Pre-Approval

Pre-Approval

Marcella Bona, Emanuele Di Marco, Joseph D. Lykken,

Paolo Meridiani, Christopher Rogan, Chiara Rovelli, Ilaria Segoni, Maria Spiropulu, Thiago Tomei, Marco Zanetti

Authors:

Gautier Hamel de Monchenault (Chairperson), Jeffrey Berryhill, Cecilia Elena Gerber

ARC committee:

EWK-08-006 Pre-Approval, April 28 2

Z+jets candle analysis• Data-driven strategy to study properties of Z+jets

production in final states with di-electrons and di-muons– Focus on LHC start-up - order of data at– Use two detector-wise orthogonal jets definitions:

calo jets and track jets

• Analysis goals:– (i) Investigate whether over

yields ratio is constant as a function of – (ii) use track-jet counting, independently from calo-jet counting,

to increase available signal statistics – (iii) Select a validated, pure Z+jets sample that can be used for

detector commissioning at LHC start-up– (iv) Assuming (i) is positive, use jet multiplicity distribution as a

probe for NP, typically responsible for an excess of events at large values of


Background control sample

Event Reconstruction and Cut-Based selection

Maximum Likelihood Fit

Tests of the Fit sPlotsFit PDF

validation

Z boson “Candle”

Yield ratio

Over

MET characterization/

correctionsProbe of NP Background

estimation

Analysis Strategy


Analysis Strategy





validation


Yield ratio

Over



estimation

single, non-isolated lepton triggers

high efficiency selection to allow maximal signal yield and background modelling

Jet clustering (calo/track) consistent with event primary vertex


Event Selection• Trigger selection

– trigger menu– HLT single non-isolated muon and electron triggers with pT

thresholds and L1 thresholds and for electrons and muons, respectively

• Z boson reconstruction and primary vertex selection– Z candidates selected requiring – fraction of events with multiple Z candidates is found to be

negligible - in this case Z with highest pT leptons chosen– Reconstructed primary vertex closest to Z candidate vertex in z

(nominal beam axis, minimum ) is chosen as event primary vertex - in the absence of PU this corresponds to the highest reconstructed primary vertex ~100%

– Choice of event primary vertex is essential for lepton selection and jet clustering


Electron selection• Electron identification

– Use PixelMatchGsfElectrons– loose electron identification criteria

(see table)

• Vertex requirements– Consistency with event primary

vertex, requiring:

• Tracker isolation– consider tracks consistent with

electron vertex in a cone of R < 0.4 with a veto cone of R > 0.015

– RequireE/Gamma POG-developed ID


Muon selection• Muon identification

– Use GlobalMuons

• Vertex requirements– Consistency with event primary vertex, requiring:

• Tracker isolation– consider tracks consistent with event primary vertex in a cone

of R < 0.5 Require


Jet clustering• For Z + jets selection, everything is done as a function of

inclusive jet multiplicity

• We consider two type of jets (SISCone algorithm):– Calo-jets: jets clustered from the calorimeter (ECAL+HCAL) cells re-

projected w.r.t. event primary vertex

– Track-jets: jets clustered from tracks consistent with the event primary vertex

• These two types of jets – –

Requiring:

Requiring:

Have orthogonal detector systematicsProbe different regions of phase space


Selection summary


Analysis Strategy





validation


Yield ratio

Over



estimation

1-D extended and un-binned maximum likelihood fit of used to determine signal and background yields

Independent fits done for each jet multiplicity

Data-driven strategy for determining line-shapes, testing and validating the fits



total number of eventsentering the fit

(i.e. extended likelihood)Z+jets:

1dim fit: P=PDF(mll)

Ni=signal and backgrounds yields

• One-dimensional unbinned and extended maximum likelihood fit based on

• ML fits to Monte Carlo signal samples indicated that the fit parameters do not depend on the jet multiplicity within precision of target luminosity

• Signal PDF held fixed (assumed from inclusive Z study) - assumption can be later validated

• 2 species yields + 2 bkg shape parameters = 4 parameter fit


• Non-Gaussian effects in signal lepton invariant mass distribution result from– mismatched signal events, where one muon is from the

Z and the other coming from jets, muons from – one of the two leptons irradiating a photon, shifting the

measured Z mass lower

• Use Crujiff function to parameterize signal:



Analysis Strategy





validation


Yield ratio

Over



estimation

Background predominantly QCD

Use control sample from data to measure line-shape for fit


QCD background control sample• Background in both electron and muon final states is dominated by QCD contribution• Shape of background PDF’s is studied using “anti-lepton” sample, obtained by inverting tracker-

isolation cut:

• Other backgrounds accounted for by floating shape parameters (distribution well-described by second-order polynomial)


Analysis Strategy





validation


Yield ratio

Over



estimation

Toy Monte Carlo experiments used the expected statistical error on the signal yields

Check for fit biases

Test confidence interval coverage


Tests of the Fit• Toy Monte Carlo experiments indicate and expected error on signal event

yields of 2% for (16% for ) • Similar errors for

• Larger statistics with track-jet counting improves precision by at least a factor of 2

• Toy MC experiments also prove ML fit is unbiased and that 68% confidence interval comuted using likelihood ratio correctly covers true number of events

• See extra slides for details


Fit Results


Fit Results


Fit Results


Fit Results


Analysis Strategy





validation


Yield ratio

Over



estimation

Test a posteriori the validity of assumed PDF’s from ML fit using sPlots technique


sPlots ML Fit Validation• The ML fit is extended to an additional variable that is uncorrelated with the dilepton

invariant mass and also discriminates between signal and background• We use the variable , which satisfies these requirements (see

backup slide)

• Allows for a posteriori validation of Z line-shape PDF using sPlots


sPlots ML Fit Validation• 1D ML fit performed with each variable to check other • the output of the fit is used to compute sWeights • the sWeights-weighted plot of the variable taken out of the fit

produces the distribution for signal events. Each event contributes to the plot proportionally to its probability of being signal

• By comparing the distribution with PDF obtained in the nominal fit we can use the other variable(s) in the fit to test our assumption on the functional form of the PDF


sPlots ML Fit ValidationNOTE: these are NOT fits


Analysis Strategy





validation


Yield ratio

Over



estimation

Using tested and validated ML fit we can study event yields as a function of inclusive jet multiplicity


over yields ratioS.D. Ellis et al., “W’s, Z’s AND

JETS”, Phys. Lett. 154B (1985) 435

From: From:CMS AN2008-091

Alpgen (CSA07) generator level study of Z+jets events

Berends et al.:


over yields ratio


over yields ratioWith candle selection:


Fit probabilities between 75% and 94%

over yields ratio


Analysis Strategy





validation


Yield ratio

Over



estimation

sPlots framework also allows for statistical background subtraction without a decrease in signal statistics

Can create ‘pure’ Z(ll)+jets dataset to use for various applications


sPlots background subtraction• An interesting feature of the sPlots is the fact that the integral

of the weighted plot of a signal distribution is equal to the number of signal events found in the ML fit

• This means that the sPlots can be used to perform the subtraction of the background component without losing statistics on signal

• Z+jets dataset weighted with computed sWeights from nominal 1-D fit corresponds to a pure, validated signal sample


MET characterization with sPlots

• Can use pure Z()+jets dataset to study calorimetric detector response and MET measurement

• With only MIP calorimetric deposits, Z() pT can be used as ‘generator level’ MET

• Allows for characterization of detector response and validation/tuning of full simulation calorimetry


MET correction with sPlots

• Using similarities between W/Z + jets topologies, we can use events to calibrate MET for with event-by-event correction

• Take a ‘W-like’ view of Z+jets events, treating one muon leg as an unobserved neutrino

• Decompose MET into two orthogonal components:

• Calibrate each component separately


NP probe with Z candle We can observe new physics in beyond the standard model scenarios that yield an excess of Z’s

Assuming to be sufficiently constant, we can use the lower jet multiplicities to predict the rates in the higher multiplicities and measure any deviations

• We consider LM4 as a benchmark scenario where Z’s are produced in the decays of neutralinos

• We perform a set of Toy MC experiments where the SM signal and background are generated in addition to LM4 events


NP probe with Z candle


background estimation for NP searches


Conclusions• We present a data-driven strategy to study properties of

Z+jets production in final states with di-electrons and di-muons that, with at

achieves the following goals:– (i) Investigate whether over

yields ratio is constant as a function of – (ii) use track-jet counting, independently from calo-jet counting,

to increase available signal statistics – (iii) Select a validated, pure Z+jets sample that can be used for

detector commissioning at LHC start-up– (iv) Assuming (i) is positive, use jet multiplicity distribution as a

probe for NP, typically responsible for an excess of events at large values of


We would like to add the MET characterization example (one paragraph) from the jet-dimuon Pt balance study with the sPlot subtraction so Figure 9 would like this:


EXTRA SLIDES


Datasets

Fall08

samples

Summer08

samples

CMSSW_2_1_X

Full Simulation

“Ideal” conditions


MET correction with sPlots• Consider an orthogonal decomposition of MET

• Use Z candle to measure same components with Z, using Z candidate and leading (2nd leading) muon

• Derive correction factors

• bin correction factors in and / , fit distributions to get

• Correct events

such that:

Can use candle sWeighted sample to calibrate MET




After Z+jets candle derived corrections, the characteristic Jacobian edge in the W MT distribution is recovered



“MHT” Variable

◉ references for 16X analyses:◎ CMS notes◉ Z+jets & W+jets Alpgen Validation: CMS AN-2008/091◉ Z(ll)+jets Candle Analysis: CMS AN-2008/092, CMS AN-2008/095◉ W(l)/Z(ll)+jets Ratio Analysis: CMS AN-2008/096, CMS AN-2008/105◎ previous EWK talks on 16X analysis:◉ summary by M.Pierini Nov 11th 2008◉ note release: E. Di Marco, C. Rogan, and I. Segoni: Oct 17th 2008

◉ references for 21X analyses:◎ CMS PAS◉ EWK-08-006◉ https://hypernews.cern.ch/HyperNews/CMS/get/EWK-08-006.html◎ CMS notes◉ W(l)+jets/Z(ll)+jets Ratio Analysis: CMS AN-2009/045◉ Z(ll)+jets Candle Analysis: CMS AN-2009/xxx◎ previous EWK talks on 21X analysis:◉ PAS candle status: M.Bona, Mar 3rd 2009◉ analysis updates: M.Pierini, Feb 6th 2009, M.Bona and I.Segoni, Jan 23rd 2009 C.Rogan, Jan 20th 2009

◉ general strategy◎ single non isolated leptonic trigger◎ Z mass window: [60, 110] GeV/c2

◎ lepton pt cut:◍ first leg -> higher than the trigger threshold: pt > 20 GeV/c◍ looser second leg: pt > 10 GeV/c◎ lepton selection: optimized for the W+jets ratio analysis and then loosen it to allow the highest efficiency for the standard candle ◎ jet clustering (SisCone with 0.5)◍ calorimeter jets: E > 30 GeV and || < 3◍ track jet: pt > 15 GeV/c and || < 2.4◎ 1D maximum likelihood fit: ◍ Z mass with two species: signal+background◎ fixing the signal shape parameters from the inclusive study and floating the background parameters◎ 2D fit and “sPlots” to verify shapes on data: both for signal and background

Vector-boson + jets: strategy

◉ lepton selection:◎ global muons or optimized electron ID◎ analysis specific selection:◉ vertex variables: xy z(-PV)◉ isolation variables: i pi

T/pT, ET

ECAL ETHCAL

◎ optimized simultaneously for the W+Njets final states where the background rejection is more critical◎ for the candle we want the highest efficiency possible◍ maintaining reliability on the sample for example: being able to calculate efficiencies if needed◍ establishing a fit procedure that is robust against background, taking advantage of the abundance of the Z signal and the separation power of the Z mass

Vector-boson + jets: lepton selection

◍xy <0.04 cm z(-PV) <0.12 cm◍ i pi

T/pT <0.15

◎ xy <0.02 cm z(-PV) <0.15 cm◎ i pi

T/pT <0.30

Z(ee) candle Z() candle

Emanuele's talk, Jan 16th 2009EWK electron meeting

Z(ee) candle final sample composition

Vector-boson + jets: electron final state

calojets trackjets

Z(ee) candle signal selection efficiencies

Z() candle final sample composition

Vector-boson + jets: muon final state

Z() candle signal selection efficiencies

Vector-boson + jets: muon event sample

Z() candle final sample composition

calojets trackjets

Christopher Rogan - VecBos + jets 15-01-09

51

◉ 1D maximum likelihood fit with the dilepton invariant mass◎ the mll shape is fixed for signal and taken from the inclusive Z study on data◎ the mll shape parameters are floated for the background

Vector-boson + jets: the fit variable

m

signal parameterizations:Gaussian function with asymmetricwidths and non-Gaussian tails

≥ 1jet

background parameterizations:second order polynomial

m ≥ 1jet


52

Vector-boson + jets: signal parameterization

m

≥ 1jet

m

≥ 2jet

m

≥ 3jet

mee

mee

mee

≥ 1jet

≥ 2jet

≥ 3jet

signal only fit

electrons muons

signal only fit


53

Vector-boson + jets: signal parameterization (II)

calojets trackjets

calojets trackjets

electrons

muons


54

Vector-boson + jets: background parameterization

≥ 1jet

≥ 2jet

≥ 3jet

m

m

m

≥ 1jet

≥ 2jet

mee

mee

background only fit

ttbar only fit

electrons

muons


55

Vector-boson + jets: background parameterization (II)

calojetselectrons

calojets

trackjets

muons


56

Vector-boson + jets: toy Monte Carlosignal pull

signal error

≥ 1jet ≥ 2jet

≥ 3jet≥ 1jet ≥ 2jet

≥ 3jet

Z(ee) candlewith calo-jets


57


signal error

≥ 1jet ≥ 2jet


≥ 3jet

Z(ee) candlewith track-jets


58


signal error

≥ 1jet ≥ 2jet


≥ 3jet

Z() candlewith calo-jets


59


signal error

≥ 1jet ≥ 2jet


≥ 3jet

Z() candlewith track-jets


60

◎ signal box cut:◍ Z mass [60,110] GeV/c2◍ second variable: angular variable |sin(MET, MHT)| < 0.85◎ for the 2-dimensional fit:◍ 1D parameterizations of the two variables for signal and background◍ Monte Carlo toys to verify the fit procedure

Vector-boson + jets: 2-dimensional fit

m

≥ 1jet≥ 1jet

sin(MET,MHT)

Z othersW+jetsttbarQCD

Z() candle

◎ the mll shape is fixed for signal and taken from the inclusive Z study on data◎ the angular variable shape parameters are floated for the signal for low jet multiplicities and fixed for the high multiplicities from the low-jet bins


61

sin(MET,MHT) ≥ 1jet

Z() signal parameterization

≥ 2jet

≥ 3jet

sin(MET,MHT)

sin(MET,MHT)

Vector-boson + jets: the second variable

≥ 1jet

≥ 2jet

≥ 3jet

sin(MET,MHT)

background parameterizationsin(MET,MHT)

sin(MET,MHT)

Z() candle


62

Vector-boson + jets: the second variable (II)

calojets trackjets

background parameterizationcalojets trackjets

Z() signal parameterization

Z() candle


63

W+jetsttbarQCD

QCD from data with the “anti-muon” sample: inverted track isolation cut

m

≥ 1jet

Vector-boson + jets: backgroundbackground composition andbackground shapes for the fit:

Z() candle

m

QCDanti-muon

sin(MET,MHT)

≥ 1jet

sin(MET,MHT)


Q&A

HN posting April 9


Conveners: Z + Jets (version 18)We are happy with the new 1-D fit using only Mll variable. The figures 2-5 are all show clean extraction of signal over the background. We continue to be not so happy with the variable sin(phi_MET,MHT). The test described in lines 126-157 is not addressing our concerns of the use of the allied variable in the W+Jets analysis.Response The physics, derivation and explanation of the variables is documented by Chris Rogan and attached at the end of the responses. On the strategy and use of the angular variables: In the most recent version of the fit for Z+jets (described in v18 of the PAS) the variable sin(MET-MHT) is no not used in the Z fit. The fit is now a 1-dim fit based on the dilepton invariant mass. Regarding the line-shape of the Z, since the pT spectrum of the boson does not depend strongly on the jet multiplicity, we don't expect to be sensitive to the dependence of the Z line-shape on jet multiplicity, within the precision from the statistics in a 100 pb-1 . We have confirmed this by both CSA07 ALPGEN and Fall08 MADGRAPH Monte Carlo productions (we note that the detector effects are different between 16x and 21x: Geant4, particle list, material, miscalibration/misalignment, thresholds etc). We assume that we can fix the line-shape to what is seen in a fully inclusive analysis (with the same pT selection on the leptons). This assumption would induce a systematic error which we can quantify using sPlots distributions to check the agreement between the parameterization and the data and to look for alternative parameterizations of the mll.For the main background, namely QCD, we use the “anti-lepton” control sample to determine the functional form (2nd order polynomial vs 1st order polynomial vs Gaussian tail vs exponential tail vs ....). The parameters are floated in the fit (i.e. the systematic error associated to the knowledge of the parameters is folded in the statistical error), so we are not requiring the MET to be understood at the level of being able to predict this distribution from the MC . This is described in the PAS, and you say you are happy with it. So, the second variable is not used to extract the physics results in the Z+jets candle PAS, i.e. the ratio Z+(n+1) jet/Z+ n j is measured with mll only.


Conveners : We remain concerned that there is no punch line in this PAS note.

Response We have fixed the presentation of the results and the punch line in the recent version. The introduction refers to three main points:1) Use the data to test the linearity of the dN/dn je (regardless of the value of the slope and regardless of the jet definition)2) Use this sample to commission the detector with physics3) Use the linearity to predict the number of events in >=3j and >=4j, since the SM predicts at most a small negative deviation from linearity, while NP can produce a large positive deviation (calling this a mild evidence of NP is like calling sin2b(b->s) = -0.5 +/- 0.1 or lepton universality breaking in Z decays a mild deviation of new physics)All these three points are covered individually after the fit is shown (sec 4, sec 5, and the LM4 plot in the conclusion). We cannot cover all the NP models with real Z’s, but we give a practical illustrative example of what will happen if NP events that contain real Z candidates are there. We made this more clear and explicit now by separating the NP discussion. We added the sPlots of Zμμ and Zνν MET prediction for Zνν as a further proof. All these plots have been shown in JetMET and SUSY PAGs . We received good feedback from both groups.


Conveners: Section 4 is too terse, and does not seem appropriate final result from the paper.

Response

In this section we demonstrate the possibility of using Z+jets events to commission MET for physics events characterized by real MET, exploiting the facts that the MET range covered is the same as for W+jets and ttbar+jets and has substantial overlap with the MET range of SUSY searches, and considering that Z+jets is the only channel allowing a crystal clean MC-truth-like determination of the MET from orthogonal parts of the detectors. This is indeed one of the three main results of this note (which stays within the EW group but whose interest goes beyond). If the language in this section is too terse or unclear we can try improve it.

•One additional plot in the section of MET commissioning with 100 pb-1 has been added showing the MET characterization using the candle.


Conveners: In section 6, lines 211-215 and figure 10, we have not even understood what is being plotted.

Response: We changed the text to better explain this plot. This is what we did:1) generated a set of toy MC experiment with SM + N SUSY events (we use LM4 since it includes real Z) 2) Fit the >=1, >=2, >=3, and >=4jets samples according to our fit model namely: -i) Z signal mll distribution assumed as input -ii) bkg distribution floated -iii) yields floated 3) determine with the toys the expected error on the four signal yields 4) predict N(Z+>=3j) and N(Z+>=4j) from N(Z+>=1j) and N(Z+>=2j) N(Z+>=3j) = N(Z+>=2j)2/N(Z+>=1j) N(Z+>=4j) = N(Z+>=2j)3/N(Z+>=1j)2 5) plot the predictions and the measurements as a function of the number of NP events (with real Z and “fake” Z from leptonic combinatoric) in the Z+>=1j sample (100 pb-1 taken as normalization)N.B.1) not all the NP events have Z (the fraction depends on the features of LM4) 2) the NP events not only produce an excess at 3 and 4, but also bias the predictions from 1 and 2. This is seen by the fact that the prediction goes up as a function of N(NP) 3) the observed excess is larger (a O(1%) correction to N(>=1j) is an O(1) correction to N(>=4j) 4) the errors are realistic estimates of 100 pb-1. i.e. there is a real potential for observing a discrepancy 5) this does not replace the dedicate searches. It just demonstrates how --NP can be probed w/o MET requirements in this case -- in the presence of NP, we are not going to have a biased (signal polluted) prediction of Z(nn) because we will see its sPlots allow also to characterize the excess (e.g. an sPlot of MET-ZpT could reveal the presence of the LSP in the >=3 and >=4 jet bins). You can also see how important is not to tune the thresholds to make track-jets equivalent to calo-jets. The excess will be different (in size) but correlated. And track-jets are more sensitive to NP because they allow lower pT threshold -> larger statistics -> better precision


Conveners: We propose the following route forward with Z+Jets portion of the analysis: Concentrate on Z+Jets and develop a robust result. For instance, the linearity and slope measurement can be made robust by cross calibrating the calo-jets and track-jets, showing that above a certain cutoff the two results agree. You can probably show that track-jets can be used below the cutoff with some degree of uncertainty that can be quantified using the MC information. Response: We have provide the plots of “equivalence” you request but for the purpose of the analysis we propose to keep the powerful handle for detector commissioning with physics and the improved statistics that the track-jets offer. By making the calo-jets and track-jets identical (in terms of genjet definition they correspond to) the robustness in terms of the orthogonality of detector effects is retained. The robustness vs the theoretical effects including a deviation from linearity is gone. These effects are related to accessible phase space, hardness of the emission and many other effects. We have the feeling that this big issue that we discussed repeatedly in the HN about the slopes being different, is in fact our fault, in that we quoted the actual numerical value of the slopes. We fixed this in the current version of the PAS (to be posted asap), replacing it with the χ2 probability of the fit (as returned by ROOT). Τhe values are between 75% and 94%, a proof of the successful test of linearity. We thus remove any perceived unintended emphasis on the numerical value of the slopes that has possibly caused a confusion.


Conveners: You can then go on to show that with new physics there will be deviation from linear fall, and quantify it at appropriate level.

Response:

This is what section 6 and fig.10 explains, see above. This has been moved to a dedicated section in the new version of the draft and we consider this already answered. We made the caption more clear. Maybe this generated your misunderstanding.

Conveners: Since the sin-variable and the splots are not needed for anything they can be simply be removed.

Response:

We have to figure out how to present this. The validation section of the "Z as candle" is important for us. And it also allow to introduce the sPlots as a tool for additional studies. Given the additional delay we have added the example of the candle Zνν plots (the update/improvement to the PTDR similar candle plots). The sPlot debate has already happened between Belle and Babar (and for many years) so we don’t feel it is necessary to repeat it but we can work with you to figure out presentationally and substance-wise how to incorporate them.


Conveners: An alternative data driven estimate of the remaining (dominantly QCD) background using isolation or some other variable will be useful

Response:

We did this test. Results from the “anti-leptons” samples were already in the PAS for muons (Fig 1).

Documents

Pre-Approval