1. Trigger Strategy and Performance of the LHCb Detector
Mitesh Patel (CERN)
(on behalf of the LHCb Collaboration)
Thursday 7th July 2005

2. Introduction LHCb experimental goals :
Precision measurements of CP Violation in B decays
Aim to (over-)constrain the unitarity triangle by making measurements in multiple channels
b production predominately at small polar angles →
LHCb optimised as single forward arm spectrometer
To meet the physics goals require a trigger which can select :
Multitude of signal channels in the LHCb experimental environment
Channels required for calibration, alignment and systematic studies
Channels that allow the efficiency of the tagging of B flavour to be evaluated
Unbiased control channels
System must be simple, robust and flexible
7th July 2005
HCP Mitesh Patel

3. The LHCb Trigger Environment
LHC Bunch crossing frequency: 40 MHz
Non empty bunches → 30 MHz
LHCb Luminosity : 2×1032 cm-2s-1
10-50 times lower than ATLAS, CMS
B decays → displaced secondary vertex,
need ~1 interaction/event
‘Visible’ interactions at 10 MHz
100 kHz bb events (800 kHz cc)
15% of bb events: all decay products of at least one B in detector
Branching ratio of interest : 10-3 to 10-7
Use information from a variety of LHCb’s detectors to reduce 10 MHz …
7th July 2005
HCP Mitesh Patel

4. Detectors in the LHCb Trigger
VErtex LOcator
primary vertex
impact parameter
displaced vertex
Scintillator Pad Detector
Charged multiplicity
Trigger Tracker
p, pT
Calorimeters
PID: e,, 0
Trigger on hadr.
Muon System
Pile-up system multiple interactions, charged multiplicity
7th July 2005
HCP Mitesh Patel

5. Trigger Strategy The physics goals of LHCb motivate : → Data mining
Exclusive triggers : ‘hot’ physics eg. BsDsh, Bsff, B0J/y KS, B0D*p, B(s)h+h-, B0K*m+m- , B0D0 K*, Bsm+m-, BsJ/y f, Bsfg
Inclusive triggers :
Inclusive single-muon sample [independent of signal type ]
Sample triggered independent of signal type – unbiased on the signal side
Signal trigger efficiencies
Inclusive di-muon sample [selected without lifetime information]
Clean mass peaks for alignment, momentum (B field) calibration
Proper time resolution using prompt J/ events
Inclusive D* sample [selected without RICH information]
Clean signal of D*D(K)
Measure PID performance as a function of momentum
→ Data mining
7th July 2005
HCP Mitesh Patel

6. Trigger Overview LHCb will use three levels of trigger : 10 MHz 1 MHz
Level 0 Trigger [4 ms] [hardware]
‘high’ pT particles in calorimeters and m detector
Pile-up System throws away busy events
Level 1 Trigger [~1 ms] [software]
Partial read-out: Vertex Detector (VeLo), Trigger Tracker (TT) and L0 summary
Find high IP tracks, estimate pT of tracks, link to L0 objects
pT of the two highest pT tracks
High Level Trigger [~10 ms] [software]
Confirmation of L1 decision then full reconstruction of event
Exclusive selections for most important physics channels
Inclusive selections
10 MHz
1 MHz
Already well developed, relatively fixed
40 kHz
Now being developed
200 Hz
1800 Hz +
L1 and HLT will be run on a single ~1600 CPU PC farm
7th July 2005
HCP Mitesh Patel

7. The Level 0 Trigger: Overview
2 highest pT muons
m CHAMBERS
Fast search for ‘high’ pT particles
Cut on global variables
L0 has 4 ms latency
“L0 OBJECTS”
Highest ET g, electron, p0, hadron candidates
CALORIMETERS
L0 decision unit
L0DU
SET
CALORIMETERS
GLOBAL VARIABLES
z and # trks in 1st, 2nd vtx
PILE-UP SYSTEM
Charged particle multiplicity
SPD, PILE-UP SYSTEM
7th July 2005
HCP Mitesh Patel

8. Level 0: Muon Trigger Search for high pT muons
Five muon stations M1-5
Variable granularity
Projective geometry
2 highest pT candidates per quadrant sent to L0 decision unit
p/p ~ 20% for b-decays
Typical Performance:
~88% efficiency on B→J/(µµ)X
7th July 2005
HCP Mitesh Patel

9. Level 0: Calorimeter Trigger
ECAL: 6000 cells, 8x8 to 24x24 cm2
HCAL: 1500 cells, 26x26, 52x52 cm2
Look for high ET candidates in the calorimeters :
In regions of 2x2 cells
Particle identification from
ECAL / HCAL energy
PS and SPD information
ET threshold ~ 3 GeV
Sent to L0 decision unit:
Highest ET candidate of each type
Global variables
Total calorimeter energy
SPD multiplicity
Typical Performance: 30-50% efficiency on hadronic channels for about 700 kHz bandwidth
ECAL
HCAL
Pre-Shower Detector (PS)
Scintillator Pad Detector (SPD) FE
SPD-PreShower
ECAL
HCAL
Validation cards
Selection crates e± g p0 p0
ETtot
SPD mult.
hadr
7th July 2005
HCP Mitesh Patel

10. Level 0 : Performance L0 Decision unit : Performance :
OR of high ET candidates
Applies cuts on global variables
Performance :
Efficiency ~50% for hadronic channels, 90% for
m channels, 70% for radiative channels
1% bb → 3% after L0
8% cc → 10% after L0
Type
Threshold
(GeV)
Rate (kHz)
Hadron
3.6
705
Electron
2.8
103
Photon
2.6
126
p0 local
4.5
110
p0 global
4.0
145
Muon
1.1
Di-muon
1.3
Global Variable
Cut
Tracks in 2nd vtx 3
Pile-Up multiplicity
112 hits
SPD multiplicity
280 hits
Total ET
5 GeV
7th July 2005
HCP Mitesh Patel

11. The Level 1 Trigger: Overview
2D (rz) tracking in VELO
Find high IP tracks (VErtex LOcator)
Confirm track / Estimate pT from Trigger Tracker
Link VELO tracks to L0-objects
L1 algorithm run on PC Farm
Average latency: 1 ms (max 50 ms)
Primary Vertex search
Allow up to 3 PV
2D track selection
0.15mm < IP < 3mm
L0 m match
Multiple routes through L1 :
Generic Line :
L1-Variable: log(pt1)+log(pt2)
pt1,2 two highest pT tracks
Muon lines :
Single muon: pT>2.3 GeV, IP >0.15 mm
Dimuons: J/ ± 500 MeV window
OR : mµµ>500MeV and IP>0.05mm
OR : mµµ>2.5GeV
Photon, electron lines :
L1-Variable (relaxed) + ECAL>3.1 GeV
3D (rfz) VELO tracking
Confirm IP
L0 m match
p, pT estimation
Use VELO-TT track + fringe B field
OR : VELO track + L0 muon
L1 decision
7th July 2005
HCP Mitesh Patel

12. Level 1: Event Reconstruction
L1 relies on the LHCb VErtex LOcator (VELO) :
Silicon tracker before the LHCb magnet
Angular coverage of full angular range of downstream detector
Sensors ~7mm away from beam, retractable (injection), in secondary vacuum
Foils protect against RF pickup from the LHC beam
21 sensor stations : 2 R- and 2 f-measuring sensors per station
Gradual increase of pitch (40 mm to 103 mm)
Si Sensors
RF foils
Interaction region
f sensor
R sensor
100 cm
7th July 2005
HCP Mitesh Patel

13. Level 1: Event Reconstruction
Fast tracking strategy :
~70 tracks/event after L0 : perform tracking
in R-z view (using only R sensors)
Primary vertex σZ ~ 60 mm, σX,Y ~ 20 mm
Select 2D tracks with 0.15 < IP < 3 mm
→ 8.5 tracks/event
3D tracking for selected tracks
pT measurement using Trigger Tracker
TT : two layers of Si detectors with 200 mm pitch
Only 0.15Tm of B field between VELO and TT
→ DpT / pT ~ 20-40%
allows rejection of low p tracks
which can fake high IP
Matched L0-m :
→ DpT / pT ~ 5%
z-vertex histogram
xy-vertex
Example: 2D tracks in 45o
sz~60mm
sx,y~20mm
The Trigger Tracker
7th July 2005
HCP Mitesh Patel

14. Level 1 : Performance L1 decision : Performance :
Take OR- of the multiple routes through L1
Tuned for retention of 4% of minimum bias L0 triggers
(40 kHz L1 output rate)
Performance :
eg. Generic L1-Variable: log(pt1)+log(pt2)
log(pt1)+log(pt2)
L0 efficiency
L1 efficiency
L0L1 eff
Channel
Generic
Single m
Di-m
J/Y
Electron
Photon
Total
Bd0→ p+p-
81.8%
1.6%
0.1%
4.3%
2.7%
82.6%
Bs0→ Ds-K+
79.8%
4.0%
0.7%
0.8%
4.4%
2.9%
80.9%
Bd0→ D0(K+p-)K*
83.5%
2.0%
0.4%
5.0%
3.3%
85.4%
Bs0→J/Y(m+m-)f
74.3%
42.8%
25.0%
45.5%
1.9%
1.7%
87.2%
Bd0→ K*g
54.9%
2.4%
0.3%
17.3%
30.9%
67.2%
Minimum Bias
0.2%
3% bb after L0 → 16% after L1
10% cc after L0 → 18% after L1
7th July 2005
HCP Mitesh Patel

15. The High Level Trigger: Overview
High Level Trigger (HLT) :
Generic Algorithm: repeat L1 then full readout of the detector
Muon Highway feeds inclusive muon modes
Form basic particles, composites, search for signatures of hot physics channels (exclusive), D* inclusive
10ms to run (in 2007) – algorithm run on same CPU farm as L1
Bs → ff
Bd → D*p
Bd → D0K*
Bd → mmK*
B → J/yX
Bs → Dsp
Bs → fg
Exclusive HLT
Inclusive di-m
Inclusive D*
Inclusive B→m
Ds→KKp
f →KK
D0→Kp, KK
Loose D0→hh
Loose dimuons
K*→Kp
muons
Photons,
electrons
p, K
Generic HLT
Muon Highway
7th July 2005
HCP Mitesh Patel

16. HLT Generic Redo “L1” with improved:
momentum resolution
muon matching
HLT generic reconstructs ~1/3 of tracks in the event and redoes L1 in ~4 ms
Reduces rate from 40 kHz input from L1 to 10 kHz:
16% bb after L1 → 38% after HLT Generic
18% cc after L1 → 27% after HLT Generic
Then have ~24 ms for further HLT selections
7th July 2005
HCP Mitesh Patel

17. HLT Exclusive HLT Exclusive being tuned for ~10 core physics channels
In these channels cuts tuned to take ~15Hz MB / channel
B mass resolutions ~30 MeV
Mass windows >±500 MeV
Bs→ DsK
Bs→Dsp
Bd→D*p
Bd→pp
Bs→KK
Ds→KKp
Bs→mm
Bd→Kp
Bs→pK
B → hh reconstructed
as B→pp
7th July 2005
HCP Mitesh Patel

18. Efficiencies w.r.t. Offline and L0xL1 selected signal
HLT : Performance
HLT still under study, efficiencies on L0L1 and offline selected events 60-90%
Philosophy to try and trigger channels in many ways
Further inclusive triggers will make us more robust to the unknown
Limited time available for online tracking → different to offline : inefficiency in high multiplicity channels – strategies being explored to resolve this …
Channel
Efficiencies w.r.t. Offline and L0xL1 selected signal
Generic
Tracking
Total Efficiencies
Excl. B mm m D*
Total
Bs  m+m-
99%
93%
91%
90%
94%
Order of 1%
98%
Bd  K*m+m-
82%
73%
62%
58%
Bd,s  h+h-
95%
88%
Order of 0.5%
Order of 2%
Bs  fg
71%
61%
Bs  Dsh
60%
Bd  D*p
48%
43%
55%
7th July 2005
HCP Mitesh Patel

19. → HLT exclusive selection efficiency e = 0.71
Bs→ff selection on L0L1 and offline selected events :
Online tracking etracking = 0.73
HLT selection cuts eselection = 0.97
→ HLT exclusive selection efficiency e = 0.71
Select only 3 of the 4 tracks (fK), use online RICH to control the background :
Online tracking etracking = 0.93
HLT selection cuts eselection = 0.94
→ HLT exclusive selection efficiency e = 0.87
1200MeV
1000MeV
Before RICH cut
3 track
4 track
After RICH cut
fK mass
ff mass
7th July 2005
HCP Mitesh Patel

20. Overall Trigger Performance
Bs  J/Y f ~70%
Bd  p+p- ~37%
Bs  DsK ~23%
Level-0
Level-1
HLT
Total 2 KHz
bb 900 Hz
cc 600 Hz
7th July 2005
HCP Mitesh Patel

21. Trigger Robustness Several scenarios considered : Event multiplicity
Noise, misalignment, resolution
Increased material
LHC beam position
LHC background
Size of the CPU farm
The performance of L0 is stable within 10%, L1 is stable within 20%
The execution time and L1 event size is stable to within 30%
7th July 2005
HCP Mitesh Patel

22. Real Time Trigger Challenge
But will it work … ?!
The Real Time Trigger Challenge :
Operate one (few) subfarms of the DAQ under realistic conditions
44 double CPU boxes
Full-speed Data Input
Long-term operation (hours)
Exercise realistic Level-1/HLT code
Exercise/evaluate realistic overheads
Establish performance of ‘modern’ CPUs compared to (today’s) standard CERN
Happening now !
7th July 2005
HCP Mitesh Patel

23. Conclusions
LHCb will use three level of trigger to deliver inclusive and exclusive samples of B decays suitable for it’s physics goals :
Exclusive selection of core physics channels
Samples to allow calibration/alignment studies
Data that allows the trigger efficiencies to be determined
The L0 and L1 triggers are well developed and performance is good
The High Level Trigger continues to evolve as our understanding of the LHCb physics potential evolves
The trigger system is robust and flexible
7th July 2005
HCP Mitesh Patel

24. Level 0: Pileup System B A Silicon r-sensors RA ZPV - ZA RB ZPV - ZB
Two planes of Silicon sensors upstream of the interaction point
Used to identify and reject multi-Primary-Vertex events :
Measure R coordinate (-4.2<<-2.9)
From hits on two planes  produce a histogram of z on beam axis
Remove hits contributing to largest peak, look for 2nd peak above threshold
L0 Decision Unit cuts on # of tracks in the second peak + hit multiplicity
Performance :
Vetoes 60% of double interactions keeping 95% of single interactions RA
ZPV’ k’ ZB ZA RB
ZPV
B A k
Silicon r-sensors
RA ZPV - ZA
RB ZPV - ZB =
k 
7th July 2005
HCP Mitesh Patel