Transcript 120514_RPPMontpellier

Search for the Higgs Boson
Rencontres de Physique de Particules
Montpellier
May 14, 2012
Dirk Zerwas
LAL Orsay
• Standard Model Higgs (low mass)
• Digression: Statistics
• Couplings
• Beyond the Standard Model Higgs
• Conclusions
The LHC
2011:
5.6fb-1
delivered,
4.6fb-1 to
4.9fb-1 for
analysis
Great
Startup
in 2012:
Gain
1TeV
2012:
conditions
becoming
more difficulty
The Standard Model Higgs at the LHC
• Signal dominated by gluon fusion
• VBF (qqH) next candidate
• VH smaller with large backgrounds
• ttH for higher energies
• low mass: tau (but background)
• photons (rare but pure)
• VH (difficult)
• ZZ, WW with low rates
H ττ
• usually 1 hadronic tau
• transverse mass (W+misID-jet)
• tau mass via LL technique
• cat 1:VBF
• cat 2: boosted (1j > 150GeV)
• cat 3: 0/1 j >30GeV
• Z decays dominate (resolution)
• VBF channel (2jets) as example
• better separation (recoil)
• MMC
• Z decays dominate
• 5x signal
• currently essentially inconclusive
W/Z+H WZ+bb
• BDT used (jj, separation….)
• order 100-150GeV pT
• 2 b-tagged jets
• missing ET/Z-mass
• dominant BG: top, Z/Wbb
• similar approach
• order by V pT
• S/B 1% (lowest pT) - 10% (highest pT)
• BG normalization from data (sidebands)
• 5x signal (ATLAS)
• difficult without subjet (Gavin Salam)
H WW lνlν
• mass information: weak (neutrinos)
• Spin correlations (lepton acoplanarity)
• up to 2 jets
• separate by jet-multiplicity and flavour (e/μ)
• at least 20GeV proj ETmiss
• BDT used!
• WW dominates….
• no significant excess
H WW lνlν
• good description of ETmiss necessary
• jet categories 0,1,2
• separated by flavor
• transverse mass final discriminant
• S/B order of 1/10
• compatible with bg only
HZZ 4l
• good lepton ID down to low pT
• 7/5GeV (electron/muon)
• ZZ main background
• Z+jets secondary background
• clean channel
HZZ 4l
Excellent description of BG:
• on-shell Z ok (m12)
• off/on-shell Z ok
High mass: width is width
Low mass: width is detector resolution
Hγγ
Excellent description of BG necessary
• vertex reco (83%) for mass resolution
• side-bands (power-law)
• different categories
• use of BDT based on the reconstructed
photons
Similar results with cut-based analysis
Hγγ
Understanding of BG:
• ABCD method for background
decomposition
• estimation from sideband
• search for deviation (bump hunting)
Irreducible BG > Reducible BG
Digression: Statistics
The frequentist approach (A can be repeated n times):
BAYES approach: subjective probability which includes a prior encoding a degree of belief
(more useful for an ensemble view)
H0: background hypothesis
H1: a signal hypothesis
Define a test statistic in variable t
Cut defines whether the background
hypothesis is accepted or not
p-value: probability, under assumption of
H0, to observe data with
equal or lesser compatibility with H0
relative to the data
(does not mean that H0 is true)
Significance related to n-sigma Gaussian
interpretation
A simple counting experiment
Counting experiment:
• n observed events
• s expected signal events
• b expected background events
Background free experiment:
• b=0
• exclude at 95% CL
• equivalent to one-sided Gaussian –∞ to +1.64σ (95% of total area)
• observe 0:
• P(0,s,0) = exp(-s)
• deduce s = 3
exp(-3)=0.0497
• observe 1:
• P(1,s,0) = s exp(-s)
• deduce s = 4.75 4.75*exp(-4.75) = 0.05
• translate limit into excluded signal cross section:
• σ = s/(ε * L)
And with background?
Measure n events
• determine a limit on S+B?
• subtract background
• Gaussian regime (extreme example):
• b = 990 (known perfectly: theoretical calculation)
• n = 900
• n-b = -90
• limit = -90+1.64*30 = -41
• would exclude all signals, but also the background model
• Define likelihood when n events are observed:
• Same thing for S+B
• Test statistic Q:
• In practice: large fluctuations of the background decrease the significance
More complexity
Real life: need to extend the simple Likelihood
• fit signal form: N bins
• introduce a signal strength parameter μ
• systematic errors: e.g. background is measured via M control measurements
• nuisance parameters: θ
• Test statistic:
CLs
In practice:
n=900
b=990
CLs+b small (exclusion)
CLb also small (3σ)
CLs increases (no exclusion)
• use the same test statistic for μ=1 (S+B) and μ=0 (B)
• find μ for which CLs = 0.05 (95%CL)
• In practice default method (PCL abandoned)
• Viewed critically by true statisticians:
• not a true confidence level
• over coverage
• systematic errors usually conservative
Application (ATLAS)
The complete picture
• LEE!
• about 2 sigma
Higgs couplings at the LHC
Define couplings as deviations from SM:
Restrict total width (LHC blindness):
• allow only tree-level deviations
• tree-level transported to loops
• no genuine deviations in loops
hep-ph/1205.2699
Future Higgs couplings
Near future (2012, <2020) 125GeV:
• 14TeV: major improvement
• loop couplings testable
• typically 20%
• portal order of 10%
Far future HL-LHC:
• portal: 5% (saturation)
• 7%-20% precision
• no luminosity scaling
3000 fb-1
Supersymmetry: neutral Higgs bosons
Higgs sector: mass of A, tanβ (vev ratio)
tanβ ↑: g(Hτ,b) ↑
D0:
• final states with τ and bbb
ATLAS and CMS:
• tau pair final states
• mA ↑ cross section ↓
• large exclusion with 4.6fb-1
SM-like h
mA up to 500GeV, tanβ down to 10
Supersymmetry: charged Higgs boson
Signature for m(H±) <m(top)
• top pair production
• increase decays of top to tau
• larger transverse mass
• no excess 
• exclude as function of BR
Interpretation in the MSSM:
Exclude down to 2%
Conclusions
• Higgs: not discovered yet
• interesting indications
• end of 2012 the SM Higgs case will be settled
• being optimistic: couplings will be measured
Statistics part based on lectures by Glen Cowan