Diapositiva 1

1
Z?bb measurement in ATLAS
Iacopo Vivarelli, Alberto Annovi
Scuola Normale Superiore,University and INFN
Pisa, University of Athens
2
Introduction

• We studied the Z decay into bb
• Signal and background generation
• LVL1 requirements (with full simulation)
• Kinematical selection
• Results (and comments)

3
Signal and background generation

• Both signal and background generated with PYTHIA
  6.203
• Signal (Z?bb) MSEL11 s 9.1 nb
• Background (Generic QCD) MSEL1 with in 6 QT
  bins (beginning with QT gt 10 GeV) s 9.45 mb.
  Cross check on the accepted cross section made
  with Alpgen generator. Results are consistent ?
  Uncertainties on the background under control (to
  our best knowledge). eb50, ec10, ej 1
• Underlying event in PYTHIA tuned on the CDF data
  (A. Moraes et. al.). Initial an final state
  radiation included. Pileup effects not included
• Simulation/reconstruction made using ATLFAST-OO
  in Athena version 8.5.0.

4
General issues
The final signal to background ratio is expected
to be about 1. To avoid the background shape
prediction with the MonteCarlo, it is necessary
to subctract the background directly from the
data, i.e., to have a lower and a higher sideband
to allow the background evaluation. Too low
thresholds have to be considered on the two
b-jets if one consider the di-jet final state.
Too large trigger rates foreseen for threshold
low enough to have safe sidebands. A different
strategy has to be followed look at boosted Z
5
3 jet selection
We require a leading non-b jet. This decreases
the LVL1 rate, and moves at low masses the
trigger  turn-on in the background invariant mass
distribution
The reason is the following requiring the
leading jet to be non-b, one strongly reduces the
contribution from direct bb production and
selects mainly gluon splitting events (mainly
gg?gg?bbg). They are characterized by low
invariant mass of the bb couple, because of the
small angle between the b-jets
?Rbb
6
3 jet selection (3)
To reduce further the LVL1 rate the trigger
turn-on mass, a hard selection on the leading jet
has been tried. A leading jet of PT gt 80 GeV is
required for the plots below. This reduces the
value of the peak in the background invariant
mass.
50 Gev
Mbb
?Rbb
7
Trigger issues

• Lowering as much as possible the LVL1 thresholds
  for jets. This could be possible thanks to
  b-tagging at LVL2
• Hypothesis
• b-tagging available at LVL2
• Few KHz available at LVL1.

My eb is 50

• In fact
• The LVL1 bandwith is limited by the LVL2
  rejection
• LVL2 rejection can be improved if one is
  interested in b-jets make use of b-tagging at
  LVL2
• The LVL2 performances for the b-tagging are
  limited w.r.t. the offline, but they dont
  introduce biases (HLT TDR).
• I assume a LVL2 rejection factor 2 coming from
  better resolution on jets (ATL-DAQ-2000-015) plus
  a factor 5 for each b-jet required.

40ltPTlt70 70ltPTlt100 PTgt100
? lt 1.5 eb60 5.9 5.1 4.8
eb70 3.6 3.9 3.1
eb80 2.3 2.8 2.4
? gt 1.5 eb60 3.7 4.9 3.0
eb70 2.6 2.9 2.6
eb80 1.5 1.6 1.7
8
Trigger issues (3)

• The final state we are interested in consists of
  one hard jets two soft jet.
• Strategy at LVL1 trigger on the hard jet and
  identify a RoI for the soft jets. At LVL2 ask for
  two b-tag (with reduced performances first, then
  with offline performances).
• What needs to be understood?
• What are the thresholds to be used on LVL1 jets
  to get a reasonable rate
• What is the efficiency w.r.t. true jets
• What is the threshold to be set for the
  efficient identification of the RoI corresponding
  to the soft jet.

9
Trigger issues (4)
Full simulated QCD di-jet events with QT gt 17 GeV
with lumi02 pileup included have been used.
Efficiencies have been computed on QCD jets. To
be repeated for the signal. Athena 8.5.0 has been
used. LVL1 jets have been reconstructed on a 8x8
window. The truth is defined as jets
reconstructed using a cone algorithm (?R0.4) on
MC particles. It is the standard truth
reconstruction for jets. (The rate of the LVL1
multijet triggers of the HLT TDR can be
reproduced with a 10 accuracy)
Use only LVL1 jets within 2.4 (it can be done),
since at further trigger levels they have to be
identified as b-jets. Check the rate for a di-jet
trigger. Consider as acceptable a LVL1 rate of 5
KHz.
10
Trigger efficiencies
Match the LVL1 jet with the MC jet if ?R between
them is less than 0.4. Fill histograms with the
ET at LVL1 for a given MC transverse energy.
Then, find the ET of the LVL1 jet to be 95
efficient for the corresponding MC energy.
11
Trigger efficiencies
The discussed threshold is 95 efficient on 190
GeV MC jets.
12
RoI ID for the soft jet
With the same definition of matching between the
LVL1 and the MC jet, a threshold at 10 GeV (at
LVL1) is 90 efficient on a 40 GeV jet
13
Trigger issues summary
A rate of 4.5-5 KHz can be obtained with a single
threshold of 190 GeV at LVL1. This corresponds
(with high efficiency) to MC jets of 190 GeV. The
soft jet (40 GeV) can be identified as a RoI if a
10 GeV threshold is applied at LVL1. LVL2 ask 2
b-tag with reduced rejection. With the present
performances, the LVL2 rate is about 100 Hz. If
one considers offline performances the rate is 10
Hz. The question now is can the LVL2 b-tag at
that rate if the tracking is made by the LVL2
itself? Which is the maximum rate that can be
analyzed at LVL2 if the tracking is done by FTK?
It needs more understanding
14
Offline selection

• The softest selection allowed by the single jet
  threshold is
• Leading jet of 190 GeV. It is required not to be
  b-tagged
• Two b-jet of 40 GeV
• The statistics available for the background is
  much lower than the expected number of QCD events
  after few tens of fb-1. The distribution is well
  shaped down to 50 GeV. An exponential function
  fits well this low statistic distribution

15
Final results
The parametrized distribution for the background
and the signal have been normalized with the
corrected statistics and then
Signal QCD
LVL1 Rate KHz - 4.9
LVL2 Rate Hz - 98-10
Kin. Sel2b fb 1098 195000
80ltMbblt100 fb 551 21100
S/B S/vB 30 fb-1
2.5 20.6
16
Conclusions

• Different trigger/selection strategies are
  available
• The signal can be triggered (good results with
  5KHz available at LVL1)
• The signal can be reconstructed
• The biggest uncertainties come from the real
  detector performances (b-tagging, low PT b-jet
  reconstruction efficiency, real efficiency of the
  trigger on the signal. So far only ATLFAST has
  been used). Nevertheless, the selection is
  flexible. Systematics on the background at LO
  under control (to our best knowledge). It would
  be nice to repeat the study at NLO (MC_at_NLO)
• Personal comment After the fullsim of LVL1 and
  the cross check with Alpgen, I think the results
  are reliable. Time to finish the note and ask for
  approval.