ADM

1
Calgo-Workshop (11-12 May)

• Short term goals
• - produce better/uniform data with Fix pass-2
• - Recommend set of cuts/data samples/
  procedures for the first round of publications
• - Improve calibrations/resolutions/error (em/jet)
• Longer term goals
• - Finalize / Certify p17 which has the tools
  (DB, new info on tmb) to allow a further step in
  reconstruction quality.
• - keep improving algorithms using this new info
• Status of the on-going efforts will
  be reviewed in Fresno

2
Calgo-Workshop (11-12 May)
tuesday
wednesday
54 talks in 2 days a few gallons of coffee
3
Calgo-Workshop (11-12 May)
General / Review ? Electron
- id/calibration   ?  Photon-Id/PreShowers 
  ? future ADM? Jet-Id 
  ? Jet Energy Scale 
? future ADM? MET and
d0correct    ? Energy Flow 
? CALOP 
  ? future ADM? ICD/MG
  ? future ADM? Too
much material ? try to concentrate on studies and
changes which are directly influencing current
analyses   
4
Electrons / Photons / PS / calibration
5
EM-Shower Shape Studies

• disagreement data/MC in Hmx8
• particularly shower shape in eta (sigZorR
  variable)
• simulation of possible Cross-talk effects

6
em longitudinal shower profile

• rescaling of em-fractions needed!
• sampling weights adjustment?

7
Shower shape in eta/phi

• 1 cross-talk in eta (read-out strips)
• 1 electronics cross-talk (per tower in depth)

Electronics crosstalk helps here
8
Electron reconstruction in phi-cracks
study of Oleg K. (03/03) crack width in data/MC
vs PT(em)
E/p study with e- from W/Z
better crack simulation needed ? less charge
collected in Run II
fid cuts
9
J/Psi calibration MC sample and fit
File ? TMBAnalyse_x-p14.05.02-d0correctv06-p14.03.
00_Stark_emid_mcp14 (direct J/y events ) The
trigger was simulated with TrigSim in p15.06.01
After geometrical corrections
10
J/Psi calibration MC sample and fit
All Files on SAM ? CSskim-JPSI.raw_p14.06.00 (
70 of the skim)
Without any corrections
The calibration constants are obtained using the
energy corrections for geometry effects for
electrons.
11
EMScale constants J/Psi vs Z ? ee
Only CC is considered, due to low statistics only
4 eta CC zones were defined Zone 1 -1.2 lt h
lt -0.6 Zone 2 -0.6 lt h lt 0.0 Zone 3 0. 0 lt h
lt 0.6 Zone 4 0.6 lt h lt 1.2
no dramatic non-linearity effect!
N EM obj/zone Zone 1 102 Zone 2 219 Zone 3
215 Zone 4 114
12
electrons to scan electronics behavior
in CC
In agreement with Jans study with e from Z but
with much less statistics
e- from W/Z
adc
adc 2
E/p gaussian fit crate 9 3000 evts/adc
diff
adc 7
13
at the BLS level (4 towers) ? tower level -gt
layer level
(use skimmed rootrees ? get cells)
E/p(BLS)-E/p mean
Preliminary
sigma2.4
bad BLS ?
384 BLS ? 4 CC crates
Z w 3.2 GeV corrected
Z w3.4 GeV uncorr
better by 6 ? resol on E improved by
sqrt(2)6 8
without latest gain/nlc calibration effort on
phi-intercalibration started also on min-biais
evts
14
PMCS
Juniie Zhu
A 20/sqrt(E)
C 3.9 0.2
If A 15/sqrt(E) C 4.2 0.2
Likelihood on the Z mass with PMCS/data
15
Reconstruction of p0?gg

• Require one good EM object (no HMx cut)
• Require at least two CPS clusters associated with
  EM
• If more than two use most energetic ones
• These CPS clusters must pass tight shower-like
  criteria
• Use CPS clusters position with respect to
  primary vertex to reconstruct photons
  trajectory. Assume energy of each photon a half
  of EM objects energy
• One can also use CPS energy for weighting

16
p0?gg and h?gg MC

• Lines indicate population of events where we
  reconstruct CPS clusters incorrectly
• Most background comes from one high energy CPS
  cluster reconstructed as two clusters

17
Invariant mass fit (plus h)
Use bMU CSG sample where we require at least one
muon-based trigger to fire ? cal-unbiased sample
18
Jet / MET / Energy Flow
19
CalJetMet LBN selection for post-shutdown data

• Summary

With T42 Without T42 of data
kept 90.44
89.20 Estimation of data kept
at least 95.53 95.11 after p17
reprocessing

• T42 helps to clean helps to clean distributions
  but no effect on serious calorimeter problems
• A large fraction of the rejected data were taken
  when there were  online  problems
• Hot cells or pedestal shifts or toroid noise
  
• At the last meeting, fraction of data kept 93
  but hot cells problem in the ICD in very recent
  data (runs 192308 ? 192365). Jets are good but
  MET is very bad.
• dq_calo BAD runs are 1.6 of the data we
  checked that 30 of the runs flagged BAD by
  dq_calo are not flagged BAD by our selection

20
ltMETBxgt and ltMETBygt after selection
ltMETBygt per LBN
With T42 Without T42
ltMETBxgt per LBN
GeV
GeV
21
ltMETBxygt and ltSETBgt after selection
ltMETBxygt per LBN
With T42 Without T42
ltSETBgt per LBN
GeV
GeV
22
lt bad jetsgt vs lt good jetsgt after selection
lt bad jets gt per LBN
With T42 Without T42
lt bad jets gt per LBN
lt good jets gt per LBN
Proportion of bad jets is reduced after T42
lt good jets gt per LBN
23
MET vs ? SET
Min bias Di-jets back to back (3CJT5 monitor)
Without T42
Slope is steeper than in run I
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
Low ET Jets properties
28
(No Transcript)
29
Jet definition (top)

• Good jet L1 confirmation Quality cuts
• 0.5 lt EMF lt 0.95
• CHF lt 0.4
• HotF lt 10
• N90 lt 2
• Bad jet Jet failing one the quality criteria
  ? Reproduced by Pythia
• Noise jet no L1 trigger confirmation
  ? Not in MC

All bad jets
Noise jets removed
Note Jets energy are smeared in top analyses
30
Bad jet study using L1 conf
h
PT
Bad jets with L1 confirmation are physics
jets very well described by Pythia
EM fraction
CH fraction
hot fraction
N90
31
Effect of bad noisy jets on MET
Z 0 good jet
No Bad jet No noise jet

• Broader MET distribution
• in events containing bad jets,
• reproduced by the MC
• Large tails due to noise jets
• Same feature observed in
• Z?ee and Z?mm channels

Bad jet gt 0 Noise jet gt 0
Bad jet gt 0 No noise jet
32
Outline

• Search for an Event Quality Variable to reject
  Noisy events
• Basis
• Comparison of the occupancy of TT for
• different Jet Multiplicity
• Event Quality Variableprovisional proposal
• Noisy Precision TT see comparatively higher
  energy than L1 TT through-outexplained later.

33
Jet ID for p14
Good Jets Noise Jets
Blue new id proposed jet-id cuts w/o n90,
Hotf, EMF lt 0.05, vs OLD
34
Conclusions

• Things are moving on many fronts
• Monitoring, Infrastructure, Calibration,
  Algorithms
• Data Quality is reaching a satisfactory status
• Still struggling with resolution/calibration, but
  maybe we are simply paying the fact that the
  calorimeter is measuring only 2/3 of the signal
  (-gt more fluctuations?, no compensation anymore?)
• Need better EM shower understanding
• Need to take into account EM-Had scale difference
• Fix02 is providing us already improved data
  (T42.5, HC) killing checkerboard fix and
  more info on tmb
• P17 ? Full use of DB, T42.0, CCMG correction

35
backup
36
Photon ID tools

• Yurii Maravin
• for photon ID group

37
New photon/electron ID tools

• CFT/SMT hits on the road to an EM object (Oleksiy
  A. and Y.M.)
• Ongoing effort to apply the algorithm in Wg and
  Zg
• Better track matching to EM object
• CPS-EM matching
• TRK-CPS-EM matching
• CPS stereo cluster quality flags
• Reconstruction of p0?gg
• Several methods of efficiency estimation
• System8 (Drew A.)
• MC studies (still in progress, but it is getting
  very close)

38
CFT/SMT hits for EM object
D0 Note 4444
Not to scale!

• Use EM objects position and pT to define roads
  in the tracker
• Unknown charge two roads (left and right)
• Calculate the number of CFT and SMT hits
  associated to the road
• 4s of hit-road resolution measured in Z ?ee sample

39
EM hits performance

• Measure performance on electrons from Z?ee
  and fake EM objects from QCD EM-jet sample

Z?ee
noise
EM
Number of hits for EM object are blue
Number of hits for a random point in space is red
40
Match probabilities

• Use hit distributions from electrons and noise to
  calculate probability functions

Prob
noise
electron
41
Overall performance

• Can increase track matching efficiency without
  big contamination from the fake EM electrons
• Can decrease the probability of electron
  mis-identified as photon by factor of 4.

42
(TRK-)CPS-EM matching
D0 Note 4414

• One can improve track matching to EM object by
  requiring EM object to be matched with CPS
  stereo cluster first, then check if a track
  matches CPS cluster.
• Use higher resolution of the CPS in f and z

Z?ee
n CPS

• How to associate CPS cluster to EM object?

43
CPS-EM matching

• Study different (6) CPS-EM matching methods,
  analyze quality of CPS clusters

44

• Best method associate the most energetic CPS
  cluster to EM object
• Assume CPS-EM matched if separation between CPS
  and EM is below 5s (e 96.6)
• can be significantly improved if CPS-CAL is
  aligned!

dz 0.22 cm
df 7.8 mrad
45
TRK-CPS-EM matching

• Resolution of TRK-CPS match depends on quality of
  central track propagation to CPS
• Propagation is done using a simple circle in
  axial view and a line in z-y plane. Empirical
  corrections need to be applied.
• Checked D0GTrackPropagator slower(?) and still
  needs empirical corrections. No gain in
  resolution
• Use Z?ee, U?ee, J/Y?ee samples to calibrate track
  propagation to CPS for the first time in data
• Will be available in p17

46
TRK-CPS-EM matching

• Higher efficiency than the standard methods
  (without CPS) at the comparable fake rate

47
Reconstruction of p0?gg
D0 Note 4446

• Natural continuation of CPS-EM matching study
  reconstruct p0?gg
• Low opening angle use CPS stereo clusters to
  reconstruct individual photons
• Neutral pions are usually produced in jets
• A lot of CPS clusters from MIPS, some of them are
  fake
• For efficient p0 reconstruction we need to
  distinguish MIPS-like and ghost-like CPS clusters
• Once we are able to separate shower-like CPS
  clusters, we can cut on occupancy of such
  clusters to reduce background from jets
  misidentified as electrons or photons.

48
Assessing CPS clusters quality

• CPS reconstruction software (DØ Note 4014)
  provided several variables to measure quality of
  CPS cluster
• matchQ spatial quality the smaller, the better
• matchEQ energy quality the smaller, the better
• In addition, a few variables can help us fight
  MIPS-like CPS clusters
• Smallest energy deposited in the SLC (single
  layer cluster)
• Smallest number of strips of the SLC
• Use di-EM samples as a source of CPS clusters
  associated with good electrons, and JESB CSG
  sample as a source of MIPS-like CPS clusters
• Require no calorimeter activity in the tracks
  vicinity (no cells in df lt 0.3)

49
matchQ and matchEQ
50
minEn, minNStrips
51
Shower-like quality of CPS clusters

• Loose shower-like CPS cluster (e 88.2
  fr3.5)
• matchEQ lt 2.0
• matchQ lt 1.0
• minNStrips gt 2
• minEn gt 0.005 GeV
• Tight shower-like CPS cluster (e 82.4
  fr2.5)
• matchEQ lt 1.5
• matchQ lt 0.7
• minNStrips gt 2
• minEn gt 0.007 GeV

52
Reconstruction of p0?gg

• Require one good EM object (no HMx cut)
• Require at least two CPS clusters associated with
  EM
• If more than two use most energetic ones
• These CPS clusters must pass tight shower-like
  criteria
• Use CPS clusters position with respect to
  primary vertex to reconstruct photons
  trajectory. Assume energy of each photon a half
  of EM objects energy
• One can also use CPS energy for weighting

53
p0?gg and h?gg MC

• Lines indicate population of events where we
  reconstruct CPS clusters incorrectly
• Most background comes from one high energy CPS
  cluster reconstructed as two clusters

54
A few words on CPS MC

• It is not in a good shape
• CPS gains are not set properly
• Some of the parameters are hardcoded, so dont be
  fooled by those RCPs
• Good news Tania Moulik is going to investigate
  and try to improve it

55
CEM(1,3.) special runs

• Use several special runs taken with low-pT Cal
  triggers
• Apply the method, estimate background by taking
  low-energy CPS clusters (2nd and 3rd)

56
CAL-unbiased data

• Use bMU CSG sample where we require at least one
  muon-based trigger to fire.
• Apply the same procedure

57
Invariant mass fit
58
Invariant mass fit (plus h)
59
Whats needs to be done next

• We can reconstruct p0?gg using CPS!
• We need to apply the method and test it on
  B-physics data sample (D processes)
• Study CPS and CellNN information
• We need to identify manpower for this task

60
More use of CPS in EM fake rejection

• Fake EM objects are normally high-pT p0 produced
  in jets
• Expect more CPS clusters around fake EM object
  than that around real one
• Many CPS clusters around EM object are clusters
  produced from a single broad electron shower
• Do not want to remove those

61
Neural Network

• Use matchEQ, minEn, and minNStrips variables as
  inputs to neural network algorithm implemented in
  ROOT v4.0/3.

62
Cut on shower-likeness

• Standard track matching criteria
• chi2prob gt 0.01
• Use chi2prob gt 0.001 cut on number of
  shower-like CPS clusters in Z?ee and fake EM
  objects
• Standard matching e 79.4 1.1, fr 1.7
  0.4
• N(CPS) lt 3 e 79.7 1.1, fr 1.3
  0.4
• Once can reduce fake EM object without much loss
  of efficiency. However, this is just a quick
  look, one needs to develop this idea further

63
Weve got tools how do we use them?

• Usage of these tools is simple for electron
  identification
• We have a clean electron sample!
• What do we do for photons???
• Can anyone please get a clean sample of H?gg?
• Work is still in the progress for photon
  identification
• System8 studies
• MC studies

64
System 8

• Require two samples enriched in signal and
  backgrounds and two sets of uncorrelated cuts

65
System 8 in Wg and fake EM samples

• Analyzed a very very long list of different cuts
• Select several cuts of interest
• Number of L1 cells
• Iso7
• sum of ET of all cells in cone dRlt0.7 minus EM
  objects ET
• HMx8
• Results from System 8
• Iso 7 lt 7 GeV e 97 Fake 5
• Ncells in L1 lt 3 e 97 Fake 3
• HM8 lt 20 e 73 Fake 12

66
Problems with system 8

• Big errors (currently 14)
• Systematic effects due to correlations
• Applying two cuts that are 95 efficient results
  in 70 efficiency
• It seems that we should use Monte Carlo
  simulation to estimate photon efficiency
• Assume the difference between photon and electron
  is described properly in Monte Carlo (not for CPS
  as of now)
• Measure efficiency in data for electrons
• Scale the results based on parameters from MC
  study

67
Monte Carlo studies

• Currently in progress

68
Summary

• Quite a lot of photon/electron tools have been
  developed over the last few months
• We see p0?gg!
• Most of these tools are documented in D0 Notes
  4414, 4444, and 4446
• Once we converge on a method to estimate photon
  ID efficiency we will produce a photon ID
  certification note

69
Backup slides