1. LHCb trigger: algorithms and performance Hugo Ruiz Institut de Ciències del Cosmos March 12 th -17 th 2009, Tsukuba, Japan On behalf of the LHCb Collaboration

2. 2 LHCb environment LHC bunch crossing rate: 40 MHz – 30 MHz with bunches from both directions @ LHCb Luminosity: 2·10 32 cm -2 s -1 – 10 to 50 times lower than @ ATLAS/CMS – Single pp collisions preferred to identify B vertices Rates : – Visible events (  2 tracks in acceptance): 10 MHz – bb:  100 kHz, one whole B decay in acceptance: 15 kHz Interesting B decays: BR ~ 10 -4 -10 -9 m B  5 GeV ~ 8 mm IP

3. 3 Detector overview VELO & PileUp System: primary vertex impact parameter secondary vertices Tracking Stations: p of charged particles Calorimeters: PID: e, ,  0 Muon System RICHes: PID: K,  separation Interaction point

4. 4 Trigger overview 10 MHz 1 MHz Calo + Muons VELO (PileUp system) L0: on custom boards hight p T candidates (optionally: veto busy events) Fully synchr. (40 MHz), 4  s latency ~ 2 KHz ~ 35 Kb/evt HLT: in a PC farm, full event, full flexibility HLT1: Regions of interest (RoIs). IP,  cuts HLT2 global event reco + selections

5. L0: 10  1 MHz 5

6. 6 Level 0: calorimeter triggers The LHCb calorimeter: – ECAL: 6000 cells, 8x8 to 24x24 cm 2 – HCAL: 1500 cells, 26x26 to 52x52 cm 2 L0: look for high E T candidates: – In 2x2 cell clusters – Particle identification from ECAL / HCAL energy PS and SPD information Decision based on a single candidate – Different E T threshold for each type – In addition, global variables: Total E T SPD multiplicity Scintillator Pad Detector (SPD) ECALHCAL

7. 7 Level 0: muon trigger The LHCb muon system: – 5 stations with variable segmentation – Projective geometry L0 muons: – Straight line search in M2-M5 – Look for compatible hits in M1 p T from a look-up table Decision based on single (and/or double) candidates (0,0,0) B  L0-  p T threshold (GeV) M1 M2 M3 B  J/  Minimum bias  (%) calorimeter  200 kHz

8. 8 Level 0 performance Performance based on global optimization on  10 benchmark channels: Note: efficiencies refer to events that would be offline selected –Offline selections rely on different cuts (p T, IP, …) “Off-diagonal” efficiencies useful for understanding trigger performance with real data h  e±e± 00 Comb’d E T > (GeV)3.51.3  > 1.5 2.62.34.5- rate700 kHz200 kHz 1 MHz  (B s  D s +  - ) 45%5%10%50%  (B s  J/  (  )  ) 20%90%5%90%  (B  K *  ) 30%10%70%

9. HLT: 1 MHz  2 kHz 9

10. Trigger & DAQ System Front-end: – detector read out – zero suppression – buffer data Readout network: – gigabit Ethernet – throughput: ~50 Gb/s Event Filter Farm: – ~ 1000 CPUs x 16 cores – In ~50 sub farms Front-end VeloCaloMuon L0 Trigger Trigger and Fast control Readout network Event Filter Farm CPU RICHTrackers trigger data fast control full data Y/N Y 10

11. HLT algorithm overview HLT1: – RoIs around L0 candidate – Confirm L0 candidate on tracker, VELO HLT2: – Search for all tracks using VELO segments as seeds – Selections based on a few tracks 11 L0   alley HLT1 1MHz  10 KHz  + trackSingle  di  … Global reco L0   alley L0 hadr Hadr alley... HLT2 2 KHz L0...

12. HLT1 example: hadron alleys 12 1.2 candidates pTpT IP pTpT Vertex quality & displacement IP Single hadron alley Di-hadron alley SinglDi Comb Rate3 kHz2 kHz5 kHz  (B s  D s +  - ) 45%75%  (B  h + h - ) 60%85% Single hadron line provides robustness

13. HLT1: confirmation at VELO and tracker VELO: a) Reconstruct primary vertices: VELO 2D tracking Use R sensors only: 30% of 3D time Primary vertex located with σ x,z ~ 17  m (vs 10  m with 3D), allows computing IP b) L0 candidate confirmation at VELO: Select VELO 2D tracks matching L0-calo Do 3D tracking for the selected 2D segments (15% of whole 3D tracking) In all alleys but muons: cut on IP 13  sensor R sensor 21 stations in 100 cm pp interactions Tracker: – Decode relevant region only: 1/8 of chambers, 15% of total decoding time – p(E) T resolution: L0 calo: 30%  Track confirmed in the tracker: 3% 40-100 μm pitch

14. HLT2: 10 kHz  2 kHz a)Reco all tracks using VELO segments as seeds b)Apply inclusive selections: – Displaced (2, 3, 4- track) vertices – Di-lepton (incl. no IP) – Single lepton / lepton + track – B  X +  – D decays – B  X + (  K + K - ) Example: displaced vertex – The more challenging: No PID requirement No tight mass cuts 14 Variables: p T, IP, invariant mass, track  2, pointing PV pp L0xHLT1HLT2 Rate (Hz) 31k750  (B  h + h - ) 40%80%  (B s  D s +  - ) 30%85%

15. HLT2: 10 kHz  2 kHz The degree of sophistication and specialization of the HLT2 selections will increase gradually as – Tracking stable, online tracking  offline tracking – LHC luminosity requires it The generic B sample: – Expect to record  500 Hz of semilept B decays from HLT2 lepton(+track) – From the accompanying B meson: ~ 10 9 fully contained, decay-unbiased B mesons / 2fb-1 15 unbiased B PV trigger 

16. Commissioning and start-up 16

17. Commissioning: L0 Cosmics: – ECAL-HCAL coincidence triggers used for alignment and synchronization of subdetectors since March 2008 – e ±, , h,  0 triggers checked by comparison with emulation – Less useful for L0-  as projective assumption does not work First LHC particles: – 2 runs of beam dumps into TED collimator 350m upstream of LHCb  70  candidates per event – Splashes from 1 st circulating beams – Both allowed to compare L0 muon and DAQ banks 17 trackercalo  stations Cosmic event 1 st circulating beams  stations tracker calo

18. Commissioning: HLT with full system test Full experiment system test: inject MC events into the DAQ network, exercise: – Online and DAQ: express online reco for calibration stream, send to the grid, data quality procedures – HLT services: Monitoring Configuration and control Propagation of alignment constants – Successfully run with  2 kHz output, will systematically use it till collisions arrive 18 Readout network Event Filter Farm CPU Front-end VeloCaloMuon L0 Trigger Trigger and Fast control RICHTrackers trigger data fast control full data Y/N Y 35% of CPUs deployed by end of 2009 – Take most profit from Moore’s Law! MC event injector

19. Start-up strategy  bb /  inel varies slowly with E CM LHCb wants L = 2·10 32 cm -2 s -1 – # of pp-interactions/bunch crossing stable since start-up – L  #bunches Strategy: fill  2 KHz from day one – With one full bunch in each sense: Unbiased events: MC tunning “Minimum bias” trigger (some activity in HCAL,  250 Hz of visible interactions), 10 8 events / week: flavour production studies – More bunches: Priority to muon triggers:  1 kHz Commission other L0 triggers (h, e ±, ,  0 ):  100 Hz/trigger 19 total hard bb x 100  (mb)

20. Conclusions 20

21. Conclusions Efficiencies on B decays of interest: 10 9 fully-contained decay-unbiased B mesons / 2fb-1 LHCb trigger commissioning well advanced: – L0 up and working with cosmics – HLT timing and performance (on MC) extensively tested, more tests to come before real data arrives LHCb trigger ready to exploit the B factory known as LHC 21 L0HLTTotal  (B s  D s +  - ) 50%80%40%  (B s  J/  (  )  ) 90%80%70%  (B  K *  ) 70%60%40%

22. Back-up 22

23. HLT1 Example: hadron alleys Rate reduction at each step: 23 Kalman fit powerful (many ghosts amongst high IP tracks) Rate (KHz)

24. L0 and L0xHLT1 efficiencies and rates L0 efficiency: L0 x HLT1 efficiency: 24

25. 3/17/2016HERA-LHC workshop. Hugo Ruiz25 MC generation PYTHIA 6.2 used – Minimum bias: hard QCD, single / double diffraction, elastic scattering – Signal: forcing B-mesons in a minimum bias event to decay into specific final state – Charged particle distributions tuned to data for  s < 1.8 TeV – Predicted cross-sections:  inel = 79.2 mb,  bb =633  b – Pileup (multiple interactions in single bunch crossing) simulated GEANT3 for full simulation of all events (minimum bias, signal) – Moving to GEANT4 in this year’s Data Challenge Additional backgrounds: off-beam muons, low-energy background at muon chambers Spillover simulated in detector response (from two preceding and one following bunch crossings).

26. 3/17/201626 Level 0: Pile-up system Pileup system: – 2 silicon planes measuring R coordinate – backwards from interaction point  no tracks from signal B L0: veto multi-PV evts – From hits on two planes  produce a histogram of z on beam axis Sent to L0 Decision Unit: height of two highest peaks + multiplicity Interaction region

27. bb cross-section vs CM energy 27

28. HLT1 di-hadron alley 28