1 Selection of non-triggering muons in J/ψ  μμ events for the calibration of the Muon System 1)Offline Selection:  use of Mass Constrained Global Fit (thanks to Stefania Vecchi for the help!) 2) Online Selection:  Modification of the single muon line in order to increase the rate G. Lanfranchi – May third, 2006

2 Main steps: 1)Pick up events from single muon line (as it is now) 2)Identify the muon which provides the trigger 3)Search for a second long track of opposite charge wrt the triggering muon 4)Do the mass-constraint-global fit imposing the J/Psi mass 5)Select events with a chi2 < cut 6)Evaluate the purity of non-triggering muons I have used (for the moment): Bd  J/Psi(mu+ mu-) Ks events - DaVinci v12r15 - HltGeneric v2r8 1) Offline selection using a mass-constrained global fit

3 Pt distribution BEFORE Pt cut N = ~ 2.3 GeV Pt (GeV/c) Pt distribution AFTER Pt cut 1) Pick up events from the single muon line  standard single muon line requires a muon with Pt> 3GeV, IPS>3  Pt distributions of triggering muon: Pt (GeV/c) Comment: Pt > 3 GeV is a quite strong cut : do we really need it? N = 9329

4 2) Identify the muon which provides the trigger Pt (trigger)-Pt (Offline) (MeV/c) N = 7778

5 4) Mass Constrained Global Fit for B  J/Psi (μμ) X (I) π π-π- SV (x +, y +, tx +, ty +, p + ) (x -, y -, tx -, ty -, p - ) (x o, y o, z 0 z M J/ ψ (x,y,t x,t y,p) PV (x PV,y PV,z PV ) By means of Lagrange-multiplier method the constrained least square is minimized and “best” parameters are found. Measured quantities: x,y,tx,ty,p for each muon (5x2 = 10) Constraints : 4 – momentum conservation (4) J/Psi, mu+ and mu- from the same vertex (3x2 =6) total = 10 Fitted parameters: J/Psi vertex position (3) J/Psi x,y,tx,ty,p (5) NdF = Constraints – parameters = 2 See S. Vecchi talk, LHCb Italia Meeting, Florence

6 χ2 <100 N = 7778 χ2 < 50 N = 6549 χ2 <10 N = 4903 μ+μ- invariant mass vs χ2 cut χ2 distribution of the global fit χ2<10 χ2<50 χ2<100 Mass Constrained Global Fit for B  J/Psi (μμ) X M( μμ) (MeV/c 2 )  the χ 2 cut cleans up the sample !  J/Psi mass peak shows up without tails

7 χ2 <100 χ2 < 50 χ2 < 10 χ2 <100 χ2 < 50 χ2 <10 Momentum Distribution for Triggering and Non-Triggering Muon Triggering muon: Non-Triggering muon: Non-triggering muon has a spectrum in the whole range of interest: Above 150 GeV there are few events but the efficiency is also high so we need less statistics to evaluate it!

8 Pt Distribution for Triggering and Non-Triggering Muon χ2 <100 χ2 < 50 χ2 <10 χ2 <100 χ2 < 50 χ2 <10 Triggering muon: Non-Triggering muon:

9 Pt and P distributions of triggering (red)-non triggering (blue) muons P (GeV/c)Pt (GeV/c)

10 Purity = N. of true muons / N. of selected muons Pt distribution for “ true” muons and “selected” muons ~ 88% for 0<p<100 GeV/c Purity of the selected calibration sample for chi2 < 10 P (GeV/c) Pt (GeV/c) “ true” “ selected” For Pt > 0.7 GeV almost all the selected muons are true muons from J/Psi

11 Purity vs P ~96% From GeV/c ~ 97% for 0.7 < p < 100 GeV/c Purity of the selected calibration sample for χ2 0.7 GeV/c Momentum distribution: “ true” “ selected” P (GeV/c)

12 1)With the two cuts χ2 0.7 GeV we get, from the single muon line, a calibration sample with a purity of 97% and momentum spectrum which covers all the interesting range. 2)I used B  J/ψ(μμ) Ks events, BUT no assumption has been made on Ks:  We can use the same selection for all the B  J/ψ(μμ) X decay modes 3) The game must be repeated with bb-inclusive sample to evaluate the S/B ratio:  BUT we can tighten the χ2 cut to recover the same purity. Comments (I) :

13 Comments (II) : The problem is the rate: 1) from generated events we endup with ~ 3800 events; 2) total efficiency Eff ~ 8% for B d  J/Psi(μμ) Ks events (acceptance x L0 x L1 x HLT single muon line x identification of triggering muon x χ2 cut x Pt cut) : Rate = σ (bb) x L x f(B d,B s,B + ) x BR(B  J/ψ (μ  ) X) x Eff ~ 500 μb x 2x10 32 cm -2 s -1 x 90% x 2x10 -4 x 8% ~ 1.4 Hz 3) I expect that the efficiency will go down if we analyze bb-inclusive and we want recover the same purity. 4) So we will endup with a rate ~ 1 Hz, probably less !!

14 A PROPOSAL FOR A MODIFICATION OF THE SINGLE-MUON LINE IN HLT ( idea already presented at LHCb Italia Meeting in Florence, April 2006 ) 1. Start from events with the standard single muon line cuts: one “online recognized muon” with pt>3 GeV/c, the cut in IPS to be defined (use also J/Psi prompt?) 2. Loop on long triggertracks and select the first one with opposite charge which comes from the same vertex (trgvertexfit χ2 < 5 ); 3.Filter the events if the Muon track and the second track give a J/Psi invariant mass between a certain range.

15 Pt distribution BEFORE Pt cut N = ~ 2.3 GeV Pt (GeV/c) Pt distribution AFTER Pt cut  standard single muon line requires a muon with Pt> 3GeV, IPS>3  Pt distributions of triggering muon: Pt (GeV/c) N = 9329

16 χ 2 distribution of the vertex fit for events after Pt and IPS cut Events with χ2<5 and 2.5 GeV<M(J/Psi)<3.5 GeV N = 7242 Eff = 7242/9329 ~ 80% The algorithm works with high efficiency But…. What about speed? χ 2 / ndf M(J/Psi ( μμ)) (MeV/c 2 )

17 Generic and Specific HLT: Use 1/4 of 1600 CPUs at 40 kHz: 10 ms/event on a 2007 CPU. Assume 60 ms on a 1 GHz Pentium III 1 GHz Pentium III ExecutionAverage VeLo Tracking 7 ms Generic HLT 19 ms Rest of Forward Tracking 15 ms5 ms PID (mainly RICH) 35 ms 12 ms Shared Resonances 12 ms4 ms D* stream 1 ms< 1 ms Exclusive stream 9 ms3 ms Total N/A50 ms HLT speed requirements P.Koppenburg, LHCC Comprehensive Review, January 2005

18 ToolSvc.SequencerTimerTool INFO ToolSvc.SequencerTimerTool INFO This machine seems about 1.01 times faster than a 2.8 GHz Xeon. ToolSvc.SequencerTimerTool INFO Algorithm (millisec) | | | min max | entries | total (s) | ToolSvc.SequencerTimerTool INFO ToolSvc.SequencerTimerTool INFO DaVinciMainSeq | | | | 456 | | ToolSvc.SequencerTimerTool INFO DaVinci | | | | 300 | | ToolSvc.SequencerTimerTool INFO OldRichPIDCnvAlg | | | | 300 | | ToolSvc.SequencerTimerTool INFO MuonIDFOI | | | | 300 | | ToolSvc.SequencerTimerTool INFO ChargedProtoPAlg | | | | 300 | | ToolSvc.SequencerTimerTool INFO PrimVtxFinder | | | | 300 | | ToolSvc.SequencerTimerTool INFO HLTGenericSeq | | | | 130 | | ToolSvc.SequencerTimerTool INFO TrgMuonRec | | | | 130 | | ToolSvc.SequencerTimerTool INFO Load from Buffer | | | | 130 | ToolSvc.SequencerTimerTool INFO Make pad from strip | | | | 386 | | ToolSvc.SequencerTimerTool INFO Reconstruct muons | | | | 386 | | ToolSvc.SequencerTimerTool INFO Store muons | | | | 130 | | ToolSvc.SequencerTimerTool INFO HltMuonRec | | | | 130 | | ToolSvc.SequencerTimerTool INFO Associate muons | | | | 130 | | ToolSvc.SequencerTimerTool INFO HltGenericSelection | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenMakeTTHit | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenVeloTT | | | | 130 | | ToolSvc.SequencerTimerTool INFO TrgMuonRecover | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenMakeTHit | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenForward | | | | 130 | | ToolSvc.SequencerTimerTool INFO TrgMuonRefine | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenErrParam | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenParticleMaker | | | | 130 | | ToolSvc.SequencerTimerTool INFO HltGenericDecision | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaJpsiLine Timer | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaMuonLine Timer | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaL1Conf Timer | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaSndVtx Timer | | | | 130 | | ToolSvc.SequencerTimerTool INFO Official Single Muon Line

19 ToolSvc.SequencerTimerTool INFO ToolSvc.SequencerTimerTool INFO This machine seems about 0.99 times faster than a 2.8 GHz Xeon. ToolSvc.SequencerTimerTool INFO Algorithm (millisec) | | | min max | entries | total (s) | ToolSvc.SequencerTimerTool INFO ToolSvc.SequencerTimerTool INFO DaVinciMainSeq | | | | 456 | | ToolSvc.SequencerTimerTool INFO DaVinci | | | | 300 | | ToolSvc.SequencerTimerTool INFO OldRichPIDCnvAlg | | | | 300 | | ToolSvc.SequencerTimerTool INFO MuonIDFOI | | | | 300 | | ToolSvc.SequencerTimerTool INFO ChargedProtoPAlg | | | | 300 | | ToolSvc.SequencerTimerTool INFO PrimVtxFinder | | | | 300 | | ToolSvc.SequencerTimerTool INFO HLTGenericSeq | | | | 130 | | ToolSvc.SequencerTimerTool INFO TrgMuonRec | | | | 130 | | ToolSvc.SequencerTimerTool INFO Load from Buffer | | | | 130 | | ToolSvc.SequencerTimerTool INFO Make pad from strip | | | | 386 | | ToolSvc.SequencerTimerTool INFO Reconstruct muons | | | | 386 | | ToolSvc.SequencerTimerTool INFO Store muons | | | | 130 | | ToolSvc.SequencerTimerTool INFO HltMuonRec | | | | 130 | | ToolSvc.SequencerTimerTool INFO Associate muons | | | | 130 | | ToolSvc.SequencerTimerTool INFO HltGenericSelection | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenMakeTTHit | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenVeloTT | | | | 130 | | ToolSvc.SequencerTimerTool INFO TrgMuonRecover | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenMakeTHit | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenForward | | | | 130 | | ToolSvc.SequencerTimerTool INFO TrgMuonRefine | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenErrParam | | | | 130 | | ToolSvc.SequencerTimerTool INFO GenParticleMaker | | | | 130 | | ToolSvc.SequencerTimerTool INFO HltGenericDecision | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaJpsiLine | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaMuonLine | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaL1Conf | | | | 130 | | ToolSvc.SequencerTimerTool INFO AnaSndVtx | | | | 130 | | ToolSvc.SequencerTimerTool INFO New Single Muon Line:

20 Comments : 1.The algorithm itself does not introduce any consistent delay (new single muon line = 135 μsec old single muon line = 132 μsec 2. The difference overall in the HltGenericDecision is μsec mainly due to the retrieving of the trigger tracks  this can be further optimized. 3. The efficiency in selecting muons from B  J/Psi X is quite high, of the order of ~80%.

21 Next Steps: 1)Reduce the IPS cut on the single muon to select J/Psi prompt and tighten the Invariant Mass window in order to maintain the bandwidth at a reasonable level. 2) Study with the Offline Procedure the purity, the efficiency, the S/B ratio of this new sample. 3) Since σ (J/ψ prompt) ~ 10 x σ (J/psi from B) we expect to have a total rate of Hz of J/Psi prompt events  fully negligible with respect to the single muon line (~ 900 Hz) and the di-muon line (~600 Hz). 4) Hans Dijkstra is fully open to accept our requests if they are reasonable.  we have to present our proposal at a T-Rec meeting within May  I am available – as soon as I am ready – to give a talk to the whole Muon Group in order to explain all the details.

22 Issues to be discussed: 1)As soon as we will have new tracks for DC06 production we must be ready to recalibrate MuonID and Pion misID as soon as possible since stripping needs PID from all the subdetectors. It could be worth to do in both ways: a) the old one : calibration of FOI and DIST using MC truth b) the new one: using selected muons for calibration and compare the results.  Erica and Miriam: have you this procedure ready?  Can you help me to do the same using calibrating muons? 2) Could we use only the DIST variable instead of the FOI+DIST? It would simplify quite a lot the whole procedure… Did you try already in the past?

23 3) We must provide an exact estimate of how many events we will need to calibrate the Offline MUONID. Do you confirm the 100k of “pure” muons? Can you provide some quantitative result? How many events do you think that we would need for calibrating the DIST variable only ? 4) Is somebody looking at the D* inclusive sample for calibration of Pion misID?

24 5) What is the meaning of “Muon” and “NotMuon” DLL values? C. Jones talk at the last Software Week 6) Other issues ??