• looking at geometry issues
  • performed cycloidal track fit - results uploaded to the wiki page; no obvious misalignment (few mrad which is probably compatible with bias in calculation)
  • looked at space point residuals - results uploaded to the wiki page; no big residual which might indicate a misalignment
  • no evidence of a misalignment


s-z cut
  • nturns algorithm assesses the number of turns; gives "real phi"
  • s is the "real phi"*radius
  • then fit straight line to s
  • Apply chi2 cut to determine if it is a good track
  • for each station, plot residual between fitted s and measured s
  • plot that against s length (which is helix radius times azimuthal change in angle for entire radius)
  • expect mean of 0 with 1 mm RMS
  • getting RMS ~ 10s mm
  • Proposed actions:
    • check vs MC; if we see it in MC it is algorithmic; if we see it in data, it is some combination of algorithm and data
    • Check for correlations between circle fit errors and s-z fit errors (are errors in circle fit pulling the s-z fit out?)
    • put a series of counters in to see where the fit fails
  • n_turns algorithm; there is a std::vector<>::push_back() and a std::vector<>[index] before the push_back(); so attempt to dereference an unallocated vector element.
  • Proposed action: keep digging and writing unit tests
  • TKU, TKD is okay(ish) in MC
  • TKU is okay(ish) in data
  • TKD is bad in data
  • Either:
    • Uncorrelated Noise/dead channels
    • Alignment
    • Correlated noise
  • Discussion of how noise can create inefficiency
    • Assertion that uncorrelated noise should not create sufficient inefficiency
    • Circle fit has too generous road cut, includes noise and pulls off radius; which effects the s-z fit?

Updated by Rogers, Chris over 7 years ago ยท 3 revisions