QCD and d the e Search rch for Exotic tic Physic ysics Holly - - PowerPoint PPT Presentation

QCD and d the e Search rch for Exotic tic Physic ysics

Holly Pacey

17/07/2019 QCD@LHC2019

2

Why?

• The physics of ATLAS pp collisions is mostly QCD.
• We produce LOADS of events with jets up to several TeV.
• BSM models like to
• Be strong: couple to quarks
• Be heavy: make high energy resonances.
• For new physics searches, being able to control and

understand the QCD background is key

3

Conte tents ts

• 1. Analyses covered
• 2. QCD background intro and modelling
• 3. General strategy
• 4. Dijet analysis
• 5. Dijet – 1 Isolated lepton analysis
• 6. Dijet – ISR Photon analysis
• 7. Conclusion

Recent ATLAS results

4

ATLAS-CONF- 2018-015

Dijet+1 Isolated Lepton

EXOT-2018-05

Dijet + 1 ISR photon

ATLAS-CONF- 2019-007

Dijet

Analyses discussed in this talk

• Too many analyses with QCD backgrounds to

discuss them all!

• Focus on recent results using di-jet

signatures: QCD dominates!

• Final states chosen to ….
• Probe models with lower Masses
• Exploit lower momentum triggering

alternatives (leptons or photons rather than jets)

• Probe models with higher masses
• Main background in all 3 is QCD multijets
• Data Driven background estimate!
• QCD MC used for validation only

BSM Models with dijet signatures

5

Techni- colour

Dark Matter

Excited Quarks

More Generic gaussian resonances! Sequential Standard Model W’

Z’

ATLAS-CONF- 2018-015

Dijet+1 Isolated Lepton

EXOT-2018-05

Dijet + 1 ISR photon

ATLAS-CONF- 2019-007

Dijet

Generic gaussian resonances

Analyses discussed in this talk

Charged Higgs

6

Dijet QCD background

• Mainly 2-2 scattering in QCD
• Dominant in t-channel →

very forward compared to possible signals

• Can place upper bound on jet

rapidity difference: 𝑧∗ = 1 2 𝑧𝑘𝑓𝑢

1 − 𝑧𝑘𝑓𝑢 2

• High dijet invariant mass Mjj

smooth & continuously falling spectrum

Mjj Log events

• Pythia 8.186 [1]
• LO NNPDF23 PDFs [2] A14

tune [3]

• Renormalisation/Factorisati
• n scales = ave. pT of 2

leading jets

• Reweight to NLOJET++

[4,5,6,7]

7

QCD Monte Carlo

Dijet+1 Isolated Lepton

Dijet

Dijet + 1 ISR photon

[See Marcello Fanti’s talk] @LO Scaled Sherpa models inclusive photon σ better than pythia in some places [21]

• Sherpa 2.1.1 [8]
• NLO ME merged with

Sherpa Parton shower [9] using ME+PS@LO [10]

• CT10 PDF [11]

• Pythia 8.186 [1]
• LO NNPDF23 PDFs [2] A14

tune [3]

• Renormalisation/Factorisati
• n scales = ave. pT of 2

leading jets

• Reweight to NLOJET++

[4,5,6,7]

8

QCD Monte Carlo

Dijet+1 Isolated Lepton

Dijet

Dijet + 1 ISR photon

• Sherpa 2.1.1 [8]
• NLO ME merged with

Sherpa Parton shower [9] using ME+PS@LO [10]

• CT10 PDF [11]

Scaled Sherpa models inclusive photon σ best [21]

(Link) NEW PUB note: Updates to ATLAS QCD MC configurations... Coming to an analysis near you!

9

Strategy – Sliding Window Fit

• Model SM bkg by fitting data Mjj distribution to a smooth function

⊚ Mainly QCD: smoothly falling background ⊚ Mjj from 2 highest 𝑞𝑈 jets ⊚ →tail: lower mass resolution/stats: wider bins. ⊚ Best function found ⊙ 𝜓2 p-value with data etc. ⊙ 𝑞5 introduced for more flexibility at low Mjj ⊙ Generally at higher Mjj:

3 param: 𝑞5 = 𝑞4 =0 or 4 param: 𝑞5 = 0

⊙ 𝑌 =

Τ 𝑁

𝑘𝑘

𝑡

Example of data & function fit shape from ISR photon analysis

10

+

+ + + + +

Data Sliding Window Estimate Function being fitted

+

• Sliding window fit to obtain

function parameters

⊚ (1) Take 𝑂 ≤ 𝑂𝑈𝑝𝑢𝑏𝑚/2 bins

(2) Fit for params (3) Set as bkg in centre bin if

good enough fit, else modify window width / function

(4) Slide along so next bin is in the centre…

⊚ Uncertainty = STDev of

poisson fluctuations of pseudodata from fit result Better than using a global fit? Better simultaneous modelling of tail AND low Mjj region

11

Strategy – Bkg Fit Validation

• Spurious signal tests

⊚ In SM BG fit – should not find

signal!

⊚ Likelihood test to compare S+B

hypothesis (MC sig+bg) to SM BG fit.

⊚ Look for bias in pull

(𝑡𝑗𝑕𝑗𝑜𝑘𝑓𝑑𝑢𝑓𝑒=0)

pull =

𝑡𝑗𝑕𝑗𝑜𝑘𝑓𝑑𝑢𝑓𝑒 −𝑡𝑗𝑕𝑓𝑦𝑢𝑠𝑏𝑑𝑢𝑓𝑒 𝑣𝑜𝑑𝑓𝑠𝑢𝑏𝑗𝑜𝑢𝑧

expect mean 0, var 1

• Signal injection tests

⊚ As above but add some given

signal events.

⊚ See how many of them can be

extracted in pull.

⊚ Ideally the fit identifies all of

the signal you added

12

Strategy – BumpHunter

• Search for local excesses in

Mjj distribution: BumpHunter! [12,13]

• Calculate p-value for bkg-
• nly hypothesis in each bin
• Use pseudo-experiments to

get global significance and most significant local excess

+ + + + +

Data Sliding Window Estimate

+ +++++++++ + + + + + + + +

Dijet Analysis /Event Selection

• 139 fb-1 data 2015-2018; 𝑡 = 13 TeV

First full run-2 dijet search! ~4x data as previous analysis @ 37fb-1

• Trigger on events with ≥1 jet with 𝑞𝑈 >420 GeV.
• Require Mjj>1.1 TeV for trigger efficiency.
• Events selected with ≥2 jets, two highest jet

𝑞𝑈s>150 GeV.

• |y*| < 0.6 to reduce forward QCD background.

13

Highest mass dijet event AntiKt4, R=0.4 Jets (backup)

Dijet Analysis /Bkg Estimate

• Sliding window fit over Mjj bins

⊚ 4 param nominal fit ⊚ Require fit to give:

(1) global chi2 p>0.05 (2) BumpHunter p>0.01

⊚ If fit fails iteratively try:

(1) 𝑞5!=0 (2) shrinking window

• Validation:

⊚ Signal Injection + spurious signal tests; robust apart

from a signal with 15% width/mass at 6 TeV – account for with a systematic uncertainty.

14

Dijet Analysis /Fit

• BumpHunter:

⊚ Biggest = 0.8 σ @ 7.052–7.326

TeV (blue lines)

⊚ Data consistent with SM

background hypothesis!

• Uncertainties:

⊚ Dominant: Choice of fit function +

params: Compare with estimate using floating 𝑞5, look at poisson variations.

⊚ Jet calibration: Jet Energy Scale:

1-3%

⊚ Mjj resolution: 2.4-2.9%

15

Dijet Analysis /Results

• Limits on q* and gaussian models used HistFitter packages [18]

⊚ 95% Confidence level on upper limit of BR x σ x acceptance ⊚ Uses CLs method with binned profile likelihood ratio [19]

16

• Exclude q* up

to 6.7 TeV! 700 GeV better than old 37fb-1 result ☺

• Also place limits on

width/mass for generic gaussian peaks

Dijet – 1 lepton /Event Selection

• First ATLAS run-2 result looking at dijet+1lepton! 79.8fb-1 data 2015-2017; 𝑡 = 13 TeV
• Add final state lepton (e or mu) to

⊚ Trigger on lepton – can probe lower Mjj>0.22 TeV ⊚ Lower QCD background! (QCD is 90-99% of bkg)

• Potential bias from leptons faking jets or signal bumps.

⊚ Study in 3 jet Control Region ⊚ Compare MC to global fit of 5param function ⊚ no deviations, function is appropriate ☺

• “looser” ID electrons more likely to be mis-reconstructed jets…

⊚ Control region: Dijet + 1 loose (and NOT tight) electron. ⊚ Use sliding window fit to compare to MC:

Simplest+best fit: 5 param <1 TeV, 3 param single fit above.

17

Dijet – 1 lepton /Bkg Estimate

• Check function is appropriate using likelihood fit

to MC QCD+Wjets+ttbar. Τ 𝜓2 𝑜𝑐𝑗𝑜𝑡 = 1.3 ☺

• Fit over Mjj

⊚ <1 TeV sliding window fit: 5 param. ⊚ >1 TeV single fit: 3 param. ⊙ Overlaps well with sliding window fit ⊙ params not constrained by low stats in

tails.

⊚ Also optimise window sizes – 20 bins best

• Validation:

⊚ Spurious signal test: 3 param function fit

less accommodating to spurious signals, equally good fit quality.

18

Dijet – 1 lepton /Fit

• Bumphunter largest deviation: 3.56 TeV

⊚ Only 0.7 σ – consistent with SM!

• Main uncertainties:

⊚ Function choice: use alternative fit

𝑞5𝑚𝑜2𝑦 → Τ 𝑞5 𝑦 <1 TeV; 𝑞5=0 >1 TeV. dominant % >4.5 TeV.

⊚ Jet Energy Scale+Resolution

dominant <1 TeV @ 1.4%

• Model dependent limits on generic

gaussian resonances for different width/mass ratios. Limits on cross section x acceptance x efficiency x BR

19

50fb @ 0.25 TeV 0.1fb @ 6 TeV

Dijet – 1 lepton /Results

• Also place model dependent limits on

W’ and technicolour models. Exclude:

20

• 1. Sequential Standard

Model W’<2 TeV

• 2. Technicolour 𝜍𝑈<0.5 TeV

1 2

Dijet – ISR Analysis /Event Selection

• First ATLAS run-2 result looking at 2 jets +1 ISR photon! 79.8fb-1 data 2015-

2017; 𝑡 = 13 TeV

• Look for lower mass events boosted by 1 ISR photon. 0.225 TeV< Mjj <1.1 TeV.

⊚ Lower mass than available with jet triggers ⊚ Higher mass than 1 large-radius jet

• 4 event categories:

⊚ 2 jets (1) any flavour, (2) b-tagged to enhance sensitivity to resonances

preferring b-quarks.

⊚ Events triggered by (A) single photon 𝐹𝑈>140 GeV, (B) combined

photon+2jet trigger with 𝑞𝑈s>50 GeV: combined has higher signal acceptance at higher Mjj

• Reduce 80% of QCD background again with |y|<0.75

21

Complements existing searches by ATLAS & CMS

Full cuts in backup

Dijet – ISR Analysis /Bkg

• Sliding window fit:

Try 5, 4, 3 param

• Signal injection tests

show can find signals with width/mass < 15%

22

Dijet – ISR Analysis /Fit

• Bumphunter shows no signs of deviations from background fit.
• Use single photon trigger for Mjj <450 GeV, combined trigger above.
• Uncertainties

⊚ Dominant:

Uncertainty

• n fit, using

2nd best function to calculate.

⊚ Signals: ⊙ Jet Energy Scale ~ 2% shift in mass distributions ⊙ In b-tagged regions: b-tagging uncertainty dominates 5-15%. ⊙ Jet Energy Resolution 3%, photon-ID 1.4-2%, trigger efficiency 3%

23

(1) (2)

Dijet – ISR Analysis/Results

• Model dependent limits placed on Z’-quark coupling in leptophobic Z’ model:

⊙ Flavour inclusive: Strong limits down to 𝑕𝑟~0.12 ⊙ B-tagged comparable

24

• Model dependent limits placed on general gaussian peaks: exclude 12fb for 400 GeV

and 5.1fb for 1050 GeV for flavour inclusive.

• Significantly extend

previous constraints by ATLAS and other experiments.

25

Dijet – ISR Analysis/Results

26

Future Outlook

26

• QCD background estimated with data-driven sliding window fit to

well-established function -> dominant uncertainties currently

• Can we improve this?
• Try different functions? Add more/fewer parameters?
• A different Data-Driven approach…
• Template method – SUSY multijets searches [22]
• ABCD method – HH → bത

𝑐bത 𝑐 [24] … with Machine Learning! – SUSY higgsino search [23]

• Weight data from Control Region – using MC and
• ther corrections – CMS dijet [25]
• Fitting the data using Machine learning, Gaussian

Processes, …

27

Summary

27

• Covered a vast portion of the Dijet mass spectrum
• QCD background dominates in Dijet events:
• Reduced by -cutting out forward regions
• adding lepton or photon to trigger on
• Estimate: -smooth continuously falling background
• data-driven sliding window fit
• Look out for more ATLAS results in the future using the full

run-2 dataset!

Dijet+ISR Dijet Dijet+1Lepton 8 Mjj TeV 6 1.1 0.22

The End

28

Thank you for listening, any questions…

ba back ckup up

Dijet+ISR analysis Signal Region Cuts

30

Jets

• Anti-Kt4 algorithm with R=0.4, using

FastJet [14][15]

• Reconstruct from calorimeter deposits.
• Calibrated Jet Energy Scale etc. to MC

“In Situ”: account for MC/data differences and detector response at different rapidities [16,17]

• Jet Vertex Tagger (JVT) for jets with

𝑞𝑈<60 GeV and |η|<2.4 to reduce impact of pileup: must be from primary interaction vertex!

31

More about BumpHunter

⊚ Accounts for look-elsewhere

effect.

⊚ Considers a range of

possible bump widths

⊚ Finds global p-value ->

Converts to significance

32

References

1. arXiv:0710.3820 2. arXiv:1207.1303 3. cds.cern.ch:1966419 4. arXiv:0110315 5. arXiv:0307268 6. arXiv:9605323 7. arXiv: 1210.0438 8. arXiv: 0811.4622 9. arXiv: 0709.1027

• 10. arXiv: 0903.1219
• 11. arXiv:1007.2241
• 12. arXiv: 0809.3781
• 13. arXiv:1101.0390
• 14. arXiv: 0802.1189
• 15. arXiv: 1111.6097
• 16. arXiv:1703.09665
• 17. arXiv: 1607.08842
• 18. arXiv:1410.1280
• 19. arXiv: 1007.1727
• 20. arXiv:1711.02692
• 21. https://doi.org/10.1016/j.physletb.201

7.04.072

• 22. arXiv:1708.02794
• 23. arXiV:1806.04030
• 24. arXiV:1804.06174
• 25. https://cds.cern.ch/record/2637847/fi

les/EXO-17-026-pas.pdf

ATLAS-CONF-2018-015: https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/CONFNOTES/ATLAS-CONF-2018-015/ EXOT-2018-05: https://www.sciencedirect.com/science/article/pii/S0370269319303612 ATLAS-CONF-2019-007: https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/CONFNOTES/ATLAS-CONF-2019-007/

33

Image credits

1. http://www.clker.com/clipart-ice-skating-3.html 2. https://www.gettyimages.co.uk/illustrations/sherlock- holmes?sort=mostpopular&mediatype=illustration&family=creative&phrase=sherlock%20holmes 3. https://www.symmetrymagazine.org/article/inside-the-large-hadron-collider 4. https://www.amazon.com/Simple-Fighter-Military-Plane-Cartoon/dp/B074KSNZLK 5. https://ya-webdesign.com/explore/borg-cube-png/ 6. https://www.the-blueprints.com/blueprints/sciencefiction/battlestar-galactica-cylon-vessels/28687/view/raider_mk- ii_fighter/ 7. https://www.amazon.com/Alliance-Millenium-Classic-Episode-Removable/dp/B01BMRCRX2 8. https://www.trekbbs.com/threads/the-enterprise-nx-01-appreciation-thread.279305/ 9. https://www.google.com/url?sa=i&source=images&cd=&cad=rja&uact=8&ved=2ahUKEwisoriX953jAhXvyYUKHXX5CRA Qjhx6BAgBEAM&url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DMUgX6b4souE&psig=AOvVaw21IZjpb1zpKla US7Nb3Ret&ust=1562421335525254 10. http://t.co/4gvi5v157v 11. https://pixels.com/featured/tardis-blue-space-koko-priyanto.html 12. https://spaceships.fandom.com/wiki/Planet_Express_Ship 13. Adultswim.com 14. https://starwars.fandom.com/wiki/Devastator/Legends 15. https://www.amazon.co.uk/Klip-Kitz-Story-Lightyear-Spaceship/dp/B0036VOIEO 16. http://papercraft.wikidot.com/printer--friendly//papercraft:firefly-serenity-spaceship

34

QCD and Searches for Exotic Physics: Signature-based searches for new physics probe the full range of the collision energy spectrum probed by the Large Hadron Collider. Often the most numerous decays,and thus those with the highest reach in new particle mass, occur in channels with at least one hadronic object and therefore requires control and understanding of the corresponding QCD

• background. This talk will cover recent search results by the ATLAS

experiment highlighting the various techniques used to suppress and understand QCD.

(Abstract)

35