detector variaon studies - indico-fnal (indico) · repeat nominal cpv analysis using full mc...

22
Detector Varia+on Studies Elizabeth Worcester (BNL) Calibra+on Task Force Mee+ng October 16, 2018

Upload: others

Post on 30-Sep-2020

7 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

DetectorVaria+onStudies

ElizabethWorcester(BNL)Calibra+onTaskForceMee+ngOctober16,2018

Page 2: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Introduc+on•  Goalistostudyimpactofvaria+onsindetectorperformance/

specifica+onsonCPVanalysis•  Listofvaria+onschosenbasedonimportanceandeaseof

simula+on:–  Deadchannels–  Dri-fieldmagnitude–  DriTfieldnon-uniformi+es–  Noise

•  Methodistore-runsimula+onchangingsomeparameterandrepeatnominalCPVanalysisusingfullMCsimula+on,nominalenergyreconstruc+onandCVNeventselec+on–  Energyreconstruc+onnotre-tuned,crudeenergyscalingisapplied

(morelater)–  CVNisnotre-trained–  CVNselec+oncutisre-op+mized–  Secondordereffects(eg:impactonsystema+cs)notconsidered

Thistalk

Simula+onplanned/inprogressTodo

2

Page 3: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Warning

•  Donotover-interpret!Thesestudiescangiveasenseofwhatspecifica+onsareimportantfortheCPVanalysis,but,withoutfullyrepea+ngtheanalysisincludingreconstruc+ontuningandnetworktrainingandwithoutincludingtheimpactonourabilitytocalibrateandconstrainsystema+cs,theydonottellanythinglikeafullstory!

3

Page 4: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

AnalysisDetails•  FullLArSoTMonteCarlosimula+on

–  GENIEeventgenerator–  GEANT4par+clepropaga+on–  Detectorreadoutsimula+onincluding“realis+c”waveformsandwhitenoise

•  Automatedsignalprocessingandhitfinding•  Automatedenergyreconstruc+on

–  Muonmomentumfromrange(contained)ormul+pleCoulombsca^ering(exi+ng)

–  Electronandhadronenergyfromcalorimetry•  Eventselec+onusingconvolu+onalvisualnetwork(CVN)•  MCtruthusedtoproduceefficiencyhistogramsandtrue-recosmearing

matricesthatareusedasinputtoGLoBESanalysis•  GLoBESanalysisiden+caltoCDR(withtheefficiencyandsmearingcoming

frompreviousstep)performedtoobtainCPVsensi+vity–  Includesnormaliza+onsystema+cschosentorepresentresidualuncertain+es

aTerNDconstraintsandsample-samplecancella+onsapplied

4

Page 5: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

ExampleEfficiency

5

Page 6: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

ExampleSmearing

6

Page 7: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

AnalysisInputDetails•  Badchannelstudy:

–  MCC10.1–  Remove5%or25%ofallchannels(Backhouse)

•  Atrandom•  Groupedbychip(groupsof16)•  Groupedbyboard(groupsof128)•  Forgroupings,~followprotoDUNEmappingscheme

•  DriTfieldstudy:–  May2018CVNupdate(improvedefficiencyrela+vetoMCC10.1)–  ReducethedriTfieldinsim/recofclfiles–  Generatenewfieldresponsehistogramsgivenreducedfield(Garfield,

thanksYichen!)

–  Comparetonominalsehngs:

7

Page 8: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

ExampleEventDisplays:BadChannels

8

5%random

25%random

Page 9: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

EnergyCorrec+onDetails•  Runonlynominalenergy

reconstruc+on,whichiscalorimetricforEMshowers–  Expecttobemissing~5%or25%of

energyforrandombadchannels–  Scalingreconstructedenergyby

1.05or1.25approximatelycorrectsbacktonominalreconstructedenergyasexpected

•  Forstudieswhereexpecta+onislessobvious(ie:alltheothers)IjustscaletoforceaGaussianfittothepeaktomatchnominal

•  Note:scalingperformedatplotlevelonly,doesnotimpacteventselec+on

9

Exampleforrandomlydistributedbadchannels:

Page 10: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

EnergyResolu+onCrosscheck•  Checkfrac+onal

energyresidualsforeachcase

•  Seebiasasexpectedinuncorrectedresiduals

•  Nominalresolu+onsimilartothatreportedbyN.Grant

•  Seesomeincreaseinresolu+onfor25%missingchannelscase

10Nominal 5%Random 25%Random

Page 11: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

EffectofDriTFieldonEreco•  EffectssimulatedinMC:

–  Largerdiffusion(driTingforlonger)–  Largerlossduetoa^achment(driTingfor

longer)•  Ofcoursethisdependsonpurity–we

simulate3mselectronlife+me–  Largerlossduetorecombina+on(Birks’

Law)•  Usingh^ps://lar.bnl.gov/proper+es/)

fordE/dx=4MeV/cm,3.6mdriTdistance,nominal(3ms)electronlife+me:–  At500V/cm:Ra=0.48,Rc=0.63–  At250V/cm:Ra=0.33,Rc=0.51

•  Calorimetricenergyreconstruc+onusesaverageRc=0.63–  Expecttoreconstructaround25%less

energyatlowerfield•  Life+mecorrec+onishandledcorrectly•  ImpactofdiffusiononErecoshouldbe

small?11

Page 12: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

PIDCutTuning•  Method:

–  SetnumuPIDcutat0.5(notuning)–  RequirenumuPID<0.5andnuePID>xfornueeventtobeselected(and

inversefornumuevents)–  ScanthroughnuePIDcutvaluesinstepsof0.05–  EvaluateCPVsensi+vityatδCP=-π/2–  Defineop+mumcutvalueastheonebeforesensi+vitystartstodecrease(so

mightmiss2ndop+mumvalueifsensi+vityisnotmonotonicaboveandbelowop+mum)

•  Resul+ngop+mizednuePIDcut(CVN)–  Nominal:0.7–  5%badchannels:0.65–  25%badchannels:0.6

•  Withmorebadchannels,prefercutwithslightlyhigheracceptancesuchthatsignalefficienciesaremoresimilarthanwithun-tunedcuts–  Balancedbyincreasingbackgroundwithloosercuts

12

Page 13: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

SummaryofCorrec+ons/Op+miza+on

Sample EnergyScaleFactor Op:mizedνeCut

MCC10.1Nominal 1.0 0.7

5%Bad–Random 1.05 0.65

5%Bad–ByChip 1.03 0.65

5%Bad–ByBoard 1.02 0.6

25%Bad–Random 1.25 0.6

25%Bad–ByChip 1.21 0.55

25%Bad–ByBoard 1.13 0.45

13

Sample EnergyScaleFactor Op:mizedνeCut

May2018Nominal 1.0 0.85

250V/cm 1.264 0.8

Page 14: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Efficiency:RandomBadChannels

14

Page 15: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Efficiency:5%BadChannels

15

Page 16: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Efficiency:25%BadChannels

16

Page 17: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Efficiency:DriTField

17

Page 18: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Sensi+vity:RandomBadChannels

18

Page 19: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Sensi+vity:BadChannelGroups

195%BadChannels 25%BadChannels

Page 20: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

Sensi+vity:DriTField

20

Page 21: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

ConclusionfromExis+ngStudies•  CPVanalysispre^yinsensi+vetobadchannels,par+cularlywhengrouped

randomly–  Seemsreasonable:byeye,differen+a+onbetweentracksandshowersisquite

easyevenwithmanymissinghits•  Largergroupsofbadchannels(eg:boards)causemoretroublethan

randombadchannels–  Againseemsreasonable:whenmissingawholeboard,significantpor+onsof

showerscouldbemissing–  Energycorrec+onmethodalsolessrobustinthiscase–overcorrec+ng

showersthatdon’thappentohavebadboardsandundercorrec+ngthoseforwhichalargefrac+onofhitsaremissing

–  Somelossofsensi+vitycouldberegainedbydetailedenergycorrec+onsand/orretrainingCVN

•  250V/cmdriTfieldproduceslargedegrada+oninsensi+vityrela+vetofull500V/cm–  Wouldnetworkretrainingrecoversomeofthiseffect?

21

Page 22: Detector Variaon Studies - INDICO-FNAL (Indico) · repeat nominal CPV analysis using full MC simulaon, nominal energy reconstruc+on and CVN event selec+on – Energy reconstruc+on

GoingForward•  Iclaimbadchannelstudiesarecomplete:conclusionisCPVanalysis

reasonablyrobusttoanyrealis+csetofbadchannelsevenunderworstcasescenarioofnoretrainingordetailedenergycorrec+ons

•  Retrainingnetworknotfeasibleformanyvaria+ons–ifveryimportantwecouldpickafewtostudymorecarefully(250V/cm?)–  Firststepwouldbetolookatsomeeventdisplays–doweseemajorvisual

differencesbetweenshowersinthedifferentsamplessuchthatwesuspectretrainingwillhelp?

•  Fieldnon-uniformitystudies–  ImplementinMCusingspacechargeeffectsimula+onmachinery(justdummy

up“spacecharge”maps)–  Inprogress–Boismakingtheini+alfielddistor+onmapsforus

•  Noisestudies–  Studyseverallevelsofincreasedwhitenoiseandadata-drivennoise

simula+on–  ToolsexistwithinMC,butworknotyetac+velyunderwaytogenerate/analyze

thesesamples

22