Novel real-time alignment and calibration, and trackingperformance for LHCb Run II

Espen Eie Bowen

on behalf of the LHCb collaboration

4th August 2015

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Overview

Introduction

The LHCb Trigger

Real-time alignment and calibration

Track reconstruction

Validation

Conclusion

2/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

The LHCb experimentI LHCb is the dedicated heavy flavour physics experiment at the LHCI Its primary goal is to look for indirect evidence of new physics in CP violation

and rare decays of beauty and charm hadronsI This requires:

1. Excellent tracking (momentum, impact parameter and primary vertex resolution)2. Excellent decay time resolution3. Excellent particle identification

]2c ) [MeV/−µ+µm( 9000 10000 11000

)2 cC

andi

date

s / (

25 M

eV/

0

2000

4000

6000

8000

10000 = 8 TeVsLHCb

JHEP 06 (2013) 064 decay time [ps]0 1 2 3 4

cand

idat

es /

(0.1

ps)

0

200

400

Tagged mixed

Tagged unmixed

Fit mixed

Fit unmixed

LHCb

New J. Phys. 15 (2013) 053021 Eur. Phys. J. C (2013) 73

3/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Vertex Locator

Tracking system (TT,IT,OT) RICH detectors

Magnet

Calorimeters

Muon system

4/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Vertex Locator

Vertex Locator (VELO)I Surrounds interaction pointI 42 silicon micro-strip stations with r-φ geometryI Two retractable halves, 8 mm from beam when closedI Closed for each fillI PV resolution of 13 µm in x/y and 71 µm in zI Decay time resolution of 45 fs for a 4 track vertex

4/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Tracking system (TT,IT,OT)

Magnet

TrackerTuricensis (TT)

I Upstream of magnetI Four planes of

silicon micro-stripsensors

T stationsI Downstream of magnetI Consists of inner and outer

detector regions(Inner Tracker (IT) andOuter Tracker (OT))

I IT: Three stations each withfour planes of siliconmicro-strip sensors

I OT: Three stations eachwith four planes of strawtubes

4/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

RICH detectorsRICH1

I Upstream of magnetI C4F10 radiatorI 2 < p < 40 GeV/cI 25 - 300 mrad

RICH2I Downstream of

magnetI CF4 radiatorI 15 < p < 100 GeV/cI 15 - 120 mrad

4/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Why is alignment important?I Spatial alignment of the detector is vital to achieve the best possible physics

performanceI Correct alignment of the VELO is crucial to be able to distinguish the secondary

vertices of the decays of b- and c-hadronsI Misalignment of the tracking system would cause deterioration of the mass

resolution

LHCb Preliminary

First alignment

σY = 92 MeV/c2

Run I LHCb Preliminary

Improved alignment

σY = 49 MeV/c2

Run I

5/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Why is calibration important?

I An exclusive selection making use of hadron identification criteria requires thecomplete calibration of the RICH detectors

No PID requirementsB0 → K+K− , Bs → K+π−

B0 → π+π− , B0 → K+π−

B0 → 3-body, Λb → pKΛb → pπ

With PID requirementsB0 → K+K− , Bs → K+π−

B0 → π+π− , B0 → K+π−

B0 → 3-body, Λb → pKΛb → pπ

6/24

The LHCb Trigger

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Trigger for Run I

Level 0 (L0)I Implemented in hardwareI High pT and ET signatures in muon and calorimeter

systemsI 1 MHz detector readout

Higher Level Trigger (HLT)I Flexible software triggersI Two stages (HLT1 and HLT2)I Track reconstruction and PV finding performed

J Simplified reconstruction w.r.t offlineJ Preliminary alignment and calibrationJ No RICH PID used

I Combination of inclusive and exclusive selections

8/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Novel idea for Run II

I We would like to be able to perform physics analysis directly on the output ofthe HLT

J Especially interesting for charm physics measurements

Ü Need to achieve offline-quality event reconstruction onlineI Extremely challenging to keep full tracking efficiency whilst staying within the

stringent timing budgetI Want to have best possible physics performance including use of RICH PID

onlineÜ Real-time alignment and calibration!

9/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Trigger for Run II

I After hardware stage (L0) and partial eventreconstruction (HLT1), selected events are bufferedto disk

J Automatic alignment and calibration is performedJ Larger farm→ larger timing budgetJ Full offline reconstruction with same alignment and

calibrationJ RICH PID now available

ImprovementsJ Increased selection efficiencies and purityJ Can perform physics analysis directly on triggeroutput

10/24

Real-time alignment and calibration

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

VELO, Tracker and Muon alignmentI Uses an iterative procedure to minimise the residuals of a Kalman fit to a sample

of reconstructed tracksI Multiple scattering and the magnetic field are taken into account and both

particle masses and information from vertices are used as global constraints

Reconstruct tracks usingcurrent alignment constants

Compute a new set ofalignment constants

minimising a global χ2

I Iterate until χ2 difference is below threshold12/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

RICH mirror alignmentI Misalignment of RICH w.r.t tracking system is observed as a shift of the track

projection point on the photodetector plane from the centre of the correspondingCherenkov ring

I Distribution of ∆Θ as a function of φ shows sinusoidal shapeI Histogram fitted to determine alignment constants

I Iterative procedure, separate for RICH1 and RICH2

Before alignment

13/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

RICH mirror alignmentI Misalignment of RICH w.r.t tracking system is observed as a shift of the track

projection point on the photodetector plane from the centre of the correspondingCherenkov ring

I Distribution of ∆Θ as a function of φ shows sinusoidal shapeI Histogram fitted to determine alignment constants

I Iterative procedure, separate for RICH1 and RICH2

After alignment

13/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Online alignment frameworkI Real-time alignment and calibration uses dedicated data samples selected by

HLT1 at the start of a fillI Computation of alignment parameters are highly parallelised using ∼1700 nodes

of the HLT farmI Details for ¬ Tracker alignment and ­ RICH mirror alignment

Analyser (multiple nodes)¬ Performs track reconstruction­ Performs photon reconstruction and fills histograms

Iterator (single node)¬ Combines output of analysers and minimises χ2 to extract alignments constants­ Fits combined histograms and extracts alignments constants

H Complete real-time alignment possible within only a few minutes of data taking!14/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Online alignment frameworkI Automatic alignment procedure runs at the start of each fillI Only update alignment constants when necessaryI VELO is opened and closed with each fill

I Expect to update constants every few fillsI Tracking system (TT, IT, OT)

I Expect to update constants every few weeksI Alignment procedure of the muon stations only used for monitoringI RICH mirror alignment also only used for monitoring

Fill3960 3980 4000

m]

µV

aria

tion

[

5−4−3−2−1−0

1

2

3

4

5x-translationy-translationx-translationy-translationLHCb VELO Preliminary

Iteration0 5 10

m]

µ X

var

iatio

n [

∆

300−

200−

100−

0

100

200

300

400

500IT1 ASideIT1 BottomIT1 TopIT1 CSide

IT1 ASideIT1 BottomIT1 TopIT1 CSide

LHCb Tracker Preliminary

15/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

RICH CalibrationI RICH automatic calibration consists of calibrating:

I RICH refractive indexI Hybrid photon detector (HPD) images

Refractive index

I Refractive index of gas radiators depends on the temperature, pressure, andcomposition so can change with timeI The difference between the reconstructed and expected Cherenkov angle isfitted to extract a scale factor

HPD images

I Correction applied to account formagnetic and electric field effectsI Calibrate anode image

I Evaluated and updated every run!16/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

OT drift time calibrationI Measured drift time may be different from that estimated due to distance of track

from wireI Due to differences between collision time and LHCb clockI Automatic procedure computes calibration using drift time residual

Commissioning                                                                          Opera.on  t=0.04ns                  t=0.1ns  

Preliminary

I Evaluated each run and updated for the next run (if difference is above threshold)17/24

Track reconstruction

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Track reconstruction

I In Run II, the full offline reconstruction will run within the triggerI In order to maintain full reconstruction efficiency, while keeping within a strict

timing budget many improvements were neededI Changes to the reconstruction chain and optimisation of pattern recognition algorithmsI Optimisation of code (e.g. vectorisation)

19/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Tracking in HLT1 for Run II

Offline VELO tracking

Velo → TT:Initial momentum estimate

TT → T-stations:Full track

Offline Kalman filter

I Improved sequence forming Velo-TT tracks asan intermediate stage

I Momentum estimate allows a preselection onthe pT of tracks

I Charge estimate allows greatly reduced searchwindows downstream of the magnet

J Vast reduction in both ghost rate (factor 4) andexecution time (factor 3)

20/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Speed up through vectorisation

I Many areas of code are heavy on vectoralgebra

I Can utilise faster implementationsexploiting vector instructions

I Large gains in speed (∼ 30%) fromvectorising areas of the code e.g.Kalman filter (takes large fraction oftotal HLT time)

Every Cycle Counts!

Gerhard Raven Nikhef & VU Amsterdam

7974M Cycles (30/11/2014)

5576M Cycles (15/01/2015)

callgrind “cycle count” Moore v23r2, Physics_September2012 Hlt1+Hlt2

21/24

Validation

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Validation with early measurements

I Offline reconstruction nowavailable online

J Use it for physics analysis!I Turbo stream

J Only write out information ofsignal candidates

J Large saving in spaceI Ideal for channels with high

signal yield (O(106))I Very quick turn around!

Measurement of differential J/ψ cross-section [LHCb-PAPER-2015-037]

]2c [MeV/ -µ +µM2950 3000 3050 3100 3150 3200

2 cC

andi

date

s pe

r 5

MeV

/

2

4

6

8

10

12

14

16

310×

-1 = 13 TeV, L =3.02 pbs

LHCb Preliminary

]c < 3 [GeV/T

2 < p3 < y < 3.5

[ps]zt-10 -8 -6 -4 -2 0 2 4 6 8 10

Can

dida

tes

per

0.2

ps

1

10

210

310

410

510 -1 = 13 TeV, L =3.02 pbsLHCb Preliminary

]c < 3 [GeV/T

2 < p3 < y < 3.5

Measurement of differential charm cross-section [LHCb-PAPER-2015-041]

Candidates

/(0.992366MeV

/c2)

0

0.02

0.04

0.06

0.08

0.1

0.12

×106

Data

Fit

Signal

Background

NSig 2171646.45± 2207.31

NBkg 736641.05± 1853.92

χ2/DoF 3.39

LHCb preliminary√s = 13TeV

m(K−π+) [MeV/c2]1.8 1.82 1.84 1.86 1.88 1.9 1.92

×103

∆/σ

-4

-2

0

2

4

Candidates

/(0.992366MeV

/c2)

0

1

2

3

4

5

6

×103

Data

Fit

Signal

Background

NSig 91956.53± 345.58

NBkg 25375.76± 229.89

χ2/DoF 1.28

LHCb preliminary√s = 13TeV

m(K−K+π+) [MeV/c2]1.92 1.94 1.96 1.98 2 2.02 2.04

×103

∆/σ

-4

-2

0

2

4

23/24

Introduction The LHCb Trigger Real-time alignment and calibration Track reconstruction Validation Conclusion

Conclusion

J First experiment of this scale to be able perform a complete real-time alignmentand calibration within only a few minutes of data taking!

J Vast improvements made to the track reconstruction sequence to allow the samereconstruction online and offline

J Full offline track reconstruction, PV reconstruction and PID at software triggerlevel

H Can perform physics analysis directly on trigger output!

24/24

Backup

Tracking alignment procedure

Reconstruct the tracks using thecurrent alignment constants

dχ2

dα= 2

∑tracks

drdα

TV−1r

d2χ2

dα2 = 2∑

tracks

drdα

TV−1RV−1 dr

dα

Compute a new set of align-ment constants (α) minimizinga global χ2:

α = α0 −(

d2χ2

dα2

)−1∣∣∣∣∣α0

dχ2

dα

∣∣∣∣α0

Iterate until the χ2-difference is below a threshold

r: tracks residuals, V: covariance matrix, R: residuals’ covariance matrix

2/3

Finite state machine

Offline

Ready

Running Paused

Alignment

configure (initialise)∼2min

start

continue

pause

stop stop

reset (finalize)new constants

3/3