TkrRecon Update
Mar 18, 2004

• Status of track fit c2
  • History: In the summer of 2002 Bill Atwood ran a complete set of "sea trials" for the Kalman Filter track fit. These trials used 10 GeV muons (to minimize other interactions) generated in the range -1 < cos(q) < 0. During the fit, the "true" energy of the muon is given to the fitter which should result in a track fit c2/dof with mean ~1. These tests were reported to the Analysis Group on August 12, 2002, the presentation can be found here.
     
  • Recent: Spurred by some desired improvements to the track fits by Bill and Johann, Bill dusted off his code to confirm the previous results with the latest code (many changes including new G4 version - remember the MS debacle, new  TDS organization, reorganization of some tracking code, new internal tracker recon geometry, etc.). He was not able to reproduce those results.
     
  • Illustration of the current situation:
    
    
    The main points to note:
    • The distribution of track fit c2/dof has a mean ~0.7, not 1.
    • Plotting c2/dof vs number of hits per track shows a slight dependency on the number of hits per track
    • Plotting c2/dof vs MC cos(q) shows a clear dependency which appears to dip to a minimum for q ~ 30°
    • Plotting c2/dof vs track f hints at some periodic structure.
       
  • Referring back to the presentation of August 12, 2002 one notes that none of these structures were present at that time.
     
  • Currently working on a process of elimination to pin down the likely culprit. At this point believe that neither the simulation (Geant4), or the propagator are the culprits since two tests have been run which do not eliminate the structure:
    • Have taken the propagator out of the track fits
    • Have turned off all muon interactions in the simulation
       
  • The following plots illustrate the comparison of the "standard" test job (blue hist and black points) versus a job with no multiple scattering in the simulation and no propagator in the track fit:
    
    
    
    The  bottom right plot is c2/dof vs track f for -0.9 < cos(q) < -0.7 which shows a clear periodic structure.
     
  • In addition, have set up a completely independent Kalman Filter track fitter using a Monte Carlo pattern recognition to cross check the default reconstruction system. Currently, both methods agree on the current situation.
     
  • Currently looking at:
    • Subtle geometry errors in the track fit geometry?
    • Assignment of spatial errors in track fit
    • ?

T.Usher Last Modified: 03/23/2004 08:12