Jim Branson 1

1
E/Gamma and b/Tau PRS(small US effort)(what you
should work on when you finish)

• US CMS Annual Collaboration Meeting May 2002
• FSU
• Jin Branson

2
ElectronPhoton main packages

• EgammaAnalysis
• Modular analyzer and analysis helpers
• Abstract writer support
• Iteration wrapper and UserCollection support
• EgammaNotification
• Notification and flow control
• EgammaH4Support
• Hbook CWN writer
• EgammaClusters
• Basic clustering algorithms
• Position and energy corrections
• Isolation and p0 rejection tools

3
ElectronPhoton main packages

• ClusterTools
• Endcap-specific reconstruction
• Preshower clustering
• Brem recovery algorithms
• EgammaL1Tools
• Level1 trigger candidate matching
• ElectronFromPixels
• Pixel matching and track seeding algorithm
• Electron track reconstruction based on pixel
  seeds
• EgammaMCTools
• Generator- and GEANT-level analysis
• EgammaTracks
• Tracking setup and helper classes

4
e/gamma Development
Working with MC
5
Basic Calorimeter Software Activities

• Calorimeter software has been stable for a few
  years.
• US is involved in upgrade program.
• There are three areas of activity
• improvements of current architecture of
  Calorimetry
• FORTRAN elimination
• using new ROU naming schema
• navigation and speedup optimization
• online/testbeam specific preparations
• splitup the Readout on two parts to read the
  online/testbeam format
• preparations for the migration to OSCAR
• isolation of what is required for hit-formatting
• first prototype of DDD usage

6
Island Clustering
Log-weighed Position Correction
7
Depth modeling

• Dependence of shower max on energy log(E) with
  energy in GeV
• Tmax AT0log(E)
• Parameterization for ECAL with A 0.89 (PbO4 rad
  length)
• Optimize T0 by finding the zero offset for the
  two half barrels (optionally one could minimize
  position resolution)
• Specific for electrons OR photons

8
Log weighting

• Linear-weighted cog produces characteristic
  s-shape
• Rather than applying ad-hoc correction, use a log
  weight

Linear weight
W0 smallest fractional energy to contribute to
position calculation
Log weight W04.2
9
Position resolution
10
Brem recovery

• Average brem loss (44) corresponds to an
  average thickness of 0.57 X0
• Need a brem recovery strategy in ECAL
• Cluster composite ECAL objects according to some
  criterion
• E.g. energy deposition from brem well aligned in
  h
• Use narrow h window
• Collect clusters along f

• Produces a SuperCluster collection of ECAL
  clusters
• Removes large tails

11
Hybrid algorithm

• In more detail
• Start if EtseedgtEthyb
• Make 1x3 domino
• If center of dominogtEwing
• Extend to 1x5
• Proceed Nstep in 5
• Remove dominoes below Ethresh
• Disconnected domino preclusters with EgtEseedare
  then reclustered in f (producing a SuperCluster)

• Use h-f geometry of barrel crystals
• Start from a seed crystal (as for island)
• Take a fixed domino of 3 or 5 crsytals in h
• Search dynamically in f

12
Optimize Hybrid Performance
13
Energy Scale

• Energy is estimated by the sum of energy deposits
• Emeas/Etrue gaussiantail, peaking at lt1
• Incomplete containment
• Unrecovered brem
• Set the energy scale such that the gaussian peak
  falls at 1
• Parameterize corrections as a function of the
  number of crystals included in the cluster
• E.g. for hybrid (barrel) clusters

Electrons 10-50 GeV
14
Energy scale performance I

• In the barrel, with hybrid clusters
• No Pt dependence
• Small residual h dependence

15
Energy resolution

• Effective width is defined as the half-width
  containing 68.3 of the distribution
• Performance on unconverted photons (using fixed
  window)
• seff/E 0.9

16
Preshower matching

• Endcap SuperCluster
• extrapolate components to Preshower
• search PS cluster in narrow road around
  extrapolated point
• correct component energy
• Recalc SuperCluster energy

17
Pixel Matching (level 2.5)
18
e/g Level 2.5
19
e/g Triggers
20
Electron Tracks

• Use standard tracking with pixel seeds from
  matching Level 2 clusters
• Fast (few tracks to reconstruct)
• In the spirit of regional reconstruction
• Special e track fitter may help.

21
Electron Position Matching in h
22
Electron Rates and Efficiency
23
HLT Algorithm Timing

• Time on (dual) 700 MHz P III
• Data access time (objectivity) excluded
• Optimization possible.

24
June Milestones
25
Tracking Photon Conversions
Efficiency still low due to seeds
26
Callibration with W?en
27
Background to H?gg after standard cuts plus
tracker and ecal isolation
28
Egamma/Jet Available
29
b/t (Tracker Group)

• Many developers and much progress.
• US not involved (?).
• Software depends on CommonDet.

30
ORCA for the Tracker

• 4 subsystems
• Tracker geometry, hit formatting, hit loading,
  digitization and persistency. Lets say
  everything up to the persistent digis. This is
  the package which has to be ready for the Monte
  Carlo productions.
• TrackerReco anything which has to do with
  reconstructed objects RecHits and Tracks. In
  principle those are not persistent, even if now
  tracks can be written to DB.
• Vertex same as above, but dealing with primary
  and secondary vertices.
• bTauAnalysis high level objects, like b and tau
  taggers. They use all the above packages.

31
Tracker

• Geometry put some detectors in the space and
  call it a Tracker
• Hit Formatting cmsim flat file to Persistent DB
  structure
• Hit Loading read back the last
• Digitizing simulate the electronics attached to
  the sensors, and apply filters to reduce the data
  volume.

32
Geometry
The number of hits a charged track can leave is
always gt 10, considered enough to allow an
efficient tracking and a reasonable combinatorial
overhead.
Number of Si hits excluding pixels
33
Digitization

• New and more reliable (from real tests in
  Karlsruhe) treatment of the Lorentz angle in
  silicon, as a function of bias, irradiation etc.

• Not yet implemented for pixels, where the
  modeling is more difficult (after irradiation,
  the depletion will not be complete) wait for
  the optimization workshop

• Code in ORCA can be adapted via configurables to
  any
• Irradiation conditions
• Temperature
• V bias
• Etc

• Lorentz angle very important for hit resolution
• Silicon tan(?L) 0.12 (6 at 4T)
• Pixel tan (?L) 0.53 (28 at 4T)

Silicon
34
RecHit Resolution
Versus r
Versus z
Mean error
RMS
35
Seed Generation

• In this step a first approximation of a track is
  constructed using some supposed clean
  information.
• You can think about different types of seeds
• Take any two silicon/pixel layers and fit a helix
  with each pair of hits fulfilling some conditions
• Use the 2/3 pixel layers
• Have a seed from outside (for example muons
  beam spot or calorimeters)
• Seed generation affects efficiency and timing
  greatly.

36
Available Seed Generators

• Currently available
• CombinatorialSeedGeneratorFromPixel the standard
  one
• SeedFromConsecutiveHits takes 2 consecutive
  layers and uses the hits to build a seed
• SeedFromSeparatedHits even more difficult!
• SeedGeneratorFromSimTrack a MC based seed
  generator with 100 efficiency. Useful for tests.

37
Pixel Inefficiencies
Different staging/Lumi scenarios
L 2?1033
L 1034
Expected Inefficiencies at 1/2/10 ? 1034
38
Seeding with Pixel Silicon
Hence, work has started to produce seeds from
pixels the first layer of microstrips. Remember
that it is 20 cm away from the IP, so you expect
a huge number of compatible RecHits and thus a
combinatorial explosion.
39
Seeds from Pixel Silicon
40
New Propagator

• AnalyticalPropagator a new implementation in
  ORCA 6. Better protected against numerical
  problems, more precise and as fast as the Gtf.
  TO BECOME THE STANDARD SOON!!!!

41
Trajectory Cleaning
Since the generation of trajectories from the
seeds is not one-to-one, we can in the end have
two or more different trajectories sharing a
great fraction of the hits and thus are not
compatible. Such ambiguities are resolved by the
trajectory cleaner, which identifies mutually
exclusive subsets and chooses one trajectory per
subset. It works by iterating over the input
trajectories, finding for each Trajectory all the
others which share more than a given number of
hits with it, and then choosing the best
trajectory in the set, where best is based on the
chi2 of the fit.
42
Trajectory Smoothing
Since the trajectory building starts with a seed,
typically close to the beam spot, and propagates
to the outer barrel. In this way, the last fit is
done when reaching the end and there all the
information is available. Close to the start,
where (by the way!) we are usually more
interested in the track parameters, we have
initial information. A smoothing algorithm
guarantees an uniform and optimal set of
parameters everywhere. In this stage, no new hits
are allowed, but some hits might be dropped if
found not compatible wrt to the full information.
43
Performances
No 2-pixels!
44
Track Parameter Resolution
45
B tagging in HLT
We can trigger on b-jets on the online farm with
performances similar to those we obtain offline!
46
Timing

• Pixel Readout PixelReconstructiondoIt
• Seed Generator PixelSelectiveSeedsseeds lt
  5
• Trajectory Builder
• CombinatorialTrajectoryBuildertrajectories
  gt80
• Trajectory Smoother
• KalmanTrajectorySmoothertrajectories lt10
• Trajectory Cleaner
• TrajectoryCleanerBySharedHitsclean 1
• Trajectory Builder CombinatorialTrajectoryBuild
  er
• ModularKFReconstructorreco
• Tagging BTaggingAlgorithmByTrackCountingisB

47
Timing
48
Tracker Material
49
Detailed Description
50
Pixel Geometry
51
Total Tracker Material
52
Total Tracker Absorption Lengths
53
Alignment Studies

• Alignment Tools they work ?
• one can still add functionality
• Mis-Alignment studies
• reconstruction is uncritical up to even 1mm/1mrad
  misalignment (10 times more than
  survey/laser-alignment accuracy)
• Trigger ?
• update documentation (done), Note (preparation)

54
Summary

• CMS is making good progress on software and HLT
  studies in both e/gamma and b/tau.
• Current production to meet June milestones still
  ambitious,
• US role in these groups is small so far.