High TanB Higgs Analysis Update

1
High TanB Higgs Analysis Update

• Andy HaasUniversity of Washington
• D0 Higgs Meeting
• December 11, 2003

2
A Quick Review

• MSSM has 5 Higgs BosonsWe search for h, H, A
• Production is roughly proportional to tan2B
• Signal is gt3 b-tagged jets, with a peak at a
  Higgs mass
• bh / bbh
• Backgrounds are
• jjj(j) 3 fake b-tags
• bbj(j) 1 fake b-tag
• bbbb - irreducible
• tt 1 fake b-tag
• Z(bb)b/j possible fake tags

3
Data and MC Samples

• Data is p13
• Triggers v10-113JT15 (_PV)
• 131/pb on tape86 is good
• Higgs multijet skim3 jet ETgt15, 1 jet ETgt20
  GeVetalt2.530.3 million events
• MC generated with p13.10
• Signals for bh and bbhmh 100, 120, 150 GeV
• Backgrounds
• ALPGEN bbj(j) and bbbb
• Pythia tt, Zjets, Zb
• Custom Rootuple makerhiggs_multijetapplies all
  known cal corrections, JES 4.2, b-id (SVX, CSIP,
  and JLIP), etc.

Good jet ET spectra of skimmed data sample
Simulated Higgs invariant mass peakfor mm 120
GeV/c2
4
Signal Production

• bh and bbh processes now both have NLO
  calculations!
• Events are generated with LO Pythia, then
  normalized in each pT bin, to NLO
• Uncertainties are comparable to other acceptance
  errors, 20

Higgs pT spectraat LO and NLOin bh production
Our Pythia bh sample(before re-weighting)
Higgs productionat tanB 30 at LOcalculated
with Hqq
h
A
b pT spectraat LO and NLOin bbh production
H
5
Masses, Widths, and Branchings

• Used M. Spiras HDECAY and Hqq packages to
  calculate(for various tanB)
• mh/H vs. mA
• Widths of h/H/A
• Branchings of h/H/A
• Branchings are verified to be 90 -gt bb (for
  all tanB gt 20)
• Widths verified to be less than detector
  resolution (20 GeV) for all mA lt 500 GeV/c2 and
  tanB lt 100

Mass
Width
H
H
A
h
H
h
h
tanB 30
b
b
tau
tau
g,s,mu
g,s,mu
hh
6
Object ID

• Jet ID efficiency
• Jyothsna has released a Data/MC scale factor
• The Monte Carlo models jet ID efficiency very
  well (!)
• Resulting uncertainty in acceptance is only 5
• B-id
• We have studied some cases of tagging efficiency
  and fake rates vs. Njets in data and Monte Carlo
• Overall b-tagging acceptance uncertainty is 15

7
Triggers

• Well understood triggersCJT(3,5)L2J(3,8)
  L2Ht(50)L3J(3,15)(2,25)_PV(Z)
• Calculate a trigger pass probability for each
  Monte Carlo event
• Acceptance uncertainty is lt 10

Level 1Turn-onStudy
Event by eventtrigger pass probability for
signalwith mh120 GeV/c2
8
QCD Background

• The ALPGEN cross-sections for bbj(j) are
  normalized to data using the double b-tagged
  events
• The correction factor is small (!)
• The shapes agree very well
• Fakes are estimated using the single b-tagged
  data
• Parameterized by pT of the jet

Invariant mass inthe double b-tag data
Fakes
bbjj MC
9
Triple b-tagged Data

• We have a good understanding of the source of the
  triple b-tagged events
• QCD Fakes (jjjj) extrapolated from data in
  the double b-tagged sample
• QCD Heavy Flavor (bbjj) events
• Small contributions from Other processes
  generated with Pythia
• Z(bb)jets
• Z(bb)b/j new MC generated last week !
• tt and bbbb
• Shapes are well modeled by the Monte Carlo, and
  the bbjj normalization is still very close to
  unity (!)
• For setting limits, we use a fit of the
  extrapolated double b-tagged data to the triple
  b-tagged spectrum, outside the 1 sigma Higgs mass
  window

Data bbbj
MC bbjj
Data jjjj
Other (Z,bbbb,tt)
Signal (mh120, tanB50)
Total Background(from data)
10
Optimization

• An optimization of signal significance is
  performed for each mh and for njmin 3 and 4
• The signal MC is compared to the bbjj MC in the
  triple b-tagged inv. mass spectrum
• No large gains are observed above the sensible
  cuts we used before
• 4gt15, 2gt35, 1gt45 GeV
• nj4
• etalt2.5

Loose initial cuts
11
Systematic Errors

• Significant work has gone into calculating errors
  conservatively
• Acceptance errors
• Luminosity
• Trigger efficiency
• Vary by - 1 sigma
• Jet energy resolution affects the width of the
  Higgs signal peak
• JES affects efficiency for passing analysis
  cuts
• Jet ID efficiency affects efficiency
• B-id efficiency affects efficiency
• Background errors are measured by
• Statistical error (times X2) on background fit to
  data
• Fit errors on extrapolation from double to triple
  b-tagged data (shape of spectrum)

Acceptance Errors ()
Background Normalization Errors ()
12
Current Results

• Using a Gaussian peak weighting of the triple
  b-tagged data, after optimized cuts, for each mh
• Limits set using Hobbss Simple Limit machinery
• njmin 3 is significantly more sensitive
• Takes advantage of both the bh and bbh production

CDF Run I (bbh/H/A only)
njmin4, h/H/A
SHWG (1999)(bbh/H/A only)
njmin3, h/H/A
Sweet spot
Excluded by LEP
13
Conclusions / Future

• This analysis has been run from beginning to end
  (!)
• We believe the p13 data is relatively well
  understood
• All MC samples we think we need have now been
  included (Zb was the last remaining)
• The production and decays of the MSSM Higgses are
  on firm ground (at NLO)
• Our sensitivity has been optimized (with square
  cuts)
• The most obvious systematic errors have been
  calculated
• Good agreement with previous studies is
  observed... (despite many changes in
  understanding of signals and backgrounds)

• D0 Note 4290
• First draft is finally done ?
• Now being reviewed by the experts
• Will be released to the Higgs group for review
  soon... then sent to an EB for Moriond, hopefully