7 Nov 2002Niels Tuning - Vertex A vertex trigger for LHCb The trigger for LHCb ….. and the use of the Si vertex detector at the first and second trigger levels Vertex2002 Hawai’i, 7 Nov 2002 Niels Tuning (CERN) (on behalf of LHCb)

7 Nov 2002Niels Tuning - Vertex LHCb A Large Hadron Collider Beauty Experiment for Precision Measurements of CP-Violation and Rare Decays Colliding beams: 25 ns 7 TeV x 7 TeV pp L = cm -2 s -1  (visible) = 68 mb  (pp  b  bX) = 0.5 mb  ~10 12 b  b / year BR( interesting channels ) ~10 -2 – Finding B-mesons: High P T decay products Large lifetime  sec.vertex Invariant mass A low multiplicity B   +  - event LHCb trigger = looking for a needle in a haystack… …every 25 ns!

7 Nov 2002Niels Tuning - Vertex LHCb Trigger – Overview L0: high P T Pile-up Veto, using VETO detector High E T calorimeter objects High P T muons L1: high P T + impact parameter High impact parameter tracks, using VELO detector High P T tracks, using TT detector and L0 info L2+L3: high P T + displaced vertex + B-mass + PID Use tracking stations and RICH Bunch crossing rate40 Mhz Non empty bc rate30 MHz Visible interaction rate~12 MHz Input to L0~10 MHz Input to L11 MHz Input to higher level40 kHz Writing to disk0.2 kHz

7 Nov 2002Niels Tuning - Vertex LHCb detector L0 vetoL1L0 trigger TT T1 T2 T3

7 Nov 2002Niels Tuning - Vertex L0 – Pile-up VETO (L0 = first trigger level) Purpose: remove multiple interactions Nominal luminosity: L = cm -2 s -1 Single : Double : Triple  16 : 4 : 1  75% : 20% : 5% Why? More difficult to find high IP tracks at L1 Reduce bandwidth for L0 Detector: 2 Si disks (4 sensors) Same sensors as VErtex LOcator (see talk J.Palacios) Only R information Use Beetle chip  OR of 4 strips: comparator output of 4 channels  1280 channels for 2 disks 320 strips 84 mm 16 mm

7 Nov 2002Niels Tuning - Vertex L0 – Pile-up VETO algorithm Calculate vertex for all combinations of 2 points a and b. Find highest peak (PV) Remove the hits and find 2 nd peak Veto if peak>threshold  (Z vtx )  2.8 mm,  (beam)  53 mm Z vtx (cm) Ra (cm) Rb (cm) combinations trueall

7 Nov 2002Niels Tuning - Vertex L0 – Pile-up VETO performance B   +  - L0 efficiency increase from 50% to 60%  the L0 P T,HADR threshold can be lowered from 4 GeV to 3.6 GeV Reduce bandwidth and enhance purity: Pileup VETO vetoes ~15% of all events Vetoed events are more likely to trigger Only small inefficiency for single interactions: ~5% Reject ~30% of multiple interactions (NB: multiple interactions include inelastic+elastic !) Same L0 output rate!

7 Nov 2002Niels Tuning - Vertex CPU SCI Scheduling network RU L1 trigger: vertex trigger (L1 = second trigger level) Implementation: Clustering in FPGA on front-end  Send data to RU (3-4 GB/s) CPU-farm:  300 – 400 CPUs  2D torus Use scheduler Prototype with 32 CPUs running at 1.24 MHz Buffer depth: 1820 events  Latency = 1.65 ms Strategy: Find 2d-tracks with R-sensors and reconstruct vertex Reconstruct high-impact parameter tracks in 3d Extrapolate to TT through small magnetic field  P T Match track to L0 Muon objects  P T and PID Select B–events using impact parameter and P T information

7 Nov 2002Niels Tuning - Vertex L1: VErtex LOcator (see talk by Juan Palacios) Si: 220  m thick, n-on-n, Pitch:  37–98  m, R 40–92  m Sens. area: 0.8 cm < R < 4.2 cm 21 stations (84 sensors) cm < Z < 75 cm 170,000 channels RF foil: Very thin “beampipe” to separate prim. and sec. vacua R sensor 2048 strips  sensor: stereo angle +10 o,-20 o R sensor: 4 inner and 2 outer  sectors  sensor 2048 strips

7 Nov 2002Niels Tuning - Vertex L1: input data Velo clusters: Clusters are found in FPGAs per groups of 32 strips. Digital (offline is analog) ~1000 clusters ~0.1% noise clusters (  200) TT clusters ~300 clusters L0 objects  3 muons Some calorimeter data Cluster resolution: (testbeam)  =14  m 1000 clusters (simulation) Pitch (  m) Resolution (  m)

7 Nov 2002Niels Tuning - Vertex multiplicity ~60 L1: track reconstruction Look only for R-clusters:  2d RZ-tracks Fast! Find “triplets” of clusters Combine triplets ~98% efficiency for B-tracks Z vtx histogramX,Y vtx VELO is being redesigned to 45 o sectors: faster L1 tracking lower noise less 2d tracks 2d tracks in a 90 o sector:

7 Nov 2002Niels Tuning - Vertex L1: primary vertex Primary vertex reconstructed with 2d tracks XY information from  segmentation Flight direction of B is forward  RZ projection of impact parameter contains most information XZ Vertex resolution:Impact parameter: 2d3d Lifetime

7 Nov 2002Niels Tuning - Vertex L1: P T information – L0 Match VELO track to Muon from L0:  PID  Momentum Efficient selection of B s  J/  (  )  B d  J/  (  ) K s Enhance  - tagged sample VELO MAGNET MUON dp/p=4.8% TT CAL

7 Nov 2002Niels Tuning - Vertex Good momentum resolution, cut on J/  mass: ~60% of events contain both muons <1% min.bias retention OR require 1 muon with high PT, high IP ~75 % efficiency ~3% retention Work ongoing: achieved 90% eff. using neural net L1: algorithm (1) Preliminary J/  mass Performance with all event info (except TT)

7 Nov 2002Niels Tuning - Vertex L1: Trigger Tracker TT = Tracking station before the magnet Design still under study Si:  m thick Wide pitch: 200  m Sensor dimensions: 7.8 x 11.0 cm 2 4 layers (x,u [30 cm gap] v,x) Stereo angle: 0 0, -5 0, +5 0, 0 0 To be optimized 836 sensors (~7 m 2 ) 141 cm 116 cm VELO TT MAGNET RICH-1

7 Nov 2002Niels Tuning - Vertex L1: P T information – TT High impact parameter 2d tracks are reconstructed in 3d and extrapolated to TT1 Magnetic field between VELO and TT:  B dl  0.1 Tm Ensures momentum information dp/p ~ 30% 30% ---  p (GeV) 30 GeV ---  dp/p

7 Nov 2002Niels Tuning - Vertex L1: algorithm (2) Preliminary Get two highest P T tracks, using TT Consider impact parameter and P T of these tracks Look in plane  IP/  (IP) vs  P T Bd+-Bd+- BsDs-K+BsDs-K+ Signal Minimum bias

7 Nov 2002Niels Tuning - Vertex Tracking+Vertexing time (ms) Events ms L1: performance - timing Remember: latency ~1.7 ms Possibly x32 more. A more flexible system is under study were CPUs from DAQ can be used for the TRIGGER and vice versa Tracking+Vertexing: < 20 ms 2007 CPUs: x8 faster Optimize algorithm+code  in the right ballpark! 2d tracking: ~70% Vertex: ~15% 3d tracking (a few) ~15%

7 Nov 2002Niels Tuning - Vertex L1: performance Efficiency vs retention: (Example B   +  - ) Expected overall trigger performance: (cumulative) Bd+-Bd+- L1+L0 info L1+P T info 40 kHz L0  L1L1  L2 L1 output rate (MHz) B d   +  - efficiency

7 Nov 2002Niels Tuning - Vertex Conclusions The pileup-VETO detector efficiently rejects multiple primary 40 MHz at L0 The VELO detector reconstructs primary vertices at L1 with excellent resolution  (Z vtx )  60  m  high impact parameter tracks can be identified The TT detector - or L0 information - enables measuring the momentum of tracks  Efficient L1 selection algorithms under study Efficiency of 70% (90%) for B   +  - ( B  J/  (  ) K s ) reachable at 4% minimum bias retention