1 st look at dE / dx in GArSoft 19 Mar 2019 Leo Bellantoni
New art Data Product • .../ReconstructionDataProduct/TrackIoniz.* • Now ../Reco/tpctrackfit2_module.cc , in the KalmanFit function, will (forward pass only): for (TPCCluster=1 to <nTPCClusters) { add some d ℓ to the track for the new cluster push d ℓ and the ADC counts onto TrackIoniz } • And in anatree_module.cc , call processIonizationData (which is now in anautil.cxx ) • Well all that does is divide the ADC counts by d ℓ and take the average for the whole track • Later, some calibration service should convert ADC counts into pC 2
• That went into git 3 weeks ago • And a little bug fix last week • I made a bunch ‘o MC • 50k default GENIE events One file stuffed NaN into all my plots I got rid of it • • Match the MCParticle to a track if PID ∈ { p ± , ! ± , " ± , e ± } • Honestly, the matching algorithm is crude – just uses direction of initial momentum • Use P TRUE not P RECO • So far, no track quality cuts • OK now some plots! 3
First look! GeV GeV 4 GeV GeV
Separation • ( ! 1 – ! 2 )/( σ 1 ⊕ σ 2 ) actually not a great statistic for non-Gaussian distributions • low momentum p hit kinematic limit (e.g. at P = 150MeV, T = 12MeV, CDSA track length = 14cm and we are not getting those yet Then my 2 nd rate track matching finds the wrong MCParticle • GeV 5
Truncated means • It is a well established technique to take the mean of only the lower ionization readings to use dE / dx for PID OPAL: 882mº Ar, 98mº methane & 20mº, 4 bar. Truncated @ 700mº. • ALEPH: 910mº Ar, 90mº methane at atmosphere. Truncated at 600mº. • ALICE: 857mº Ne, 95 mº CO 2 , 48mº N 2 , at atmosphere. Truncated @ 700mº. • PEP-4: 800mº Ar, 200mº methane, 8.5 atmospheres. Truncated at 650mº. • • I just tried cutting at 0.7 – obviously, some optimization study possible 6
Second look! 7 GeV
Why the p do that? • Scanning gives plausible reasons why there might be abnormally low ionization: Missing hits (sector boundaries) Splinter tracks stole my ionization! 8
Possible next steps • Keep trying to figure out why the p do that (possible event display upgrade) • Select a cut and put it into the analysis • Fix matching MCParticle ↔ charge deposits in GEANT ↔ hits ↔ tracks (gonna have to do this anyway) • Build a calibration service (gonna have to do this anyway) • A better algorithm: figure out how to correct for the sector boundaries • optimize the 70% • build a log-likelyhood discriminant based on distance from end • 9
Extra slides • Level 1 Level 2 • • Level 3 10
Extra slides • Level 1 Level 2 • • Level 3 11
Extra slides • Level 1 Level 2 • • Level 3 12
Recommend
More recommend
