1. Di-jets & MT2 for early SUSY discovery
Alan Barr (Oxford) Claire Gwenlan (UCL)
Why is MT2 any good as a discriminating variable?
Is the MT2 dijets analysis useful for discovery in early data?
Do we have the appropriate Monte Carlo to study this channel?
6 June 2007
Barr & Gwenlan

2. Topologies of interest
LSP q
squark q _ q _ q
BACKGROUND topology (QCD)
(and similar)
LSP
Concentrate on small number of high-pT jets
Large signal cross-section
Large control statistics
Relatively well known SM backgrounds
Relatively model independent
Does not rely on leptonic cascades
Does not rely on hadronic cascades
Use kinematics rather than “business of event” to pick out SUSY
SIGNAL topology
6 June 2007
Barr & Gwenlan

3. Reminder of MT2
Used in events where two identical particles decay semi-invisibly
Hard Scatter
J.Phys.G29: , Phys.Lett.B463:99-103,1999
“Try all possible directions for the neutralinos and find the minimum heavy sparticle mass”
Variable has a good history…
MT2 first used for sleptons (Cambridge)
Squark masses from dijets in DC1 (Milano)
Sleptonic channel further investigated…
Masses (Pavia)
Spins (UCL)
Di-jets analysis further studied for CSC (Belgrade)
6 June 2007
Barr & Gwenlan

4. MT2 for discovery?
It is a property of the variable that MT2(χ=0) → 0 if:
ET → 0
ET parallel to either jet
Either jet ET → 0
Expect small MT2 for backgrounds from
decays of “light” semi-invisible particles
events with small ET
mis-measurement of a single jet energy
This includes: WW, ttbar, QCD fakes, neutrinos in jets, …
No a-priori reason to expect MT2 to be small for e.g. MET coming from Z￫ + associated ISR
Expect this to dominate at larger MT2
Very nice features! This is why MT2 is a useful discriminator for discovery
6 June 2007
Barr & Gwenlan

5. Simple Analysis MT2 already “does the job” of traditional cuts:
-> Dππ (δφ) cut
-> MET cut
-> Transverse sphericity cut
(small if MET parallel to either jet)
(small if MET → 0)
(small if 2-jets back-to-back)
Go for “simple approach”
Cuts
At least two jets with:
ETJet1,2 > 150,100 GeV
|Jet1,2| < 2.5
(simple two-jet cuts only)
plot MT2
NB: We don’t claim you’d just plot MT2 and publish(!) but a simple selection can be easier to e.g. calculate systematics, and so speeds up the whole process
6 June 2007
Barr & Gwenlan

6. Dijet cuts only 1 fb-1 Two-Jet Dijet inclusive: - No lepton veto
- No b-jet veto
- No multi-jet veto
6 June 2007
Barr & Gwenlan

7. Dijet cuts only Two-Jet Dijet inclusive: - No lepton veto
low mass region (SU4)
Dijet inclusive:
- No lepton veto
- No b-jet veto
- No multi-jet veto
bulk region (SU3)
coannihilation (SU1)
6 June 2007
Barr & Gwenlan

8. 1fb-1 mT2 > 200 GeV mT2 > 400 GeV
Dijet cuts only
Two-Jet
low mass region (SU4)
1fb-1
mT2 > 200 GeV
mT2 > 400 GeV
Model
S/sqrt(B)
S/sqrt(S+B)
SU1 24 23 80 36
SU2
1.3
1.9
1.8
SU3 52 45
110 44
SU4
180
120 64 32
SU6 15 54 28
SU8.1 22 20 74 34
bulk region (SU3)
bulk region (SU3)
bulk region (SU3)
coannihilation (SU3)
coannihilation (SU3)
coannihilation (SU3)
6 June 2007
Barr & Gwenlan

9. Di-jet analysis without MT2?
More traditionally (since TDR), Meff distribution used as a discriminator
Cuts
at least two jets with:
ETJet1,2 > 150,100 GeV
|Jet1,2| < 2.5
Meff = Jets pTi + MET
6 June 2007
Barr & Gwenlan

10. No MT2 Dijet cuts only Meff alone gives large QCD BD Barr & Gwenlan
6 June 2007
Barr & Gwenlan

11. Di-jet analysis without MT2?
More traditionally (since TDR), Meff distribution used as a discriminator
Cuts
at least two jets with:
ETJet1,2 > 150,100 GeV
|Jet1,2| < 2.5
Meff = Jets pTi + MET
But with addition of some other cuts…
 MET > 100 GeV
 cuts based on i = |(Jet,i)-(MET)|):
R1 = (22+(-1)2) > 0.5 rad
R2 = (12+ (-2)2) > 0.5 rad
no jet with i < 0.5 rad
MT2 basically “does this job” implicitly
6 June 2007
Barr & Gwenlan

12. No MT2 Dijet cuts + MET + 
Two-Jet
6 June 2007
Barr & Gwenlan

13. 1fb-1 meff > 350 GeV meff > 700 GeV No MT2 Dijet cuts + MET + 
Two-Jet
1fb-1
meff > 350 GeV
meff > 700 GeV
Model
S/sqrt(B)
S/sqrt(S+B)
SU1 27 24 55 37
SU2
0.8
1.4
SU3 58 47
110 61
SU4
320
160
400
130
SU6 14 13 29 23
SU8.1 22 20 46 33
6 June 2007
Barr & Gwenlan

14. Useful for early discovery?
Need:
Some understanding of ET and energy scale
Degree needs to be determined
Some lepton ID
Estimate Z→νν from Z→ μμ
Some idea of ttbar background
“Do not need”:
B-tagging
Only if needed to measure ttbar background
Detailed understanding of jet resolution tails
In the limit where only one jet per event fluctuates
Missing ET tails from multi-jets
Need to validate the above statements:
Need 2-parton Alpgen to validate against 2->2 MC
QCD and Drell Yan backgrounds
Study effect of extra jet mis-calibration/resolution
6 June 2007
Barr & Gwenlan

15. Simulation and Trigger?
Monte Carlo
Currently have to use CSC Pythia/Herwig MC
Existing SUSY Alpgen has 4-jet truth filter
Probably not ideal for these types of cuts
Z+2 jets, W+2 jets not well modelled
Would be good to cross-check against 2 ￫ n parton MC
Trigger?
“SUSY” 4-jet trigger no good
Existing jet/MET triggers probably sufficient
Including pre-scales for lower threshold jets
Needs to be confirmed
6 June 2007
Barr & Gwenlan

16. 3-jet also sensitive No MT2 Tri-jet cuts + MET +  Three-Jet
Hundreds of signal events rapidly
6 June 2007
Barr & Gwenlan

17. 3-jet also sensitive 1fb-1 meff > 350 GeV meff > 700 GeV
No MT2 Tri-jet cuts + MET + 
3-jet also sensitive
Three-Jet
1fb-1
meff > 350 GeV
meff > 700 GeV
Model
S/sqrt(B)
S/sqrt(S+B)
SU1 30 21 40 23
SU2
1.6
1.5
2.0
1.9
SU3 61 34 80 36
SU4
200 69
250 68
SU6 18 14 24 16
SU8.1 27 19 35
Hundreds of signal events rapidly
6 June 2007
Barr & Gwenlan

18. Summary Dijets and trijets + MET have good SUSY sensitivity
Large cross-section
High S/√B than multi-jet at low lumi
Goods stats signal & control regions
Should be part of early search strategy
MT2 ‘does the job’ of several traditional cuts
Expected from its properties
Combination of {MET, sphericity, }
Reduce # cuts  simplfy analysis?
Needed:
Monte Carlo to better cover this region
Geant and ATLFAST
Same true for lepton(s) + 2/3 jets + MET channels?
Not just “4 or more jets”!
6 June 2007
Barr & Gwenlan

19. Backups
6 June 2007
Barr & Gwenlan

20. V12 BACKGROUND SAMPLES USED (SO FAR)J1
trig1_misal1_csc J1_pythia_jetjet.recon.AOD.v * J2
trig1_misal1_csc J2_pythia_jetjet.recon.AOD.v * J3
trig1_misal1_csc J3_pythia_jetjet.recon.AOD.v * J4
trig1_misal1_csc J4_pythia_jetjet.recon.AOD.v * J5
trig1_misal1_csc11_V J5_pythia_jetjet.recon.AOD.v * J6
trig1_misal1_csc J6_pythia_jetjet.recon.AOD.v * J7
trig1_misal1_csc J7_pythia_jetjet.recon.AOD.v * J8 - T1
trig1_misal1_mc T1_McAtNlo_Jimmy.recon.AOD.v *
TTbar
trig1_misal1_mc TTbar_FullHad_McAtNlo_Jimmy.recon.AOD.v *
Zee
trig1_misal1_mc McAtNloZee.recon.AOD.v *
Zmumu
trig1_misal1_mc McAtNloZmumu.recon.AOD.v *
Ztautau
trig1_misal1_csc PythiaZtautau.recon.AOD.v *
Znunu
trig1_misal1_mc pythia_Znunu_qg_Nj2.recon.AOD.v *
Wenu
trig1_misal1_csc JimmyWenu.recon.AOD.v *
Wmunu
trig1_misal1_csc JimmyWmunu.recon.AOD.v *
Wtaunu
trig1_misal1_csc pythia_Wtaunu_qg_Nj2.recon.AOD.v *
6 June 2007
Barr & Gwenlan

21. SUSY POINTS Point m_0 (GeV) m_1/2 (GeV) A0 (GeV) tan() sign()  (pb)
Coannihilation (SU1) 70
350 10 +
7.43
Focus Point (SU2)
3550
300
4.86
Bulk (SU3)
100
-300 6
18.59
Low Mass (SU4)
200
160
-400
262
Funnel (SU6)
320
375 50
4.48
Coannihilation (SU8.1)
210
360 40
6.44
Coannihilation (SU8.2)
215
6.40
Coannihilation (SU8.3)
225
6.32
6 June 2007
Barr & Gwenlan

22. Dijet cuts only
MET distribution
6 June 2007
Barr & Gwenlan

23. Do not believe the backgrounds!
4-jets + MET + dφ
Four-Jet
Do not believe the backgrounds!
6 June 2007
Barr & Gwenlan

24. 1fb-1 meff > 350 GeV meff > 700 GeV 4-jets + MET + dφ Four-Jet
Model
S/sqrt(B)
S/sqrt(S+B)
SU1 19 12
SU2
2.5
2.3
SU3 40 18
SU4
120 34
110
SU6 15 10
SU8.1 11
Do not believe the backgrounds!
6 June 2007
Barr & Gwenlan

25. Cuts MT2 plot  at least two jets with: ETJet1,2 > 150,100 GeV
3-jet plot
 at least three jets with:
ETJet1,2,3 > 150,150,100 GeV
|Jet1,2,3| < 2.5
 MET > 100 GeV
Dijet MEff plot
 at least two jets with:
ETJet1,2 > 150,100 GeV
|Jet1,2| < 2.5
 MET > 100 GeV
 cuts based on i = |(Jet,i)-(MET)|):
R1 = (22+(-1)2) > 0.5 rad
R2 = (12+ (-2)2) > 0.5 rad
no jet with i < 0.5 rad
 cuts based on i = |(Jet,i)-(MET)|):
R1 = (22+(-1)2) > 0.5 rad
R2 = (12+ (-2)2) > 0.5 rad
no jet with i < 0.5 rad
 |Jet1| < 1.5
6 June 2007
Barr & Gwenlan

26. 100 fb-1 Plots and significances
6 June 2007
Barr & Gwenlan

27. Dijet cuts only
Two-Jet
6 June 2007
Barr & Gwenlan

28. 100 pb-1 mT2 > 200 GeV mT2 > 400 GeV Dijet cuts only Two-Jet
Model
S/sqrt(B)
S/sqrt(S+B)
SU1
7.7
7.1 25 11
SU2
0.42
0.59
0.57
SU3 16 14 35
SU4 57 39 20 10
SU6
4.8
4.6 17
8.9
SU8.1
6.9
6.5 23
6 June 2007
Barr & Gwenlan

29. dijet + MET + dφ
Two-Jet
6 June 2007
Barr & Gwenlan

30. 100 pb-1 meff > 350 GeV meff > 700 GeV dijet + MET + dφ Two-Jet
Model
S/sqrt(B)
S/sqrt(S+B)
SU1
8.6
7.6 17 12
SU2
0.25
0.43
SU3 18 15 35 19
SU4
100 49
130 42
SU6
4.4
4.2
9.2
7.2
SU8.1
7.1
6.4 14 10
6 June 2007
Barr & Gwenlan

31. Some MC Issues/Questions
1. Different samples used for Z->nunu backgrounds in v11 and v12, giving rather different results at low MT2. Comparing JO’s:
 v11, sample used:
csc PythiaZnunu.recon.AOD.v *
- has lower pTZ cut of 50 GeV
(also has *no* line which turns off FSR: ‘pydat1 parj ’ – hopefully dosen’t make a difference?)
 v12, sample used: trig1_misal1_mc pythia_Znunu_qg_Nj2.recon.AOD.v *
- no lower pTZ cut
- requires 2 jets at truth level (Cone4) with pT1 > 80 GeV, pT2 > 40 GeV
(I am weighting this sample by a “cross section x filter efficiency” of 58 pb, which I got from: However, I can’t find any other information anywhere else (e.g. on AMI) to help me verify that this is the correct number for PYTHIA, and haven’t yet had time to check it ourselves “by hand”.
NOTES on Jets Samples used: So far, have just used the PYTHIA samples. In the v12 plot the J1-J6 samples have the have 30 micron step length while J7 suffers from 1mm bug (as no samples available with 30 micron yet) – this should be a small effect presumably. Also in the v12 plot, the J8 sample has not been included as couldn’t get any samples, which had MET_RefFinal (which is what I used for J1-J7) – however, inclusion or not of J8 sample in the plot gives only small differences .
ALPGEN samples study is now in progress but it is difficult to perform full study with current available samples e.g. there are not yet any 2-Jet ALPGEN QCD samples, and not sure that there are all the necessary Drell-Yan samples either e.g. WmunuNp2, ZnunuNp2, etc.)
Other samples that might be useful: Znunu? (already using for Zee, Zmumu)
6 June 2007
Barr & Gwenlan