VBF H inv L1 Prefire Impact S. Narayanan , on behalf of the VBF H - - PowerPoint PPT Presentation

▶
vbf h inv l1 prefire impact
SMART_READER_LITE
LIVE PREVIEW

VBF H inv L1 Prefire Impact S. Narayanan , on behalf of the VBF H - - PowerPoint PPT Presentation

Jun 16, 2023 398 likes •596 views

VBF H inv L1 Prefire Impact S. Narayanan , on behalf of the VBF H inv group Higgs PAG, 30/07/2018 S. Narayanan (MIT) VBF H inv 30/07/2018 1 / 17 Overview of problem -1 36 fb (13 TeV) L1 paths may prefire due to mis-timed

slide-1
SLIDE 1

VBF H→inv L1 Prefire Impact

  • S. Narayanan, on behalf of the VBF H→inv group

Higgs PAG, 30/07/2018

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 1 / 17

slide-2
SLIDE 2

Overview of problem

◮ L1 paths may prefire due to mis-timed

ECAL trigger primitives (TPs) in endcap

◮ Prefire ⇒ next event (interesting

physics) cannot fire L1A

◮ Reported weeks ago, heavily discussed

in L1, PPD, JME, etc meetingsab

◮ What is the impact on this analysis?

◮ High pT and forward jets can set off

forward ECAL L1!

◮ Mis-measured jet energy can bias pmiss

T

ahttps://indico.cern.ch/event/737391/ bhttps://indico.cern.ch/event/729127/

5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 1000 2000 3000 4000 5000 6000 7000 8000 9000

Events

Data W+jets [QCD] W+jets [EWK] Top Z+jets [EWK] Z+jets [QCD] Diboson QCD Data W+jets [QCD] W+jets [EWK] Top Z+jets [EWK] Z+jets [QCD] Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred.

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 2 / 17

slide-3
SLIDE 3

Determining prefire inefficiency

  • 1. Select “un-prefirable” events

◮ Exploit trigger rules by looking for L1A at bx0 and bx-3 ◮ N. Smith has kindly provided a filter to select such events1

  • 2. Compute rate of L1IsoEG30 TP prefire in this set of unbiased events

◮ Start with this L1 path since it is unprescaled ◮ Most plots will use the following definition of efficiency:

ǫIsoEG30 bx-1 = N(jets matched to L1IsoEG30 in bx-1) N(all jets)

◮ This is the easiest observable to translate into analysis 1https://github.com/nsmith-/PrefireAnalysis/

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 3 / 17

slide-4
SLIDE 4

Frequency of L1IsoEG30 prefire in JetHT

◮ Averaging over 2016 ◮ Events pass pmiss T

filters

◮ Central jets pass ID ◮ Only ∼ 0.2% of data is

un-prefirable

◮ Not enough data to

focus on VBF-like event topologies

◮ However, effect should

factorize to jets

◮ Hot tower much more

likely to preire

100 200 300 400 500 600

[GeV]

T

p

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

| η |

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

L1IsoEG BX=-1 eff

JetHT

CMS CMS CMS CMS CMS CMS

Efficiency

3 − 2 − 1 − 1 2 3

φ

5 − 4 − 3 − 2 − 1 − 1 2 3 4 5

η

2 4 6 8 10 12 14 16 18 20 22 24

3

10 ×

Number of BX-1 jets

JetHT

CMS

Number of jets

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 4 / 17

slide-5
SLIDE 5

Removing the spike

◮ Turn-on modified slightly with spike removed

◮ Reject any jet within 0.2 of (−2.81, 2.07) ◮ Will remove events containing such a jet in the analysis 100 200 300 400 500 600

[GeV]

T

p

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

| η |

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

L1IsoEG BX=-1 eff

JetHT +2.81|>0.4 η

  • 2.01|>0.4 or |

φ |

CMS CMS CMS CMS CMS CMS

3 − 2 − 1 − 1 2 3

φ

5 − 4 − 3 − 2 − 1 − 1 2 3 4 5

η

200 400 600 800 1000 1200 1400

Number of BX-1 jets

JetHT +2.81|>0.4 η

  • 2.01|>0.4 or |

φ |

CMS

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 5 / 17

slide-6
SLIDE 6

PD dependence

◮ Looking in slice 2.75 < |η| < 3 ◮ Both MET and SingleMuon have a weaker inefficiency than JetHT at low pT ◮ Caused by correlations between event topology, selected triggers, and L1 seeds

100 200 300 400 500 600

[GeV]

T

p

0.2 0.4 0.6 0.8 1 1.2 1.4

IsoEG30 BX=-1 eff

MET SingleMuon JetHT

CMS

100 200 300 400 500 600

[GeV]

T

EM p

0.2 0.4 0.6 0.8 1 1.2 1.4

IsoEG30 BX=-1 eff

MET SingleMuon JetHT

CMS CMS

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 6 / 17

slide-7
SLIDE 7

PD dependence in MET PD

◮ Prefire decreases pmiss T

in events with real pmiss

T ◮ ECAL pulse attributed to bx-1 can zero-suppress part of jet in bx0 ◮ Both online and offline pmiss T

can be biased

E C A L D e p

  • s

i t Z ν ν H C A L D e p

  • s

i t s p

T mi s s

= ⇒

Z ν ν H C A L D e p

  • s

i t s p

T mi s s

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 7 / 17

slide-8
SLIDE 8

PD dependence in JetHT PD

◮ Prefire decreases pT and HT in multijet events (most of JetHT) ◮ ECAL pulse attributed to bx-1 can zero-suppress part of jet in bx0 ◮ Both online and offline jet pTs can be biased

E C A L D e p

  • s

i t H C A L D e p

  • s

i t s

= ⇒

H C A L D e p

  • s

i t s p

T m i s s

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 8 / 17

slide-9
SLIDE 9

SingleMuon and FinOR

◮ A mis-timed ECAL TP lowers probability that MET/MHT L1/HLT will fire on

real-pmiss

T

events

◮ These jets can have weird energy fractions ⇒ cleaned from MHT

◮ Even offline pmiss T

can be biased due to zero suppression

◮ Same is true for any trigger that uses ECAL ◮ SingleMuon triggers should be largely unbiased ◮ Compare to global trigger “Final OR” decision ◮ Efficiency definition changes slightly because FinOR cannot be matched to specific

  • bjects

ǫFinOR bx-1 = N(events w/ exactly 1 endcap jet and FinOR bx-1) N(events w/ exactly 1 endcap jet )

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 9 / 17

slide-10
SLIDE 10

L1EG30Iso vs L1EG40 vs GT FinOR

◮ Non-isolated TTs can show up in L1EG40 (not prescaled) ◮ In addition, ECAL deposits can prefire other triggers, so compare to global trigger ◮ Inefficiency increases slightly, but generally shows the same PD dependence

◮ Larger ǫ because more triggers to prefire

100 200 300 400 500 600

[GeV]

T

p

0.2 0.4 0.6 0.8 1 1.2 1.4

IsoEG30 BX=-1 eff

MET SingleMuon JetHT

CMS

100 200 300 400 500 600

[GeV]

T

p

0.2 0.4 0.6 0.8 1 1.2 1.4

IsoEG30,EG40 BX=-1 eff

MET SingleMuon JetHT

CMS

100 200 300 400 500 600

[GeV]

T

p

0.2 0.4 0.6 0.8 1 1.2 1.4

FinOR BX=-1 eff

MET SingleMuon JetHT

CMS

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 10 / 17

slide-11
SLIDE 11

Final correction

◮ SingleMuon measurement has large stat uncertainties above 300 GeV ◮ Use SingleMuon below 300 GeV and JetHT above 300 GeV ◮ Apply correction from FinOR inefficiency

◮ Should capture all effects

◮ Lose up to 20% of signal in tails

100 200 300 400 500 600

[GeV]

T

p

0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

| η |

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

L1IsoEG BX=-1 eff

SingleMuon +2.81|>0.4 η

  • 2.01|>0.4 or |

φ |

CMS CMS CMS CMS CMS CMS

SF =

  • j∈jets

(1 − ǫ(pj

T, ηj))

500 1000 1500 2000 2500 3000 3500 4000 100 200 300 400 500

Events/400 GeV

JetHT L1

ε × VBF H(inv) VBF H(inv)

JetHT L1

ε × VBF H(inv) VBF H(inv)

CMS

(13 TeV)

  • 1
  • 1 fb

500 1000 1500 2000 2500 3000 3500 4000 [GeV]

jj

m

0.8 1 1.2

Data/Pred.

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 11 / 17

slide-12
SLIDE 12

SM backgrounds

◮ We have always seen the dip at 2.5-3 in leading jet η ◮ In principle, this is correlated between all regions

5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 2 4 6 8 10

3

10 ×

Events

Data VBF H(inv) Z+jets [QCD] W+jets [QCD] Z+jets [EWK] W+jets [EWK] Top Diboson QCD Data VBF H(inv) Z+jets [QCD] W+jets [QCD] Z+jets [EWK] W+jets [EWK] Top Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred. 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 1000 2000 3000 4000 5000 6000 7000 8000 9000

Events

Data W+jets [QCD] W+jets [EWK] Top Z+jets [EWK] Z+jets [QCD] Diboson QCD Data W+jets [QCD] W+jets [EWK] Top Z+jets [EWK] Z+jets [QCD] Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred. 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 100 200 300 400 500 600 700

Events

Data Z+jets [QCD] Z+jets [EWK] Top W+jets [EWK] W+jets [QCD] Diboson QCD Data Z+jets [QCD] Z+jets [EWK] Top W+jets [EWK] W+jets [QCD] Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred.

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 12 / 17

slide-13
SLIDE 13

Correcting SM backgrounds (FinOR)

◮ Region between 2.5-3 closes better after correction

5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 2 4 6 8 10

3

10 ×

Events

Data VBF H(inv) Z+jets [QCD] W+jets [QCD] Z+jets [EWK] W+jets [EWK] Top Diboson QCD Data VBF H(inv) Z+jets [QCD] W+jets [QCD] Z+jets [EWK] W+jets [EWK] Top Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred. 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 1000 2000 3000 4000 5000 6000 7000 8000 9000

Events

Data W+jets [QCD] W+jets [EWK] Top Z+jets [EWK] Z+jets [QCD] Diboson QCD Data W+jets [QCD] W+jets [EWK] Top Z+jets [EWK] Z+jets [QCD] Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred. 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 100 200 300 400 500 600

Events

Data Z+jets [QCD] Z+jets [EWK] Top W+jets [EWK] W+jets [QCD] Diboson QCD Data Z+jets [QCD] Z+jets [EWK] Top W+jets [EWK] W+jets [QCD] Diboson QCD

CMS

(13 TeV)

  • 1

36 fb 5 − 4 − 3 − 2 − 1 − 1 2 3 4 5 η Jet 1

0.5 1 1.5

Data/Pred.

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 13 / 17

slide-14
SLIDE 14

Effect on transfer factors

◮ Transfer factors essentially unchanged ◮ Effect is correlated between all regions ◮ Expect fit results to be similarly unchanged

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 14 / 17

slide-15
SLIDE 15

Shape analysis fit

◮ No significant tension introduced in fit ◮ As expected, backgrounds were being fixed even without a priori correction

Events / GeV

3 −

10

2 −

10

1 −

10 1 10

2

10

3

10

4

10

Data ) (QCD) ν ν Z( ) (QCD) ν W(l ) (EW) ν ν Z( ) (EW) ν W(l Top VV Other inv → qqH inv → ggH

(13 TeV)

  • 1

35.9 fb

CMS

Data / Pred. 0.5 1 1.5

[GeV]

jj

m

1000 2000 3000 4000 5000

Unc. (Data-Pred.) 2 − 2

Events / GeV

3 −

10

2 −

10

1 −

10 1 10

2

10

3

10

4

10

Data ) (QCD) ν µ Post-fit W( ) (QCD) ν µ Pre-fit W( ) (EW) ν µ W( Other backgrounds

(13 TeV)

  • 1

35.9 fb

CMS

Data / Pred. 0.5 1 1.5

[GeV]

jj

m

1000 2000 3000 4000 5000

Unc. (Data-Pred.) 2 − 2

Events / GeV

4 −

10

3 −

10

2 −

10

1 −

10 1 10

2

10

3

10

Data Post-fit Z(ee) (QCD) Pre-fit Z(ee) (QCD) Z(ee) (EW) Other backgrounds

(13 TeV)

  • 1

35.9 fb

CMS

Data / Pred. 0.5 1 1.5

[GeV]

jj

m

1000 2000 3000 4000 5000

Unc. (Data-Pred.) 2 − 2

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 15 / 17

slide-16
SLIDE 16

Ratios and results

)+jets ν Z(ll)+jets / W(l

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22 )+jets Data ν Z(ll)+jets / W(l )+jets MC ν Z(ll)+jets / W(l

(13 TeV)

  • 1

35.9 fb CMS [GeV]

jj

M

500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Data / Pred. 0.5 1 1.5

Transfer factor in data still well-described 20% uncertainty on correction to cover PD dependence Translates to ∼ 3% effect on signal and minor SM processes Shape analysis: observed (expected) limit goes from 0.28 (0.21) to 0.33 (0.25). C&C analysis: from 0.53 (0.27) to 0.58 (0.30).

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 16 / 17

slide-17
SLIDE 17

Conclusions

◮ VBF H→inv is one of the analyses most affected by L1 prefiring ◮ Inefficiency is computed and validated using data and propagated to analysis ◮ Results in ∼ 15% degradation of sensitivity ◮ Documentation:

◮ Changes are reflected in paper v13 and AN 2016/418 v9 ◮ Modified paper text is indicated in red ◮ We were in the middle of post-FR when this issue was discovered

◮ Paper contains combination of 2016 channels

◮ Include the statement “While the observed limit is slightly higher than the one presented

in Ref. [14], the median expectation shows a relative improvement of about 15%.”

◮ Modifications wrt pre-prefire paper are minimal. We think this is a sufficient description

  • f the effect

◮ However, it may be useful to quote the efficiency loss due to prefire. Thoughts?

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 17 / 17

slide-18
SLIDE 18

BACKUP

  • S. Narayanan

(MIT) VBF H→inv 30/07/2018 18 / 17