170 likes | 379 Views
Status PPR chapter 5. ALICE Performance Offline tracking HLT – online tracking Vertexing – primary and secondary vertices Charged Particle Identification Photon Identification. Tracking Performance. Development since last ALICE week Parallel TPC tracking – further improvements
E N D
Status PPR chapter 5 • ALICE Performance • Offline tracking • HLT – online tracking • Vertexing – primary and secondary vertices • Charged Particle Identification • Photon Identification Karel Safarik: Status PPR chapter 5
Tracking Performance • Development since last ALICE week • Parallel TPC tracking – further improvements • Efficiency approaches dangerously 100% • and is even faster • but has to be optimized for dE/dx • See details in M.Ivanov talk • Secondary vertexing during the tracking • Have to be applied for TRD • There are ideas also for ITS • New muon tracking – Kalman filter • Supplied with smoother (A.Zinchenko) • Improve efficiency Karel Safarik: Status PPR chapter 5
Only primaries Karel Safarik: Status PPR chapter 5
TPC and ITS raw data • For ITS and TPC raw data generated from simulation • check was done that reconstruction works without having access to simulated data • ITS is starting to use new geometrical modeler Karel Safarik: Status PPR chapter 5
New geometry SSD Cone: top view Karel Safarik: Status PPR chapter 5
New geometry SSD Cone: 3D view Karel Safarik: Status PPR chapter 5
Muon tracking • more sophisticated cluster finder • maximum likelihood expectation maximization (MLEM) • improvement in efficiency • new tracking – Kalman filter • similar approach as in other detectors • improvement in precision • smoother with different cuts applied • further development • position error parameterization – till now constant value used Karel Safarik: Status PPR chapter 5
Results (cont.) 2500 UpsilonsNew cluster finder “Loose” - 8s, 2max =100; “Tight” - 4s, 2max = 25Window_y < 5cm
Results (cont.) 2500 UpsilonsNew cluster finder
Combined PID • Our large problem – misunderstanding • Bayes theorem: P(H/m) = norm * P(m/H) * P(H) (probability that hypothesis H is true after we observed m is proportional to the probability that we measure m under condition of H times the prior probability of H) There is no doubt in this statement – this is not the cause of the infinite discussion between Baysians and Frequentionists • The dispute is on the interpretation of P(H/m) in the case when H is in principle true or false • frequentionist in this case would say that P(H/m) has no sense • all what you measure is contained in P(m/H) because, strictly speaking you have no access to prior P(H) (purist point of view) Karel Safarik: Status PPR chapter 5
Combined PID • However, there is a huge difference between P(H/m) and P(m/H) • Examples: • LEP have measured “probability to obtain measured data under assumption of SM with Higgs mass in interval … is 95%” • and not (how it was sometimes interpreted) “probability of standard model Higgs mass being in interval … after LEP measurement is 95%” • consider: hypothesis H = women, measurement m = pregnant • P(m/H) = 0.03; however, P(H/m) = ?? Karel Safarik: Status PPR chapter 5
Combined PID • Our problem is that we want to measure P(H) itself • Frequentionist, however, will not allow you to talk about P(H/m) and P(H) for given track, because it is definitely of some type • Luckily, even he would agree to use it on ensemble • For each j : SiP(Hj/mi) = P(Hj) = Si [P(mi/Hj) * P(Hj) / SkP(mi/Hk) * P(Hk)] 1 = Si [P(mi/Hj) / SkP(mi/Hk) * P(Hk)] Karel Safarik: Status PPR chapter 5
Combined PID Bayes theorem and the combined PID • N – number of tracks in the selected set of tracks. • Mi= {mi(ITS), mi(TPC), …} – vector of the PID signals in the detectors (ITS, TPC, TRD, …) by the i-th track. • C(s) – percentage of the particles of s-type in the selected set of tracks. Depends on the selection and does not depend on the detector properties ! • R(Mi|s) – probability to observe the vector of signals Mi, if the detectors register a particle of s-type. Depends on the detector properties and does not depend on the track selection ! This is the result of the “PID combining” and this is what goes to the ESD. • P(s|Mi) – probability to be a particle of s-type, if the vector of signals in the detectors was Mi . This is what we want to get (the PID weight). R(Mi|s)C(s) R(Mi|t)C(t)t = e, , , … s = e, , , … i = 1, 2, …, N P(s|Mi)= ALICE PPR meeting 07 Feb 03Karel Safarik: Status PPR chapter 5
Combined PID Combining detectors’ PID signals • Probabilities R(mi|s) can be calculated for the “detector response functions” of any shape (not only Gaussian). • The “detector response functions” do not have to be the same for each of our detectors. • If, in some case, the resolution of some of the detectors is so bad that particles can not be identified at all, this formula is still valid. The contribution of such a detector simply cancels out. • This formula allows combining PID signals, first, only within a subset of the detectors and then (if it becomes possible) one can include the signal from an additional detector (or another subset of detectors). • This is the responsibility of the detector D to provide the final PID procedure the probabilities R(mi|s). These probabilities can be calculated if the detector knows its “response functions” for the particles of the s-type (s = e, , , …). R(mi(D)|s) D D = ITS, TPC, TRD, … s = e, , , … R(Mi|s)= R(mi(D)|t) t D ALICE PPR meeting 07 Feb 03Karel Safarik: Status PPR chapter 5
Combined PID “Two step” PID procedure • Step 1 : ESD preparation (“once and forever”) For each track and for each of the detectors a) calculate the PID signal (done by the detector trackers). b) calculate the R(mi|s) (done by detector specific PID classes). For each track c) calculate R(M|s) (done by some class responsible for the combining). Write out the ESD. • Step 2 : Data analysis (as many times as needed) (done by the analysis macros/classes) a) Select the subset of ESD tracks under interest. b) Estimate the particle composition ( C(s), s=e,,,… ) of the selected track subset. c) For each of the selected ESD tracks calculate the PID weights P(s|M) (by means of Bayes’ formula). d) Proceed with the analysis using the PID weights… ALICE PPR meeting 07 Feb 03Karel Safarik: Status PPR chapter 5
Combined PID Current status • AliESDpid: done • ITS,TPC,TRD: AliXXXpid.h/AliXXXpid.cxx are provided (AliRoot v3-09-07) • All this is very nice but: • A lot of job is still to be done by the detector experts. • The combined PID will not be significantly better than that given by the TPC stand-alone… We desperately need the TOF ! • TOF: no “production quality” reconstruction available • We have combined PID prototype without meat in it ALICE PPR meeting 07 Feb 03Karel Safarik: Status PPR chapter 5
Conclusions towards chapter 5 • section tracking – done – text practically finalized • efficient TPC tracking, including secondary tracks • improvement in ITS tracking • TRD tracking on the way • online tracking – still some text editing • section vertexing – work done – text exists – needs final editing • primary vertex • secondary vertex • ‘kink’ finder for charged K decays in TPC • section charged particle identification • stand-alone detector performance for • ITS, TPC and TOF – work done – text exists • on the way for • TRD, HMPID • combined particle identification – problem !!! • section p0/photon id – work done – text needs final editing • new method for p0/photon separation in PHOS Karel Safarik: Status PPR chapter 5