Treatment of the Correlations in b-Tagging Systematics in ATLAS and CMS

Introduction

  • Top physics at LHC has entered the realm of precision physics for both experiments ATLAS and CMS
    • gain in precision by combining the results of both experiments
  • Correct treatment of the uncertainties important
  • Flavor tagging is one of the dominant systematics uncertainties, therefore compare for ATLAS and CMS
    • the correlations between flavor tagging algorithms and calibration techniques,
    • the sources of uncertainty and provide procedures for the combination
  • b-jet identification (tagging) is a key ingredient of many analyses
    • so far no correlation has been considered
  • the two collaborations use different approaches regarding every aspect of b-jet identification:
    • b-tagging algorithms and working point definition
    • calibrations samples and methods
    • combination strategy
    • source of systematics considered and their treatment
  • we compared the different approaches, and identified a list of common sources of uncertainty:
    • treatments of each uncertainty compared to understand how it's effect is correlated in the flavour tagging
    • size of the uncertainties has been found to be in reasonable agreement across the whole pT spectrum of jets from top decays
  • a proposal is advanced for the treatment of b-tagging correlations for future top physics combinations at LHC

Contact Persons

Documentation

  • ATLAS
    • Measurement of the b-tag Efficiency in a Sample of Jets Containing Muons with 5 fb-1 of Data from the ATLAS Detector, ATLAS-CONF-2012-043
    • Measuring the b-tag efficiency in a ttbar sample with 4.7 fb-1 of data from the ATLAS detector, ATLAS-CONF-2012-097
    • Calibration of b-tagging using dileptonic top pair events in a combinatorial likelihood approach with the ATLAS experiment, ATLAS-CONF-2014-004
    • b-jet tagging calibration on c-jets containing D* mesons, ATLAS-CONF-2012-039
    • Measurement of the Mistag Rate of b-tagging algorithms with 5 fb-1 of Data Collected by the ATLAS Detector, ATLAS-CONF-2012-040
  • CMS
    • Identification of b-quark jets with the CMS experiment, 2013 JINST 8 P04013
    • Performance of b tagging at sqrt(s)=8 TeV in multijet, ttbar and boosted topology events, CMS-PAS-BTV-13-001

Flavour Tagging Algorithms at ATLAS / CMS

  • b-jets can be identified by the distinctive signatures of the B hadrons from the b-quark hadronization
    • large life times lead to:
      • tracks with large impact paramenters (wrt primary vertex)
      • displaced secondary vertices
    • high semi-leptonic decay branching ratios
      • presence of soft leptons inside the jet cone
      • charateristical signatures from muons and electrons
  • flavour tagging algorithms
    • provide a discriminator value for each jet
    • most performing algorithms using multivariate approaches to combine information from displaces tracks and secondary vertices
  • best performing algorithms for ATLAS
    • MV1: combination of the above information with a neural network to build a discriminator
    • operation points: fixed b-jet efficiencies (usually 60, 70 and 80%)
  • best performing algorithm for CMS
    • CSV: combination of secondary vertices and track-based lifetime information to build a likelihood-based discriminator
    • operation points: fixed mis-identification rate for light jets at 0.1, 1 and 10%
  • top physics analyses in both experiments are using mainly the medium operation point (MV1 at 70% (ATLAS) / CSV at 1% (CMS))
    • other tagging algorithm are available and calibrated

Flavour Tagging Calibrations

  • Discriminator values from the b-tagging algorithms not perfectly described by Monte Carlo simulations
    • calibration of the MC performance needed
    • done as a function of kinematic variables (e.g. pT or eta)
  • measure b-jet content of a b-quark enriched sample before and after flavour tagging requirement
  • data-to-MC scale factors for b-jet tagging efficiencies and for light jet mis-identification rates
    • b-tag efficiency SFs provided as a function of jet pT
    • eta dependence also studied and found to be flat in both experiments
    • c-jet calibration: independent analysis at ATLAS, b-jet calibration with inflated uncertainties for CMS
  • Measurement of the systematic uncertainty introduced by flavour tagging
    • as important as the measurement of the scale factors
    • central ingredient for the total systematic uncertainty of a measurement using flavour tagging
  • Calibration with di-jet events: jets with a soft muon
    • exploiting the high semileptonic decay branching ratio of B-hadrons
    • often looking at the away jet to enrich the sample in b-bbar content
    • independent data sample from top selection
    • muon requirement can have a bias to the impact parameter based flavour tagging algorithms
  • Calibration with top-pair events: jets from top quark decays
    • BR(t⟶Wb) ~100%
    • isolated leptons from W decays to reduce the background contamination
    • single lepton and dilepton decay channels providing two orthogonal samples
    • very pure and well known b-jet sample
    • using the same data set as for top-physics measurement
      • systematic uncertainties must be treated carefully

Calibration methods

  • Di-jet based calibrations:
    • ATLAS:
      • pTrel: template fit of the muon pT w.r.t the jet axis (only for 7 TeV data)
      • System8: equations with 8 unknowns, using two samples (with different purities) and two weakly correlated taggers (tagger under study and the muon pTrel)
    • CMS:
      • pTrel and system8 method as well
      • extending pTrel method up to 800 GeV looking at muon IP3D
      • Lifetime tagger (LT) method: template fit of a reference discriminator (usually JP which is calibrated in data)
  • top-pair based calibration:
    • ATLAS:
      • tag counting: fit b-jet efficiency on tagged jets multiplicity (only for 7 TeV data)
      • Kinematic selection and tag&probe: based on sample composition estimates
      • combinatorial likelihood PDF approach (only 8 TeV data), most precise measurement
      • kinematic fit: use a kinematic fit to reconstruct the final states and increase the purity of the sample
    • CMS:
      • Tag counting techniques (inclusive in pT)
      • bSample method (kinematic based, inclusive in pT)
      • LT method on dilepton events (only for 8 TeV data)
  • Final usage: combination of calibrations
    • ATLAS:
      • various combinations of the calibrations provided among di-jets (pTrel, S8) and ttbar calibraions
      • default calibration is a combination of di-jet (S8) and ttbar (PDF-method)
      • the choice of the combination is up to the individual analysis teams
      • global fit with all systematic uncertainties as nuisance parameters which can shift the mean data/MC SF
      • systematic uncertainties can be fully correlated or uncorrelated in each single kinematic bin or across kinematic bins
    • CMS:
      • all measurements on jets with a soft muon and top based LT analysis
      • combination also provided without top-pair based calibrations
      • using the least squared BLUE method
      • source of uncertainties common between two or more methods are taken as (anti-)correlated
      • systematics considered correlated across the bins

Correlations between Systematics among ATLAS and CMS

  • Two types of correlations of the systematic uncertainties of flavour tagging calibrations
    • correlations with other parts of top analyses
      • sources of uncertainties on flavour tagging performance measurements also considered in the physics analysis
      • example: Jet Energy Scale in a top quark mass measurement
    • correlations between the two experiments
      • common sources of uncertainty
      • related to general physics modeling of the calibration sample
  • correlations affecting different categories of systematics
    • General physics modelling
      • e.g. ISR/FSR, parton showering
      • correlated to analysis and between experiments
    • Specific physics modelling
      • e.g. pT spectrum of soft muons, light/charm ratio
      • uncorrelated to analysis but correlated between experiments
    • Detector description
      • e.g. JES, pile up
      • correlated to analysis but uncorrelated between experiments
    • Method specific
      • uncorrelated to analysis and between experiments
  • Proposed treatment of systematics:
    • Systematic uncertainties uncorrelated between experiments will be added together in an uncorrelated systematics category
    • for the time being, the correlations between different part of top physics analyses will not be considered
      • experiments must decide internally how the deal with them
    • the main sources of systematic uncertainties correlated between the experiments will be quoted separately in the systematic breakdown
    • the systematics which come from the same source will be treated with a correlation factor of one in the combination
    • the very small correlated systematics will be merged to an existing similar category

Physics Modelling Systematics

  • The sources of systematic uncertainties correlated between the experiment are related to the general physics modelling in the simulated samples
  • Physics modelling of muon jets
    • production of b- and c-quarks (fraction of gluon splitting)
    • decays of B-hadron
    • b-quark fragmentation
    • ratio of charm-to-light jets
    • simulation of the pT spectrum of the muon
  • Physics modelling of top-pair events
    • top-pair MC generator
    • description of the parton shower
    • ISR / FSR
    • underlying events

Source treatment in ATLAS Treatment in CMS Corr.
b/c production b,c --> gg scale by 50% b,c --> gg scale by 50% YES
B decay reweight acc. to BR neglected NO
b-quark frag. av. B hadron energy fraction varied by +/- 5% av. B hadron energy fraction varied by +/- 5% YES
c/l ratio c/l ratio scaled by factor 2 l/c ratio scaled by 20% YES
muon pT pT spectrum reweighting vary cut on muon pT YES
top generator compare POWHEG to MC@NLO compare fit to templates for QCD NO
parton shower compare HERWIG to PYTHIA comoare HERWIG to PYTHIA YES
ISR / FSR use AcerMC+Pythia varying Q2 scale and ME-PS threshold YES
underl. event negligible varing parameters NO

Split of Systematics

  • Six major correlated sources identified
    • b/c production, muon pT, charm-to-light ratio, b-fragmentatin, PS, ISR/FSR
    • contributing at the level of 0.2 - 1.3%
    • to be provided as separate uncertainties
    • charm-to-light systematics equally small for ATLAS and CMS (up to 0.2%)
      • very small uncertainty
      • add to b-fragmentation, not consider separately
  • Analyses should propagate in their analyses all the six uncertainties as a function of pT
  • Analyses should merge the charm-to-light systematics with the b-fragmentation systematics into one systematics
    • we have then only five systematics left for the combination: b/c production, muon pT, charm-to-light ratio + b-fragmentatin, PS, ISR/FSR
  • Resulting five uncertainties of final result can be combined taking into account the correlations among the experiments
  • Summary table (for all six systematcs):
source size at ATLAS size at CMS
b/c prod. low pT: 0.1% - 0.2%, high ph: 1.2% - 2.0% low pT: 0.1% - 0.3%, high ph: 0.5% - 1.3%
mu pT low pT: 0.8% - 2.7%, high ph: 0.2% - 0.8% low pT: 0.1% - 1.1%, high ph: 0.1 - 0.9%
c/l ratio <0.1% - 0.2% <0.1% - 0.2%
b-frag 0.2% - 2.7% 0.2% - 0.8%
PS 0.1% - 4.1% 0.3% - 0.6%
IFSR 01.% - 4.1% 0.3% - 0.6%

Systematic Breakdown

  • The breakdown of the systematic is provided for the tagger mostly used in top analyses, and for different data taking and calibration periods
    • please, contact us know if your analysis is using
  • ATLAS breakdown of systematics provided for MV1 and MV1c in the CalibrationDataInterface for the 8 TeV data :
  • CMS breakdown for CSVL and CSVM are provided for the PromptReco and Winter13 reprocessing of 8 TeV data:

-- RobertoChierici - 13 Jun 2014

Edit | Attach | Watch | Print version | History: r12 | r10 < r9 < r8 < r7 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r8 - 2014-09-25 - MartinZurNedden
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    LHCPhysics All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback