1

Un-ki YangUniversity of Chicago

Joint CDF/D0 Top Mass Workshop, FNAL, Oct. 11, 2005

Modeling Uncertainties on Mtop(CDF summary)

CDF

2

Signal modeling

Hard Scattering

PDFs, scales, qq vs gg, LO vs NLO

))/(((),(),( 21 Λ→⊗⊗ uttqquxfuxf sασ)

3

Signal modeling

Hard Scattering

),(),/()/( uxfuyxPu qgqs ⊗Λ →α

PDFs, scales, qq vs ggLO vs NLO

ISR/FSR showeringb

b

))/(((),(),( 21 Λ→⊗⊗ uttqquxfuxf sασ)

Hadronization & jets

PDFs, scales, extra jets, modified b & W daughter jets due to FSR

Jet response (light, b-jets)See Florencia & Pekka’s talks

4

Signal modeling

Jet response(not in this talk)

Hadronization(fragmentation)

ISR & FSRhard-scatteringQ2 Scale(fac/ren)

ISR & FSRParton-showering

extra hard jets

ISR showering

hard-scatteringNLO

hard-scatteringPDFs

EffectsSources

5

Top Signal MCs

NoYesYesNOSpin corr

cluster model

LUND string

Hadronization

Beam-beam remnants.

MPI(Jimmy)

beam-beam

remnants, MPI

Underlying evts

coherent branching

LO tt

HERWIG

Pythiaor Herwiginterface(double counting

problem, can be fixed by the

CKKW)

LO tt+njet

ME

Herwigv.6.505

coherent branching

PS shower (ISR/FSR)

NLO ttLO ttHard scatt.

MC@NLOPYTHIA

Default: HERWIG 6.505 LOSyst.: PYTHIA, MC@NLO, ALPGEN

6

CDF Methodology

ISR & FSRhard-scatteringQ2 Scale(fac/ren)

ISR & FSRParton-showering(ISR/FSR)

extra hard jets

ISR showering

hard-scatteringNLO

hard-scatteringPDFs

EffectsSources

Data driven ISR/FSR syst.: ISR/FSR shower algorithm itself , but also different ISR/FSR due to PDFs and scale, a part of the NLO effect

7

CDF Methodology

ISR & FSRhard-scatteringQ2 Scale(fac/ren)

ISR & FSRParton-showering(ISR/FSR)

extra hard jets

ISR showering

hard-scatteringNLO

hard-scatteringPDFs

EffectsSources

Data driven ISR/FSR syst.: ISR/FSR shower algorithm itself , but also different ISR/FSR due to PDFs and scale, a part of the NLO effect PDFs: reweighting method at hard-scattering level

8

CDF Methodology

ISR & FSRhard-scatteringQ2 Scale(fac/ren)

ISR & FSRParton-showering(ISR/FSR)

extra hard jets

ISR showering

hard-scatteringNLO

hard-scatteringPDFs

EffectsSources

Data driven ISR/FSR syst.: ISR/FSR shower algorithm itself , but also different ISR/FSR due to PDFs and scale, a part of the NLO effect PDFs: reweighting method at hard-scattering levelNLO: MC@NLO vs HERWIG LOGenerator difference (including hadronization) HERWIG vs PYTHIA

9

Summary on syst. errors

MEMETemplateDileptonLepton+JetsSystematic

1.3 (2.8)0.50.30.10.50.20.30.60.40.6

(2.5 )

3.20.2

0.20.60.30.50.50.40.63.0

1.5(bkgd)MC statistics

0.5b-jet modeling

3.6Total0.6Method

b-tagging0.8Bkgd shape1.0Generators1.1PDFs0.5FSR0.5ISR

2.5JES

No additional syst. due to the NLO effectTemplate vs ME in lepton+jets: very similar

10

Side effect of the imperfect ME calculation(Matrix Element Method)

Imperfect ME calculations

• No initial state radiation (pt(ttbar)=0)

• Final state: lepton & jet directions are well measured

• No extra FSR jets (lepton+jet)

• Jets are b-jets (dilepton)

Final error is scaled up by 1.4 (dilepton ME)

Pull width: dilepton ME

11

New approach• ISR is governed by DGLAP eq.: Q2, ΛQCD,

splitting functions, PDFs• Use DY data (no FSR): study Pt of the dilepton, Njets

for different mass regions (~different Q2 )

µ+

µ-

qq -> tt(85%)

How to control ISR in ttbar system?ISR syst. in tt system?

12

<Pt (dilepton)> as a function of M2(ll)

A good logarithmic Q2 (~M2) dependence is observed.Conservative ISR syst. (plus/minus) for ttbar are established• <Pt(dilepton)>: sensitive to only total size of ISR, but missing Njet• Extrapolate to top pair production region using LO MC

13

ISR uncertainty

Q2max: K PARP(67)

Qmin : PARP(62)

Kt2 = PARP(64)(1-z)Q2 : αs,PDFΛQCD = PARP (61) :αs

4.0(Tune-A)

PARP(67)(D=1.0)

4.00.25PARP(64)(D=1.0)

0.0730.292(5 flavour)

PARP(61) (D=0.146)

ISR lessISR more Pythia

14

FSR syst. is based on the ISR syst. studies• Both ISR and FSR have same parton shower process (DGLAP),

except that the PDF evolution is only involved in ISR showering.

How to control ISR in ttbar system?FSR syst. in tt system

2.08PARP(71)D=4

0.0730.292(5fl: LO)

PARP(72)

FSR minus(Top std)

FSR plus (Top std)

Pythia

Default FSR syst:No change in color singlet

(including resonance decay)Constrained by LEP data

0.0730.292PARP(72)

2.08PARP(71)

0.1450.580(4fl: LO)

PARJ(81) D=0.290

FSR minus(Top cntrl)

FSR plus (Top cntrl)

Pythia

Even allow variationin color singlet(conservative)

15

NLO effect on top pair production

Use MC@NLO: Parton shower MC (like PYTHIA, HERWIG), but the partonic hard processes are calculated up to the full NLO QCD corr.– Only sensible way to compute K factor event by event.– Pythia and HERWIG LO MC are poor for multi-jets– ME (ALPGEN, MadGraph) are fine for multi-jets, once double counting

problem is removed using CKKW or MLM, but still LO accuracy (K factor still necessary)

Size of the extra hard gluon emission due to NLO?- How much is the NLO effect covered by the ISR/FSR syst?

Different production rate for qq vs gg, but gg has a higher acceptance (more ISR jets)

- CTEQ5 LO (5.5% gg fraction) : - CTEQ5 NLO (14.3% gg)

16

A brief summary on the comparison of MC@NLO and HERWIG

MC@NLO (Mtop=175 GeV, no spin-corr) vs HERWIG (spin-corr on)NLO vs ISR/FSR syst.? – Pt(ttbar), Njets(loose): most of effects seem to be covered by the

ISR/FSR syst. Comparison at HEPG level– NLO effect at the Tevatron is small.– Pt(ttbar): good agreement, but harder at Pt~100 GeV by 10%– Njets (loose jets): increased by 10% at Njet~7.– Pt(top): good agreement except very low Pt and high Pt regions– Pt of W,b and leading jet Et: all good agreement– More tops from gg channel are produced in plug region

17

Pt of tt : NLO vs ISR syst.

Pt(ttbar) for 1tagT+2tag

18

Njets (Et>12 GeV at HEPG particles)NLO vs ISR syst.

19

Njets (Et>12 GeV ~ 8 GeV at L4)NLO vs ISR syst.

Njets for 1tagT+2tag:Et(L4)>8 GeV

20

Pt of b-parton and W

21

How can we improve the ISR/FSR/NLO syst.?

Update on the ISR studies– More DY data are added. (total 340pb-1)– Njet, δφ(ll) distributions are included: good agreement with Pythia

and data– Q2 dependences on Njet, δφ(ll) are under studies.– Low-pt DY data need to be added

Future improvement– Comparison of the ttbar data (with 1fb-1 data) with various MCs

( MC@NLO, CKKW ALPGEN, MadGraph ttbar + Njets )– Especially helpful in pinning down FSR syst.– FSR uncertainty (outside cone) is a double counting with another

jet syst. (out-of-cone correction)– Study with Pythia 6.3

22

Njets Distribution

Pythia 6.2

Dimu data

4 < ET < 15 GeV(Soft Jets)

Entries 355

Mean 1.034

RMS 1.074

jetsN

0 1 2 3 4 5 6 7 8 9 100

20

40

60

80

100

120

140

Entries 355

Mean 1.034

RMS 1.074

Entries 5015

Mean 1.108

RMS 1.131

40 < M < 76 ( low )

Entries 355

Mean 1.034

RMS 1.074

Entries 355

Mean 1.034

RMS 1.074

Entries 5015

Mean 1.108

RMS 1.131

jetsN

0 1 2 3 4 5 6 7 8 9 100

200

400

600

800

1000

1200

1400 Entries 52957

Mean 1.177

RMS 1.178

Entries 3807

Mean 1.199

RMS 1.272

Entries 3807

Mean 1.199

RMS 1.272

76 < M < 106 ( Z )

Entries 3807

Mean 1.199

RMS 1.272

Entries 3807

Mean 1.199

RMS 1.272

Entries 52957

Mean 1.177

RMS 1.178Pythia

CDF IIEntries 124

Mean 1.032

RMS 1.07

jetsN

0 1 2 3 4 5 6 7 8 9 100

10

20

30

40

50

Entries 124

Mean 1.032

RMS 1.07

Entries 1647

Mean 1.213

RMS 1.231

106 < M < 200 ( high )

Entries 124

Mean 1.032

RMS 1.07

Entries 124

Mean 1.032

RMS 1.07

Entries 1647

Mean 1.213

RMS 1.231

40<M<76 (low) 76<M<106 (Z) 106<M<200 (high)

ET > 15 GeV(Hard Jets)

Entries 355

Mean 0.1549

RMS 0.4195

jetsN

0 1 2 3 4 5 6 7 8 9 100

50

100

150

200

250

300

Entries 355

Mean 0.1549

RMS 0.4195

Entries 5015

Mean 0.1567

RMS 0.408

40 < M < 76 ( low )

Entries 355

Mean 0.1549

RMS 0.4195

Entries 355

Mean 0.1549

RMS 0.4195

Entries 5015

Mean 0.1567

RMS 0.408

Entries 3807

Mean 0.1579

RMS 0.4282

jetsN

0 1 2 3 4 5 6 7 8 9 100

500

1000

1500

2000

2500

3000

3500

Entries 3807

Mean 0.1579

RMS 0.4282

Entries 52957

Mean 0.1711

RMS 0.4297

76 < M < 106 ( Z )

Entries 3807

Mean 0.1579

RMS 0.4282

Entries 3807

Mean 0.1579

RMS 0.4282

Entries 52957

Mean 0.1711

RMS 0.4297Pythia

CDF II

jetsN

0 1 2 3 4 5 6 7 8 9 100

20

40

60

80

100Entries 1647

Mean 0.1779

RMS 0.4246

Entries 124

Mean 0.2823

RMS 0.5471

Entries 124

Mean 0.2823

RMS 0.5471

106 < M < 200 ( high )

Entries 124

Mean 0.2823

RMS 0.5471

Entries 124

Mean 0.2823

RMS 0.5471

Entries 1647

Mean 0.1779

RMS 0.4246

40<M<76 (low) 76<M<106 (Z) 106<M<200 (high)

Updated studies by Sasha Rahlin (Chicago REU student)

23

∆φ(µµ) Distribution

Pythia 6.2

dimu data

φ∆0 0.5 1 1.5 2 2.5 3

-1

1

10

210Entries 6360Mean 2.882RMS 0.327Entries 458Mean 2.858RMS 0.3792

Entries 458Mean 2.858RMS 0.3792

40 < M < 76

Pythia

CDF II

of Dimuonφ∆

φ∆0 0.5 1 1.5 2 2.5 3

-1

1

10

210

310Entries 65110Mean 2.962RMS 0.2375Entries 4590Mean 2.955RMS 0.2379

Entries 4590Mean 2.955RMS 0.2379

76 < M < 106

φ∆0 0.5 1 1.5 2 2.5 3

-1

1

10

210 Entries 1990Mean 2.99RMS 0.2267Entries 159Mean 2.975RMS 0.2

Entries 159Mean 2.975RMS 0.2

106 < M < 200

Updated studies by Sasha Rahlin (Chicago REU student)

24

PDFs systematic

-0.08+-0.03MRST72-CTEQ5L

+0.20+-0.0920 EigenCTEQ6M

+0.22+-0.03MRST75-MRSR72

∆Mtop

(lepton+jet)Syst.

)/()()( 2211 ttggqqxfxf →⊗⊗ σ

Reweighting method: reweight the hard-scattering cross sections with new PDFs (f1, f2)

– Lepton+jet: ∆Mtop = 0.3 GeV (Template)– Dilepton: ∆Mtop = 1.1 GeV (ME): due to a limited CPU power

25

• W+heavy flavor jets(bb,cc,c)– HF fraction from ALPGEN MC– Normalized to data

• W+jets(mistag)– Use measured mistag rate,

applied to the data• Multijet:fake-W (jet->e, track->µ)

– Estimated from data• Single top, dibson (WW,WZ)

– Estimated from MC

Backgrounds (template: lepton+jets)

160.7+-0.2

2tag

2510.2+-1.7

1tagLno est7.6+-1.2bkgds

4057Data

1tagT 0tag

>=1-btag

controlsignal

26

Background Shapes

Lepton+jets: major uncertainties by W+HF & W+4P modeling by ALPGEN – For W+HF/W+4P(mistag): replace entire

background shape by individual major background, take ½ of largest difference as a syst. = 0.3 GeV

– In addition, shapes due to diff. Q2 scales (4Mw2, Mw2, ¼Mw2, Mw2 + PtW2) is 0.4 GeV – ½ the largest difference

– Non-W by W+4P(mistag) or non-iso data or multi-jet fake electron data: 0.1 GeV

Dileptons major uncertainties by DY & Fake– Drell-Yann: 0.4+-0.6 GeV (ME)– Fake: 0.7 GeV (ME)

More data and tunning Q2 scale– Q2=a*Mw2 + b*Pt_w2

Non-W

27

Summary on syst. errorsDilepton

MELepton+jetsTemplate

Systematic

1.3 (2.8)…0.50.20.30.60.40.6

(2.5 )0.5b-jet modeling

3.6Total…….0.8Bkgd shape1.0Generators1.1PDFs0.5FSR0.5ISR

2.5JES

Improves with more CPU power

Improves with more data and smarter algorithm

Not much…

28

Things to do

Signal modelingBetter constraints on ISR/FSR syst.Tasks with MC@NLO

• More validations on MC@NLO MC• Extra jets:- Extra jets from MC@NLO (mainly ISR) are covered

by the ISR syst.- FSR jets from b-quark or light quark (from W): this is not

handled by MC@NLO but by Herwigh PS: need to check with CKKW ALPGEN or MadGraph

• Q2 scale dependence? (Q=0.5Mtop to 2*Mtop, never done)Background modeling• Better understanding of W+HF, W+4P, DY+2P ALPGEN MCs

and tuning with data ( Q2 scale etc)

29

Leading Jet and Sum Jet Ets

30

Eta of top (qq vs gg)