Jet ID experience at D

1
Jet ID experience at DØ

• Jet ID
• Choose energy to be clustered into jets
• Choose clustering algorithms and parameters
• Choose sets of selection criteria
• Prepare optimize jet ID tools
• Choose a basic set of cuts
• Fancier optional cuts (e.g. jet-vertex matching)
• Measure reconstruction and selection
  efficiencies
• For this talk Ill concentrate on choosing the
  basic set of selection criteria (3b)

• Amnon HarelUniversity of Rochester

2
Jet selection timeline
Development of jet ID tools
First comprehensive optimization study
practical choices, driven by specifichardware
problems
First (preliminary) results with jets
This data published with these cuts, no further
work expected.
3
Geometry
The No EM Gap Lots of CH with less calorimetry
than normal before it...
CH2
CH3
CH1
FH
CH1
FH
EM
EM
FH
CH1
Beam collision area
4
How we choose jet criteria

• Choosing samples enriched and depleted in
  unphysical jets
• We used the independent level 1 trigger readout
  to verify which selections really enhance jets
  with precision-electronics noise
• We assumed those samples would also be enriched
  in other unphysical jets
• Identify and study unphysical jets
• Used brand new, and as we later found out, very
  good data quality criteria ? identified only one
  type of unphysical jets in our samples
• Study selection efficiencies for physical jets
• We identified unmotivated inefficiencies
• Optimize the overall selection
• Arbitrary decision is a jet with 30 noise a
  bad jet? What about 70?
• Minimize bias, e.g. nice to have a similar
  efficiency for quark and gluon jets
• Purity will vary greatly from analysis to
  analysis
• Separate into ? and pT regions when selecting
  cuts
• We aimed at 98-99 efficiency and removal of all
  identifiable noise
• Caveat JES procedure required the cut EMFlt0.95
• Note no MC events were used

5
The samples good jets

• Physical-Jet enriched sample tag probe di-jet
• Basic cuts
• A jet trigger (required for all samples)
• P.V.Z. lt 30cm (required for all samples)
• Dijet, tag jet passes jet ID
• 2 GeV track match (for ?lt2.0)
• ?F gt 3.0

• Fancier details
• MET lt 20 GeV
• METT lt 8 GeV
• METT for the dijets is the (size of the) MET
  component perpendicular to the leading jet.
• pT balance (?pT/ltpTgt lt 0.5)

These jets just have to contain physical energy
consider them as signal
6
The samples too many jets

• Physical-Jet depleted sample
• 6 jets with ?lt2.5 (implied pT gt 12 GeV)
• choose a jet without a track match
• pT gt 0.5 GeV
• ?R lt 0.5
• Loose track quality
• Fancy detail MET gt 20 GeV (when ?lt1.5)

Without jet ID, this strongly enhances the sample
in calorimetry noise (and top pairs)
And this selects the suspicious jets in the sample
No enhancement in end caps
Noise enhancement
7
The samples 3rd jet

• Physical-Jet depleted sample
• 2 great, balanced jets
• ?lt2.5
• pass jet ID
• 2 GeV track match (for ?lt2.0)
• ?F gt 3.0
• A third suspect jet
• At least ?F gt 1.2 with both of them
• Has no 0.5 GeV track match
• Fancy detail METT lt -5 GeV (when ?lt0.8)

Tag jet
3rd jet
Tag jet
METT for the 3rd jets is the MET component in
their direction. In a perfectly clean event with
a perfectly fake 3rd jet
Noise enhancement
In the CC it can greatly enhance the sample in L1
rejected jets, though it throws out many
interesting suspicious events. In other regions
the enhancement, if it exists, is not worth the
statistics.
8
The samples kinematics

• The jet kinematics differ greatly between the
  samples!
• cuts optimized in ? and pT regions, which
  largely overcomes this problem
• also played with reweighting the good jet sample
  to match the bad jet kinematics

dijets too many jets 3rd Jets
Inter cryostat region is noisier, with inferior
energy resolution
?
Raw pT GeV
All pT and E are rawJES is from 2 at low pT
to 1.2 at high pT
BTW as the NS asymmetry hints, this study was
done before the final calorimetry calibration was
in
9
Noise studies
Identified a type of fake jets in busy events
coarse hadronic noise is gathered into wide hard
jets. But this only happens in busy events where
a seed is available (jet algorithm has a veto
against CH seeds). This conclusion is supported
by the pT dependent CHF, ?det, f-width, P.V.Z.,
n90 and MET distributions. Since physical ICR
jets with a high CHF are narrow, this can be used
to discriminate against this background.
0.8lt?lt1.5
1.5lt?lt2.5
20ltpTlt40

• Looked suspicious
• Proven to be noise
• Reject with CHFlt0.6

dijets too many jets 3rd Jets
n90
CHF
CHF
More work required at low pT
A measure of jet width
CHFlt0.4 (CHFlt0.6 8.5lt?detlt12.5 n90lt20)
10
Selection efficiency
We set cuts on minimal EM fraction and L1
confirmation to achieve a selection efficiency
(in the Good jets sample) of 99.7-99.9 per cut.
0.8lt?lt1.5, L1Conf, .4gtchf, pTgt40, emflt.05
Requiring EMF gt 5 is too inefficient in the no
EM gap, pT dependence checked, and is negligible
(high pT?narrow jets?more inefficiency but in
same place) There is separation between the
samples in EMF even for narrow jets in the gap
region, but it is too weak to be
useful. Furthermore, the bad jets in these
regions look like single particles, rather than
unphysical energy
? width
0.8lt?lt1.5, L1Conf, .4gtchf, pTgt40
No minimal EMF in this region
? width
Looser EMF cut in this region
?det
11
Closing comments

• Jet selection criteria are not a good thing, as
  they imply that perfectly physical jets are
  ignored. Typical rate o(1). ? What about MET?
• Most new physics contain a dark matter candidate
  ? MET signatures
• The fake MET caused by ignoring physical jets can
  be simulated only if fake rates are well
  simulated ? use data as much as possibleReached
  o(1) accuracy with 1fb-1 of data
• Vetoing any event with a bad jet?
• sometimes a good solution
• hard to evaluate the efficiency of the veto
  without a simulation of the noise

Jet production
New physics?
12
Conclusions

• Mostly data driven choice of basic jet ID
  criteria
• The data shaped the studies
• Tracking helps
• It really helps to have a reliable ID for at
  least one kind of noise
• Correlations with MET can help identify
  unphysical jets (at high enough pT)
• Details of noise characteristics and detector
  geometry are relevant
• With the tools in place, quick turn around is
  possible
• Both ID tools (variables) and optimization macros
• We had a two week time window from the time the
  (partially calibrated) data was available to full
  collaboration approval official implementation.

13

• Back Up Slides

14
?
b