140 likes | 279 Views
More on Trigger, Retention, Efficiency with the Minimal Layout S. Blusk Nov 4, 2009. Introduction. Goal is come up with a sensible trigger strategy that can be implemented when we have a new detector description and reconstruction.
E N D
More on Trigger,Retention, Efficiency withthe Minimal LayoutS. BluskNov 4, 2009
Introduction • Goal is come up with a sensible trigger strategy that can be implemented when we have a new detector description andreconstruction. • Victor has shown with a various set of selection requirements, he can ~double the TRIGGER efficiency in a few channels. • The signal efficiencies are evaluated for offline selected events.(more on this later)
Intro - 2 • Implement Victor’s cuts, except • I use ‘Offline tracks’: Expect lower ghost rate in upgraded trigger than in current HLT. Assume we can get ‘offline’ quality, hope for better..(lower ghosts) • Assume comparable slope resolution between Velo and upgraded Velo. • Degrade pT resolution by smearing the reconstructed pT bys(1/pT)/(1/pT)=(15.0 - 0.76pT + 0.86pT2)%, pT in GeV • I have not included L0 CAL trigger (pT>2 GeV in CAL), could be used to throttle (but maybe tracking info is better ?) • Look at efficiencies and retention rates vs Luminosity.DC06 vs MinLayout. V. CocoLHCb-INT-2009-008 CAL cut rejects ~ 50%of MB at 1e33..
Trigger Flow (for results in this talk) • Select VELO tracks with IP to nearest vertex > 120 mm (N1) • Get pT of tracks passing #1, cut on pT>1 GeV and IPS>3.0 (N2) • Form N2(N2-1) pairs and require: • DOCA < 80 mm • 2.0 < (Zsv-Zpv) < 100 mm, Rsv<5.8 mm • pT(L)>1.25 GeV, pT(c)> 1.0 GeV • Require polar angle q >23 mrad, both tracks • Point < 0.3 (see last talk for explanation of this cut) • Track c2/DOF < 4.5 If any 2-body combination passes all these cuts, I call this a “triggered event”. This is meant to imply that the event or candidate is passed on to a higher-level trigger that looks for specific exclusive or inclusive decays. I can then ask & answer: - Composition of triggered events, b, c, or u,d,s,g…? - Was the event ‘triggered’ by a 2-body in which both tracks were MC matched? - How often do ghost track pairs lead to a trigger (with no real track pairs) ?
MC Samples • DC06 – Bsff, Offline Selected Eventsat various L • MinBias and Bsff from “Minimal Layout” from recent MC Production, also at several L values. • Thanks Tomasz!
Track Multiplicity in VELO – Minimal Layout 2e32 5e32 10e32 20e32 Yes, we do seeincreased multiplicityas expected All Mean #’s: 43, 53, 74, 110 MC Matched Mean #’s: 4.8, 8.3, 19.5, 69.3 NOT MC Matched
pT Resolution How does a ~15-20% pT resolution impact the fraction of events passing the trigger? Evaluate using “Minimal Layout” pT smearing s(1/pT)/(1/pT)=(15.0 - 0.76pT + 0.86pT2)%, • At 1e33, about a factor of 500 rejection: 30 MHz ~50 kHz • Victor obtained ~30 kHz, but I do not have CAL pT > 2 GeV (~50% eff, see backup) • ~90% of the events passing are either b or c, not light ! Negligible impact of pT resolution on trigger efficiency for Bsff as well…
Trigger efficiencies in % this analysis (no CAL Cut)Current LHCb Detector using new trigger scheme,except use nominal pT Bsff Trigger Efficiencies on Offline selected events(DC06) L = 5e32 10e32 20e32 -- factors out the efficiency in reconstructing the event… more in a bit on this. V. CocoLHCb-INT-2009-008 1st number: Any “di-hadron trigger” 2nd number: All daughters MC matched to recon. long tracks 47/43 58/53 40/30 Probably due to B daughters accidentallypointing to a PV • Input files: Offline Selected Bsff events • 5e32: /castor/cern.ch/user/d/dijkstra/Selections-Upgrade/phiphi-lum5.dst • 10e32: /castor/cern.ch/user/d/dijkstra/Selections-Upgrade/phiphi-lum10.dst • 20e32: /castor/cern.ch/user/d/dijkstra/Selections-Upgrade/phiphi-lum20.dst
Bs ff “Actual” Efficiency Run over “Minimal Layout” Events. They only have DecProdCut applied, which is ~20% efficient for all 4 daughters to be in LHCb acceptance. 1K events, so stat error ~ 1% (on #’s in first 3 rows) Without and with VeloTT track req. DC06 Offline Selected MinimalLayout Most of loss in efficiency trigger? Vertex tracks “accidentally” pointing toa PV? * Assumes an “offline” analysis selection efficiency of 65% (see backup); will try and apply offline selection cuts Efficiency in LHCb light TDR is ~0.5% (esel listed as 39% there) If want to assume this efficiency, reduce efficiencies in bottom row by 40%
MinBias Rejection Comparison vs Lumi Minimal Layout • Rejection scales roughly with the #inelastic collisions / crossing… very good! • Must be reconstructing PVs efficiently & rejecting primary tracks… good • But fraction of non-empty x-ing increasing also.. • Probability of getting b or cper crossing increases as expected: • % b per X-ing: 1.3 2 3.2 7 % • % c per X-ing: XX 15 23.2 40 % • Trigger still dominated by heavy flavor, even at highest lumi… very good! • But lots of it !! • CAL pT cut can be used as a throttle, but should explore track-based selections..
Summary • Looking at offline selected events, my #’s consistent with Victors • ~15-20% pT resolution does not appear to be a problem • Looked at “Minimal Layout” for Bsff: • Lose about a factor of 2 in trigger efficiency in going from 2e32 2e33 • In Minimum Bias • Rate of events passing trigger increases ~ as expected, from10 kHz at 2e32 to 150 kHz at 2e33) • Most triggers are from real heavy flavor, not light quarks or g. • Other cuts to reduce the rate should be explored, ECAL ET cut, or/and other tracking cuts • The big issue is not really whether we can maintain the efficiency, it’s the CPU cost. • If upgraded detector provides comparable tracking performance to current offline tracks, ghosts do not swamp this first level trigger. • It would be nice to have “offline-selected” samples from the MinLayout, and to know what the efficiencies were for those selections. • To do’s: • Look at impact on other signal samples • Explore other cuts to reduce MinBias rate
V. Coco, LHCb WeekFeb 26, 2009 CAL cut removes ~50% of min bias events at 1e33. But, the reduction at Step 1 will likely remove more events than after Step 15(since after Step 15, one is guaranteed to have 2 tracks with pT>1.0, 1.25 GeV) Anyhow, 50% probably still not too unreasonable.
Bsff Offline Selectionsfrom LHCb-2007-047 • Pre-selection: • IPS > 2 to PV • DLL(K-p) > -2 • M(KK)<1050 MeV • Vertex c2(f)<100 • Vertex c2(Bs)<36 • 4 < M(Bs) < 7 GeV • Mostly loose cuts, but the PID goes as e4 • Taking 95%/kaon epid~80% Analysis selection The efficiency of these offline criteria are ~ 4663/5843 ~ 80% So, offline selections + PID are ~ 65% efficient