eduardo rodrigues nikhef seminar at cppm, marseille, 4 th may 2007

Eduardo RodriguesEduardo RodriguesNIKHEFNIKHEF

Seminar at CPPM, Marseille, 4Seminar at CPPM, Marseille, 4thth May 2007 May 2007

LHCb: a stroll from Trigger to LHCb: a stroll from Trigger to PhysicsPhysics

LHCb: a stroll from Trigger to Physics 2/54

Eduardo Rodrigues CPPM, 4th May 2007

ContentsContentsI.I. LHCb: the must knowLHCb: the must know

II.II. Detector and simulationDetector and simulation

III.III. Trigger systemTrigger system

IV.IV. Reconstruction and PIDReconstruction and PID

V.V. Analysis: tagging and event selectionAnalysis: tagging and event selection

VI.VI. Physics sensitivity studiesPhysics sensitivity studies

Disclaimer: no complete review of all aspects; emphasis put on personal work



9 km diameter

GenevaGenevaJuraJura

CERN



I. LHCb: the must knowI. LHCb: the must know

LHCbLHCb

Goal: B-physics studies CP violation rare B-decays

Acceptance: 1.8 < < 4.9

Luminosity: 21032 cm-2

s-1

Nr of B’s / year: 1012

How forward is LHCb?How forward is LHCb? B hadrons mainly produced in forward B hadrons mainly produced in forward region(s)region(s)

both B’s tend to be correlatedboth B’s tend to be correlated

Unique forward spectrometer

at high energy!

Excellent tracking

Excellent PID



B-hadronsB-hadrons are heavyheavy and long-lived !long-lived !

LHC(b) environmentLHC(b) environmentLHC environmentLHC environment

pp collisions at Epp collisions at ECMCM = 14 TeV = 14 TeV

ttbunchbunch = 25 ns = 25 ns bunch crossing rate = 40 MHz bunch crossing rate = 40 MHz

<L> = 2x10<L> = 2x103232 cm cm-2-2 s s-1 -1 @ LHCb interaction region@ LHCb interaction region

10-50 times lower than for ATLAS/CMS

~ 7x1012 pairs~ 700 kHz~ 3.5 mb (c-cbar)

~ 100 kHz

~ 12 MHz

Event rate

~ 60 mb visible

~ 0.5 mb

~ 100 mb

Value

~ 1012 pairsb-bbar)

total

Yield / yearPhysical quantity

Expected B-signal ratesExpected B-signal rates branching ratios ~ 10branching ratios ~ 10-9-9 – 10 – 10-4-4

10 – 1010 – 106 6 events / year ?events / year ?

Cross sectionsCross sections



II. LHCb detectorII. LHCb detector

(side view)

pp collision

Acceptance

10 mrad

250/300 mrad

« Tracking » detectors« Tracking » detectors Muon SystemMuon System

CalorimetersCalorimetersRing Imaging CherenkovRing Imaging Cherenkov



LHCb LHCb caverncavern



LHCb Monte Carlo simulation software:

Pythia, EvtGen, GEANT4 and Gaudi-based reconstruction– Detailed detector and material description (GEANT)– Pattern recognition, trigger simulation and offline event selection– Implements detector inefficiencies, noise hits, effects of events from the previous

bunch crossings

SimulationSimulation



III. The LHCb III. The LHCb TriggerTrigger

Trigger overview and strategyTrigger overview and strategy

Level-0: first level triggerLevel-0: first level trigger

HLT: High Level TriggerHLT: High Level Trigger



HLT

Level-0 custom hardware high ET particles partial detector information

CPU farm -> software trigger high ET / IP particles full detector information

Trigger overviewTrigger overview

Storage

pp collisions

12 MHz

1 MHz

~2 kHz

event size ~35 kb



Two-level TriggerTwo-level Trigger

L0L0 high ET / PT particles

hardware trigger, sub-detector specific implementationhardware trigger, sub-detector specific implementation

pipelined operation, fixed latency of 4 pipelined operation, fixed latency of 4 ss

(minimum bias) rate reduction ~12 MHz -> 1 MHz(minimum bias) rate reduction ~12 MHz -> 1 MHz

HLT:HLT: high ET/PT & high Impact Param. particles & displaced vertices & B-mass & …

algorithms run on large PC farm with ~1800 nodesalgorithms run on large PC farm with ~1800 nodes

several trigger streams to exploit and refine L0 triggering informationseveral trigger streams to exploit and refine L0 triggering information

software reconstruction on part/all of the datasoftware reconstruction on part/all of the data

tracking / vertexing with accuracy close to offline

selection and classification of interesting physics eventsselection and classification of interesting physics events

inclusive / exclusive streams

rate reduction 1 MHz -> 2 kHzrate reduction 1 MHz -> 2 kHz

estimated event size ~ 30kbestimated event size ~ 30kb

Trigger StrategyTrigger StrategyBe alert !Be alert !



select high Eselect high ETT / P / PTT particles particles

hadrons / electrons / photons / 0’s / muons

reject complex / busy eventsreject complex / busy events

more difficult to reconstruct in HLT

take longer to reconstruct in HLT

reject empty eventsreject empty events

uninteresting for future analysis

L0 thresholds on ET / PT of candidates

global event variables

Level-0 Trigger strategyLevel-0 Trigger strategy

Pile-up systemPile-up system CalorimeterCalorimeter Muon systemMuon system



L0 Pile-up system: 2 silicon planes upstream of nominal IP,

part of the Vertex Locator (VELO)

Identifies multi-PV events:- 2-interactions crossings identified with

efficiency ~60% and purity ~95%

L0 Calorimeter : Identify high-ET hadrons / e’s / ’s / 0’s

Electromagnetic and hadronic calorimeters- large energy deposits ET in 2x2 cells

Scintillator Pad Detector (SPD) and Preshower (PS)- used for charged/electromagnetic nature of clusters, respectively

L0 Muon System: M1 – M5 muon stations (4 quadrants each)

σp/p ~ 20% for b-decays

Level-0 sub-systemsLevel-0 sub-systems

Scintillator Pad Detector (SPD)

Pre-Shower Detector (PS)

ECAL HCAL



Pile-up system

total multiplicity # tracks in second vertex

Muon system

2 candidates per each of 4 quadrants

Calorimeter

total ET in HCAL SPD multiplicity highest- ET candidates:

h, e, , 2 0‘s

L0 Decision unitL0 Decision unit cuts on global event variables cuts on global event variables thresholds on the Ethresholds on the ETT candidates candidates

L0DU reportL0DU report

Level-0 Decision UnitLevel-0 Decision Unit

1 MHz



1302.6Photon

1104.5o local

1504.0o global

280

1002.8Electron

7007003.6Hadron

1501.3Di-muon160

1101.1Muon

Approx. rate (kHz)Threshold (GeV)Trigger

~ 13

~ 8.3

3Tracks in 2nd vertex

112 hitsPile-up multiplicity

280 hitsSPD multiplicity~ 7

5.0 GeV ET

Rate (MHz)CutGlobal event cuts

… and then cuts on the EETT / P / PTT candidates candidates

Global event variablesGlobal event variables applied first …

Di-muon trigger is specialDi-muon trigger is special

- not subject to the global event selection- PT

= PT1 + PT

2 with PT2 = 0 possible

- “tags” clean B-signatures

Decision Unit Decision Unit VariablesVariables

Redundancy:Sub-triggers overlap

Level-0 bandwidth divisionLevel-0 bandwidth division

Overall optimization given benchmark channelsOverall optimization given benchmark channels



Level-0 performance studiesLevel-0 performance studies

45

105

50

20

5

90 90

30

60

10

70

0102030405060708090

HadronicChannels

MuonicChannels

ElectromagneticChannels

HCALECALMuonTotal

Dedicated sub-triggers most relevant for each « channel type »Dedicated sub-triggers most relevant for each « channel type »

Bs → Ds B →

B → J/(Ks

B → K* B → K* B → J/(eeKs

Typical Level-0 performanceTypical Level-0 performance

N.B.: efficiencies with respect to events selected offlineN.B.: efficiencies with respect to events selected offline



StrategyStrategy

Independent alleysIndependent alleys::Follow the L0 triggered candidate

Muon, Muon + Hadron,

Hadron, ECal streams

~2 kHz

Summary Information:decision, type of trigger fired, info on what triggered

Partial Reconstruction: A few tracks selected per alley (cuts e.g. on PT, Impact Parameter, mass)

full reconstruction done at the end of the alleys

High Level TriggerHigh Level Trigger



Trigger FarmTrigger Farm Event Filter Farm with ~1800 nodes

(estimated from 2005 Real-Time Trigger Challenge)

Sub-divided in 50 sub-farms

Readout from Level-0 at 1 MHz 50 Gb/s throughput

Scalable design possible upgrade

FE

Readout network

CPU

FE FE

L0 trigger

Timing and Fast Control

CPUCPUCPU CPUpermanent

storage

HLT algos CPU time tested on a real farn

will fit in the size of the farm foreseen



Muon Alley - StrategyMuon Alley - Strategy

L0- Entry

MuonPre-trigger

MuonTrigger

~200 kHz

~20 kHz

~1.8 kHz

Muon PreTrigger Standalone reconstruction: σp/p ~ 20%

VELO tracks reconstruction Primary vertex reconstruction Match VELO tracks and muons: σp/p ~ 5%

Muon Trigger Tracking of VELO track candidates in the

downstream T stations: σp/p ~ 1%

Refine identification:

match long (VELO-T) tracks and muons



Muon Alley – PerformanceMuon Alley – Performance

Muon PreTrigger bμ ~11% Signal efficiency: ~88%

Muon Trigger Single muon

• PT> 3GeV and IPS> 3

• B content 60% Dimuon

• mass >0.5GeV and IP>100m

• J/mass>2.5GeV (no IP cut!) Signal efficiency: ~87%

J/Ψ

~20 kHz

~1.8 kHz

dimuon mass (MeV)dimuon mass (MeV)

< 1s of LHCb< 1s of LHCb



Exclusive SelectionsExclusive Selections

Bs→DsBs→ DsK

4 tracks

Exclusive selections– Use common available reconstructed and

selected particles (Ds, D0, K*, Φ,..)

– Wide B-mass windows (typically ~ 500 MeV)

– Efficiency: e.g. ~90% for Bππ~200 Hz

Bs →

Bs →

Ds→KK

→KK

, K

Bs → Ds

Off-line B → On-line B →



IV. Reconstruction and IV. Reconstruction and PIDPID

Tracking: pattern recognition and fittingTracking: pattern recognition and fitting

VertexingVertexing

RICH, Calo. and Muon reconstructionRICH, Calo. and Muon reconstruction

Particle Identification (PID)Particle Identification (PID)



Tracking strategyTracking strategy

A typical event:A typical event: 26 long 26 long tracks tracks best quality for physics: good P & IP resolution best quality for physics: good P & IP resolution 11 upstream tracks 11 upstream tracks lower p, worse p resol., but useful in the RICH1 pattern recognition lower p, worse p resol., but useful in the RICH1 pattern recognition 4 downtream tracks 4 downtream tracks improves K improves Kss finding efficiency (good p resolution) finding efficiency (good p resolution) 5 seed/T tracks 5 seed/T tracks used in the RICH2 pattern recognition used in the RICH2 pattern recognition 26 VELO tracks 26 VELO tracks used in primary vertex reconstruction: good IP resolution used in primary vertex reconstruction: good IP resolution

|B| (T)

Z (cm)

B-field measured to ~ 0.03% precision!B-field measured to ~ 0.03% precision!

A set of PR algorithmsA set of PR algorithms

Multi-pass strategyMulti-pass strategy



Pattern recognition - Pattern recognition - strategystrategyVELO tracking

(VELO seeds)

Forward tracking(long tracks)

Matching(long tracks)

Seeding(T seeds)

VELO-TT(upstream tracks)

Clonekilling

Set of “best” tracks

Ks tracking(downstream tracks)



Pattern recognition - Pattern recognition - performanceperformance

Eff

icie

ncy

Eff

icie

ncy

p (GeV)

Performance on B-signal samples for “long” tracks: Efficiency ~ 95 % for p > 10 GeV Ghost rate ~16.7 % Dip due to beam-pipe material



Tracks- pattern recognition attributes detector measurements to a track- modelled with straight-line segments, the (track) states

Track fitting- determine the best track parameters and corresponding covariance matrix given the measurements on the track- take into account multiple scattering and energy losses in the detector material

“banana-shaped” Outer Tracker modules

Track fittingTrack fitting

Description of a realistic detector with mis-alignments- introduction of trajectories to model “measurements”

(concept adapted from BaBar)- trajectory = 3D curve in space (of arbitrary shape)

- detector shapes locally described with trajectories- a track state (straight-line segment) can also be

modeled as a trajectory

- any trajectory can provide a parabolic approximation of its shape at any point along itself closest approach track state – measurement becomes a general point-of-closest-approach problem!



LHCb uses the Kalman filter technique:

Mathematically equivalent to least 2 method Adds measurements recursively starting from a initial track estimate Reconstructs tracks including multiple scattering and energy losses We use a “bi-directional” fit …

Track fitting « à la Kalman »Track fitting « à la Kalman »

direction of the filter track predictionfiltered track

direction of the reverse filter

track prediction filtered track

Bi-directional fitBi-directional fit

Filter in both directions … smoothed result = weighted mean of both filters



Fitting from a seed (track) stateFitting from a seed (track) state

• Measurements(hit & error)

• Seed State(position, direction &

momentum)

• Prediction(adaptive 5th order

Runge-Kutta method solves motion in the inhomogeneous B)

• Kalman Filtered(update prediction with

Measurement info)



Fitter: prediction and Kalman filterFitter: prediction and Kalman filter

Velo TT

IT&OT



Fitter: Kalman smootherFitter: Kalman smoother

Velo TT

IT&OT



Bi-directional fittingBi-directional fitting

Velo TT

IT&OT

filtered state

weighted mean

filtered state



Track fitting performanceTrack fitting performance

p/p = 0.5%

Momentum resolution for “long” tracks p/p = 0.35 – 0.55 %



Vertexing performanceVertexing performance

bt

Bs K

K

,K

Ds

Primary vertex

p p

Bs vertex resolution

Typical resolutions

z,core = 168 m

Primary vertex resolutionsx = 8 m z = 47 m



RICH reconstructionRICH reconstruction

Algorithm: Input: all types of tracks available Search for hits assuming different hypotheses Global ring fit Cherenkov angle resolutions in the range 0.6 – 1.8 mrad

RICH1 RICH2

nc

1)cos(



RICH performaceRICH performace

K

Particle Momentum (GeV/c)

K K or p

K or p

0

20

40

60

80

100

e, or

K e, or

Particle Momentum (GeV/c)

0

20

40

60

80

100

identification K identification

Particle identification needed over momentum range 1-100 GeV/c

Ex.: need to distinguish Bd from other similar topology 2-body decays



Calorimeter reconstruction (1/2)Calorimeter reconstruction (1/2)

electrons: Identification with ~95% based on

- 2 of match calo. cluster energy – position of extrapolated track- Energy deposited in the preshower detector- Energy deposited in the hadronic calorimeter- Bremsstrahlung correction

ECALMagnet

EPS(MeV)

True Electronsbackgrounds

example



Calorimeter reconstruction (2/2)Calorimeter reconstruction (2/2)

Neutral pions: Resolved 0 :

- reconstructed from isolated calo. photon clusters- mass resolution ~10 MeV, ~30%

Merged 0 :- Several clustes merged into single (high energy) cluster- mass resolution ~15 MeV , ~20%

all π0s

Merged π0

π0 from B

π0 mass (Mev/c²)

Resolved π0

all π0sπ0 from B

Bd→ π+π-π0 events

π0 mass (Mev/c²)

Merged clusterResolved cluster



Muon reconstructionMuon reconstruction

~ 94 % and ~ 1.0 % for tracks in Muon detector acceptance

~ 10 MeV/c²

Typical performancesfor b J/() X

eff ~ 94 %

misid ~ 1.0 %

Extrapolate well-reconstructed tracks to the Muon system Look for compatible hits in a “field of interest” around the extrapolated tracks



Particle Identification (PID)Particle Identification (PID)

invariant mass K invariant mass

With PID With PID

invariant mass

No PIDHadrons: RICH1 and RICH2

Electrons & neutrals: Calorimeter system

Muons: Muon system



V. Physics analysisV. Physics analysis

Flavour taggingFlavour tagging

Event selection & background estimationEvent selection & background estimation

- example of the decay B- example of the decay Bss D Dss h h



Purpose / need:Purpose / need: Determination of flavour of B-mesons at production (at primary vertex)Determination of flavour of B-mesons at production (at primary vertex)

tagging gives (best estimate of) the flavour by examining the rest of the eventtagging gives (best estimate of) the flavour by examining the rest of the event

time-dependent analyses require flavour taggingtime-dependent analyses require flavour tagging

Quantifying the tagging performance :Quantifying the tagging performance :

tagging efficiencytagging efficiency

wrong tag fractionwrong tag fraction

CP asymmetries dilutedCP asymmetries diluted

with dilution factorwith dilution factor

Flavour tagging (1/2)Flavour tagging (1/2)

R Wtag

R W U

N N

N N N

Wtag

R W

N

N N

( ) ( )obsCP CPA t D A t 1 2 tagD



Flavour tagging (2/2)Flavour tagging (2/2)

bb

ss

uu

t =0Bs

0

PV

z/c = t recKDs

t picoseconds after leaving the primary vertex, the

reconstructed B

decays.

t picoseconds after leaving the primary vertex, the

reconstructed B

decays.

K+

l (e+, )

l (e, )

K

Uses flavour conservation in the hadronization around the Brec D2 1% (B0) , 3% (Bs)

Same-side tag

ll

X b

Wc s

ll

WAssume:

Opposite side tag D2 5%

b-hadron



Event selection: example of BEvent selection: example of Bss D Dss h h (1/3)(1/3)

bt

Bs K

K

,K

Ds

Primary vertex

Bs → Ds-

Bs -> Ds∓ K±

Requirements Trigger on the B-decay of interest : high-PT tracks and displaced vertices

efficient trigger Select the B-decay of interest while rejecting the background

good mass resolution Tag the flavour of the B-decay

efficient tagging and good tagging power (small mistag rate) Time-dependent measurements

good decay-time resolution




Bs→DsK : Bs reconstructed mass signal and main background

Bs→Dsπ : Bs reconstructed mass signal and main background

Bs mass resolution ~14 MeV

LHCb note 2007-017LHCb note 2007-017




Event yields for 2fb-1 (defined as 1 year)

Bs→Ds- π+ 140k ± 0.67k (stat.) ± 40k (syst.)

Bs→DsŦ K± 6.2k ± 0.03k (stat.) ± 2.4k (syst.)

Results for B/S ratios, no trigger applied

Channel B/S at 90% CL

(bb combinatorial)

B/S at 90% CL

(bb specific)

Bs→Ds-π+ [0.014,0.05]

C.V 0.027±0.008

[0.08,0.4]

C.V 0.21±0.06

Bs→DsŦ K± [0,0.18]

C.V 0.0

[0.08,3]

C.V 0.7±0.3

(central values used for sensitivity studies)

LHCb note 2007-017LHCb note 2007-017



VI. Physics sensitivity VI. Physics sensitivity studiesstudies

The example of BThe example of Bss D Dss h h

MotivationsMotivations

Sensitivity studies with RooFitSensitivity studies with RooFit



Motivations (1/2)Motivations (1/2)

B-decay channels Bs→Ds-π+ and Bs→Ds

ŦK±

• Topology is very similar

• Physics is different:

– Bs→Ds-π+ can be used to measure Δms , Δs

• CDF measurement Δms = 17.77 ± 0.1(stat.) ± 0.07(syst.) ps-1

– Bs→DsŦK± can be used to extract the CP angle +s

• Standard Model prediction ≈ 60°



Motivations (2/2)Motivations (2/2)Bs Ds mode:

Flavour-specific decay(2 decay amplitudes only)

Bs Ds K mode: 4 decay amplitudes of interest:

Bs, Bs -> Ds+K-, Ds-K+ -> 2 time-dependent asymmetries

for the 2 possible final states

Ratio of amplitudes of order 1-> large interference and asymmetry expected Bs -> Ds

∓ K±

Bs → Ds-



Physics sensitivity studiesPhysics sensitivity studiesLHCb Sensitivity studiesLHCb Sensitivity studies

Measurement of Measurement of ++ss with combined B with combined Bss D Dss and Bs and Bs D Dss K samples K samples

Complete fit allows to obtain Complete fit allows to obtain mmss, mistag rate, etc., mistag rate, etc.

RooFit implementation for Toy MC:RooFit implementation for Toy MC: Simultaneous BSimultaneous Bss D Dss and Bs and Bs D Dss K fit: correlations taken into account K fit: correlations taken into account

All decay rate equations included All decay rate equations included (see next page)(see next page)

Using latest 2004 data challenge selection results (LHCb-note 2007-017)Using latest 2004 data challenge selection results (LHCb-note 2007-017)

Toy in propertime and mass includes:Toy in propertime and mass includes:

propertime acceptance function (after trigger & selection)propertime acceptance function (after trigger & selection)

per-event propertime errorper-event propertime error

smearing due to mis-taggingsmearing due to mis-tagging

background (rough estimate/description)background (rough estimate/description)

Fit with tagged events, and also with untagged Bs Fit with tagged events, and also with untagged Bs D Dss K events K events



Decay rates with RooFitDecay rates with RooFit

tmStmCt

Dte

q

pAt sfsff

t

fffBsincos

2sinh

2cosh

21

22

2

tmStmCt

Dte

At sfsff

t

fffB sincos2

sinh2

cosh2

122

21

Re2

f

ffD

21

Im2

f

ffS

2

2

1

1

f

f

fC

Final state f : Ds-π+ or Ds

-K+

For charge conjugate final states:

B → B , f → f , λf → λf , Af → Af p/q → q/p



Description in propertimeDescription in propertime

ss DB KDB ss

Acceptance function(after trigger & sel.)

Combined sample, tagged events, 5-year statistics

Proper time per-event error distribution

mean value 33fs

most probable value 30fs



Description in massDescription in mass

Signal & background, 5-year statistics



Results with untagged eventsResults with untagged events

Sensitivity to Error on + s (scaled to 1 year) ~ 12°

• 2 Dsπ equations, 4 DsK equations, simultaneous fit. • DC04 input, collected data from 400 “experiments” of 5y tagged data,

scaled results to 1y• Fit a Gaussian to the fitted values distribution, pull distribution

Parameter Δms (ps)-1 ωArg(λf) rad Arg(λ f ) rad |λf| +s ° ΔT1/T2 °

input value 17.5 0.328 1.047 -1.047 0.37 60 0

fitted value 17.5 0.328 1.06 -1.04 0.37 60.14 0.28

resolution 5y 0.003 0.001 0.17 0.14 0.03 5.47 5.37

resolution 1y 0.007 0.003 0.26 0.32 0.06 12.2 12.0



Thank youThank you

eduardo rodrigues nikhef seminar at cppm, marseille, 4 th may 2007

Documents