search for wz + zz production with missing transverse

Search for WZ + ZZ productionwith missing transverse energy and b-jetsat CDF

Cornell University LEPP Journal Club : 2011-01-28

Stephen Poprocki (Cornell University)for the CDF collaboration

1

Stephen Poprocki – Dibosons with MET+bbLEPP Journal Club 2011-01-28

Outline

Introduction / MotivationOur b-taggerBackgroundsThe fitterSystematic uncertaintiesResults

2

Motivation• Cross section for WW+WZ+ZZ in the

+ jets final state recently measured at CDF

- Phys. Rev. Lett. 103, 091803 (2009)

• No one has measured WZ+ZZ with jets

• Use b-tagging to reduce WW contribution

- W doesnʼt go to 2 bʼs like Z

- WZ+ZZ → + 2 h.f. jets enhanced

• Associated Higgs production (WH, ZH) is important for low mass Higgs searches at the Tevatron

- Observation of WZ+ZZ will be a major milestone

3

H

W

p

b b

v

Z

p

b b

v v

H

l

Z

W

p

b b

v

Z

p

b b

v v

Z

l

WH, ZHWZ, ZZ

p p

pp

/ET

/ET


Tevatron & CDFp pbar collider operating at 1.96 TeVTevatron will run until September 2011Currently have ~8 fb-1 data acquiredHoping to collect 10 fb-1 or more by end of Run II

4

CDFDelivered: 9.0 fb-1

Acquired: up to 8.0 fb-1

Analyzed: up to 6.7 fb-1

Method overview• Look for events with- large missing transverse energy

- 2 jets consistent with B hadron decays

• Fit the dijet mass distribution from data using 3 templates- Electroweak background

(mainly from W,Z+jets, from Monte-Carlo)

- Multijet background (MJB)(QCD jet production, estimated from data)

- Signal (WZ+ZZ Monte-Carlo)

⇒ # of signal events⇒ cross section

5

Signal events:1516 ± 239 (stat) ± 144(sys)

Data – background

Without b-tagging, measured WW+WZ+ZZ cross section

6

Measured cross section: 18.0 ± 2.8(stat) ± 2.4(sys) ± 1.1(lum) pbTheoretical cross section: 16.8 ± 0.5 pb

The Prequel

• WW dominatesσWW x BR=8.2pbσWZ x BR=2.5pbσZZ x BR=1.0pb


Outline


7

b-tagger• B hadrons often travel a measurable

distance before decaying• Typical b-taggers look for tracks forming a

secondary vertex- Use decay length in transverse plane (L2D)- Displaced tracks often have large impact

parameter (d0)

• We developed a new b-tagger to further exploit individual track information

8

How Do We Identify b-quarks?

! b quarks hadronize into B mesons and baryons! Typical lifetime is O(10^-12 ps)

" B hadron will travel a couple of millimeters before decaying, enough that its decay products (tracks) can be reconstructed into a secondary vertex

! Also: invariant mass of B-tracks; # of tracks in a jet

• Track “bness”: distinguish daughter tracks of Bʼs from non-daughters

• Jet bness: Combine track bness with jet variables

Two levelneural network:

Tune a cut on jet bness to achieve desired b tagging efficiency/mistag rate

b-tagger: Track bness inputsInputs to the track bness NN:- Signed impact parameter (d0), z position,

and their significances (value/uncertainty)- Track momentum in transverse plane (pt)- Track momentum transverse to jet axis (pperp)- Rapidity with respect to the jet axis (Y)

9

Track Bness Separation

! Putting these variables into a NN, we can separate tracks from B's vs. tracks not from B's – and then feed the output discriminator into a B-jet discriminator

NN

Distributions

Black = non-B tracksRed = B tracks

pt pperp Y signedd0

z0 sz0sd0

Distributions


pt pperp Y signedd0

z0 sz0sd0

Distributions


pt pperp Y signedd0

z0 sz0sd0

B tracksNon-B tracks

Track bness

d0d0sig. z0

z0sig.

Take advantage of track displacement

Take advantage of high B momentum}

b-tagger: Jet bness inputs

10

Inputs to the Jet bness NN- Top 5 track bnesses (“bness 1”, etc.)- # of tracks of bness > 0 & invariant

mass of those tracks- If secondary vertex found, L2D

significance- Muon likelihood- # of Ks candidates

bness 1 bness 2

# of tracksbness > 0 mass of tracks

L2D sig.

B jets Non-b jets

4.2 b Tagger Validation 9

!"#$%&"''() (*+, * *+, )*

*+*,

*+)

*+),

*+-

*+-,

*+.

*+.,

!"#$!"%&'(

!"%&'(

/01$23&$44$56"7898&:6;

:6<8#6:6;$3&8#'

Figure 6: Jet bness distribution from our electroweak MC, for jets matched to b quarks (red)

vs. jets not matched to b quarks (black)

as described in [3]. We use Joint Physics scale factors to account for the trigger efficiency

and lepton ID efficiency in MC. Because we see an additional disagreement in the number of

muon events between data and MC, we calculate the ratio of data and MC events in a simple

Z + 0 jet selection for each lepton category pair, and apply this additional scale factor to

the Z + 1 jet samples (see Table 3). The observed agreement after this correction is good.

We apply the same scale factor to the MC in the tt selection, using the CMUP-CMUP scale

factor for events with a CMUP muon, and the CMX-CMX scale factor for events with a

CMX muon.

4.2.2 Z + 1 jet Selection and the Mistag Rate

To investigate the behavior of our b tagger on non-b jets, we look for events in which one

jet is reconstructed along with a Z boson. We follow a similar analysis as was described

in [3]: we search for events with two reconstructed, opposite-sign leptons (from the lepton

categories listed in Table 3), whose reconstructed dilepton mass is consistent with the Z

boson mass. Cuts that we apply are described in Table 4. Figure 7 shows the reconstructed

dilepton mass in data and MC, and Table 5 the number of events in data and MC. After

including the additional scale factors described above, we see very good agreement between

data and MC.

Figure 8 shows the jet bness distribution for jets in this region. Here, the distributions’

shapes roughly match each other, but do not agree particularly well in the ranges [−0.8,−0.5]

NN


Validity / Uncertainty of b-tagger

• Neural networks are trained on Monte Carlo samples

• How do they actually perform in data?

• 2 things you need to know to characterize a b-tagger:

• What fraction of b-jets does it correctly tag? “efficiency” or “tag rate”

• What fraction of non-b jets does it incorrectly tag? “mistag rate”

• Also need way to quantify differences between data and MC to take as systematic uncertainty on b-tagger

• Compare data and MC jets in two control regions similar to our data sample:

• Z+1 jet: Mostly non-b jets; obtain mistag rate

• t-tbar: Mostly b-jets; obtain tag rate

11

Mistag rate (Z + 1 jet)• Typical Z→leptons selection,

• only one jet with ET > 20 GeV

• Data and MC agreement good

• Still, large number of bʼs

• Subtract their contribution when calculating the mistag rate in data

• For a given bness cut in data, slide the bness cut in MC to match the mistag rate in data

• Similarly use the uncertainties for the b-tagger systematic uncertainty

12

Efficiency, Lower bness Jets!

9/24/10!

Highest Jet bnessnd

2

-0.4 -0.2 0 0.2 0.4

Eff

icie

ncy

0.4

0.5

0.6

0.7

0.8

0.9

1

-1L = 4.8 fb!CDF Run II Preliminary,

Data

Uncertainties

Monte Carlo

21!

MR(b) ≈ N(non-b) with bness > b

N(non-b)

16 4 B TAGGING

bness-1 -0.5 0 0.5 1

Even

ts /

0.05

uni

ts

1

10

210

310

410

bness-1 -0.5 0 0.5 1

Even

ts /

0.05

uni

ts

1

10

210

310

410


Data

Z/ZZ±W

-W+W

tt

+ b-jets-"+" #Z

+ b-jets-µ+µ #Z

+ b-jets-e+ e#Z

-"+" #Z

-µ+µ #Z

-e+ e#Z

Figure 8: A comparison of the jet bness in data and MC in the Z + 1 jet selection region.

Overall, the agreement between data and MC is good, but not extremely well modeled.

There was more significant mismodeling in the region [0.6, 0.8] before increasing σZ+bb by a

factor of two, a procedure suggested by [8].

Sample bness Cut in Data Equivalent MC Cut

−1σ Central Value +1σ

Non-b Jets 0.0 −0.114 −0.0795 −0.052

0.85 0.805 0.8325 0.861

b Jets 0.0 0.0275 0.1225 0.2675

0.85 0.8465 0.876 0.903

Table 8: Alternative bness cuts applied to Monte Carlo samples, chosen to match the mea-

sured mistag rates and tagging efficiencies in data. The uncertainties are determined using

the calculated uncertainties on the mistag rates and tagging efficiencies, displayed in Table

7.

Highest Jet bness in tt Selection!

9/24/10!

Highest jet bness

-1 -0.5 0 0.5 1

Even

ts / 0

.05 u

nit

s

0

100

200

300

400


Data

All MC

MC Matched to b-quarks

15!

Highest bness jet in t-tbar selectionEfficiency (t tbar)• First application at CDF of this method for

obtaining the b-tag efficiency

• Typically use tag & probe method in generic dijet events, but difficult to get a high purity sample of b jets

• Typical t-tbar→lepton + jets selection

• Data and MC agreement good

• High b-jet purity in high bness region

• Subtract the small non-b contribution when calculating efficiency

• For a given bness cut in data, slide the bness cut in MC to match the efficiency in data

• Similarly use the uncertainties for the b-tagger systematic uncertainty

13

Efficiency, Lower bness Jets!

9/24/10!

Highest Jet bnessnd

2

-0.4 -0.2 0 0.2 0.4

Eff

icie

nc

y

0.4

0.5

0.6

0.7

0.8

0.9

1


Data

Uncertainties

Monte Carlo

21!

16 4 B TAGGING

bness-1 -0.5 0 0.5 1

Even

ts /

0.05

uni

ts

1

10

210

310

410

bness-1 -0.5 0 0.5 1

Even

ts /

0.05

uni

ts

1

10

210

310

410


Data

Z/ZZ±W

-W+W

tt

+ b-jets-"+" #Z

+ b-jets-µ+µ #Z

+ b-jets-e+ e#Z

-"+" #Z

-µ+µ #Z

-e+ e#Z







Non-b Jets 0.0 −0.114 −0.0795 −0.052

0.85 0.805 0.8325 0.861

b Jets 0.0 0.0275 0.1225 0.2675

0.85 0.8465 0.876 0.903




7.

16 4 B TAGGING

bness-1 -0.5 0 0.5 1

Even

ts /

0.05

uni

ts

1

10

210

310

410

bness-1 -0.5 0 0.5 1

Even

ts /

0.05

uni

ts

1

10

210

310

410


Data

Z/ZZ±W

-W+W

tt

+ b-jets-"+" #Z

+ b-jets-µ+µ #Z

+ b-jets-e+ e#Z

-"+" #Z

-µ+µ #Z

-e+ e#Z







Non-b Jets 0.0 −0.114 −0.0795 −0.052

0.85 0.805 0.8325 0.861

b Jets 0.0 0.0275 0.1225 0.2675

0.85 0.8465 0.876 0.903




7.


Outline


14

Basic event selection• 2 or more central jets with ET > 20 GeV

- Sort jets by bness (instead of ET)

- Get dijet mass from 2 highest bness jets

• Jet 1 bness > 0.85, Jet 2 bness > 0.0

- Call this the “two-tag” channel

- Also exploit a “no-tag” channel for events which fail this cut

• Missing ET > 50 GeV

• 40 GeV < dijet mass < 160 GeV

15

2 b-jets

neutrinos

define signal region

Multi-jet background

• Before selection the diboson signal is swamped by backgrounds– Rejecting QCD multijet events is a major

challenge

• Generic jet production via QCD• QCD ~9 orders above WW+WZ+ZZ

• Rare fluctuations × huge rate = large background

• Difficult to model with Monte Carlo• Use a data-driven method instead

1624


1) Δφ(met, closest jet) > 0.4

2) MET-significance > 4• Uses the jet energy uncertainties to estimate how likely the MET is due to mis-

measurement (low significance) or neutrinos (high significance)

3) Δφ(calorimeter MET, tracker MET)• Two nearly independent ways to detect neutrinos

• MET: energy imbalance in calorimeter (use towers)• Track MET (trkMET): momentum imbalance in tracker (use tracks)

• Small for MET from neutrinos, large for MET from mis-measured jets (MJB)

17

Mis-measured jet aligned with MET: ⇒ small Δφ(met, jet)

Jet 1

Jet 2

Jet nLost/mis-measured jet

Fake MET

Three handles on Multi-jet background:

Dibosons with MET+bbLEPP Journal Club 2011-01-28

Data driven Multi-jet background estimate

18

There is an excess of data in the region Δφ(MET,track MET)>1 which we take to be from MJB.

MJB = (data - MC) dijet mass distribution for events in this region.

Scale up to account for events in the region Δφ(MET,track MET)<1.

• Correction factor = 1.66 ± 7%

Very few MJB events in the 2-tag channel; poor statistics

⇒ use shape from no-tag channel

MJB enhanced regionQCD Mjj Template Comparison for Tagged Region!

9/24/10! 34!

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

0

2

4

6

8

10

12

14

16

18

310!

jet 1 bness > 0.85, jet 2 bness > 0.0

jet 1 bness > 0.85

no-tag channel shape

shapes"1±

-1CDF Run II Preliminary, <L>=5.5 fb

Checking Background Model: Key MJB plots

• Great agreement in no-tag & 2-tag regions; little MJB in 2-tag 19

no-tag two-tag

29

>5 GeV)T

(MET,closest jet E!"0 0.5 1 1.5 2 2.5 3

Even

ts /

bin

0

10

20

30

40

50

60

70

310#

0 0.5 1 1.5 2 2.5 30

10

20

30

40

50

60

70

310#WZ+ZZWWQCD

$%&W$µ&W$e&W

Single Toptt

$$&Zll&Z

Analysis region


>5 GeV)T

(MET,closest jet E!"0 0.5 1 1.5 2 2.5 3

Even

ts /

bin

0

100

200

300

400

500

0 0.5 1 1.5 2 2.5 30

100

200

300

400

500WZ+ZZWWQCD

#$%W#µ%W#e%W

Single Toptt

##%Zll%Z

Analysis region


MET significance0 2 4 6 8 10 12 14 16 18 20

Even

ts /

bin

0

5

10

15

20

25

30

310!

0 2 4 6 8 10 12 14 16 18 200

5

10

15

20

25

30

310!WZ+ZZWWQCD

"#$W"µ$W"e$W

Single Toptt

""$Zll$Z

Analysis region


MET significance0 2 4 6 8 10 12 14 16 18 20

Even

ts /

bin

0

50

100

150

200

250

300

350

400

0 2 4 6 8 10 12 14 16 18 200

50

100

150

200

250

300

350

400 WZ+ZZWWQCD

!"#W!µ#W!e#W

Single Toptt

!!#Zll#Z

Analysis region


Figure 23: Left: no-tag region. Right: 2-tag region. Upper row: min(∆φ(E/T , jet)) distribu-tion for events that pass all of the analysis cuts except for the min(∆φ(E/T , jet))-cut. Lowerrow: E/T -significance distribution for events that pass all of the analysis cuts except for theE/T -significance cut. Points are data, red hatched histograms are QCD background, othercolors are various EWK contributions. The MC uncertainties are only statistical .

!"#$%%&' (' )'' )(' *''

'+*

'+,

'+-

'+.

)

!/!012!* !)3+((!/!*45678!! !'+.,39 !)+).4!!)4+,.!8!!!!!!!! !'+33)-!!(+,*(!: !'+''.,).!!'+3,,(!

!/!012!* !)3+((!/!*45678!! !'+.,39 !)+).4!!)4+,.!8!!!!!!!! !'+33)-!!(+,*(!: !'+''.,).!!'+3,,(!

;09<=>?>!6#@?70!

!"#$%&'$(($)*+,-.-'/*01$234$5$67689:

!"#$%&'( )( *(( *)( +(((

(,+

(,-

(,.

(,/

*

$0$123$+ $4,-++$0$*.5678$$ $(,9.-+: $(,).49!$-.,/*$8$$$$$$$$ $(,;+9+!$),+..$< $(,((..;)!$(,99*-$

$0$123$+ $4,-++$0$*.5678$$ $(,9.-+: $(,).49!$-.,/*$8$$$$$$$$ $(,;+9+!$),+..$< $(,((..;)!$(,99*-$

=1:>?@A@$6&BA71$

!"#$%&'$(($)*+,-.-'/*01$234$5$67689:

Figure 24: Trigger efficiency as a function of Mjj (left) and E/T (right). The dashed lines inthe E/T plot show the statistical uncertainty.

t-tbar rejection cuts

• Njet(Et>10 GeV) + Nele+ Nmu + Ncrk < 4• Nele + Nmu + Ncrk < 2• Nmuon < 2• Nelectron < 2

• Odd looking combination of cuts, but it works quite well• A neural network offered no

improvement

• t-tbar a large background in 2-tag channel

• t-tbar should give more leptons and jets than signal

• Leptons may be misreconstructed as jets

20

Full event selection• 2 or more central jets with ET > 20 GeV

- Sort jets by bness (instead of ET)

- Get dijet mass from 2 highest bness jets

• Jet 1 bness > 0.85, Jet 2 bness > 0.0

- Or fail these cuts: no-tag channel

• Missing ET > 50 GeV

• 40 GeV < dijet mass < 160 GeV

21

2 b-jets

neutrinos

Cuts to reduce:


ttbar

define signal region

• Njet(Et>10 GeV) + Nele+ Nmu + Ncrk < 4

• Nele + Nmu + Ncrk < 2

• Nele < 2, Nmuon < 2

• Missing ET significance > 4

• Δφ(Missing ET, closest jet) > 0.4

Electroweak backgrounds

• Z+jets- Z→ee, Z→µµ, Z→ττ- Z→νν

• W+jets- W→eν- W→µν- W→τν

• ttbar• single top

Pythia Alpgen MadEvent+Pythia

22

• Model EWK background shape with Monte-Carlo

• In W,Z+jets, can get bʼs from gluon splitting

• No lepton requirement

• Missing ET > 50 GeV makes events witha neutrino the dominant EWK backgrounds


Outline


23

24

The Fitter: Channels• Fit the dijet mass distribution

in data using templates for the signal and backgrounds

• Also allow backgrounds to vary in the fit

• Simultaneous fit in 2 channels:

• Two-tag channel:Events with 2 b-tags

• No-tag channel: Events without 2 b-tags

38 8 SIGNAL EXTRACTION

Note that in the case of the WZ/ZZ → bb + E/T measurement, the signal template isbroken in up into events with 2 b’s matched to jets (signal) and the rest (background). Thesetemplates are shown in Fig. 32.

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000

40000

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000

40000 WZ/ZZ (bb)WZ/ZZ (non-bb)WWQCDSingle Toptt

EWK

-1CDF Run II Preliminary, <L>=5.5 fbNo-tag Channel

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140

160

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140


EWK

-1CDF Run II Preliminary, <L>=5.5 fb2-tag Channel

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700

800

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700

800 WZ/ZZ (bb)WZ/ZZ (non-bb)WW


Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

2

4

6

8

10

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

2

4

6

8

10WZ/ZZ (bb)WZ/ZZ (non-bb)WW


Figure 32: Top row: templates for mclimit csm.C, for the no-tag channel (left) and two-tagchannel (right). Bottom row: just the diboson samples, stacked.

8.2 Systematic Uncertainties

Besides the nuisance parameters in the fit associated with the rate uncertainties as describedabove, there are parameters for EWK shape, QCD shape, JES, and b-tagging efficiency un-certainties. The shape uncertainties are handled by supplying 2 more templates, correspond-ing to upward or downward fluctuations of the nuisance parameter. Template morphing ishandled by mclimit csm.C, and we allow extrapolation beyond the supplied shapes.

• EWK shape: The central value template used is the average of the EWK MC and theγ+jets template from Section 7, with those templates used as the shape uncertainties.



Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000

40000

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000


EWK


Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140

160

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140


EWK


Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700

800

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700



Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

2

4

6

8

10

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

2

4

6

8







Just the dibosons

b-tagging indeed reduces WW

25

The Fitter: Channels

• Systematics go into the fitter:• B-tagging uncertainty (already explained)• Jet energy scale uncertainty• EWK shape uncertainty• MJB shape uncertainty

• Fit in both channels simultaneously:- Signal (WZ+ZZ)

- WW

- Single top + t-tbar

• Allow to float separately in the two channels:

- EWK (W,Z+jets): donʼt trust the modeling of the b-quark content

- MJB: donʼt know the b-quark content

• Constrain the backgrounds based on their cross sections, but

• Let EWK float in the fit unconstrained (donʼt trust the overall normalization)

• Let signal float unconstrained; weʼre measuring it!

will explain soon}

• To optimize the analysis, need an a priori estimate of the measurementʼs sensitivity

• Run many pseudo-experiments and calculatefor pseudo-data generated

• with signal+background hypothesis, and

• with background-only hypothesis.

• Obtain the probability of a 3-sigma measurement:

• Find the Δχ2 where only ~0.3% of background-only PEs lie below

• Prob 3-sigma = fraction of S+B PEs below this.

The Fitter: Sensitivity

∆χ2 = χ2S+B − χ2

B

26

signal+backgroundpseudo-data

background-onlypseudo-data

Example:

• We can now optimize the cuts, specifically: bness

• Scan over jet bness cuts in MC,

• Get the probability of 2 sigma for each set of cuts(more accurate than 3 sigma for a given number of PEs)

• Choose Jet 1 bness > 0.85, Jet 2 bness > 0.0

27

The Fitter: Optimization

Optimization of bness Cuts for Best Sensitivity in Double-Channel Fit!

9/24/10! 49!

0.46

0.48

0.5

0.52

0.54

0.56

0.50548 0.51146 0.50296 0.49924 0.51104 0.50714 0.49184

0.54058 0.53954 0.54546 0.5415 0.54696 0.53966 0.53702

0.55814 0.54992 0.55698 0.56112 0.54902 0.55554 0.54408

0.5442 0.54928 0.55578 0.55418 0.55496 0.5487 0.54454

0.54762 0.5548 0.55496 0.5641 0.55868 0.55368 0.5536

0.54824 0.54562 0.55358 0.55152 0.55234 0.5445 0.54316

0.5393 0.54398 0.54652 0.5431 0.55192 0.54854 0.54624

0.53612 0.54064 0.5443 0.5457 0.53594 0.52904 0.52288

0.5118 0.51032 0.52092

jet 1 bness0.6 0.65 0.7 0.75 0.8 0.85 0.9

jet

2 b

ne

ss

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

!Prob. 2-1CDF Run II Preliminary, <L>=5.5 fb


Outline


28


MJB shape uncertainty

29

• Need some way of assessing the uncertainty on the MJB estimate• Look just outside the signal region (3 < MET-significance < 4, whereas the cut is at

MET-sig > 4)• Obtain shape uncertainty by comparing Mjj in two different regions:

Δφ(MET,track MET) > 1 (MJB enhanced region)Δφ(MET,track MET) < 1 (EWK dominated region)QCD Mjj Template!

9/24/10! 33!

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

0

2

4

6

8

10

12

14

16

18

310!

Central Shape

shape"+1

shape"-1


no-tag channel


Jet Energy Scale Systematic

30

Vary the jet energies according to their uncertainties

Dijet mass peak in WZ/ZZ and WW will shift

+1σ−1σ


EWK uncertainty via photon+jets

31

We use photon+jets data to assess the systematic uncertainty on our W/Z+jets background from Monte-Carlo.Idea is that jet kinematics in photon+jets events should be similar to jet kinematics in events with other gauge bosons (W,Z)

Z, gamma have same interactions; W slightly different

Some cuts modified: MET > 50 GeV → MET+photon > 50 GeV

As photon+jets and W/Z+jets aren't identical (photon is massless), we need a correction

Weight photon+jets data by the ratio of the dijet mass distribution of our W/Z+jets MC to photon+jets MC

This should compensate for the physical differences, in a MC independent fashion

MC uncertainties- PDFs, radiation, etc. - should cancel out in the ratio

MET (MET+jets ) MET

photon

MET+photon = “MET”

j j

photon

j j

W/Z

≈


EWK MC compared to pho+jets

Excellent agreement

but still quantify the difference as a systematic

32

no-tag two-tagEWK Mjj Templates in No-Tag Region!

9/24/10! 35!

) 2 (GeV/cjjM

40 60 80 100 120 140 160

1/N

dN

/(8 G

eV

)

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

No-Tag Electroweak MC Template

+Jets Data Template!No-Tag

-1CDF Run II Preliminary, <L> = 6.0 fb

EWK Mjj Templates in 2-Tag Region!

9/24/10! 36!

) 2 (GeV/cjjM

40 60 80 100 120 140 160

1/N

dN

/(8

Ge

V)

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

Double-Tag Electroweak MC Template

+Jets Data Template!Double-Tag


Stephen Poprocki – Dibosons with MET+bbLEPP Journal Club 2011-01-28 33

Use a combination of MET triggers to get the most dataMET trigger efficiency: 93.9±0.3%5.5 fb-1 of data becomes effectively 5.2 fb-1

Cross checked using Z→µµMjj and Δφ(MET,track MET) are not sculpted

!"#$$%"&'()#%*)#%+&

•! Final integral efficiency is

96.2%±2.2%

–! Take 90% efficiency for MET>120 ! 2% effect ! assign additional

2% uncertainty

,-..-/0& 12+32&4"5*657&8#$$+&9%%:*$& ,&

MET

!

Eff =c

1+ e

a"x

b

No sculpting - no effect on QCD

Trigger efficiency & Luminosity


Outline

Introduction / MotivationOur b-taggerThe fitterHandling the multi-jet backgroundSystematic uncertaintiesResults

34

35

Signal Region Predictionsno-tag

two-tag

5.1 Background Simulation/Modelling 21

Sample Description No-tag channel Two-tag channel

Z→ee 13.5 0.3

Z→µµ 1108.1 9.0

Z→ττ 2538.2 12.1

Z→νν 25097.4 204.0

W→eν 34889.1 128.6

W→µν 24299.4 143.4

W→τν 61885.9 216.9

tt 495.2 154.8

single top 1337.4 200.0

WW 2679.8 6.8

WZ 814.9 23.8

WZ (bb) 58.0 20.6

ZZ 332.3 21.2

ZZ (bb) 50.2 19.6

WZ+ZZ 1147.1 45.0

WZ+ZZ (bb) 108.2 40.2

Non-QCD background 154343.8 1075.8

QCD estimate 73853.5 58.4

Table 8: Expected contribution of different processes, for 5.5 fb−1.

The underlying assumption of how QCD enters the analysis is that either jets are mis-

measured, or a leading charged track, π0 or a γ is lost in an uninstrumented region of the

detector. We expect the dominant effect to be jet mismeasurement. Most of the QCD multi-

jet background is suppressed by the E/T -significance and min(∆φ) cuts described in detail in

Section C. To estimate the remaining QCD contribution, we construct a new variable, P/T , to

compliment the traditional calorimeter based E/T . The P/T is defined as the negative vector

sum of tracks with pT >0.3 GeV/c. Tracks used in calculation of P/T have to pass minimal

quality requirements (χ2COT

<6) and be within ±4σzvx window from the primary vertex. The

σzvx depends on a type of the tracking algorithm which was used to reconstruct a track. For

example, σzvx=1.2 cm for COT-only tracks and σzvx=0.6 cm for outside-in tracks (majority

of tracks) which use both COT and Silicon hits.

When comparing the azimuth angle (φ) for E/T and P/T , alignment is expected between

the two quantities in the case of true E/T (e.g., for diboson signal and EWK backgrounds).

We will call the difference between these two angles as ∆φMET . The expected shape (slowly

falling spectrum) of the ∆φMET distribution for the QCD background is illustrated at Fig. 17.

CDF Preliminary L=5.5/fb

CDF Preliminary L=5.5/fb

QCD tamed even in no-tag channel;W+jets the largest background

36

no-tag two-tagControl regions (Signal region blinded)

53

A Control Region Plots

Comparisons of MC and data are shown in both the no-tag channel and the two-tag channelin Figs. 43,44,45. The bness distributions are shown in 46. Comparisons for ∆φ(jet, E/T ) andE/T -significance were shown in 23.

Dijet mass [GeV]0 5 10 15 20 25 30 35 40

Even

ts /

8 G

eV

0

2

4

6

8

10

12

14

16310!

0 5 10 15 20 25 30 35 400

2

4

6

8

10

12

14

16310!

-1CDF Run II Preliminary, <L>=5.5 fbWZ+ZZWWQCD

"#$W"µ$W"e$W

Single Toptt

""$Zll$Z

Dijet mass [GeV]0 5 10 15 20 25 30 35 40

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140

0 5 10 15 20 25 30 35 400

20

40

60

80

100

120

140WZ+ZZWWQCD

!"#W!µ#W!e#W

Single Toptt

!!#Zll#Z


Dijet mass [GeV]160 180 200 220 240 260 280 300 320 340 360

Even

ts /

8 G

eV

0

0.5

1

1.5

2

2.5

3

3.5310!

160 180 200 220 240 260 280 300 320 340 3600

0.5

1

1.5

2

2.5

3

3.5310!

WZ+ZZWWQCD

"#$W"µ$W"e$W

Single Toptt

""$Zll$Z


Dijet mass [GeV]160 180 200 220 240 260 280 300 320 340 360

Even

ts /

8 G

eV

0

5

10

15

20

25

30

35

40

45

160 180 200 220 240 260 280 300 320 340 3600

5

10

15

20

25

30

35

40

45 WZ+ZZWWQCD

!"#W!µ#W!e#W

Single Toptt

!!#Zll#Z


Figure 43: Dijet mass plots for the no-tag channel (left column) and two-tag channel (rightcolumn). The top row shows the predictions in the signal region, while the bottom two rowsshow comparisons with data outside the signal region.

Good agreement between data and MC

4

[GeV]jjM40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

10

20

30

40310!

[GeV]jjM40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

10

20

30

40310!

DataWZ+ZZMulti-jetWW

and single tttEWK

[GeV]jjM40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

50

100

150

[GeV]jjM40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

50

100

150

DataWZ+ZZMulti-jetWW

and single tttEWK

Dijet mass [GeV]40 60 80 100 120 140 160

-200

0

200

400Data - backgrounds

WZ+ZZ

Standard Model

Dijet mass [GeV]40 60 80 100 120 140 160

-20

0

20

Data - backgrounds

WZ+ZZ

Standard Model

FIG. 2. Result of the fit to data for the double fit to all of WZ/ZZ. Left column is the no-tag channel; right column is the2-tag channel. Bottom row is data−backgrounds.

as nuisance parameters and are incorporated into the fit

by use of a Bayesian marginalization technique (see [13]).

We optimize our selection, in particular for the jet

bness thresholds, based on the probability of obtaining

a result with significance of 2σ. The optimization points

to a broad region where the sensitivity is maximized and

we choose our bness thresholds in that region.

Process(es) Fit Nevents (no-tag) Fit Nevents (2-tag)

EWK 153300 ± 3000 694 ± 48tt and single t 1700 ± 140 313 +24

−26

Multi-jet 72300 ± 2800 54.6 ± 7.3WW 2720 ± 190 8.3 +1.8

−1.9

WZ/ZZ 1160 ± 620 39.9 ± 20

TABLE I. Fit number of events from the 2-channel fit forWZ/ZZ, with all systematics applied.

Fig. 2 shows the results of the fit, and Tab. I. shows

the number of fitted events.

To translate the result of our fit to the data to bounds

or limits on the true cross section of WZ/ZZ produc-

tion, we construct modified Feldman-Cousins bands by

analyzing the distribution of fitted (i.e., measured) cross

sections in pseudo-experiments generated with a variety

of scale factors on the input signal cross section [18].

When running pseudo-experiments, we consider the ef-

fect of additional systematic uncertainties that affect ouracceptance. These include, in order of increasing signif-

icance: jet energy resolution, E/T modeling, initial and

final state radiation, parton distribution function, and

luminosity and trigger efficiency uncertainties.

Based on a Monte Carlo calculation, the acceptance

times efficiency for the WZ and ZZ production is 4.1%,

and 4.6%, respectively. In the calculation of the com-

bined diboson cross section, we assume that each signal

process contributes proportionally to its predicted SM

cross section: 3.6 pb for WZ and 1.5 pb for ZZ obtained

using the mcfm v5.4 program [17] with CTEQ6.1M

PDFs [11]. Our measured result, using the 1 σ bands

from the modified Feldman-Cousins analysis, is σ(pp →WZ,ZZ) = 5.0+3.6

−2.5 pb. We set a limit on σWW,ZZ at

13 pb(2.5× σSM ) with 95% CL.

We thank the Fermilab staff and the technical staffsof the participating institutions for their vital contribu-

tions. This work was supported by the U.S. Department

of Energy and National Science Foundation; the Italian

Istituto Nazionale di Fisica Nucleare; the Ministry of

Education, Culture, Sports, Science and Technology of

Japan; the Natural Sciences and Engineering Research

Council of Canada; the National Science Council of the

Republic of China; the Swiss National Science Founda-

tion; the A.P. Sloan Foundation; the Bundesministerium

fur Bildung und Forschung, Germany; the World Class

University Program, the National Research Foundation

of Korea; the Science and Technology Facilities Coun-

cil and the Royal Society, UK; the Institut National de

Physique Nucleaire et Physique des Particules/CNRS;

the Russian Foundation for Basic Research; the Minis-

terio de Ciencia e Innovacion, and Programa Consolider-

Ingenio 2010, Spain; the Slovak R&D Agency; and the

Academy of Finland.

[1] All jets in the region |η| < 3.6 are considered.[2] T. Aaltonen et al. (CDF Collaboration), Phys. Rev. Lett.

37

Fit results: double fit for WZ/ZZno-tag two-tag

Result of Double Channel Fit!

9/24/10! 51!

69

CDF Run II Preliminary,�L = 5.5 fb−1

Process(es) Fit # events (no-tag) Fit # events (2-tag)

EWK 153312 694.4

tt 517.3 129.3

single top 1184.4 183.2

QCD 72342.6 54.6

WW 2721.6 8.3

WZ/ZZ 1156.3 39.9

!"#$%&"'()&*$$($")+*,(-(./0012!(

σ = 5.0+3.6−2.5 pb


signal events, the first bin (at a measured scale factor of 0) may contain a very large numberof events; however, we do not treat it differently than any other bin, and if it is included inthe range, its entire contents contribute towards the calculation of coverage. This method isa slight variation of the method proposed in [10], due to the strict boundary on the measuredscale factors (as well as the input scale factors), but retains the properties that it avoids flip-flopping, and aims for coverage as close as possible to (but always as much as) the statedvalue.

Figure 41 shows the results of our Feldman-Cousins analysis in the double channel fit.Our measured result, using the 1σ bands from the Feldman-Cousins plot, is then σmeasured =0.99+0.7

−0.5 × σSM . We set a limit on σmeasured at about 2.5× σSM with 95% CL. Using

σSM = σWZ + σZZ = 5.08 pb

we then calculateσmeasured = 5.0+3.6

−2.5 pb,

with a 95% CL limit σ < 13 pb.Similarly, Figure 42 shows the results of our Feldman-Cousins analysis in the 2-tag only

channel fit. Our measured result is 0.12 × σSM , which corresponds to a limit of about2.4 × σSM . However, here our signal is composed of only those events in which a Z bosondecays to a bb pair, thus

σSM = σWZ × Br(Z → bb) + σZZ × 2× Br(Z → bb) = 0.975 pb

where we assume the branching ratio of Z → bb is 15.1% [11]. Thus, our measured crosssection is

σmeasured = 0.12 pb,

with a 95% CL limit σ < 2.3 pb.

σ = 0.99σSM


Significance of double fit for WZ/ZZ

Significance of ~1.5σ

38

8.4 Results 49

2!"-30 -25 -20 -15 -10 -5 0

Pseu

do-e

xper

imen

ts

1

10

210

310

410

510

WZ/ZZ Significance

Test hypothesis

Null hypothesis0.13% 2.3%

Data

-1CDF Run II Preliminary, <L>=5.5 fbWZ/ZZ Significance

Figure 39: The ∆χ2 distributions for null and test hypothesis for the double-channel fit toall of WZ/ZZ. In data, ∆χ2 = −2.51, and 7.0% of the null hypothesis PEs have a ∆χ2 lessthan this. The dashed lines show the ∆χ2 values for which the fraction of the null hypothesisdistribution with smaller ∆χ2 is 0.135% and 2.28%.

2!"-25 -20 -15 -10 -5 0

Pseu

do-e

xper

imen

ts

1

10

210

310

410

510

WZ/ZZ (bb) Significance

Test hypothesis

Null hypothesis

0.13% 2.3%Data

-1CDF Run II Preliminary, <L>=5.5 fbWZ/ZZ (bb) Significance

Figure 40: The ∆χ2 distributions for null and test hypothesis for the 2-tag only fit forWZ/ZZ with bb. In data, ∆χ2 = −0.020, and 68% of the null hypothesis PEs have a ∆χ2

less than this. We essentially have no sensitivity in this channel.


Cross section limit for double fit for WZ/ZZUse a modified Feldman-Cousins method for determining an upper limit on the cross sectionPerform pseudo-experiments with signal scaled from 0 to 3 times the SM

39

Coverage Bands for Double Channel Fit!

9/24/10!

SM! / Measured!0 1 2 3 4

SM

! /

!

0

1

2

3

Measured Result

68% Coverage Bands

95% Coverage Bands

-1L = 5.5 fb"CDF Run II Preliminary,

54!

!"#$%&"'()*+*,-(.(/(012.3!(Measured limit: σ < 2.5σSM at 95% CL

40

Bonus: WZ+ZZ to MET + bb

• Can also try to measureWZ/ZZ to bb- One step closer to WH/ZH to

MET + bb

• Fit to only the two-tag channel

• Break WZ/ZZ template into bb (signal) and non-bb (background) templates

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000

40000

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000


EWK


Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700

800

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700





Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000

40000

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

5000

10000

15000

20000

25000

30000

35000


EWK


Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140

160

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

20

40

60

80

100

120

140


EWK


Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700

800

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

100

200

300

400

500

600

700



Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

2

4

6

8

10

Dijet mass [GeV]40 60 80 100 120 140 160

Even

ts /

8 G

eV

0

2

4

6

8







41

Fit results: single fit for WZ/ZZ→MET+bbData - backgrounds:two-tag

This measurement has virtually no significance

Result of 2-Tag Channel Only Fit!

9/24/10! 55!

70 D TABLES FOR BLESSING


Process(es) Fit # events (2-tag)

EWK 710.7

tt 134.3

Single top 189.7

QCD 54.5

WW 8.8

WZ/ZZ (l.f.) 5.7

WZ/ZZ (bb) 4.4

!"#$%&"'()&*$$($")+*,(-(

./0123!(

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

Ge

V0

20

40

60

80

100

120

140

160

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

Ge

V0

20

40

60

80

100

120

140

160


WZ+ZZ (lf)

QCDWWSingle Top

ttEWK

Data

[GeV]jjM40 60 80 100 120 140 160

Eve

nts

/ 8

GeV

0

20

40

60

80

100

120

140

160

180

[GeV]jjM40 60 80 100 120 140 160

Eve

nts

/ 8

GeV

0

20

40

60

80

100

120

140

160

180


WZ+ZZ (bb)WZ+ZZ (lf)QCD

WWSingle Top

ttEWKData

[GeV]jjM40 60 80 100 120 140 160

Ev

en

ts / 8

Ge

V

-20

-10

0

10

20

WZ+ZZ (bb)

Standard Model

Data - Backgrounds



9/24/10! 55!




EWK 710.7

tt 134.3

Single top 189.7

QCD 54.5

WW 8.8

WZ/ZZ (l.f.) 5.7

WZ/ZZ (bb) 4.4

!"#$%&"'()&*$$($")+*,(-(

./0123!(

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

Ge

V

0

20

40

60

80

100

120

140

160

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

Ge

V

0

20

40

60

80

100

120

140

160


WZ+ZZ (lf)

QCDWWSingle Top

ttEWK

Data

[GeV]jjM40 60 80 100 120 140 160

Ev

en

ts / 8

Ge

V

0

20

40

60

80

100

120

140

160

180

[GeV]jjM40 60 80 100 120 140 160

Ev

en

ts / 8

Ge

V

0

20

40

60

80

100

120

140

160

180



WWSingle Top

ttEWKData

[GeV]jjM40 60 80 100 120 140 160

Ev

en

ts /

8 G

eV

-20

-10

0

10

20

WZ+ZZ (bb)

Standard Model

Data - Backgrounds



9/24/10! 55!




EWK 710.7

tt 134.3

Single top 189.7

QCD 54.5

WW 8.8

WZ/ZZ (l.f.) 5.7

WZ/ZZ (bb) 4.4

!"#$%&"'()&*$$($")+*,(-(

./0123!(

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

0

20

40

60

80

100

120

140

160

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

0

20

40

60

80

100

120

140

160


WZ+ZZ (lf)

QCDWWSingle Top

ttEWK

Data

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

0

20

40

60

80

100

120

140

160

180

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

0

20

40

60

80

100

120

140

160

180



WWSingle Top

ttEWKData

[GeV]jjM40 60 80 100 120 140 160

Even

ts / 8

GeV

-20

-10

0

10

20

WZ+ZZ (bb)

Standard Model

Data - Backgrounds


8.4 Results 49

2!"-30 -25 -20 -15 -10 -5 0

Pseu

do-e

xper

imen

ts

1

10

210

310

410

510

WZ/ZZ Significance

Test hypothesis

Null hypothesis0.13% 2.3%

Data

-1CDF Run II Preliminary, <L>=5.5 fbWZ/ZZ Significance

Figure 39: The ∆χ2 distributions for null and test hypothesis for the double-channel fit toall of WZ/ZZ. In data, ∆χ2 = −2.51, and 7.0% of the null hypothesis PEs have a ∆χ2 lessthan this. The dashed lines show the ∆χ2 values for which the fraction of the null hypothesisdistribution with smaller ∆χ2 is 0.135% and 2.28%.

2!"-25 -20 -15 -10 -5 0

Pseu

do-e

xper

imen

ts

1

10

210

310

410

510

WZ/ZZ (bb) Significance

Test hypothesis

Null hypothesis

0.13% 2.3%Data

-1CDF Run II Preliminary, <L>=5.5 fbWZ/ZZ (bb) Significance

Figure 40: The ∆χ2 distributions for null and test hypothesis for the 2-tag only fit forWZ/ZZ with bb. In data, ∆χ2 = −0.020, and 68% of the null hypothesis PEs have a ∆χ2

less than this. We essentially have no sensitivity in this channel.


Cross section limit for single fit for WZ/ZZ→MET+bb

Still able to set a decent limit though

42

Coverage Bands for 2-Tag Channel Only Fit!

9/24/10!

SM! / Measured!0 2 4 6

SM

! /

!

0

1

2

3

4

Measured Result

68% Coverage Bands

95% Coverage Bands

-1L = 5.5 fb"CDF Run II Preliminary,

57!

!"#$%&"'()*+*,-(.(/(012.3!(Measured limit: σ < 2.4σSM at 95% CL


Conclusions

43

We measured the cross section for WZ+ZZ in the MET plus 2 b-jets enhanced channel

The key: identify b-jets to reduce WW and measure just WZ+ZZ

Developed a custom b-tagger

Using 5.2 fb-1 of data,

WZ+ZZ

• Measured a cross section of , consistent with the SM (5.1 pb)

• Set a limit of σ < 13 pb (2.5 σSM) at 95% CL

WZ+ZZ→MET+bb

• Set a limit of σ < 2.3 pb (2.4 σSM) at 95% CL

Final states for WZ+ZZ the same as WH+ZH in MET + jets channel

Our techniques can be used in a future Higgs search

σ = 5.0+3.6−2.5 pb

search for wz + zz production with missing transverse

Documents