Topics for the TKR Software Review Tracy Usher, Leon Rochester

1
Topics for the TKR Software ReviewTracy Usher,
Leon Rochester

• Progress in reconstruction
• Reconstruction short-term plans
• Simulation
• Calibration issues
• Balloon-specific support
• Personnel and Schedule

2
Tracker ReconWhere were we at the last workshop?

• Status at last workshop
• Pattern recognition track fit linked together
  within the context of the Kalman Filter.
• Tracker reconstruction incorporated within the
  Centella framework
• Tracker reconstruction used in the analysis of
  the Test Beam data and Monte Carlo.
• ? Thanks Jose!

• Plan at that time
• Separate the track fit from the pattern
  recognition.
• Create
• Track Extrapolator
• Pattern Recognition
• Track Fit (Kalman Fit)
• Standard HEP Solution
• Personnel changes
• Jose departs
• Tracy and Leon arrive
• ?Tracker Recon in the shop

3
Tracker ReconWhat is the goal of the
reconstruction?

• To reconstruct the the direction and energy of
  gamma ray conversions in the tracker.
• Must find and reconstruct the trajectories of the
  e e- pair
• Must determine the energy of the e and the e-
  (individually)
• Must find the common point of origin of the e e-
  pair
• From this information, determine the energy and
  trajectory of the incident gamma ray
• Find and reject background

4
Tracker ReconNew kids have a few misconceptions

• How hard can it be?
• What happened to the magnetic field?
• This lead stuff is not helping the tracking!
• What happened to the stereo layers?
• Isnt the Monte Carlo supposed to also give you
  the answer?
• Electrons dont get along well with others
• Learning c will be easy!
• Ok, this is not your standard tracking problem
• Where/how do we get started?

5
Tracker ReconReality according to Tbsim
6
Tracker ReconHow do the new kids proceed?

• Need to
• Learn (enough) c to be able to understand and
  modify the code ultimately write new code
• Learn the GLAST system
• Learn how the existing code works
• Get the full appreciation for the problem
• Make progress along the path set forth at the
  last workshop

• Approach
• Take on the task of separating the pattern
  recognition from the track fit
• Work within the Test Beam / Centella framework
• Implement a new pattern recognition (link and
  tree) which is independent of the Kalman Filter
• Attempt to perform a simple pointing resolution
  study comparing fit tracks to the first links

7
Tracker ReconPattern Recognition Kalman Filter
approach

• Kalman Filter
• Particle trajectories are straight lines
• Changes in trajectory due to multiple scattering
  are gaussian in nature
• Pat Rec looks for gammas (vees), then for
  particles
• But
• Multiple scattering is not entirely a gaussian
  process
• Bremsstrahlung results in many low(er) energy e
  and e- tracks along principal path
• Leading to large scattering angles for principle
  e or e-
• And confusing the pattern recon

8
Tracker ReconPattern Recognition simple
approach

• Implement a Link and Tree algorithm
• Simplest algorithm to dive into the code
• Allows one to follow pre-shower development
• Longest, Straightest branch is trajectory of
  the primary e or e-
• Can be projected to calorimeter for initial
  clustering
• Passed to Kalman Filter for track fit
• Other branches can (hopefully) give more
  information on energy of primary e or e- pair
• As with Kalman Filter, Pattern Recognition runs
  in 2-D
• Association to 3-D done after initial 2-D
  tracking finding
• Strategy is to find individual tracks first
• Then put tracks together to form/find gamma
  conversions

9
Tracker ReconPattern Recognition simple
approach

• Algorithm
• Links formed between all pairs of clusters in
  adjacent layers
• Beginning with the top most layer containing
  cluster hits, links are combined to form a tree
  structure
• Links are not allowed to be shared
• Clusters are allowed to be shared
• Trees sorted by longest and straightest for
  association to 3-D

10
Tracker ReconPattern Recognition simple
approach
11
Tracker ReconPattern Recognition simple
approach

• Current Status
• Link and Tree algorithm in 2-D operational
• Rudimentary association of 2D tracks to 3D
  operational
• Tracks start in same layer
• Tracks have same length
• Straightest tracks associated
• Longest, straightest tracks are fit by Kalman
  Filter
• Only looking at single charged particles at this
  point
• Not quite ready for gammas

12
Tracker ReconSome initial results

• Look at TBsim e runs
• Positrons incident normal to the first tracker
  layer
• Energies 0.1, 0.25, 0.5, 1.0, 2.0, 5.0, 10.0,
  20.0 GeV
• 3-D Track Reconstruction of e requirements
• Track must be 12 or more layers in length
• Must start in first tracker layer
• Look at
• Track recon parameters
• Number reconstructed and passing above cuts
• Length of tracks
• Etc.
• Pointing at start of track comparing
• Compare between fit parameters at first hit and
  first link

13
Tracker ReconSome initial results

• Track Accounting
• ? 951/1000
• 95.1
• Number tracks/event
• ltNgt 1.6
• Length of tracks
• ltLgt 11.1

14
Tracker ReconSome initial results
15
Tracker ReconSome initial results

• 3-D Pointing Resolution
• Take mean value
• 2-D Pointing Resolution X
• Fit Gaussian
• 2-D Pointing Resolution Y
• Fit Gaussian

16
Tracker ReconSome initial results

• Resolution in 3-D pointing for Kalman Fit
  and for first link is approximately the
  same

17
Tracker ReconSome initial conclusions

• Link and Tree Pattern Recognition
• Simple algorithm implemented within the Centella
  context
• Shows promise for
• Finding primary e and e- tracks
• Keeping track of pre-shower development aid in
  helping to keep track of energy loss of the
  primary tracks
• Providing initial pointing into the calorimeter
• Dont need to know the energy before getting the
  track
• Track finding calorimeter track fit
  calorimetry track fit -
• More careful studies needed before really saying
  anything about pointing resolution
• Needs refinement ( rewrite) if really want to
  proceed
• New kids are getting to be conversant in c
• New kids have learned a (small) part of the GLAST
  system
• New kids have a much greater appreciation of
  the problem

18
Tracker ReconShort term plans

• Balloon flight needs tracking soon!
• Tested tracking exists within the centella
  framework
• Move the existing code
• The first goal is to move the existing TB_recon
  into the Gaudi framework
• Allows us to stop working on legacy code.
• Connect to New Geometry
• This will allow us to develop code which can be
  used for all GLAST configurations.
• Write new data converters for
• Test beam data and MC
• Glastsim output
• Cosmic data / Balloon flight
• Continue looking at Pattern Recognition
  alternatives

19
Calibration Issues

• There are three parts to each problem below the
  calibration algorithm, the database, and the
  automated production process
• Bad Strips
• Hot/dead strips
• Common mode failures Chips, ladders, towers
• Alignment
• Current status
• Whats ultimately needed
• TOT (Time-over-Threshold)
• Calibration signal?

20
Bad Strips

• Currently, the bad strips are recorded in an
  ASCII file, by layer and strip number. These are
  used by the reconstruction to kill bad strips and
  to join clusters separated only by bad strips.
• For the full detector
• The production database will record bad strips
  at different levels for example, chips,
  detectors, ladders, layers, and (shudder!)
  towers.
• Since the state of the strips will need to be
  monitored regularly, we particularly need a
  reliable automatic system to detect bad elements
  and update the database.

21
TKR Alignment

• An alignment was done on the BTEM using test beam
  data, first with entire layers, and then with
  individual ladders.

22
Finding Residuals
Method Find the a track. Fix a line through the
clusters in planes 8 15. Calculate the
residuals with respect to that line.
Fitting planes
23
Layers and Ladders

• The original residuals were as big as 200 µm. (s
  ? 40 µm)
• Layers were shifted to minimize the residuals.
• Two layers can be fixed (or the overall change of
  position and slope can be set to zero) because
  there are two degrees of freedom in the original
  problem
• The resulting residual distribution has s ? 25 µm.

24
Layers and Ladders (2)

• To improve resolution, the positions of
  individual ladders were adjusted with respect to
  ladders above and below, using normally-incident
  tracks, withthe final s ? 15 µm.

25
Alignment the New Frontier
A complication In the full detector, many tracks
will cross ladders and towers. Slanted tracks
allow the alignment of adjacent ladders and
towers. This is more complicated because now all
the elements are tied together with springs, and
there are six parameters per object x, y, z,
and 3 rotations. Solving this problem usually
leads to big matrices!
26
Time-over-Threshold (TOT)

• For each layer, the TOT is measured by combining
  all the fast-ORs for each event.
• The TOT measures the width of the pulse at some
  fixed pulse height, and is thus roughly
  proportional to the largest charge deposited on
  any strip in the layer.

Distribution of TOT values for a 20 GeV positron
run (normal incidence) with a Landau fit
overlaid. (Test beam data)
27
TOT (2)

• In the test beam, the TOT was sensitive to the
  photon conversion point. But this was at normal
  incidence. Will this still work for angled
  tracks? We have test beam data to answer this
  question! (I think)

• How do we calibrate the TOT?
• What is the correct level?

28
Plans for Simulation

• Glastsim/GEANT4 outputs MC truth
• Digis produced from MC hits
• Digis and hits can be read by Recon
• Upgrades to Generation
• Realistic Geometry
• Fluctuations (for TOT)
• Upgrades to Digitization
• Charge sharing
• Dead strips/chips/SSDs
• Overlay of background
• Model
• Real data
• TOT
• Common Geometry

29
Charge Sharing, Fluctuations and all that
Low
Calculated TOT response is sensitive to details
of the generation and digitization.
Medium
High
30
Balloon-specific Support

• Certain aspects of the balloon-flight data may
  require special support.
• Special reconstruction algorithms
• Dealing with high backgrounds
• Picking out photons in hadronic showers
• Analysis
• Projecting to active targets
• Finding interaction vertex
• Calibration
• Dead/hot strip list
• Probably no alignment required to reconstruct
  tracks, but we may want to demonstrate that we
  can do it. A use for the expected 107 protons?
• Same for TOT

31
A Preliminary Personnel/Task Inventory
Tasks Port of Recon to Gaudi Development of
Recon Simulation Calibration
Institutions Pisa Santa Cruz Santiago de
Compostela SLAC
People 2 3 1 (through mid-Feb)
2
32
A Preliminary Personnel/Task Inventory
And tentative set of matches...
Tasks Port of Recon to Gaudi Development of
Recon Simulation Calibration
Institutions Pisa Santa Cruz Santiago de
Compostela SLAC
People 2 3 1 (through Mid-Feb)
2
33
TKR Software Schedule

• Near-term (FebMar)
• Port recon to Gaudi
• Read TB data/MC -- verify port
• Refine recon algorithms
• Implement digitization
• Read/recon cosmic data
• Specify calibration databases
• Medium-term (AprMay)
• Refine digitization
• Read/recon Glastsim/GEANT4 digis
• Implement calibration databases
• Implement single-tower alignment algorithm
• Test alignment code with cosmics
• Implement hot/dead strip calibration
• Write special code for balloon flight
• Refine TKR-specific GEANT4 code

• Long-term (June )
• Refine balloon recon
• Perform balloon analysis
• Connect new geometry
• Implement full detector geometry
• Develop multi-tower alignment algorithm
• Demonstrate full detector capability
• gamma detection and measurement
• background rejection
• optimized PSF
• automated calibration

34
Thats All Folks!