1. Search for invisible Higgs in tth0 production
Motivation
Review of preliminary analysis
Event generation:
Comparison of Pythia
Ricardo Gonçalo, RHUL

2. Introduction Motivation: Previous analyses:
Several scenarios for physics BSM predict a significant invisible branching ratio:
The simplest is the Higgs decay to neutralinos (probably disfavoured by the MSSM)
New scenario proposed: D.Cerdeño, A.Dedes and T.Underwood (hep-ph/ ) which includes a light, invisibly decaying Higgs)
Complementary to the Vector Boson Fusion and associated ZH production with invisible Higgs in case of suppressed coupling of Higgs to vector bosons (MSSM “intermediate regime” mA < 500 GeV; small tan)
Aim to contribute to the Invisible Higgs CSC note
Previous analyses:
Within Atlas (that I’m aware of): I must review these
Paper by E.Richter-Was, B.Kersevan and M.Malawski
M.Malawski - MSc Thesis
Dietrich Schroff - PhD Thesis
Frank Meisel – Diplomarbeit
My own preliminary analysis presented in Higgs WG meeting (3 Nov.04) and in local meetings (1 Nov.04, 22 Apr.05): overview and main results shown below
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

3. Overview of preliminary analysis
Cut-based analysis
Event simulation with ATLFAST
Only signal and tt background
Difficulties:
Two components of missing momentum (from Higgs and neutrino)  can’t reconstruct tbl without further assumptions
tt is the most significant background and is similar to signal: most other backgrounds reducible
ttZ irreducible but cross section small (well…comparable to signal, must be included!)
Signal/Background ~10-3
Process
xBR
tth
330 (*) fb tt fb
bbW, W  l fb
bbZ, Z  l+l- fb
ttW, W  l
420 fb
ttZ, Z  
190 fb
(*) =520 fb in reference analysis, with SM couplings
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

4. Signal and tt background
Transverse mass, ET and ETmiss are good discriminating variables
mT has sharp edge for tt background at ~mW
But background xsection ~1000 times higher than signal
Tails of background distributions very large
ttbar
tth
mT(GeV)
ETmiss (GeV)
ET (GeV)
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06
No cuts; Lumi = 1fb-1

5. Preliminary analysis: cuts
Reference analysis cuts:
Leptons:
1 electron (pT>25GeV; ||<2.5)
or 1 muon (pT>20GeV; ||<2.5)
Veto on additional electron (pT>10GeV)
or additional muon (pT>6GeV)
Jets:
2 b-tagged jets
2 or more un-tagged jets
tbjj reconstruction:
|mjj-mW|<15GeV; ||<2.0 for jets in Wjj
|mbjj-mt|<25GeV
tbl reconstruction:
mT > 120 GeV (using lepton and full ETmiss to calculate mT)
Other:
Missing ET > 150GeV
Scalar sum of pT of reconstructed l j j b b: ET>250GeV
In reconstructed W  jj: Rjj =(2jj+2jj) < 2.2 (to reject lep-tau decays) 
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

6. Preliminary analysis: comparison with reference analysis (Richter-Was et al)
Cumulative efficiency of cuts:
~10-20% agreement with reference analysis for signal (see “cross check” column)
Slightly worse agreement wrt tt (factor 1.35 – 2.8) for cuts after mT cut
Testing all tbjj combinations against mW and mt gives efficiency ~30% better wrt “cross check” (“this analysis” column) both for signal and background  (small) net gain in significance
Cut
Reference tth
Cross check tth
this analysis tth
Reference tt
Cross check tt
this analysis tt
Lepton
22%
23.6%
23.7%
Jets/b jets
5.0%
5.7%
4.9%
5.01%
5.02%
t bjj
2.6%
2.9%
3.4%
2.4%
2.60%
2.98% mT
0.87%
1.12%
1.33%
0.041%
0.055%
0.063%
ETmiss
0.41%
0.51%
0.67%
2.0x10-5
3.8x10-5 ±0.6
4.9x10-5
ET
0.40%
0.50%
3.7x10-5 ±0.6
4.6x10-5
Rjj
0.28%
0.33%
0.44%
7.5x10-6
2.1x10-5 ±0.5
2.8x10-5
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

7. Preliminary analysis: initial results
Most background comes from lep-tau and lep-lep decays of tt, as concluded by E.Richter-Was et al.
 decays increase the missing ET
Wjj reconstructed from ISR/FSR jets in lep-tau and lep-lep events
For L=30 fb-1, scaling tth cross section to SM value (as in reference analysis):
tth: = 520 fb; 44.3 events
tt: = 490x103 fb; 812 events
S/B = 1.55
lep-lep
lep-tau
tau-tau
lep-had
had-had
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

8. Preliminary analysis: kinematic fit
Tried something different:
Assume pTmiss comes from  and h0 only
Build grid of points in pT and (l,) and calculate pZ
pZ can be found from pT and plep assuming W on mass-shell
From pb , p and plep, calculate mt for each point
Propagate errors in mt from grid spacing to obtain mt and calculate 2 mW W b t
e/  mt
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

9. Preliminary analysis: kinematic fit
Kinematic fit ~works for signal: (pTh) ~ 85GeV
Will try to use fit results for discrimination against tt background
The hope was that this would allow other cuts to be relaxed
Correlations to mT and ETmiss may be important
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

10. New variables to use in selection?
Higgs pT from the kinematic fit
Signal tt background
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

11. Cut on l,Etmiss
Naively expect lepton to be mainly away from ETmiss direction in signal and close to ETmiss direction in background (lepton and  from W decay)
In fact, most tt background which passes standard cuts comes from lep-tau with ETmiss from both  and W decay and one detected lepton
ttbar
tth
Truth from angular distance between true lepton and missing ET
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

12. Preliminary analysis: optimisation
Varied several cuts over a few threshold values to optimise significance = S/(S+B)
All cut combinations considered
Cuts varied:
mT (standard cut: > 120 GeV)
ETmiss (standard cut: > 150GeV)
Scalar sum of pT of reconstructed l j j b b: ET (standard cut: >250GeV
Added pTH obtained from kinematic fit
Added  between reconstructed lepton and ETmiss
Also tried varying electron and muon pT cuts
Significance increased from ~1.6 to ~3
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

13. Preliminary analysis: optimisation
Some cuts have more impact than others
Results point to optimal thresholds:
mTW cut ~ 115GeV
ETmiss cut ~ 220 GeV
l,Etmiss cut < 2.8 rad
No clear solution for ET or pTh,fit
But: correlations between variables are large
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

14. CSC analysis
Aim to re-do and extend preliminary analysis for Invisible Higgs CSC note
Signal simulation:
Fully simulated tth0 (h0inv.) sample requested – to be simulated with Pythia 6.4
Using Pythia and forcing invisible decay of Higgs: h0  ZZ  4
Background sample must be produced independently on the grid
Background rejection factor ~10-5  need O( M) events or efficient Generator level filtering
New PtmissAndOrLepton filter written and in
Based on Pauline Gagnon’s PtmissPlusLeptonFilter
Request missing ET and/or one lepton (configurable)
Different pT thresholds for e and 
Expected efficiency for tt ~10% : 1 e (pT>12.5, ||<2.7) or 1  and missing ET > 100 GeV
It may have problems: will have to revisit this carefully
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

15. Study of Pythia versions/PS
Pythia parton-showers (PS) and multiple-interactions changed dramatically since Pythia 6.2
Using validation samples (generated by Bruce Mellado, thanks!) to evaluate the differences
Single lepton filter with pT>10GeV and ||<2.7
Pythia
ggtth0 + qqtth0
Nr.events
Passed filter
Filter efficiency
Equivalent luminosity
6.323, old PS
350.6 fb
20000
11618
58.1%
57.0 fb-1
6.323, new PS
350.8 fb
11486
57.4%
6.403 (new)
351.4 fb
30000
16879
56.3%
85.4 fb-1
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

16. Comparison Pythia 6.3/6.4/new/old PS
Following plots compare Pythia and with “new” and “old” parton shower (PS) and multiple interactions (MI) algorithms
The label “new PS” is used for pT-ordered parton showers and “interleaved” PS-MI model (PS and MI compete for available phase space)
“old PS” refers to the Pythia < 6.3 algorithm; still available but not default in 6.3 and 6.4
Bjorken x calculation uses first daughters of interacting protons in ev.record using barcode: not always right!!
Code in: Generators /GenAnalysisTools /CBNT_Truth/src/CBNT_SpclMC.cxx
Different parton shower algorithms clearly changed the ordering of outcoming partons
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

17. Observable quantities: Missing ET truth
Differences don’t look that huge (well...not compared to VBF!)
But the cuts on ETmiss and ET will be in poorly understood regions
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

18. tth truth: transverse momenta
pTtth probably shows the largest differences between versions
The tth system recoils against ISR and multiple interactions
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

19. Conclusions First steps under way to generated data for CSC analysis
Signal will be produced with Pythia 6.4 (new PS)
Background (ttbar, ttZ (Z), ?…) will be done in own production
New generator filter to provide necessary ttbar background sample
Looking at differences between Pythia versions for this channel
Differences not huge, but may be significant
Could provide estimate of systematic uncertainty?
Should try other generators?
Signal sample requested
Will be produced with Pythia 6.4 (new PS)
A lot of work to be done to revive this analysis…
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

20. Backup slides
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

21. The truth! At parton level: tops back-to-back in background
more “mercedes star”-like in signal
Most missing pT from Higgs decay (especially in had-had channel)
(,ETmiss) 
(h,ETmiss)
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

22. Preliminary analysis: simulation
Channel: Higgs (mh = 120 GeV) decaying to neutralinos in MSSM
(tan= 5, mA= 1 TeV; M1= 44 GeV, M2= 220 GeV, M3= 1 TeV)
PYTHIA for signal and background
Generated 60 M tt + 10 M tth
Atlfast simulation, ATLAS release 7.0.2
Low luminosity setting
Cone jets (Rcone = 0.4)
Jet tagging: b jets 60%; c mistag 10%; u,d,s, mistag 1%
CTEQ5L PDFs
mtop = 175 GeV
Interfaced own code to Atlfast within Athena to produce dedicated ntuple
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

23. Azimuthal angle difference between 3-momenta of tops and Higgs
Small differences between Pythia 6.3/6.4 and PS versions (right)
Compare to previous plot from Pythia 6.2 (left)
Labels:
t1 = leading top (highest pT)
t2 = sub-leading top
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

24. Observable quantities: Missing ET methods
ETmiss estimated using different methods
Reconstruction/MissingET/src/CBNT_MissingETtruth.cxx
Plotting angle between ETmiss estimates…not sure I understand the differences between methods
But they are similar for all Pythia/PS versions
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

25. Observable quantities: lepton pT & isolation
Looks ok
Left plot: with isolation
Right plot: with sub-leading lepton (note lepton filter on leading lepton)
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

26. Observable quantities: transverse mass
Plot on the right: events with a single lepton (e or )
Compare with plots below
Cut on mT at ~115GeV
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

27. tth truth: angles between true momenta
Some difference between versions; easier to see in polar angle
Pythia 6.4 seems to give an intermediate result between 6.3 new/old PS
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

28. Optimised cuts Signal… … and tt background Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

29. Sum ET Signal… … and tt background Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

30. Transverse W mass Signal… … and tt background Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

31. Lepton and jet variables
Signal… … and tt background
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

32. Not using Rjj as selection variable
Signal… … and tt background
Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

33. Higgs pT from fit Signal… … and tt background Ricardo Gonçalo
ATLAS-UK - Durham Sep.06

34. l,Etmiss Signal… … and tt background Ricardo Gonçalo
ATLAS-UK - Durham Sep.06