GSI Forum
GSI Helmholtzzentrum für Schwerionenforschung

Home » PANDA » PandaRoot » Bugs, Fixes, Releases » MVD covariance matrix = 0.
MVD covariance matrix = 0. [message #12189] Mon, 04 July 2011 16:50 Go to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi MVD experts,
I am running some tests and I see several crashes in genfit, but after looking at the problem I think that this is caused by the sds covariance matrix, when it happens to have all the elements equal to 0:

-I- PndSdsRecoHit::PndSdsRecoHit: Wrote a hit with
(x,y) = (0.2535,-0.2925).
Covariance Matrix is
2x2 matrix is as follows

     |      0    |      1    |
-------------------------------
   0 |          0           0  
   1 |          0           0  


this turns out in the crash

#10 0xb2454dd5 in GFTools::invertMatrix (mat=..., inv=...)
    at /home/lavezzi/test_dev/trunk/genfit/GFTools.cxx:288
#11 0xb244b438 in GFKalman::chi2Increment (this=0x8ab4208, r=..., H=...,
    cov=..., V=...) at /home/lavezzi/test_dev/trunk/genfit/GFKalman.cxx:200
#12 0xb244c4b5 in GFKalman::processHit (this=0x8ab4208, tr=0x12bc0090, ihit=3,
    irep=0, direction=1)
    at /home/lavezzi/test_dev/trunk/genfit/GFKalman.cxx:313
#13 0xb244b27d in GFKalman::fittingPass (this=0x8ab4208, trk=0x12bc0090,
    direction=1) at /home/lavezzi/test_dev/trunk/genfit/GFKalman.cxx:169
#14 0xb244aacb in GFKalman::processTrack (this=0x8ab4208, trk=0x12bc0090)
    at /home/lavezzi/test_dev/trunk/genfit/GFKalman.cxx:77
#15 0xafb6c461 in PndRecoKalmanFit::Fit (this=0x8ab41d8, tBefore=0xfd188a0,
    PDG=13)
    at /home/lavezzi/test_dev/trunk/GenfitTools/recotasks/PndRecoKalmanFit.cxx:282


since at that point, in the invertMatrix function, there is 1./mat[0][0] and mat[0][0] is equal to 0.
I am using rev 12447.

Could you please fix this? Or is it fixed in the latest revision?
Thank you in advance,
Lia.

Re: MVD covariance matrix = 0. [message #12190 is a reply to message #12189] Mon, 04 July 2011 17:14 Go to previous messageGo to next message
Tobias Stockmanns is currently offline  Tobias Stockmanns
Messages: 489
Registered: May 2007
first-grade participant
From: *ikp.kfa-juelich.de
Dear Lia,

which simulation chain are you using which causes these crashes?

Ciao,

Tobias
Re: MVD covariance matrix = 0. [message #12191 is a reply to message #12190] Mon, 04 July 2011 17:50 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Tobias,
I am using the macros from macro/pid: run_sim_sttcombi_pgun, run_digi_sttcombi.C, run_reco_sttcombi.C

I generate muons with momentum in the range [1, 1.5] GeV/c and theta in the range [21, 133] deg and I commented out all the detectors in digi and reco except for stt and mvd.
Lia.
Re: MVD covariance matrix = 0. [message #12193 is a reply to message #12191] Tue, 05 July 2011 11:07 Go to previous messageGo to next message
Tobias Stockmanns is currently offline  Tobias Stockmanns
Messages: 489
Registered: May 2007
first-grade participant
From: *ikp.kfa-juelich.de
Dear Lia,

I simulated 1000 events with the pid macros given by you without any problems.
All covariance matrizes have at least one value different from zero.

Can you please have a look at the Dx, Dy, Dz values for the MVDHitsPixel / MVDHitsStrips if there are any zeros. They are the basis for the calculation of the cov matrizes.

How often does this crash happen?

Cheers,

Tobias
Re: MVD covariance matrix = 0. [message #12195 is a reply to message #12193] Tue, 05 July 2011 11:50 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Tobias,

I checked the Dx, Dy, Dz and I have some of them equal to 0, exactly in the event where the crash happens, for example in one test I have from the cbmsim->Scan, looking for small dx, dy, dz (below 1e-8):

***********************************************************
*    Row   * Instance * MVDHitsSt * MVDHitsSt * MVDHitsSt *
***********************************************************
*      110 *        0 * 0.0086602 *         0 *         0 *
*      220 *        0 * 0.0086602 * 7.178e-17 * 3.347e-17 *
*      361 *        6 * 0.0086602 *         0 *         0 *
*      876 *        0 * 1.952e-17 * 1.512e-17 * 0.0086602 *
***********************************************************

and the crash is exacly in event 110.

It happens quite often: I made some tests with 2000 muons each, at low momenta (around 0.2 GeV/c) and higer momenta (around 1 GeV/c) and on 7 tests, 3 crashed this way (2 crashed in the mvd riemann track finder, I will post the crash on the forum too).

I scanned the cbmsim also for the files where the riemann track finder crashed and also there I found errors equal to 0. I post here and example:
***********************************************************
*    Row   * Instance * MVDHitsSt * MVDHitsSt * MVDHitsSt *
***********************************************************
*      423 *        0 * 0.0086602 * 1.575e-16 * 3.061e-17 *
*      511 *        0 * 0.0086602 *         0 *         0 *
*     1243 *        0 * 0.0086602 *         0 *         0 *
*     1481 *        0 * 0.0086602 * 4.763e-17 * 4.442e-17 *
***********************************************************

I guess this means that if the mvdriemann didn' t crash it could crash in genfit as in the other cases...

Moreover I looked also in the files where reconstruction went fine to the end and I have errors equal to 0 also there:
***********************************************************
*    Row   * Instance * MVDHitsSt * MVDHitsSt * MVDHitsSt *
***********************************************************
*     1317 *        7 * 0.0086602 * 1.335e-19 * 9.140e-20 *
*     1535 *        1 * 0.0086602 *         0 *         0 *
***********************************************************

My idea is that it did not crash here because this hit is not assigned to any track and so it does not enter in kalman calculation... I can check this if you think it is useful.

One more information: I get the same results on openSuse and Scientific Linux Cern.

Cheers,
Lia.


Re: MVD covariance matrix = 0. [message #12196 is a reply to message #12195] Tue, 05 July 2011 12:48 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *to.infn.it
Which trunk version?
There was a bug in mvd digitization that now should be fixed.
Could it be this the reason?
Re: MVD covariance matrix = 0. [message #12205 is a reply to message #12196] Tue, 05 July 2011 14:38 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Stefano,
I am using an updated trunk, but for sds, mvd, PndTools and pnddata I am using revision 12447, the one chosen for the last grid release.
If all the problems are solved now, I can update everything and rerun the tests with the last version of trunk... what do you think?
Ciao,
Lia.
Re: MVD covariance matrix = 0. [message #12207 is a reply to message #12205] Tue, 05 July 2011 15:03 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi again,
I updated my whole trunk to the last revision and rerun digitization and reconstruction but I still get the crash Sad
Lia.
Re: MVD covariance matrix = 0. [message #12209 is a reply to message #12205] Tue, 05 July 2011 15:36 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *to.infn.it
Hi,
I think it is fine.
The other thing I could suggest is to use the may11 release, instead of the trunk. In reality there could be still some problems with the new base.
Re: MVD covariance matrix = 0. [message #12210 is a reply to message #12209] Tue, 05 July 2011 15:38 Go to previous messageGo to next message
Tobias Stockmanns is currently offline  Tobias Stockmanns
Messages: 489
Registered: May 2007
first-grade participant
From: *ikp.kfa-juelich.de
Are you sure that this is the right topic or should your reply be part of "Number of entries in tree"?

Cheers,

Tobias
Re: MVD covariance matrix = 0. [message #12211 is a reply to message #12207] Tue, 05 July 2011 15:39 Go to previous messageGo to next message
Tobias Stockmanns is currently offline  Tobias Stockmanns
Messages: 489
Registered: May 2007
first-grade participant
From: *ikp.kfa-juelich.de
Dear Lia,

actually I do not know how to solve this problem. Maybe Ralf can have a look. These zeros are comming from the strip part of the MVD.

Cheers,

Tobias
Re: MVD covariance matrix = 0. [message #12212 is a reply to message #12210] Tue, 05 July 2011 16:18 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *to.infn.it
I was speaking about this topic.
Re: MVD covariance matrix = 0. [message #12219 is a reply to message #12212] Wed, 06 July 2011 12:07 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi,
I downloaded the may11 release and run the macros there, but I still get all the crashes.
I thought that it might be something wrong in the macros (maybe I changed something and did not remember), so I took the macro/run/tdrct/eta_c macros for the stt and just changed the generator part to simulate single muon tracks instead of etac events, but I still get the crashes.
I run six tests and only one arrived to the end without problems.

Ralf, can you please have a look into this, as Tobias suggested?

An alternative solution could be to put inside genfit some protection against the division by 0 in the invertMatrix function of GFTools.cxx, but the strip errors equal to 0 would be still there. What do genfit people think about this?
Cheers,
Lia.

Re: MVD covariance matrix = 0. [message #12224 is a reply to message #12219] Wed, 06 July 2011 13:16 Go to previous messageGo to next message
Ralf Kliemt is currently offline  Ralf Kliemt
Messages: 507
Registered: May 2007
Location: GSI, Darmstadt
first-grade participant

From: *pool.mediaWays.net
Lia Lavezzi wrote on Wed, 06 July 2011 12:07


Ralf, can you please have a look into this, as Tobias suggested?


Hi,

I was already (silently) looking at the code. However I found no sign where that could happen, yet. Tomorrow I'll run some tests.

Ralf

PS: Could you pleas post that scan command you used?

[Updated on: Wed, 06 July 2011 13:18]

Report message to a moderator

Re: MVD covariance matrix = 0. [message #12225 is a reply to message #12224] Wed, 06 July 2011 13:30 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Ralf,
thank you very much!

Here is the command:
cbmsim.Scan("MVDHitsStrip.fDx:MVDHitsStrip.fDy:MVDHitsStrip.fDz", "MVDHitsStrip.fDx < 1e-8 || MVDHitsStrip.fDy < 1e-8 || MVDHitsStrip.fDz < 1e-8")

Cheers,
Lia.
Re: MVD covariance matrix = 0. [message #12237 is a reply to message #12189] Mon, 11 July 2011 14:10 Go to previous messageGo to next message
Ralf Kliemt is currently offline  Ralf Kliemt
Messages: 507
Registered: May 2007
Location: GSI, Darmstadt
first-grade participant

From: *pool.mediaWays.net
Hi Lia,

The issue should be fixed now.
Ralf
Re: MVD covariance matrix = 0. [message #12238 is a reply to message #12237] Mon, 11 July 2011 19:13 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Ralf,
I run some tests on the files previously crashing and I can confirm that the problem is no longer there.
Thank you very much! Smile
Lia.
Re: MVD covariance matrix = 0. [message #12239 is a reply to message #12238] Mon, 11 July 2011 19:27 Go to previous messageGo to next message
StefanoSpataro is currently offline  StefanoSpataro
Messages: 2736
Registered: June 2005
Location: Torino
first-grade participant

From: *10-79-r.retail.telecomitalia.it
Trying to understand,
does it mean that now , with the latest tunk, the code stt+mvd tracking gives the same results as before, w/o crashes? because in may11 we are still using the old version before the time stamps upgrade.
Re: MVD covariance matrix = 0. [message #12242 is a reply to message #12239] Tue, 12 July 2011 10:49 Go to previous messageGo to next message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
Hi Stefano,
I did not check the results yet, just checked that the crash was gone... now I can run a couple of tests at different momenta and compare them to our previous results to confirm that everything is fine.
Ciao,
Lia.
Re: MVD covariance matrix = 0. [message #12249 is a reply to message #12242] Wed, 13 July 2011 11:46 Go to previous message
Lia Lavezzi
Messages: 291
Registered: May 2007
Location: Torino
first-grade participant

From: *pv.infn.it
After the tests I confirm that the momentum resolution are still good after the last changes.
Lia.
Previous Topic: RiemannTrackFinder crash.
Next Topic: Bug in GenfitTrack2PndTrack
Goto Forum:
  


Current Time: Sat Oct 05 04:56:43 CEST 2024

Total time taken to generate the page: 0.01077 seconds