (red) hits.
Along with this comes wrong track information, as you can see with the red and blue circles, which indicate the first and last track parameters, respectively.
Tobias and me had a closer look at the new track finder and found out, that the spurious hit assignment happens when more than one MC point belong to a hit. The quick fix introduced by us is to simply ignore those hits. This came to the PndMCMatchNewLinks class with r27667 in the trunk.
My conclusion here is, that for now this is okay but maybe someone should have a detailed look at this.
Track Reconstruction with Kalman Task
After resolving the spurious GEM hits, the reconstructed track parameters still looked quite odd. I did some comparison and found out, that in a few cases the Kalman task messes things up, both for genfit 1 and 2. The following screenshots all show the result of the new ideal track finder (PndMCMatchNewLinks):
For this event, the Kalman filter produces strange tracks, independent of the genfit version. Most other events I had a look at seemed ok with genfit 2, while genfit 1 sometimes causes charge flips for instance (as visible in the screenshot above).
Ideally, I would leave the Kalman filter out, especially because for ideal tracking it seems a bit odd to use it. But this doesn't work for the PID, which seems to require the covariant matrices or something else filled by the Kalman.
So concluding, I have two questions: Any ideas, what could cause the Kalman to produce these results? And secondly, why is it required to have it in the first place, shouldn't ideal PID be based on MC information?
[Updated on: Mon, 06 July 2015 14:12] Report message to a moderator
|
|
|
Re: Ideal Tracking Bugs(?) when using FairLinks [message #18213 is a reply to message #18211] |
Tue, 12 May 2015 18:57 |
André Zambanini
Messages: 17 Registered: February 2012 Location: FZ Jülich
|
occasional visitor |
From: *netcologne.de
|
|
What the GEM hit assignment is concerned, I tried to run with the newest PandaRoot version again. As it turns out, the old ideal track finder (PndSttMvdGemTrackingIdeal) seems to assign all GEM hits correctly now. I have tried ~20 events and the assignment problem didn't occur, previously the effect appeared every 2nd or 3rd event. A crosscheck with the new track finder (PndMCIdealTrackFinderNewLinks) shows, that this still requires the mentioned modification by skipping GEM hits with more than one point.
Something I found earlier as well for the PndSttMvdGemTrackingIdeal algorithm but was not at my focus: the STT hits are not always assigned, even though they match quite nicely between MVD and GEM hits of a track. I attached a screenshot showing this on an example event (seems to happen every 2-3 events for my channel).
The same event with PndMCIdealTrackFinderNewLinks looks fine.
About the Kalman filter: I still find some events where the genfit 1 propagation results in last track parameters with different charge sign (as the example in my first post), but the genfit 2 propagation is much better. All the events I looked at have a quite good match between ideal track and ideal track after Kalman.
Still, the question remains: why do we need a Kalman filter for ideal tracks?
[Updated on: Mon, 18 May 2015 15:43] Report message to a moderator
|
|
|
Re: Ideal Tracking Bugs(?) when using FairLinks [message #18214 is a reply to message #18213] |
Tue, 12 May 2015 23:51 |
StefanoSpataro
Messages: 2736 Registered: June 2005 Location: Torino
|
first-grade participant |
From: *ip71.fastwebnet.it
|
|
Can you provide a set of macro (.dec sim digi reco pid and event display) so that one can reproduce what you are seeing?
Quote:Still, the question remains: why do we need a Kalman filter for ideal tracks?
Kalman over ideal pattern recognition is to have ideal efficiency but realistic momentum reconstruction.
Why are you using the master of FairRoot?
[Updated on: Tue, 12 May 2015 23:58] Report message to a moderator
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Re: Ideal Tracking Bugs(?) when using FairLinks [message #18260 is a reply to message #18259] |
Wed, 20 May 2015 15:40 |
Lia Lavezzi
Messages: 291 Registered: May 2007 Location: Torino
|
first-grade participant |
From: *to.infn.it
|
|
I agree with Stefano. The fixes we made in these days were in the pattern recognition stage, not in the Kalman.
Let's hope genfit2 has that bug fixed (but I never tried it).
Anyway, glad to hear that at least before the Kalman everything is fine now.
Cheers,
Lia.
|
|
|
Goto Forum:
Current Time: Mon Oct 07 06:29:31 CEST 2024
Total time taken to generate the page: 0.01202 seconds
|