Electron Reconstruction, Selection and Identification a bremsstrahlung story

1
Electron Reconstruction, Selection and
Identification (a bremsstrahlung story)

• S. Baffioni, C. Charlot, F. Ferri, D. Futyan, P.
  Meridiani,
• I. Puljak, C. Rovelli, R. Salerno, Y. Sirois
• Analysis Note at http//cms.cern.ch/iCMS/jsp/openf
  ile.jsp?tpdraftfilesAN2005_062_v1.pdf
• Analysis Note at http//cms.cern.ch/iCMS/jsp/openf
  ile.jsp?tpdraftfilesAN2005_065_v1.pdf

2
Outline

• eSuperClustering
• eTracking
• eClasses
• eEscale corrections and eMomentum
• eIdentification
• eConclusions

3
Reminder benchmark channel

• Main motivation for revisiting of electron
  reconstruction tools

• Low pT electrons are really important in some
  main Higgs discovery channels for intermediate
  mass (ZZ, WW)
• Low statistics, so keep efficiency as high as
  possible

pT(1)
pT(2)
pT(3)
pT(4)
softest electron (Z in case one off-shell Z)
H?ZZ?4e- pT spectra
4
eSuperClustering

• Use clustering algorithms as used for HLT
• Reconstruction strategy needs brem cluster energy
  recollection to be optimized at earliest stage
• Loosening of seed cluster threshold for
  efficiency down to lowest pT

HLT value 4 GeV Offline value 1 GeV
Back to back electrons with fixed pT. No pile-up
5
eSuperClustering

• Reconstruction strategy needs brem cluster energy
  recollection to be optimized at earliest stage
• Extended phi road for brem recovery to cover low
  pT cases

Back to back electrons with fixed pT. No pile-up

• Default ? roads
• 0.2 rad (endcaps)
• 10 crystals (barrel equiv to 0.17 rad)
• Increased to
• 0.3 rad (endcaps)
• 17 crystals (barrel equiv to 0.3 rad)

6
eGSFTracking

• Starts inward-outward with seeds from pixels
• Cluster driven seeds from pixel match
• As done for HLT and H?ZZ?ee??
• Regional seeding also possible
• A la FKF tracks, but reducing combinatory using
  regional seed reconstruction based on ECAL
  superclusters (?????????????????????
• Build trajectories using Bethe Heitler energy
  loss
• Very loose ?2 cut, a min of 5 hits required
• Efficient collection of track hits up to the ECAL
• GSF fit of recontructed trajectories
• See AN2005/011, essentially here a reminder
• Some complementary material (pT30 GeV)

7
Pixel Matching Procedure

• Start from energy weighted mean position of
  super-cluster
• Propagate back through field to pixel layers and
  search for compatible hits
• 1st search with loose requirements on ? and z
• ? window 40mrad (-25mrad,15mrad for e-,
  -15mrad,25mrad for e)
• z 15cm
• Search for 2nd pixel hit to
• complete the seed with
• tighter requirements
• ? window 1mrad
• z 0.05 cm

8
Varying width of search windows

• Back to back electrons with fixed pT. No pile-up
• Loosen values used in offline reconstruction
• 1st ? window width 40 mrad ? 200 mrad (could be
  made pT dependent)
• 2nd ? window width 2 mrad ? 10 mrad
• 2nd z window 0.05 cm ? 0.07 cm

AN2005/065
AN2005/065
pT 7 GeV
pT 10 GeV
9
eTrack Hits collection
PTDR
e-, pT10 GeV
e-, flat p5-100 eGSF tracks (pixel match)
10
eTrack pT

• Use most probable value rather than usual
  weighted mean of all the components
• Therefore better estimation for low radiating
  tracks, more biased estimation for the others

30 GeV pT
10 GeV pT
11
eTrack direction at impact point
10 GeV pT
30 GeV pT
??????????
??????????
10 GeV pT
30 GeV pT
??????????
??????????
??????????
12
eTrack outermost parameters

• Meaningful momentum at last point
• pin-pout as a measurement of the true bremed
  energy
• Possibility of supercluster-track match at the
  rear using outermost track parameters and seed
  cluster energy and position

pin-pout
Eseed/pout
brem fraction (pin-pout)/pin
true bremed energy
13
Why electron classes?

• Different track-supercluster patterns are
  observed as a consequence of bremsstrahlung in
  the tracker material
• Corrections need to be adapted to the different
  cases
• Different measurement errors depending on the
  topologies

a big brem
two goldens
two showering electrons
14
Electron classification

•  Golden  electrons
• No brem sub cluster
• Low brem fraction
• Good E/pin
• Good phi geom. matching
•  Big brem  electrons
• No brem sub cluster
• Bad brem fraction
• Good E/pin
•  Narrow  electrons
• No brem sub cluster
• Intermediate brem fraction
• Good E/pin
•  Showering  electrons
• Identified brem sub-cluster or bad E/pin

AN2005/062-PTDR
E5-100 GeV

• Barrel/Endcaps/Cracks/separation
• Cracks just identified here

15
Algorithmic (energy scale) corrections
Ecorr Esc . F(Ncry) . f(?)

• F(Ncry) containment, class independent
• f(?) energy lost, residual ? dependance, one
  function for showering and one for the three
  other classes

AN2005/062-PTDR
Barrel
AN2005/062
fixed E10, 30, 50 and flat E5-100
16
More on energy scale corrections

• Exactly same treatment for endcaps
• Eendcaps EpreshEcorr

Hybrid clustering
Island clustering
Algorithmic corrections ultimately to be tuned
on Z?ee data, see CMS AN2005/054
17
Electron ECAL measurement
AN2005/062-PTDR
AN2005/062-PTDR
EB
AN2005/062
AN2005/062
AN2005/062
AN2005/062
EE
Uncorr
Corr
18
Electron momentum estimation
AN2005/062-PTDR
AN2005/062-PTDR
Tracker estimate
ECAL estimate

• From the analysis of Ecorr/Etrue and of
  prec/Etrue vs Ecorr/prec
• If E/p gt 12? pcomb Ecorr
• Else combine E and p pcomb wE Ecorr wp prec

e- in the barrel E5-100 GeV
19
E-p combination
AN2005/062-PTDR

• Take direction from track fit
• pcomb from Ecorr or weighted mean of Ecorr and
  prec
• p weights from track fit errors
• E weights from error parametrisation
• According to classes

AN2005/062-PTDR
ECAL
ECAL Tracker Combined
e- p5-100 All classes, barrel only
20
Application on H?ZZ?4e

• Higgs and Z mass spectrum for mH120 GeV, before
  and after correction

21
Electron reconstruction efficiency
AN2005/065

• Apply loose geometrical match between eTrack and
  eSuperCluster
• ?? ?? 0.1
• ???corrected for z vertex
• ???with ??from track at vertex and extrapolated
  to the supercluster position assuming helix

AN2005/065

• Efficiency fraction of generated electrons with
  a reconstructed track matching a supercluster and
  with same charge and matching in direction with
  MC truth within ????0.05

22
Electron basic preselection
AN2005/065-PTDR

• eGSF track and eSuperCluster with geometrical
  match as previously defined
• Loose constraint on E/pin (lt 3.)
• Loose constraint on H/E (lt 0.2)

AN2005/065
AN2005/065-PTDR
HCAL threshold
H/E
electrons vs preselected jets pT25-50
23
Isolation, selection of primary electrons
AN2005/065-PTDR

• Physics case H?ZZ?4e, mH150
• Track based isolation
• Use normalized sum of pT inside a cone around
  electron track
• Transverse impact parameter
• IP/?IP

AN2005/065-PTDR
AN2005/065-PTDR
24
Application on H?ZZ?4e
After impact parameter
After preselection
After isolation
25
Electron identification

• Many already existing work. Aim here what can be
  said in addition from
• New possibility to match outermost track params
  with seed cluster ?
• Electron classification ?

AN2005/065-PTDR (all 4)
preselected jet sample, pT25-50
e-, pT5-50
preselected jet sample, pT25-50
e-, pT5-50
26
Class dependent identification

• Why?
• In many of the previously used eId variables,
  preselected jets looks like  showering
  electrons 
• and 83 of preselected jets pT25-50 fall in
   showering  class 
• but if ones wanted to use only  golden 
  electrons, factor 10 better separation can be
  obtained with simple class dependent cuts

AN2005/065-PTDR
AN2005/065-PTDR
AN2005/065-PTDR
e- pT5-50 vs preselected jets pT25-50, barrel
case
27
Class dependent identification

• Rear matching distributions sensitive to brem
• With narrow and big brem appearing as tail parts
  of the single cluster distributions with clear
  separation from jets
• Remember, golden vs narrow vs big brem mainly
  differing by the brem fraction (track based brem
  measurement)

AN2005/065-PTDR
AN2005/065-PTDR
e- pT5-50 vs preselected jets pT25-50, barrel
case
28
Conclusions

• These two notes describe the new electron
  reconstruction tools put in place in connection
  with H?ZZ?4e analysis, in short
• Retuned clustering in ECAL
• Completely new tracking with eGSF tracks
• retuned pixel match seeding
• E classes definition
• so not to mix  torchons with  serviettes 
  when correcting, selecting electrons
• and define electron quality (errors) that can be
  used in any physics analysis
• Retuned E scale corrections and combined E-p
  momentum estimation
• based on e classes and weights
• Class dependant electron id variables
• new rear match variables
• showering much more difficult to separate from
  jets

29
Answers to referees questions

• Merge the two notes in one note on electron
  reconstruction
• Yes OK

30
Changes in figures since the drafts

• AN2005/062
• Fig.3,4,6,8 different statistical sample
• Fig.3,4,6,8 very small refinement in
  classification
• Fig.3 slightly changed f(?) correction for
  endcaps, now exactly as for barrel
• AN2005/065
• Fig.6 was without preselection, is now with
  preselection described in the note pTgt5 GeV/c
• Fig.7,8 E was with Egamma old corrections, being
  replaced by new corrections

31
Electron ECAL measurement
draft
new
32
Electron ECAL measurement
draft
new
33
Electron ECAL measurement
draft
new
34
Electron ECAL measurement
new
draft
35
new
draft
36
E-p combination
new
draft
37
E-p combination
new
draft
ECAL estimate
38
E-p combination
new
draft
39
new
draft
40
Class dependent identification
new
draft
41
Class dependent identification
new
draft
42
Class dependent identification
new
draft
43
Class dependent identification
new
draft
44
Class dependent identification
draft
new
45
Class dependent identification
draft
new
46
Other figures
AN2005/062
47
Other figures
AN2005/065-PTDR
AN2005/065-PTDR
48
Backup slides
49
Electron ECAL measurement
RMSeff/? golden 2.6 showering 7.4
RMSeff/? golden 2.5 showering 6.3
Corr
Uncorr
barrel, Elt25GeV
50
E-p combination endcaps
ECAL estimate
Tracker estimate
Combined
51
E-p combination e classes
golden
big brem
narrow
showering
52
Seeding
53
Seeding
54
eGSF tracks

• Gaussian sum filter instead of KF for the fit
• Non linear approximation of Bethe Heitler using
  gaussian mixture
• Also used in the forward fit to get full info at
  last hit
• Use most probable value rather than usual
  weighted mean of all the components

??????????
??????????
55
Weighting electrons eGSF errors
WM
MD

• rescale errors from mode estimate
• being investigated with Tracker experts

56
electrons in H-gtZZ
mH150
57
Electron id HOverE
Single e- flat p
HCAL thresh 0.5GeV
Elt15 GeV gt only flat contribution from cracks
HCAL threshold 0.5 GeV
58
Electron id shower shape (barrel)
59
Electron id matching (barrel)
60
Electron id shower shape (endcap)
61
Electron id matching (endcap)