amnon harel aharel@fnal university of rochester

International Workshop on Top Quark PhysicsLa Biodola, Isola d’Elba, Italy

18-24 May, 2008

MC for top backgroundat the Tevatron

Amnon HarelAmnon [email protected]@fnal.gov

University of RochesterUniversity of Rochester

Thanks to: Kirsten Tollefson, Lisa Shabalina, Christopher Neu, Tom Junk, Ann Heinson, Jason Nielsen, Michael Begel, Ulrich Husemann, Kevin Lanon, Gustavo Garzon, Gerald Grenier, and apologies to those whose names I accidentally left out

Amnon HarelAmnon Harel 22Top2008Top2008 5/22/20085/22/2008

OutlineOutlineBackgrounds Matched MC

The cost of matching BTWs:

• multijets

• Detector simulation

Reweighting the MC

Measurements

ccbb


The backgrounds – top pairThe backgrounds – top pairHard to simulate

misreconstructions Data driven estimation

A.k.a.: Z+jets

Small A rough simulation suffices

Haven’t discovered that one yet

Dilepton channelDilepton channel

Lepton+jets channelLepton+jets channel


V+jets – top pairV+jets – top pair

Dilepton channelDilepton channel

Lepton+jets channelLepton+jets channel

• Z(ee/μμ)+jets - main background in ee and μμ channels, Z()+jets in eμ • Very important at the early stages of event selection• After final selection Z+jets contribution is small • Most measurements do not use b tagging

• Flavor composition usually not important

• This channel provides the most precise measurements • W+jets - main background • Very important at all stages of selection • Most of measurements use b-tagging• Knowledge of flavor composition is a limiting factor for precision measurements


The backgrounds – single topThe backgrounds – single top

BTW: WH Higgs search is also sensitive to W+jets H.F. fractions

CDF

H.F. fractions of W+jets crucial

W+light

Multijet


V+jets cross sectionsV+jets cross sections


CDF preliminary from Monday (1.9fb-1):

Harsh b-tagging and MET (>25GeV) selections

Jet definitions:

Jet selection:

Result:

Alpgen prediction: 0.78pb

, &

WW++bbbb measurements measurements

DØ preliminary from 2005 (382pb-1): pbWbbpp 64.GeVpbT 20jet 52.jet b 750. bbR ..%@ LC95 , , &

pb

lWBRbWppb

)syst(.)stat(..

jetsjets

420270742

GeVE jetT 20 02.jet

40.coneR 40. bjR

, &

What have we measured? Though conclusions from W+HF measurements not yet applied to top analyses


WW++cc measurements measurements

GeVp jetcT 20

52. jetc

DØ Data: Alpgen+Pythia:

jetsWpp

jetcWppR

GeVp jetcT 8

03. jetc

lWBRjetcWpp

Phys. Rev. Lett. 100, 091803 (2008)

arXiv: 0803.2259v1[hep-ex]

Recent MCFM calculations (John Campbell)for DØ cuts yield R(LO)=3.7, R(NLO)=3.4

CDF Data:

MCFM:


W+jets measurementW+jets measurementDedicated W+jets measurement. 0240 .,. jetjet yR

SMPR = CKKW matched Madgraph + PythiaMLM = MLM matched Alpgen + HerwigMCFM = NLO predictions

will be a recurring theme

....

41jetsWMC

data


• MLM matched (pT>15GeV) within each

class of events• cc / bb pairs within the same parton jet are taken only from the PS MC; those in different jets taken only from the ME MC

Generate 15 samples:• W+bb+Nlp, with N=0,1, or ≥2

• W+cc+Nlp, with N=0,1, or ≥2

• W+c+Nlp, with N=0,1,2, or ≥3

• W+Nlp, with N=0,1,2,3, or ≥4

• light partons jets are MLM matched• with pT>8GeV

• MLM stable to the chosen pT

• Discard events with additional heavy quarks from the PS MC

• done in post processing

Generate 14 samples:•W+bb+Nlp, with N=0,1,2, or ≥3•W+cc+Nlp, with N=0,1,2, or ≥3•W+Nlp, with N=0,1,2,3,4, or ≥5 (includes W+c+jets massless c quarks)

Matched MCMatched MCFor top physics, we usually simulate W+jets (also Z+jets) using MLM matched Alpgen. Implementations differ:

cc bb

cc


Alpgen qfac=ktfac=1Alpgen qfac=ktfac=0.5,could be a JES issue

ZZ+jets Simulation+jets SimulationZ+jets appears at a lower rate (~×10), but has much less background a good process for tuning the simulations.

NormalizationUsually it suffices to normalize according to cross sections predicted from MCFM or NNLO calculations.• Dependency on kinematic cuts!• Some analyses require normalization from data, e.g., CDF’s top FCNC search in Z+jets

KinematicsCan be tuned using dataExample: ResBos described DØ’s Z pT data well.

DØ is starting to use it as a surrogate to the data and to re-weight ALPGEN+PYTHIA to match the same spectrum. The same re-weighting is carried over to W+jets.

arXiv:/0712.0803


ZZ+jets Shape: Pythia & Sherpa - I+jets Shape: Pythia & Sherpa - I

Normalized to data without parametrizing in Njet

Pythia is a parton shower generator not enough radiationJES uncertainties dominate systematics (and are a bit conservative)

1950 fbLdt .

Sherpa uses CKKW matching

T


ZZ+jets Shape: Pythia & Sherpa - II+jets Shape: Pythia & Sherpa - II

1950 fbLdt .

Again Pythia spectrum is too soft


ZZ+jets Shape: Pythia & Sherpa - III+jets Shape: Pythia & Sherpa - III

1950 fbLdt .

Again Pythia spectrum is too soft


ZZ+jets Shape: Pythia & Sherpa - IV+jets Shape: Pythia & Sherpa - IV

1950 fbLdt .

We’ll return to the simulation of dijet angles…


ZZ+jets vs. NLO+jets vs. NLODedicated Z+jets measurement. 123070 .,,. jetjet

Tjet yGeVpR

Phys. Rev. Lett. 100, 102001 (2008)

MCFM hiding behind data

points


•multijets estimated from data samples with leptons that pass only a looser selection (mostly looser isolation)

• “loose-tight”

•several approaches, typically:• W+jets normalized to data before b tagging

• or fitted in signal samples• W+HF fractions normalized to data after b tagging

Normalizing W+jetsNormalizing W+jetsW+jets and W+HF are normalized to data, after other backgrounds (multijets, dibosons, etc.) are subtracted.

•multijets estimated from data samples with leptons that pass only a looser selection

• “anti-electrons”, ”non-isolated”

• Studies with dijet & multijet events• “Method 2” inherited from Run I:

• W+jets normalized to data before b tagging• W+HF fraction fitted to data after b tagging

Assumptions: Kbb=Kcc, and Kc=1 (consistent with studies)


W+jets Normalization @ CDFW+jets Normalization @ CDF•Three parameter fit to Bottom-Charm-Light templates of jet-flavor separating distributions (NN output, SecVtx mass) in W+2 jet data

•yields KHF=1.4±0.4 • relative to Alpgen H.F. fraction

•Light flavor yield with prediction from•per jet fake b tag rates (estimated from inclusive multijet data) •either W LF MC or data, pre b tag yields


First data driven normalization: single-top evidence analysissingle-top evidence analysisW+jets Normalization @ DØW+jets Normalization @ DØ

For W+jets:• Normalize to pretag data• Channel dependence?

• single-top vs. top-pair analysis• same in e+jets & μ+jets within uncertainties

• Njet dependence?

For W+HF:• Normalize to data after b-tagging

• in 0 b tag bin negligible signal• Defined relative to Alpgen H.F. fraction

PRL 98, 181802 (2007)


W+jets Normalization @ DØ - IIW+jets Normalization @ DØ - II

• Sensitive to extra backgrounds included

• Sensitive to selection cuts on jets – single top ||<3.4; ttbar ||<2.5

• Consistent between e/μ channels• Studied several cuts on b tag output

• Measured HF scale factor (uncertainty from all deviations observed in the studies):

KHF = 1.170.18

• Used for summer results

Procedure refined for later ttl+jets measurements• tighter selection cuts• normalize MC to match the fraction of 0 tag events in 1 and 2 jet bins

Studies of systematic uncertainties• Separated Wc for W+lp and varied

by 20% (not enough?)• Dependence on the multijet

background contribution• Tried to extract corrections to Wcc

and Wbb separately

Results• Switched from Alpgen 2.05 to 2.12

– Several bug fixes including one in Wcc / Wbb generation

– Compared shapes: the only noticeable difference is R(j,j)

– Cross sections differ by a factor of 2!

• New KHF factor ~1.9


Normalizing W+jets to data with the same jet multiplicity as signal is not trivial for BYSM searches, or even for cross section measurements.

Examples• searches for resonant top pair production

– Iterative procedure in CDF’s– Analytical work around in D0’s

• Charged Higgs search

W+jets Normalization for BYSMW+jets Normalization for BYSM


W+jets: Other ReweightingsW+jets: Other ReweightingsSome discrepancies between Alpgen and data are showing up with the increasing statistics.

Corrected for / treated as systematic uncertainties in recent measurements.

E.g. possible missimulation of the angular distributions of the jet with the 2nd highest pT

Plot from latest single top analysis

22 jjr


DØ experience with matchingDØ experience with matchingAt the time, Alpgen was the only matched MC that CDF & DØ could: run, integrated with their software, mass produce.It allows us to produce physics results!It allows us to produce physics results!

Large scale use + cutting edge technology = painOutlining the DØ experience to identify lessons.

• Must add together the parton-jet bins with the correct weights. Unfortunately, weights are sample dependent

• Must freeze samples• Post processing in DØ due to overlaid zero bias data & HF removal• Book keeping nightmare (ttbar, Z+jets in HF

& MZ* bins, systematic variations)Error prone

• Large relative weights+ necessary for multiple (additional) jets– but complicates statistics (some physics required)


DØ experience with matchingDØ experience with matchingUsing matched Alpgen extensively for the last couple of years:• We’ve made several mistakes, e.g.,

– Random seeds outside legal range– Imperfect HF removal

• Study Interplay with MC tunes• Some bugs found along the way, had to find workarounds

– Best case: need to look at the right plot in the right channel– Recent case: “why does that matching weight look odd?”

• Slow turn around times– Alpgen release DØ release Production (large 0lp samples, slow 5lp samples) Postproduction Analyses– 6-12 months

• Limits our ability to generate sufficient samples to study systematics


BTW1: MC in multijet modelingBTW1: MC in multijet modelingThough multijet modeling is data driven, it is (mostly) based on samples with 3 jets which are reconstructed as lepton + 2 jets.Can check the methodology using simulated multijets• Lepton ID cuts• Lepton ID efficiencies and their parametrization• MET (“triangle”) cuts• Sample composition

Example of bad agreement from recent DØ studies:

Work in progress – Example of failed model


BTW2: Detector simulationBTW2: Detector simulationAn ubiquitous problem, often taken for granted.Data based modeling for:• misreconstructed leptons, i.e., multijet background• b tagging rates• relative jet energy scale

This talk focused on the generator parts of simulation rather than on detector simulation.• Modeling of multijet background is covered in analysis presentations• the last two items will be covered in two talks about “tools for top” later today (CDF @ 15:30, D0 @ 16:00).


ConclusionsConclusionsSimulation of W+jets and Z+jets• is not trivial• can be analysis dependent:

• One size can almost fit all• Don’t always need the most sophisticated treatment• Several approaches to estimating the heavy flavor contribution• Can fit and/or normalize W+jets• New physics can complicate matters

Using matched Alpgen extensively for the last couple of years+ Able to meet all our physics needs!

• Higgs / Top pairs / single top / W+HF / Z+HF• Lepton+jets+MET / di-lepton final states (Z+jets important)

- possible inaccuracies (e.g. Z pT, 2nd jet ΔR)

- difficulties adapting to this technology- Technical lesson: avoid any post processing that can break the matching

Other generators seem promising- but have received much less scrutiny


Back up slides


To estimate PDF uncertainties we reweight our MC after the fact, rather than regenerate it. Is this compatible with MLM matching?Yes [private communications with Mangano]

We simulate additional collisions by overlaying “zero-bias” data (i.e. free of trigger biases) onto simulated collisions. Since instantaneous luminosities are still rising, we “update” the simulation by giving more weight to simulated events whose overlaid data event had a high inst. lumi.

29

Other weightsOther weights


30

Matching @ CDFMatching @ CDF


Calibrating the SimulationCalibrating the SimulationWe verify each aspect the simulation, mostly on appropriate data samples, and correct the simulation as needed.

• Jets• Relative energy scale • Energy resolution (ET smearing)• Reco. & ID Efficiencies

• b-tagging rates• Corrections for gluon splitting

• Instantaneous Luminosity• Primary Vertex Z-coordinate• Electrons

• Resolutions (ET smearing)• Reco. & ID efficiencies

• Muons• Resolutions (pT smearing)• Reco., ID and isolation efficiencies


In several double-top l+jets “property” analysesIn several double-top l+jets “property” analyses• W+HF normalized to data as in previous slides

• W+jets fit to data – topological likelihood separates top pair from W+jets

Normalizing to Data @ DØ Normalizing to Data @ DØ

Top PairBackgrounds (mostly W+jets)

Example from PRL 100, 062004 (2008)


Another example: Higgs search in HZHiggs search in HZbbX channelbbX channel• Heavy Flavor in (W/Z)+jet Alpgen+Pythia predictions were

multiplied by K-factors (NLO/LO) calculated with MCFM

• (W/Z)+jets normalized to data before b-tagging

Normalizing W+jets @ DØ HZNormalizing W+jets @ DØ HZ

After After bb-tagging-taggingBefore Before bb-tagging-tagging


TerminologyTerminologyDØ is trying to converge on terminology for normalization factors:• K-factor: normalizes LO to match (N)NLO• K’-factor: normalizes MC to match (N)NLO• S-factor: normalizes MC to match “pretag” data• SHF: normalized heavy-flavor MC to match b-tagged data

At DØ:For Z+lp and Z+HF: • Using K’-factors, normalize to theory• Some analyses (e.g. ) override with S-factors

For W+lp: Using S factors

For W+HF: Using S & SHF factors

Xeebbtt


Know-how IKnow-how I

• Generate 14 samples:– W+bb+Nlp, with N=0,1,2

or ≥3– W+cc+Nlp, with N=0,1,2 or

≥3– W+Nlp, with N=0,1,2,3,4

or ≥5 (includes W+c+jets)

• Individual samples can not be used any more!

Add all parton-jet bins together with weights Fi


Know-how IIKnow-how II• Sample has to be frozen• Large relative weights• Complicates statistics• Post processing:

– Data quality selection due to zero bias overlay– Discard events with additional heavy quarks created by Pythia

• Generate 14 samples:

– W+bb+Nlp, with N=0,1,2 or ≥3

– W+cc+Nlp, with N=0,1,2 or ≥3

– W+Nlp, with N=0,1,2,3,4 or ≥5

cc

cc bb

No skimming

Book keeping is a nightmare!Book keeping is a nightmare!


From generated MC to dataFrom generated MC to data

• Generate multi-parton MEs with Alpgen

• Add showering and hadronization from Pythia – b-fragmentation model – underlying event model

• Run full D0 detector simulation and reconstruction

• Add zero bias events to match luminosity profile in data

• Apply to simulated events: – JES– Jet removal– Smear jets, electrons and

muon – Propagate to missing ET

• Correction factors:– Trigger efficiency– Electron and muon ID

efficiency

• For b-tagging:– Taggability RF– Tag rate function

• Reweightings: lumi, z vertex, etc


Z+jets @ CDF - IZ+jets @ CDF - I

FCNC analysis used this for a systematic uncertainty.

negligible – doesn’t appear in the tables

Alpgen qfac=ktfac=1 Alpgen qfac=ktfac=0.5 Alpgen qfac=ktfac=2.0


MCFM Z+jet @ DØMCFM Z+jet @ DØAn example:

Results for 60-75 GEV


MCFM W+jets @ DØMCFM W+jets @ DØMethod• All calculations come from the MCFM author (John Campbell)• K-factor = sigma(NLO)/sigma(LO)• Parameters:

– PDF: CTEQ6L1(LO), CTEQ6M (NLO)– Factorization scale = Renormalization scale = MW

Conclusions• MCFM calculations show that K-factor for W+light is stable • K-factors for Wc, Wcc, Wbb decrease as jet pT increases • MCFM does not support any additional HF scale factor in

addition to W+light k-factor• The last conclusion contradicts our observation from data


DØ ZDØ Z+jets Normalizations+jets NormalizationsWhen estimating the W+jets background in Lepton+jets+MET channels, we predict Z+jets from the simulation + studies on di-lepton channels.

For Z+lp and Z+HF:• Typically weigh Alpgen+Pythia to MCFM / NNLO

• NNLO from Hamberg et.al., Nucl.Phys.B359 + Martin et.al. hep-ph/0308087• Agree to within 10%

• Recently calculate the K-factors with MCFM (is shaping needed?)• K-factors in the theoretical sense: NLO/LO• Cuts crucial in W+2HF (e.g. )• Evaluate effect of quark masses at LO: 10-40%

bbR ,


2nd jet angles @ 1 fb2nd jet angles @ 1 fb-1-1

Clearly there’s nothing to be excited about in this plot.Very preliminary analysis of later data shows similar trends.


DØ ZDØ Z+jets Shape Corrections+jets Shape Corrections

Starting to use ResBos as a surrogate to our data• Reweighting Alpgen+Pythia MC to fit the ResBos spectrum• Applying the lesson from Z+jets to W+jets…

Other approaches• Scale directly to data as a function of Njet

• Consider other event source• Measure & reweight as a function of:

•ZpT (reconstruction vs. particle level)• jet observables

arXiv:/0712.0803


WW++cc measurements measurements

GeVp jetcT 20

52. jetc

DØ Data: Alpgen+Pythia:

jetsWpp

jetcWppR

GeVp jetcT 8

03. jetc

lWBRjetcWpp

Phys. Rev. Lett. 100, 091803 (2008)

arXiv: 0803.2259v1[hep-ex]

Recent MCFM calculations (John Campbell)for DØ cuts yield R(LO)=3.7, R(NLO)=3.4

CDF Data:

MCFM:

(4:4§ 0:3(PDF ))% 9.8§ 2:8(stat:)+1:4¡ 1:6 (syst:) § 0:6(lumi:) pb

11:0+1:4¡ 3:0pb

(7:1§ 1:7)%


TitleTitle• Text

amnon harel aharel@fnal university of rochester

Documents