SATrack Fitter Validation

1
SATrack Fitter Validation

• Sergei Avvakumov
• MINOS Collaboration Meeting
• September 16, 2004

2
Overview

• Fixes, Updates
• Validation Sample
• Comparison plots SR/SA fitters
• Conclusions

3
NtpSR modifications

• CandFitTrackSR data members
• fNDOF, fNiterate, fCPUtime, fVtxQPError,
  fVtxUError, fVtxVError, fVtxdUError, fVtxdVError
• Moved up the class hierarchy to CandFitTrack
• CandNtupleSR has been modified to fill these from
  any candidated derived from CandFitTrack
• Fit error matrix - ?

4
Fitter Improvements

• Eliminated C arrays -gt std vectors, lists
  (boundaries checking)
• Utilise ROOT new linear algebra facilities
  TMatrixDSym. More ROOT improvements expected
  (CBLAS )
• More accurate strip checking to eliminate hits
  that do not belong to the track

5
Validation

• Far Beam MC sample 9k events(December 2003MC
  before MDC)
• Processed with SR/SA fitters -gt NtpSR
• Check plots were made to compare all fit related
  reconstruction quantites script producing a set
  of histograms added to Contrib/RecoCheck
  repository

6
EVD
7
Comparison Plots
Test (SA Fitter)
Reference (SR fitter)
Reference and Test
Test/Reference-1
8
Truth Q/P
9
Truth Uvertex
10
Truth cosu(?)
11
Truth Vvertex
12
Truth cosv(?)
13
Reco Ntracks
14
Reco Track Length
15
Reco Prange
16
Reco Nplanes
17
Reco Q/P
18
Reco ?(q/p)
19
Reco reduced ?-square
20
Reco Q/P residual
Residual - res(q/p) (q/pfit q/pmc)/sigma(q/p)fi
t For a well behaving fitter ltres(q/p)gt
0. RMS(res(q/p)) 1. by construction
21
Reco CPU
Beam MC is a worst case for SA fitter -many
long (gt200 planes) tracks. Practical limit is
200x200 matrices.
22
Reco Uvertex
23
Reco ?(Uvertex)
24
Reco Uvertex residual
25
Reco cosu(?)
26
Reco ?(cosu(?))
27
Reco cosu(?) residual
28
Reco Vvertex
29
Reco ?(Vvertex)
30
Reco Vvertex residual
31
Reco cosv(?)
32
Reco ?(cosv(?))
33
Reco cosv(?) residual
34
Conclusions

• SA fitter faster, less bias
• SR fitter slightly more efficient