Tracking,calibra�on&alignment,anddataprocessing
intheLHCbupgrade
BarbaraStoraciOnbehalfoftheLHCbCollabora�on
The25thInterna�onalWorkshoponVertexDetectors
Friday,30thSeptember2016 1
TThhee LLHHCCbb eexxppeerriimmeenntt
2
LHCbisthededicatedheavyflavorphysicsexperimentatLHC
Singlearmforwardspectrometer,coveringapseudorapidityrangeuniqueamongtheLHCdetectors
ItsprimarygoalistolookforindirectevidenceofnewphysicsinCP-viola�onandraredecaysofbeautyandcharmshadrons
J.HighEnergyPhys.06(2013)064
NewJ.Phys.15(2013)053021 Eur.Phys.J.C73(2013)2431
3
TThhee LLHHCCbb eexxppeerriimmeenntt
Int.J.Mod.Phys.A30(2015)no.07JINST9(2014)09007JINST9(2014)no.01,P01002JINST10(2015)no.02JINST8(2013)P02022Eur.Phys.J.C73(2013)2431
VELO:Dis�nguishprimaryand
secondaryver�cesστ-decay~45fsforBmesons
σIP~20μm
RICH:Par�cleIden�fica�on
εk-ID~95%MisIDrateforπ:~10%
TrackingSystems(TT,IT,OT)εtracking>96%
Δp/p~0.5–1.0%
MuonSystem:εμ-ID>97%with<2.5%π
mis-IDrate
Calorimeters:Energymeasurement,e/γ
iden�fica�on,π0massresolu�on~10MeV/c2OtherLHCbtalks:
1. LHCbsilicondetectors:Run2opera�onexperience
2. TheLHCbVELOforphase1upgrade3. TheLHCbUTforphase1upgrade
B !
PrecisionofmanyphysicsmeasurementatLHCbwillbesta�s�callylimitedattheendofRun2
4xmoreluminosity:L=2x1033cm-2s-1
→ Expected5�-1ofdataperyear
LLHHCCbb uuppggrraaddee:: tthhee cchhaalllleennggee
Manyhadronicchannelssaturate,duetoenergycutsinthehardwaretrigger
4CERN-LHCC-2014-001
Fortheupgradewewillneedtomovefrom1MHz→40MHzread-outrate
LHCbUpgrade
]2 [MeV/cπππm1800 1850 1900
2C
andi
date
s/M
eV/c
0
5000
10000
15000
20000
25000
= 8 TeV DatasPreliminaryLHCb Trigger
+PID+�ghterPID
Triggerswilldiscriminatebetweendifferentsignal
classes
5
TThhee MMHHzz ssiiggnnaall eerraa
]2 [MeV/cKKKm1800 1850 1900
2C
andi
date
s/M
eV/c
1
10
210
310
410
= 8 TeV DatasPreliminaryLHCb Trigger
+PID
Samereconstruc�onasofflineandcompletealignmentandPIDcalibra�onallowstoapplya�ghterselec�ononkinema�csquan��es
RICHcalibra�onallowstousehadronpar�cleiden�fica�oninselec�on,e.g.boostefficiencyforCabibbosuppresseddecayswhilekeepingtheratelowbypre-scalingtheCabibbofavoredcounterpart
Cabibbo-Favoured:D+->2π+π-BR=(3.29±0.20)10-3
DoubleCabibbo-Suppressed:D+->2K+K-BR=(9.0±2.1)10-5
6
Run2usedastest-bench
PPllaannss ffoorr tthhee UUppggrraaddee
Movefrom1MHz→40MHzread-outrate Requiresupgradeofmanydetectorsandfront-endelectronicstoprocessat40MHz
Fully-so�waretrigger Real-�mealignmentandcalibra�on Offline-likereconstruc�onrunonline
Mostoftheanalysesdoablefromtriggerreconstructedoutput
RRuunn 22 eexxppeerriieennccee:: HHLLTT
Run1limita�ons: Simplifiedreconstruc�onrespecttooffline Preliminaryalignmentandcalibra�on RICHPIDinfomarginallyused(not-fullycalibratedsystem) Notfullexploita�onofthemachinedead-�me
Processingofthedataattheendoftheyearwiththelatestconstants 8
RUN1 RUN2
30 MHz inelastic event rate (full rate event building)
Software High Level Trigger
2-5 GB/s to storage
Full event reconstruction, inclusive and exclusive kinematic/geometric selections
Add offline precision particle identification and track quality information to selections
Output full event information for inclusive triggers, trigger candidates and related primary vertices for exclusive triggers
LHCb Upgrade Trigger Diagram
Buffer events to disk, perform online detector calibration and alignment
IIddeeaass ffoorr tthhee uuppggrraaddee:: HHLLTT
Fullso�waretrigger:drivingcriteria Muchmoreinforma�onattriggerlevelneededtocategorizethemorecomplexeventtopology
Essen�altobeableto: Havethesametrackingsequence
onlineasonline Performthereal-�mealignment
andcalibra�on Challenging�meconstraintsfor
reconstruc�on: SingleCPU�mingbudgetfor
reconstruc�ontobeO(10)ms Tobestudiedthepossibilityto
op�mizedthereconstruc�onasafunc�onofthedecaychannels
9
RReeccoonnssttrruuccttiioonn cchhaaiinn
Severalalgorithmsreconstruc�ngtrackstraversingthefully(longtracks),orpar�allythedetector
Strong�meconstraintsrequiredadeepop�miza�on
Theeventtopologyismorecomplexattheupgradecondi�ons:
Moreprimaryver�ces:2to7.6 Highertrackmul�plicity:70to180
Requirements: Suppressfaketracksrate Lowprocessing�meinthetrigger Maintainhightrackingefficiency,
i.e.>90%forp>5GeV/c Excellentmomentum,�meand
vertexresolu�on
WecanstartfromRun2reconstruc�onchain Introduc�onofmachinelearningtechniquetoboostthe
performance(ghostratereduc�on,efficiencyincrease) Massiveuseofparalleliza�on,e.g.alsoconsideringnew
architectureslikeGPU:– Workinprogressfor:
Velotracksreconstruc�on(VeloPixelalgorithmforGPU,re�naalgorithm) T-trackswithautomaton KalmanFilter
Furtherimprovementsonthealgorithms– Workinprogressontheforward,seeding,matchingandPV-
reconstruc�onalgorithms
IIddeeaass ffoorr tthhee uuppggrraaddee:: ttrraacckkiinngg
12
Manyprogressessincethe
upgradeTDRsin2014.
CERN-LHCC-2014-016CERN-LHCC-2014-001
Gaininspeedbychangingthedatastructureandallowedvectoriza�on
13
SSIIMMDD ffoorr vveeccttoorriizzaattiioonn
NeededSingleInstruc�on,Mul�pleData(SIMD)
paradigm
Appliedinseveraltrackingalgorithm(e.g.VeloUT/VeloTT,Forward,VeloPixel,…)
Dataorganizedin“arrayofstructure”(x,y,z)1,(x,y,z)2,…areintui�ve,butbadforparalleliza�on
Vectoriza�on-friendlyifmovedto“structureofarrays”(x1,x2,…),(y1,y2,...)
14
Sensor-levelparallelismimpliesprocessing~100hitsata�me.ThisworkswellforCPUs,wherevectorunitscanprocess4-8opera�onsata
�meForGPUshowever,O(2000)opera�onsneedtohappenatthesame�meTherefore,inordertoexploitmany-coreparallelism,manyeventsneedto
beprocessedinparallel
EExxaammppllee ooff ppaarraalllleelliizzaattiioonn:: VVeellooPPiixxeell
PPaarraalllleelliizzeedd VVeellooPPiixxeell ((CCPPUU vvss GGPPUU))
15
Novel�es: Eachsensorisprocessedinparallel:
Seedingandforwardingperformed48�mes Seedsaretriplets
Onlythebesttripletsiskept Savedfirstonlythe>=4hitstracks,thencheckedfor3-hitstracks Wri�eninOpenCL:itcanrunCPUandGPU
Idea:buildingstraighttracksintheVertexdetector Twosteps:
Seeding:buildingaseed(orforming-track) Forwarding:lookingforanewhittobeaddtotheforming-track
VanillaVeloPixelcode OpenCLVeloPixelcode
Comparableperformancerespecttothevanillaversionintermsofefficiency EvenlessclonesfortheOpenCLversion
16
PPeerrffoorrmmaannccee ccoommppaarriissoonn
Physicsperformancereliesonthespa�alalignmentofthedetectorandtheaccuratecalibra�onofitssubcomponents:1. AccuratealignmentoftheVELOessen�alforprimaryver�ces
discrimina�on,excellentimpactparameter(IP)andproper�meresolu�on
IImmppoorrttaannccee ooff tthhee aalliiggnnmmeenntt ((II))
FirstalignmentσIP(highpT)=14.0μm
LatestalignmentσIP(highpT)=11.6μm
IPxresolu�onvs1/pT
Int.J.Mod.Phys.A30(2015)1530022 18
IImmppoorrttaannccee tthhee aalliiggnnmmeenntt ((IIII))
Physicsperformancereliesonthespa�alalignmentofthedetectorandtheaccuratecalibra�onofitssubcomponents:2. Be�eralignmentofthetrackingsystemimprovesthemass
resolu�on
Firstalignmentσϒ=92MeV/c2
Be�eralignmentσϒ=49MeV/c2
Invariantmassdistribu�onforϒèμ+μ-
LHCbPreliminary LHCbPreliminary
19
Completecalibra�onoftheRICHdetectorsneededforexclusiveselec�onusinghadronpar�cleiden�fica�oncriteria
IImmppoorrttaannccee ooff tthhee ccaalliibbrraattiioonn
JHEP1210(2012)037 20
GeneralStrategy– Automa�cevalua�onatregularintervals,e.g.perfillorper
rundependingonthetask– Dedicateddatasampletoperformalignmentorcalibra�on
collectedwithspecifictriggerselec�onlineforeachtask– Computethenewalignmentorcalibra�onconstantsinfewminutes
– Updatetheconstantsonlyifneeded– Thesamenewalignmentandcalibra�onconstantswillbeusedbothbythetriggerandtheofflinereconstruc�on
EExxppeerriieennccee ffrroomm RRuunn22:: FFUULLLL ddeetteeccttoorr RReeaall--ttiimmee AAlliiggnnmmeenntt aanndd CCaalliibbrraattiioonn
21
HLT1selectsdedicatedeventsforeachtaskatthestartofthefill,processedinparallelon~1700nodesandsinglenodecomputeupdate
alignmentconstants.
RReeaall--ttiimmee AAlliiggnnmmeenntt iinn pprraaccttiiccee
22
Trackingalignment Analyseronmul�plenodes:massive
paralleltrackreconstruc�onbasedonalignmentfromiterator
Iteratoronasinglenode:combinedanalyseroutput,χ2minimiza�onandupdatealignmentconstantsfornextitera�on
RICHalignment Analyseronmul�plenodes:track,
photonreconstruc�onandhistogramfilling
Iteratoronasinglenode:histogrammergingandfi�ngtoextractupdatedalignmentconstants.
Itera�veproceduretominimizingtheresidualsofaKalmanfitonasampleofalreadyreconstructedtracks.
23
RReeaall--ttiimmee AAlliiggnnmmeenntt:: KKaallmmaann ffiitt
AppliedtoVELO,Tracker,andmuons!
Nucl.Instrum.Meth.,A712:48–55,2013.Nucl.Instrum.Meth.,A600:471–477,2009.
RReeaall--ttiimmee aalliiggnnmmeenntt:: ttrraacckkiinngg ssyysstteemmss
Trackingsystems: Alignmentofabout700elements
Runautoma�callyforeachfill Timerequired~7minuteseach
tasks(2independenttasks,1forVELO,1forthetrackers)
Updates: EachfewfillsfortheVELO A�ereachmagnetpolarity
switchforthetracker Muons:onlya�erhardware
interven�on
Alignment number [a.u.]50 100
m]
µV
aria
tion
[
20−
15−
10−
5−05
101520 x-translation
y-translationLHCb VELOPreliminary
Empty markers = no update 23/04/2016 - 08/09/2016
Alignment number [a.u.]50 100
Z V
aria
tion
[mm
]∆
1−0.8−
0.6−
0.4−
0.2−
00.20.40.60.8
1IT1 ASideIT1 CSideIT1 TopIT1 Bottom
LHCb TrackerPreliminary
Empty markers = no update 06/05/2016 - 08/09/2016 Fullyautoma�c: Automa�crun(sincesummer2015) Automa�cupdateoftheconstants
(sinceautumn2015) 24
25
Refrac�veindex(2constants): Fitthedifferencebetweenexpectedand
measuredCherenkovangle.Scalefactorascorrec�on
Accountfortemperature,pressureandgasmixturevaria�on
Hybridphotondetector(1940constants): UsedtodetectCherenkovphotons Correctfortheeffectofmagne�candelectric
fieldsontheanodeimage UseofSobelfiltertodetectedgesandthenfit.
RReeaall--ttiimmee ccaalliibbrraattiioonn:: RRIICCHH ssyysstteemmss
Fullyautoma�cprocedures,evaluatedrunbyrun Maintainedstablecondi�onsduringRun2datataking
Selectedexample
26
RRuunn 22 ppeerrffoorrmmaanncceess
Momentum (MeV/c)0 20 40 60 80 100
310×
Efficiency
0
0.2
0.4
0.6
0.8
1
1.2
1.4) > 0π LL(K - ∆
) > 5π LL(K - ∆
K→K
K→ π
= 13 TeV 2016sLHCb Preliminary 1. Sameperformanceas
offlineinRun1obtainedonlineinRun2
2. PIDperformancebe�erthaninRun1nowalreadyavailableonline
TTuurrbboo ssttrreeaamm HLTreconstruc�onasofflineone IntroducedTurbostreamtosaveHLT
candidatesonly: Reducedeventsize:
Fullevent~70kB Turboevent~5kB
Noneedofofflinereconstruc�on:possibletoperformanalysisimmediately
Dedicated~2.5kHzoftheoutputstream(10KHzforthefullstream)
Usedforearlymeasurementcross-sec�onAnalysisfind~106candidatesdirectlyfromthetriggerNofurtherreconstruc�on,allnecessaryinfoispersistedfromthetriggerResultspresented1weeka�ertheiracquisi�on!
JHEP10(2015)172
J/ψproduc�oncross-sec�on
28
Ofthe420HLT2lines2016physicsprogramme,150chooseTurbo:~60newlineswithrespectto2015datataking
Newfeaturesin2016:persistencyofthereconstruc�oninfoforthefullevent,higherlevelvariables(i.e.hitsinaconeregionaroundthetrack)availableforfewlines:possibletodocharmspectroscopy! Morespaceneedednow,but
essen�alprepara�onstudiesfortheupgrade
TTuurrbboo ssttrreeaamm:: pprreeppaarriinngg tthhee uuppggrraaddee
29
IntheupgradeTurbowillneedtobemoreflexibleintermsofeventformat,sothateachHLTlinecanchoosehowmuchoftherawandreconstructedeventtopersist.
InRun2wehavedemonstrated:– Possibletohavethesameofflinequalityattriggerlevel
Real-�mealignmentandcalibra�on Offline-likereconstruc�onchain
– PossibletodophysicsanalysisdirectlyontheHLToutput In2015newcross-sec�onresultspublishedinaweek�mefromthedata
acquisi�on– Enhancedsignalsefficiencybyexploi�ngthefullinforma�onattrigger
level Upgradeprepara�on:– Real-�mealignmentandcalibra�onalreadyfullyrunning
Tobere-adaptedtotheupgradeframework– Manyideasforthereconstruc�onop�miza�ontobefurther
inves�gate(e.g.paralleliza�on)– DirectdevelopmentofthenewideasneededforTurbostream
possibleonrealdata.
CCoonncclluussiioonnss
30
Revolu�onarychangesappliedinRun2bytheLHCbexperiment:thetrampolinefortheUpgradedevelopments
33
RReessuullttss ccoommppaarriissoonn ffoorr vvaanniilllliiaa && OOppeennCCLL vveellooppiixxeell aallggoorriitthhmm
Theresultsarenotgoingtobethesame!
PPaarraalllleelliizzeedd VVeellooPPiixxeell ((CCPPUU vvss GGPPUU))
34
Novel�es: Eachsensorisprocessedinparallel:
Seedingandforwardingperformed48�mes Seedsaretriplets
Onlythebesttripletsiskept Savedfirstonlythe>=4hitstracks,thencheckedfor3-
hitstracks Wri�eninOpenCL:itcanrunCPUandGPU
Foreachpairofunusedhits,athirdhitissearched
Thefirstonefoundcompa�bleiskept[hitsarepreorderedbyX]
Youkeepforwardingbasedonlast2hitsoftheforming-track
Lookingfortripletsinconsecu�vesensorsonthesameside
Forwardingalltheforming-tracksonthenextsensor(usingthelast2hits)
VanillaVeloPixelcode OpenCLVeloPixelcode
OOppttiimmiizzaattiioonn ooff tthhee rreeccoonnssttrruuccttiioonn cchhaaiinn
Examples Iden�fiedhotspotsbyprofiling
Vectorisa�on(trackfit,magne�cfield)
Caching(materialdescrip�on) Fastapproxima�ons Algorithmstuningandre-implementa�on
=>Possibletogainorderof~30%inseveralalgorithms
1. Op�miza�onofthecode(e.g.vectoriza�on):
OOppttiimmiizzaattiioonn ooff tthhee rreeccoonnssttrruuccttiioonn cchhaaiinn
Example
VelotrackscanbeextendedtotheTTdetector
Advantages:
TheTTdetectorisalreadyinthemagne�cfield:possibletoes�mateq/p(resolu�on~15%)
Momentumes�mateusedtopre-selecttracksatanearlystage
Chargees�mateallowsgreatlyreducedsearchwindowsdownstreamofthemagnet
2. Changestothereconstruc�onchainandop�miza�onofthepa�ernalgorithm:
OOppttiimmiizzaattiioonn ooff tthhee rreeccoonnssttrruuccttiioonn cchhaaiinn
Newreconstruc�onchain: ~3�mesfaster Be�erorequivalentperformanceasinRunI TrackswithapT>500MeValreadyavailableatHLT1
level(inRunIpT>1.3GeV)withoutanyIPrequirementsneeded.
Gains>50%signalefficiencyforcharmphysics Enablelife�meunbiasedtriggersforhadronicfinalstates(worldfirst!)
VELO-TTperformance: Morethan97%efficientfor
trackswithpT>200MeVthathavehitsinatleastthreelayers
~5%be�erperformancethanpreviouscode(neverusedsinceitwastooslow)
LHCbsimula�on
LHCbsimula�on
NNoovveellttiieess iinn tthhee aallggoorriitthhmmss
Forwardalgorithm:– Codespeedup(almostafactorof2.5)– Efficiencyimprovement(+2.5%forlongfromB,p>5GeV)and
ghostratereduc�on(factorof2)– BenefitfromMVA,vectoriza�on,codeop�miza�on
Seeding(newreplacingTDR-one):– 3xfaster– Boostofphysicsperformanceforlong-livedpar�cle(+25%)and
longfromBtracks(+10%)– Ghostratereducedbyafactorof2– Futuredevelopments:usageofMVAforghostrejec�on
Matchingalgorithm(newreplacingTDR-one):– ImplementedNeuralNet(hugeimprovementinperformance)
39
[c/GeV]T
1/p0 0.5 1 1.5 2 2.5 3
m]
µ) [ X
(IPσ
0
10
20
30
40
50
60
70
80
90
100
LHCb VELO Preliminary
T = 11.6 + 23.4/pσ2012 Data:
T = 11.8 + 23.7/pσ2015 25 ns Data:
LHCb VELO Preliminary
T = 11.6 + 23.4/pσ2012 Data:
T = 11.8 + 23.7/pσ2015 25 ns Data:
2012 Data2015 25 ns Data IPresolu�oncomparablewith
2012data Stablebetweendifferentfills Compa�bleresultswith50and
25nsdata
TThhee nneeww cchhaaiinn aatt wwoorrkk
]c [GeV/p50 100 150 200
ε
0.75
0.8
0.85
0.9
0.95
1
1.05
LHCb preliminary
Data EM 2015
Data 2012
Multiplicity of reconstructed tracks20 40 60 80
x [m
m]
ΔRe
solu
tion
0
0.01
0.02
0.03
0.04
Online Run II
Offline Run I
LHCb Simulation
Trackingefficiencycomparablewith2012data
Iden�calresolu�ontoRunIforthePVreconstruc�onwith70%smallerfrac�onoffakeprimaryver�ces
612elementsaligned elements.ITBoxes("TxTzRz")12boxes elements.TTLayers("Tz")4layers elements.TTModules("TxRz")128modules=>Rzremoved elements.OTCFrames("TxRz")-alignlayersinA-sideandC-
side24layers elements.OTCFrameLayers("Tz")-alignlayers12layers elements.OTModules("TxRz")=>Rzremoved432modules
41
EElleemmeennttss AAlliiggnneedd:: ttrraacckkiinngg
Numberofrunsconsidered:77– ITNumberofupdates:36-25fromautoma�c
alignmentprocedure(run160059). mainlyresponsibleforupdates:mostlyITBox.Tz
– TTNumberofupdates:13-only4fromautoma�calignmentprocedure(run160059) mostlyTTLayers.Tz,some�mesTTModules.T(x|y)
– OTNumberofupdates:16-11fromautoma�calignmentprocedure(run160059). mostlyFrames/Module.T(x),some�mesT(z)/T(y)
42
UUppddaatteess iinn 22001155
Mirrorpairstoalign:– RICH1:16– RICH2:94
1090alignmentconstants Re-op�miza�onofthecodeandHLT-
selec�onofthehighmomentumtrackstoberunintheonlineenvironment
Sameframeworkasthetrackingalignment:– Analysers:photonreconstruc�on
doneinparallel– Iterator:fitoftheΔθdistribu�onona
singlenode Usedasmonitoring(ifenough
sta�s�csavailablepoten�allyeachfill)
RRIICCHH mmiirrrroorr aalliiggnnmmeenntt:: aatt wwoorrkk Beforealignment
A�eralignment
43
Technicali�es:– Constantsexpectedtochange:updatedinreal-�me(each
fill,run,…)– Constantsexpectedtobestable:monitoring
Tookadvantageofthefarmcompu�ngpower
RReeaall--ttiimmee AAlliiggnnmmeenntt aanndd CCaalliibbrraattiioonn((IIII))
Calibra�onFarmtasks: OnlineFarmtasks:
RICHrefrac�veindexandHPDimagecalibra�on
Calorimetercalibra�on OT-t0calibra�on
VELOandtrackeraligment Muonalignment RICHmirroralignment Calorimeterπ0calibra�on
44BarbaraStoraci,UniversityofZurich
RunI Runofflinefew�mesperyear Timeneededtorunoffline~severaldaysontheGRID MinimalusageofRICHinforma�onatHLTlevel(notfully
calibratedsystem) Newalignmentupdatedofflineforthere-strippingofthedata
RRIICCHH mmiirrrroorr aalliiggnnmmeenntt:: RRuunnII vvss RRuunn IIII
RunII Re-op�mizedcode Runassoonasenoughsta�s�csisaccumulated(typically1-2fills) Timeneeded~10minuteseachsystem RICHinforma�oncanbeusedinHLT:fullycalibratedsystem Usedasmonitoring
45BarbaraStoraci,UniversityofZurich
RunI Runofflinefew�mesperyear(andupdatedonline) Timeneededtorunoffline~1h Datatakenwithapreviousversionofthealignment->not
bestcondi�onsusedinHLT Newalignmentupdatedofflineforthere-processingofthe
data
VVEELLOO,, TTrraacckkeerr aanndd MMuuoonn aalliiggnnmmeenntt:: RRuunnII vvss RRuunn IIII
RunII Runeveryfillandconstantsautoma�callyupdated Timeneeded~7minutes HLTalwayswiththemostupdatedconstants Offlinesynchronizedwithonlineparameters
46BarbaraStoraci,UniversityofZurich
RReeaall--ttiimmee aalliiggnnmmeenntt:: RRIICCHH ssyysstteemmss
Cherenkovphotonsfocusedonphoton-detectorplanebysphericalandflatmirrors
AnimaginarytrackreflectedthroughtheRICHmirrorsshouldbeinthecenteroftheCherenkovring
Thedistribu�onoftheΔθagainstϕresultsinasinusoidaldistribu�onincaseofmisalignmentofthemirrors
Fitthedistribu�ontocalculatealignmentconstants
RICHsystems: Alignmentof110mirrorpairs Runasmonitoringeachfewfills(no
updateexpected) Timeneeded~10mineachsystem RICHinforma�oncanbeusedin
HLT:fullycalibratedsystem
BeforealignmentA�eralignment
47Alignment number0 5 10
Y V
aria
tion
[mra
d]∆
0.3−
0.2−
0.1−
0
0.1
0.2
0.3
LHCb RICHPreliminary
24/05/2016 - 24/08/2016
primary mirror 0primary mirror 1primary mirror 2primary mirror 3
DOWN UP DOWN
InLHCbtwotechniquesusedtomeasuretheluminosity:– Tradi�onalVanderMeerscans:withLHCmoving
thebeamswithrespecttoeachotherinsmallsteps
– BeamGasImagingMethodwithSMOGtoreconstructbeamshapeattheinterac�onpoint
Thecombina�onofthetwotechniquesallowedinRunItoachievethebestprecisemeasurementoftheluminosityatLHCwith1.1%ofuncertainty[JINST9(2014)12,P12005]
OnlybeamgasimagingmethodavailablefortheEM�mescale
Deliveredaluminositymeasurementwith3.8%ofuncertaintyinfewweeks(measurement10thofJune,firstresultspresentedatEPSthe22thofJuly)
Furtheranalysistocombinethetwomethodsongoing
LLuummiinnoossiittyy mmeeaassuurreemmeenntt
OnlyHLT1isusedforthismeasurement:thenewHLTsplitallowedtowriteveryfastdatatodiskincreasingtheprecisionofthe
measurement(~1%)
48BarbaraStoraci,UniversityofZurich
RunI Runofflinefew�mesperyear Timeneededtorunoffline~severaldaysontheGRID MinimalusageofRICHinforma�onatHLTlevel(notfully
calibratedsystem) Newalignmentupdatedofflineforthere-strippingofthedata
RRIICCHH mmiirrrroorr aalliiggnnmmeenntt:: RRuunnII vvss RRuunn IIII
RunII Re-op�mizedcode Runassoonasenoughsta�s�csisaccumulated(typically1-2fills) Timeneeded~30minutes RICHinforma�oncanbeusedinHLT:fullycalibratedsystem Usedasmonitoring
49BarbaraStoraci,UniversityofZurich
RICHautoma�ccalibra�onconsistsofcalibra�ng:1. RICHrefrac�veindex:
Dependsonthegasmixture,temperatureandpressure Determinedbyfi�ngthedifferencebetweenreconstructedandexpectedCherenkovangle
– Hybridphotondetector(HPD)images Affectedbythemagne�candelectricfieldsdistor�ons AnodeimagescleanedandSobelfilterisusedtodetecttheedges
RRIICCHH ccaalliibbrraattiioonn:: llooggiicc
50BarbaraStoraci,UniversityofZurich
Beforealignment A�eralignment
RICH/RiCKResLongTight/Rich2Gas/ckResAll
Entries 1.451864e+07
Mean 05−2.706e−
Std Dev 0.001322
0.002− 0.001− 0 0.001 0.00250
60
70
80
90
100
310×RICH/RiCKResLongTight/Rich2Gas/ckResAll
Entries 1.451864e+07
Mean 05−2.706e−
Std Dev 0.001322
Rich2Gas Rec-Exp Cktheta | All photons
LHCbPreliminary LHCbPreliminary
RICHautoma�ccalibra�onconsistsofcalibra�ng:2. Hybridphotondetector(HPD)images
Affectedbythemagne�candelectricfieldsdistor�ons AnodeimagescleanedandSobelfilterisusedtodetecttheedges
– RICHrefrac�veindex: Dependsonthegasmixture,temperatureandpressure Determinedbyfi�ngthedifferencebetweenreconstructedandexpectedCherenkovangle
RRIICCHH ccaalliibbrraattiioonn:: llooggiicc
51BarbaraStoraci,UniversityofZurich
RRIICCHH ccaalliibbrraattiioonn:: aatt wwoorrkk
Fullyautoma�cprocedure,evaluatedrunbyrun Incaseofatooshortrun,thesameconstantsofthe
previousrunareused Calibra�onparametersaremonitoredbyPVSStrendplots Maintainedstablecondi�onsduring2015and2016data
taking
Alignment number0 5 10
Y V
aria
tion
[mra
d]∆
0.3−
0.2−
0.1−
0
0.1
0.2
0.3
LHCb RICHPreliminary
24/05/2016 - 24/08/2016
primary mirror 0primary mirror 1primary mirror 2primary mirror 3
RICH1primarymirrorsiny
Alignment number 0 5 10
Y V
aria
tion
[mra
d]∆
0.25−
0.2−
0.15−
0.1−
0.05−
0
0.05
0.1
0.15
0.2
0.25
LHCb RICHPreliminary
24/05/2016 - 24/08/2016
RICH2primarymirrorsiny
Absolutecalibra�ons:– π0calibra�on:
Selectedphotons(3x3clusters)andfixseed(central)cell– Computedi-photoninvariantmass– Findthecoefficientwhichwouldmovethemeasurementclosertothe
nominalπ0mass
– Cssourcescan: Scanoftheresponsewiththesource.DoneeachTS
Rela�vecalibra�ons:– Rawoccupancymethod:
Comparisonoftheperformanceofeachcellwithareference– LEDmonitoringsystem:
TodetectageingofthePMTs
CCaalloorriimmeetteerr ccaalliibbrraattiioonn:: llooggiicc ImportanttohavestableL0rates Itallowstoaccountforageingeffects Thecalibra�onadjustdirectlytheHVse�ngs
53BarbaraStoraci,UniversityofZurich
Global�meoffset(t0)relatedtothesynchroniza�onbetweentheOT�meandthecollision�me->ashi�of0.5nsleadstotrackinginefficiencyof~0.25%
Calculatet0fortheen�reOT(real-�meupdate) Theoffsetisevaluatedbyusingthereconstructedtracks,and
comparingthedri��mewiththeexpecteddri��me(evaluatedusingtheTR-rela�on)
OOTT--tt00 ccaalliibbrraattiioonn:: llooggiicc
drift time residuals [ns]-20 -10 0 10 20
Goo
d tra
cks/
(0.4
ns)
02468
1012141618
610×
LHCb OT
t = tmeas t(r)
LHCb
JINST9(2014)P01002
Executedeveryfill,updateexpectedeveryfewweeks Automa�cevalua�onandupdateoftheconstants
workedfromdayone!!! Addedformonitoringpurposethet0-calibra�onper
chip(4-eachmodule).
OOTT--tt00 ccaalliibbrraattiioonn:: aatt wwoorrkk
Calibration number [a.u.]0 200 400 600
[ns]
0t∆
-0.5-0.4-0.3-0.2-0.1
00.10.20.30.4 Appliednew
0t Not appliednew
0tLHCb OT Preliminary
Commissioning Early Measurements Stable running
nsthr = 0.1
nsthr = 0.04
55
Samereconstruc�onasofflineandcompletealignmentandPIDcalibra�onallowstoapplya�ghterselec�ononkinema�csquan��es
RICHcalibra�onallowstousehadronpar�cleiden�fica�oninselec�on,e.g.boostefficiencyforCabibbosuppresseddecayswhilekeepingtheratelowbypre-scalingtheCabibbofavoredcounterpart
SSeelleeccttiioonn ppoowweerr aatt ttrriiggggeerr lleevveell
56BarbaraStoraci,UniversityofZurich
]2 [MeV/cKKKm1800 1850 1900
2Ca
ndid
ates
/MeV
/c
1
10
210
310
410
= 8 TeV DatasPreliminaryLHCb
]2 [MeV/cπππm1800 1850 1900
2Ca
ndid
ates
/MeV
/c
0
5000
10000
15000
20000
25000
= 8 TeV DatasPreliminaryLHCb Trigger
+PID+�ghterPID
Trigger+PID
Technicali�es:– Constantsexpectedtochange:updatedinreal-�me(each
fill,run,…)– Constantsexpectedtobestable:monitoring
Tookadvantageofthefarmcompu�ngpower
RReeaall--ttiimmee AAlliiggnnmmeenntt aanndd CCaalliibbrraattiioonn
Calibra�onFarmtasks: OnlineFarmtasks:
RICHrefrac�veindexandHPDimagecalibra�on
Calorimetercalibra�on OT-t0calibra�on
VELOandtrackeraligment Muonalignment RICHmirroralignment Calorimeterπ0calibra�on
57BarbaraStoraci,UniversityofZurich
RReeaall--ttiimmee ccaalliibbrraattiioonn
5858
2015
2012–nocorrec�ons2015:greenline–π0basedcalibra�on;bluelines–LEDbasedcorrec�ons
2012
LEDaverageover456cellsintheverycentralpartofECAL
Calibra�onoftheRICHandOTsystemtoguaranteestablecondi�onduringdatataking
IntroducedinRunIItheLEDcorrec�onsforECALandHCALareautoma�callyperformed/updatedeveryfillStableL0condi�ons LHCbpreliminary
LHCbpreliminary
Selectedexample
TThhee LLHHCCbb eexxppeerriimmeenntt
59
LHCbisthededicatedheavyflavorphysicsexperimentatLHC
Singlearmforwardspectrometer,coveringapseudorapidityrangeuniqueamongtheLHCdetectors
ItsprimarygoalistolookforindirectevidenceofnewphysicsinCP-viola�onandraredecaysofbeautyandcharmshadrons
PerformanceinRun1andRun2:1. Excellenttracking(p,IP,σPV)
σIP~20μm εtracking>96% Δp/p~0.5–1.0%
2. Excellentdecay�meresolu�on: στ-decay~45fsfprBmesons
3. Excellentpar�cleiden�fica�on εk-ID~95% εμ-ID~97%
J.HighEnergyPhys.06(2013)064
Referencestobeadded
Comparableperformancerespecttothevanillaversionintermsofefficiency
EvenlessghostandclonesfortheOpenCLversion
60
PPeerrffoorrmmaannccee ccoommppaarriissoonn
Ofthe420HLT2lines2016physicsprogramme,150chooseTurbo:~60newlineswithrespectto2015datataking
Newfeaturesin2016:persistencyofthereconstruc�oninfoforthefullevent,higherlevelvariables(i.e.hitsinaconeregionaroundthetrack)availableforfewlines:possibletodocharmspectroscopy! Morespaceneedednow,but
essen�alprepara�onstudiesfortheupgrade
TTuurrbboo ssttrreeaamm:: pprreeppaarriinngg tthhee uuppggrraaddee
61
IntheupgradeTurbowillneedtobemoreflexibleintermsofeventformat,sothateachHLTlinecanchoosehowmuchoftherawandreconstructedeventtopersist.
612elementsaligned elements.ITBoxes("TxTzRz")12boxes elements.TTLayers("Tz")4layers elements.TTModules("TxRz")128modules=>Rzremoved elements.OTCFrames("TxRz")-alignlayersinA-sideandC-
side24layers elements.OTCFrameLayers("Tz")-alignlayers12layers elements.OTModules("TxRz")=>Rzremoved432modules
62
EElleemmeennttss AAlliiggnneedd:: ttrraacckkiinngg
Numberofrunsconsidered:77– ITNumberofupdates:36-25fromautoma�c
alignmentprocedure(run160059). mainlyresponsibleforupdates:mostlyITBox.Tz
– TTNumberofupdates:13-only4fromautoma�calignmentprocedure(run160059) mostlyTTLayers.Tz,some�mesTTModules.T(x|y)
– OTNumberofupdates:16-11fromautoma�calignmentprocedure(run160059). mostlyFrames/Module.T(x),some�mesT(z)/T(y)
63
UUppddaatteess iinn 22001155