240 likes | 402 Views
W → e n selection and background rejection. By Damien Prieur Alessandro Tricoli Monika Wielers. ATLAS RAL Physics Meeting, 13 th November 2007. Very preliminary!!. Overview. W inclusive cross section measurement (CSC W/Z incl. cross-section note)
E N D
W→en selection and background rejection By Damien Prieur Alessandro Tricoli Monika Wielers ATLAS RAL Physics Meeting, 13th November 2007
Very preliminary!! Overview • W inclusive cross section measurement (CSC W/Z incl. cross-section note) • Study for early (1031 lumi) and later (1033 lumi) running. • Trigger-aware selection of W->en events against background: • Selection ‘a la TDR’ (mainly for comparison): cut based technique • Data-driven selection procedure: fit and subtract background Alessandro Tricoli, RAL
Data Sets • Signal (W->en) and Backgrounds generated with PYTHIA • Reconstruction ATHENA v12.0.6 • Offline Analysis: AOD/AAN-tuples • use panda to submit analysis jobs All sets with correct G4 range (30 mm) Z→tt and top background negligible => not considered here Alessandro Tricoli, RAL
W Event Selection: ‘a la TDR’ • W Selection: • Trigger menu e25i applied: (optimised for 1033 cm-2s-1, Monika is working on extraction of 1031 menu in 12.0.6 AOD) • one isolated e±, tuned for efficiently select e± with ET> 25 GeV • L1, L2 and EF • only exception QCD, due to poor stat.: no trigger sel. • Electron Identification: • Tight isEM ==0 • Medium (isEM & 0x3FF)==0 • Loose (isEM & 0x7)==0 • cracks removal h=1.37-1.52 and |h|<2.4 • Electron ET>25 GeV • Missing-ET >25 GeV • Jet Cuts: Jets PT<30 GeV, Event Recoil PT< 20 GeV loose track-cluster matching + had leakage+ shower shapes in 2nd EM sampling ‘standard’ track-cluster matching + hadronic leakage + shower shapes in 1st and 2nd sampling cuts on b-layer hit + cuts on TR ratio Alessandro Tricoli, RAL
MET_Truth missETTruth - missETRefFinal MET_RefFinal Missing ET in W->en events • Study of missET after e25i + isEM + ETele > 25GeV • Missing ET Truth (MET_Truth) • Standard ATLAS reconstructed Missing ET (MET_RefFinal) from cells • Fit differences missETTruth – missETRefFinal Loose ele-cuts Loose ele-cuts Fit with Loose ele-cuts missETTruth – missETFinal: Mean= -0.15 ± 0.02 GeV s = 5.58 ± 0.02 GeV 2/NDF = 2984/ 37=81 Fit with Medium ele-cuts missETTruth – missETFinal: Mean= -0.15 ± 0.02 GeV s = 5.45 ± 0.02 GeV 2/NDF = 2602/ 35=74 Fit with Tight ele-cuts missETTruth – missETFinal: Mean= 0.02 ± 0.02 GeV s = 5.46 ± 0.02 GeV 2/NDF = 2187/ 35=62 Alessandro Tricoli, RAL
Cumulative Bkg After ele-ID only After Missing ET > 25 GeV After Missing ET > 25 GeV after e25i + isEM + ETele >25GeV after e25i + isEM + ETele >25GeV + Miss ET >25GeV after e25i + isEM + ETele >25GeV + Miss ET >25GeV W -> tn QCD Z -> e-e+ W -> en After ET recoil < 20 GeV after e25i + isEM + ETele >25GeV+ ETjet <30GeV + ETrecoil<20GeV Distributionsafter loose selection cuts Bkg contamination can be rather small after cuts, but large uncertainties on QCD background (yellow). Alessandro Tricoli, RAL
Event Selection and Bkg Rejection a la TDR Lumi = 50 pb-1 Bkg contamination can be rather small after cuts, but large uncertainties on QCD background. Alessandro Tricoli, RAL
Jet Pt vs Recoil Cut • Jet Pt and Recoil cuts redundant on QCD rejection: • applied together do not improve B/S for QCD • Recoil cut used by MW measurement to reduce low MT tail Shall we keep only one of them? Which one? Lumi = 50 pb-1 Swapped Shall we keep Jet ET Cut only? Or do we want to be consistent with MW group? Alessandro Tricoli, RAL
Data-driven W Event Selection • For early running we want to • Minimise our dependence on MCs • Measure backgrounds from data • Possibly minimise our selection biases Try a data-driven approach For background rejection • Problems with Sel ‘a la TDR’: • difficult estimation of QCD background after all cuts: • not enough MC stat., but plenty of stat. when data come • amount and shape under W peak difficult to estimate: • ETmiss and Jet Cuts remove most of backgrounds, but leave ‘some’ under W peak => can only rely on MC estimate Alessandro Tricoli, RAL
Data-driven W Event Selection • Proposed W Selection: • Trigger menu e25i • Electron Identification • cracks removal h=1.37-1.52 and |h|<2.4 • Electron ET>25 GeV • Missing-ET >25 GeV • Jet Cuts: Jets PT<30 GeV, Event Recoil PT< 20 GeV Etmis and Jet cuts remove most of backgrounds, loose knowledge of shapes • Z->ee Background Removal: • cut on Mee invariant mass • QCD Background Removal: • fit QCD spectrum independently • subtract QCD spectrum away Replaced by: Left over W->tn not yet studied here Alessandro Tricoli, RAL
Z->ee Removal • Select a well identified 1st electron: isEM == loose/medium/tight + ET > 25 GeV Look for opposite charged, loosely identified 2ndelectron (|h|<2.5) isEM == loose Cut on invariant mass: 65 GeV< Mee <130 GeV Not all Z events are picked up, so try to extend acceptance: • If no e-e-pair falls in the above Mee window For the 2nd electron legphoton container • photon isEM == loose • 65 GeV< Meg <130 GeV • If no e-g-pair falls in the above Meg window Look for 2nd electron leg injet container: • select EM-like jets (se next slide) • EM-like-jet: 2.5 <|h|<3.2 and 3.2 <|h| • 65 GeV< Me-jet <130 GeV Alessandro Tricoli, RAL
Jet w/o matching ele Jet w/o matching ele Jet w/ matching ele Jet w/ matching ele Log(HAD/EM) Log(HAD/EM) < 0.04-0.1 HAD ___ EM All jets All jets Log(HAD/EM) Log(HAD/EM) Forward Z->ee removal In Z->ee removal we want to • Maximise the rejection of Z events • Minimise the rejection of other backgrounds (mainly QCD) • In forward regions 2.5<|h| • use fraction of HAD/EM Energy Sampling in calorimeters(Jet Constituents) to select 2nd electron leg within Jet container(FCAL0 considered as EM) Z->ee Sample Z->ee Sample 2.5<|h|<3.2 3.2< |h| Jet w/o matching ele QCD Sample QCD Sample Good electron- Jet separation 2.5<|h|<3.2 3.2< |h| Alessandro Tricoli, RAL
Z->ee Removal: MT distribution after e25i + isEM + ETele >25GeV + 65<Minv<130 GeV after e25i + isEM + ETele >25GeV Cumulative W -> tn QCD Z -> e-e+ W -> en Before Minv cut After Minv cut Medium isEM • ele-ele Pair Minv cut rejects most of the Z->ee: BZee/SWenu from ~25% to ~7% • ele-g Pair Minv cut rejects few Z->ee: BZee/SWenu from ~7% to ~6% • ele-jet Pair Minv cut rejects more Z->ee: BZee/SWenu from ~6% to ~2.7% Cut –Away Events Pair Minv (GeV) Little amount of rejected QCD, Negligible amount of rejected W->en & W->tn Alessandro Tricoli, RAL
Robust Event Selection • Z->ee reduced from ~25% to ~2.7% • W->enu affected by <2% • QCD reduced by ~2-4% Lumi = 50 pb-1 • 1st Leg: • electron isEM == loose/medium/tight • electron ET>25GeV • 2nd Leg: • electron: opposite charged ele w/ isEM==loose • photon: isEM==loose • EM-like jet: 2.5<|h|<3.2, 3.2<|h| Alessandro Tricoli, RAL
Distortion on MT distribution shapes Ratio of MT distributions (areas normalised to unity) After / Before the Mee Cut W->enu QCD MTAfter / MTBefore MTAfter / MTBefore MT (GeV) MT (GeV) No Shape distortions for QCD Small effect only on tails for W->enu Alessandro Tricoli, RAL
QCD background extraction • Original Plan: • After e-IDtry to fit background MT shapefrom orthogonal QCD sample • After missing ET cut, background fully within MT from We spectrum, no way to fit • Use lower electron MT spectrum to normalise the distributions After cut on ET, crack removal and medium IsEM Alessandro Tricoli, RAL
Choice of QCD sample to fit • Use EM candidates from photon container after calo shape cuts • Quite a different MT spectrum • Problem very likely the different kinematics from direct photon production, photon brem etc. • Looked at match to ‘real’ photons from truth: most EM candidates have matched photon! • Will look into more detail using newly reconstructed JF17 ntuples which have correct truth information • Same problem using EM candidates from electron container: electrons passing calo shape cuts, but fail the medium IsEM cuts (failed electrons) Detailed investigation needed Alessandro Tricoli, RAL
ETmiss [GeV] Very preliminary!! Alternative QCD sample to fit:ETmiss Look at ETmiss distr. in events with a reconstructed photon Photon ETmiss fit: Try exponential(simplest choice) Ratio Electron/Photon close to 1 on most ETmiss range Lin(y) QCD Log(y) Possible reason: left-over Ws/Zs in JF17 (QCD) DS (to be investigated) Problem: for ETmiss>25 GeV stat is poor, but clearly ETmiss electrons > ETmiss photons Alessandro Tricoli, RAL
Cumulative MC Subtracted “data” W -> tn Z -> e-e+ W -> en Fit Normalisation and Subtraction Normalise to first bins of ETmiss In electron sample Subtract Normalised QCD fit from “Data” ETmiss [GeV] Underestimation of high tails Causes under-subtraction Normalisation dominated by QCD Alessandro Tricoli, RAL
QCD Cumulative MC Fit Subtracted “data” W -> tn Z -> e-e+ W -> en W Transverse Mass “Reweighted” MT, i.e. after QCD subtraction, Comparison: QCD (ele sample) And Fit (after normaliation) See problem of high tails Additional cut ETmiss>12 GeV ETmiss [GeV] MT [GeV] Convert bin-by-bin subtraction Into event weight: (Ele - Fit) / Ele Event Weights Weighst: =0 ETmiss < 12.5 GeV (by fit/norm definition) ~1 ETmiss > 80 GeV Alessandro Tricoli, RAL
Conclusions • Standard W Selection (a la TDR): • confirmation of previous estimates for W->tn, Z->ee rejection • large uncertainties on amount and shape of QCD background • proposal of removing recoil cut (find agreement with W mass group) • Proposed Robust W selection for early running (data-driven): • replace ETmiss and jet veto cuts with • explicit Z->ee removal by cutting on e-e pair invariant mass • fit of QCD MT distribution • good rejection of Z->ee background is possible from 25% to <3% • QCD background extraction • Needs detail study: • investigate left-over Ws/Zs in JF17 sample • improve parameterization (polynomial?) • try different distributions? (i.e. recoil, electron ET…) • test bin-by-bin subtraction and event-reweighting Alessandro Tricoli, RAL
EXTRAS Alessandro Tricoli, RAL
isEM==medium isEM eta cut Mee cut Et cut Alessandro Tricoli, RAL
Use EM objects from electron container • Use EM candidates from electron container: select electrons which pass calo shape cuts, but fail the medium IsEM cuts • Not too much stat here, but the QCD spectrum looks quite similar • Around 3x more stat available, which we’ll add • Then fit the spectrum using the events with failed medium IsEM and try to do the normalisation at low MT Medium IsEM calo IsEM, no medium IsEM Alessandro Tricoli, RAL