80 likes | 246 Vues
A bug has been identified in the calculation of chi2 values for TPC tracks, where the mean chi2 increased from approximately 1 to 3 in recent productions. This issue was caused by not accounting for R distortion in the calculation. A fix has been committed to ensure correct calculations within the AliTPCseed module, specifically in the GetPredictedChi2 function. Observations indicate that the average chi2/Ncls has a strong dependence on phi and eta, which suggests a variation in cluster association and gains since the introduction of lower gas gain settings. Suggestions for improvements in the GlobalQA metrics are also offered.
E N D
bug #48097: Incorrect calculation of chi2 for TPC tracks 2009-03-13 21:16, comment #1: Bug fix: Take into account R distortion in the calculation of chi2. The correction was applied in the AliTPCseed::Update (since october) but forgotten in AliTPCseed::GetPredictedChi2 Committed revision 31478. Marian Ivanov <miranov> 2009-03-13 11:43, original submission: chi2 of the fit for the TPC track is anomalously large, the mean went up from ~1 in previous productions to ~3 in the most recent one (LHC09a4). In addition the average chi2/Ncls for the tracks has a strong dependence on phi and eta. Adam Kisiel <akisiel>
Chi2 - Comparison • Left – Alien production • Right – corrected
Cluster ratio • Number of clusters to number of cluster in fiducial volume (fTPCncls/fTPCnclsF) (eta -0.9,0.9)
Number of clusters • Number of clusters
http://savannah.cern.ch/bugs/?47500 A year ago at the physics week in Prague, the GlobalQA (which blindly takes almost everything from ESD) was also showing ~7%. And, this was the old OCDB and no cuts on the number of assigned clusters. It looks like, with this latest data, we are getting less TPC clusters per track, in average. Do you also see it ? Is this because of the lower gain, or anything else ? 2009-03-16 23:00, comment #8: The fraction of the clusters associated with tracks was reduced introducing the lower gas gain (october 2008) For short drift length we find about 90% of clusters for long drift length about 80 % (diffusion and electron attachment) For my studies I used 2 additional cuts which you do not have in the Global QA - abs(eta)<0.9 and nCLSTPC>120 If possible I prefer to add this two variables to the GlobalQA histogram. ~ 10 bins in eta, 4-8 bins in TPCncls 2009-03-14 12:28, comment #4: Small production with offline $ALICE_ROOT OCDB entry and AliRoot v4-16-Rev06 was repeated. For MIP particles dEdx resolution ~5.8-6% was obtained (usign the $ALICE_ROOT/PWG1/AliTPCtaskPID class) OCDB entry $ALICE_ROOT/TPC/Calib/ClusterParam/Run0_999999999_v0_s2.root should be propagated to the Alien OCDB. Marian Ivanov <miranov>