Unfolding jet multiplicity and leading jet p T spectra in jet production in association with W and Z...
-
Upload
donald-foster -
Category
Documents
-
view
219 -
download
3
Transcript of Unfolding jet multiplicity and leading jet p T spectra in jet production in association with W and Z...
Unfolding jet multiplicityand leading jet pT spectra
in jet production in association with W and Z Bosons
Christos LazaridisUniversity of Wisconsin-Madison
on behalf of the V+Jets group
November 28, 2011
Christos Lazaridis, University of Wisconsin-Madison
Outline
• Analysis flow• Unfolding overview• Unfolding methods• Validation • Unfolding data• Error propagation• Final results• Conclusions
November 28, 2011Unfolding
2
Christos Lazaridis, University of Wisconsin-Madison
Analysis Flow
November 28, 2011Unfolding
3
Electron SelectionA
LL
EVENTS Jet
Selection
ZCandidat
es
Signal yields vs. # jets Unfold jet
multiplicity and leading jet
pT
Fit distributions
Correct yields for
reconstruction efficiency
Ratio plots• σ(Ζ+n jets) / σ(Ζtotal)• σ(Ζ+n jets) / σ(Ζ+(n-1) jets)
Christos Lazaridis, University of Wisconsin-Madison
Unfolding Overview• Measured distributions get “smeared”
– Due to detector resolution and efficiency effects
– “True” (particle-level) distribution differs from measured
• Jet distributions are unfolded– “Response matrix” created
based on Monte Carlo– Correlates generated with
reconstructed quantities• Number of jets• Leading jet pT
– Matrix is inverted and applied to data• Used Singular Value Decomposition
method to unfold data• Bayesian method also evaluated
– Used for systematic studiesNovember 28, 2011
Unfolding4
Responsematrices
# jets
Leading jet pT
Unfolding5 Christos Lazaridis, University of Wisconsin-Madison
Unfolding MethodsSingular Value Decomposition
– Unfolding resembles a Fourier expansion • Low frequencies systematic differences between MC and data• High frequencies statistical fluctuations in data• Regularization parameter effectively determines up to which
frequencies the terms in the expansion are kept
– Factorizing A = USVT
• U(mxm), V(nxn) : Orthogonal matrices– Columns of U, V : left & right singular vectors
• S(mxn) : Diagonal matrix with non-negative diagonal elements – Sii ≥0 : singular values
– Regularization parameter kSVD
• Small value may bias the unfolding result towards MC truth• Large value may give a result dominated by unphysically
enhanced statistical fluctuations
November 28, 2011
Unfolding6 Christos Lazaridis, University of Wisconsin-Madison
Unfolding MethodsBayes
• Iterative method– Starting with an initial set of
probabilities pi
– Obtaining an improved estimate via
• Probability an event is observed in bin i in terms of response matrix R and prior probability pi
– Regularization parameter determines number of iterations
November 28, 2011
Christos Lazaridis, University of Wisconsin-Madison
Studying unfolding methods : SVDZ+Jets leading jet pT
November 28, 2011Unfolding
7
Method: SVD; kTERM = 5 (optimal) Method: SVD; kTERM = 10
Unfolding8 Christos Lazaridis, University of Wisconsin-Madison
Studying unfolding methods : BayesZ+Jets leading jet pT
• s
November 28, 2011
Method: Bayes; #iterations: 2 (optimal) Method: Bayes; #iterations: 4
For >3 iterations we start getting increasing
disagreement
Unfolding9 Christos Lazaridis, University of Wisconsin-Madison
Validation of unfolding
• Three types of tests to verify procedure– Unfolding distribution using the same signal
MC used to derive the Response Matrix– Unfolding distribution of a signal MC
different than the one used to derive the RM
– Unfolding distribution obtained in a data-like mixture of MC signal and background samples that should reflect the corresponding mixture in data• Background subtraction and efficiency
corrections are applied before unfolding
November 28, 2011
Christos Lazaridis, University of Wisconsin-Madison
Validating jet multiplicity unfoldingZ+Jets
• Closure test performed to verify that unfolding works as expected:– Response matrix from the Z+Jets, Z2 Tune MadGraph Monte Carlo– Tests performed with:
• Z2 Tune, MadGraph MC – different event set
• D6T Tune, MadGraph MC• Z2 Tune, Pythia 6 MC
November 28, 2011Unfolding
10
MadGraph Z2, SVD (5) MadGraph Z2, Bayes (4) Pythia Z2, SVD (5)GeneratedReconstructedUnfolded
GeneratedReconstructedUnfolded
GeneratedReconstructedUnfolded
Reconstructed/Generated
Unfolded/Generated
Reconstructed/Generated
Unfolded/Generated
Reconstructed/Generated
Unfolded/Generated
• Unfolding performed on exclusive jet bins• Ratio is comparison of reconstructed events
before and after unfolding with the generated MadGraph n-jets distribution
11Unfolding
Christos Lazaridis, University of Wisconsin-Madison
Validating jet multiplicity unfoldingW+Jets
November 28, 2011
• Closure test performed to verify that unfolding works as expected• Response matrix derived from MadGraph Z2 W+Jets sample
GeneratedReconstructed
Unfolded
GeneratedReconstructed
Unfolded
Reconstructed/GeneratedUnfolded/Generated
Unfolding MadGraph Z2 W+Jets Unfolding Pythia Z2 W+Jets
Christos Lazaridis, University of Wisconsin-Madison
Validating leading jet pT unfoldingZ+Jets
• Same procedure as with number of jets unfolding• To select optimal bin width, the jet resolution was studied
– Bin sizes correspond ~2σ of the jet resolution in that pT region
– Minimizing bin-to-bin migrations
• Best results given by SVD with kterm= 5
November 28, 2011Unfolding
12
MadGraph Z2, SVD (5) MadGraph D6T, Bayes (5) Pythia Z2, SVD (5)GeneratedReconstructedUnfolded
GeneratedReconstructedUnfolded
GeneratedReconstructedUnfolded
Reconstructed/Generated
Unfolded/Generated
Reconstructed/Generated
Unfolded/Generated
Reconstructed/Generated
Unfolded/Generated
Christos Lazaridis, University of Wisconsin-Madison
Unfolding Exclusive Jet MultiplicityApplication to data : Z(ee) + Jets
November 28, 2011Unfolding13
Exclusive jet multiplicity• Response matrix from Z+Jets, Z2 Tune MadGraph Monte Carlo
• Data yields corrected for selection efficiency
• Improved agreement after unfolding
Ratio with MadGraph Z2 Tune
Generated MCReconstructed Data
Unfolded Data
Reconstructed/GeneratedUnfolded/Generated
Exclusive jet multiplicity
Christos Lazaridis, University of Wisconsin-Madison
Unfolding Leading Jet pTApplication to data : Z(ee) + Jets
November 28, 2011Unfolding14
Leading jet pT• Corrected leading jet pT
• Response matrix from the Z+Jets, Z2 Tune MadGraph Monte Carlo
• Unfolding leads to better agreement
• Indication that Monte Carlo underestimates in the low pT region
Generated MCReconstructed Data
Unfolded Data
Reconstructed/GeneratedUnfolded/Generated
Leading Jet pT
Ratio with MadGraph Z2 MC
Unfolding15 Christos Lazaridis, University of Wisconsin-Madison
Error propagation in unfolding
• Unfolding is performed on the uncorrelated n-jet bins – n=0-3, n>=4
• Unfolded exclusive jet rates are used to compute the inclusive rates
• Uncertainties are divided in three categories:– Statistical (from the fit)– Systematics uncorrelated across bins (lepton efficiency)– Systematics correlated across bins (jet counting)
• The unfolding procedure is run multiple times to determine final values with proper uncertainty estimate:– Using statistical errors only– Using statistical + uncorrelated systematics– Using central values shifted by correlated systematics– Using unfolding alternatives in algorithm, response matrix, w/o PU
November 28, 2011
Christos Lazaridis, University of Wisconsin-Madison
Final cross section ratiosσ(Ζ+n jets) / σ(Ζtotal)
November 28, 2011Unfolding16
• σ(Ζ+n-jets) / σ(Ζ+≥0-jet) ratio– Luminosity uncertainty cancels out– Event selection uncertainty reduced
• Data points– Error bars correspond to statistical
errors
• Systematic uncertainties– Jet counting
• Yellow band– Unfolding
• Blue striped band
• Good agreement between data and MadGraph
• PYTHIA fails to describe data– Result of the Parton Shower
mechanism for higher-order corrections
Ratio with Monte Carlo
Inclusive Jet Multiplicity
0
Christos Lazaridis, University of Wisconsin-Madison
Final cross section ratiosσ(Ζ+n jets) / σ(Ζ+(n-1) jets)
November 28, 2011Unfolding17
• σ(Ζ+n jets) / σ(Ζ+(n-1) jets) ratio– Reduces jet energy scale
uncertainty
• Data points– Error bars correspond to
statistical errors
• Systematic uncertainties– Jet counting
• Yellow band
– Unfolding• Blue striped band
• Good agreement between data and MadGraph
• PYTHIA does not model data as well as expected
Inclusive Jet Multiplicity
Ratio with Monte Carlo
18Unfolding
Christos Lazaridis, University of Wisconsin-Madison
Unfolded Leading Jet pT Spectrum• Transverse momentum
spectrum of leading jet– Contents of each bin scaled by
bin size
• Pythia Monte Carlo does not model leading jet pT spectrum well– Tuning PYTHIA Parton Shower
parameters can improve this
• Z2 Tune agrees more with data than D6T tune– Underlying event description not
optimal– Tunes developed based on
Tevatron data• Re-tuning based on LHC data
November 28, 2011
Ratio with Monte Carlo
Even
ts/G
eV
Christos Lazaridis, University of Wisconsin-Madison
Conclusions
November 28, 2011Unfolding
19
Backup slides
Christos Lazaridis, University of Wisconsin-Madison
Samples• December 22 reprocessed data
• Used only certified data • Corresponding to 36.1 pb-1
• Monte Carlo samples:
• Z+Jets• MadGraph, Tune Z2• MadGraph, Tune D6T• Pythia 6, Tune Z2
• Backgrounds:• W+Jets, Tune Z2 (MadGraph)
• ttbar + Jets, Tune Z2 (MadGraph)
• EM enriched QCD, Tune Z2 (Pythia)• BCtoE QCD, Tune Z2 (Pythia)
• Samples include PU corresponding to the latest 2010 collision runs
November 28, 2011Unfolding
21
MadGraph samples normalized by MCFM NLO
cross sections
systematicstudies
Christos Lazaridis, University of Wisconsin-Madison
V+Jets unfolding plots
November 28, 2011Unfolding22
Closure test Unfolding exclusive jet multiplicity