Cal Moments Analysis

1
Cal Moments Analysis

• Have implemented an updated version of Cal
  Moments Analysis which was default up until
  switch to Gaudi (I think)
• New Version implement as CalMomentsAnlaysis class
• Accepts a vector of data objects containing
• Position
• Weight
• Distance to an axis
• Implements an iterative analysis
• Step 1 Determine moments
• Step 2 Throw out points furthest from axis
• Step 3 iterate until points within a defined
  tolerance
• Now at the head of CalRecon

2
Cal Moments Analysis

• First quick look
• 50K allgamma run
• Really around 45k since hit the time limit in
  batch queue
• Analyze output ntuple
• Minimal cuts applied
• Standard GltWord gt 3 TkrNumTracks gt 0 from
  System Tests
• Try to look at Cal pointing
• Look for problems with Moments analysis
• First problems
• Nothing in ntuple to indicate when moments
  analysis fails
• Or I dont know of anything yet
• Special categories of events
• Moments analysis fails in recon
• Moments analysis gets wrong cluster axis?

3
Pointing Resolution(Low Statistics)
Poor Mans PSF vs MC Energy Profile plot of PSF
vs energy (which plots the mean of data within
range of histogram)Really just to compare, not a
real PSF (as we know)
4
Cal Moments AnalysisAnalysis Failure

• Look for something which tells us when moments
  analysis succeeded

Convert Transverse Moment to a scaled transverse
rms mT transverse moment mT,rms
v(mT / Cal Energy) Now a measure of width of
hit crystals transverse to axis
Distribution of this scaled transverse rms shows
a spike at mT,rms 0.
Scaled Cal Transverse RMS
5
Cal Moments AnalysisAnalysis Failure

• Look for something which tells us when moments
  analysis succeeded

Moments Analysis failure signalled in ntuple (is
there another way to see this?) when the
tranverse moment is identically zero.
Normalized Transverse Moment
Still have some failures of moments analysis even
when reasonable fraction of radiation
lengths.Stay tuned for more on this
Moments Analysis Fails hereEvent did not pass
through the calorimeter, or Tracker does not
point back to Cal, etc.This is the bulk of the
failures
Cal CsI Radiation Lengths
6
Cal Moments AnalysisAnalysis Failure

• Effect of Moments Analysis Failures?

Calculate Angle between incident Monte Carlo
gamma and Moments Axis in Calorimeter
Calorimeter PSF
When Moments Analysis Fails it returns (0.,0.,0.)
for an axis in ntuple which gives an MC to Cal
angle of 90º
54 Events (of 3125 in plot) have an angle of
exactly 90º to MC.
Still a long tail resulting from bad axes from
moments analysis what are these?
No way to save these events, cut them out for
rest of this study
MC-Cal Angle (radians)
7
Cal Moments Analysis

• Look at Moment Analysis problems/resolution in
  more detail
• 50K 1 GeV Normal Gamma run
• All events same energy and direction (straight
  down)
• Only launch position different
• Uniformly populates top face of LAT
• Although I dont know if orbit/LAT pointing in
  this source?
• See messages about SAA in log file
• Use to look at pointing problems
• In particular, look at events in the far tail
  region of the Cal PSF plot
• Minimal cuts applied
• Standard GltWord gt 3 TkrNumTracks gt 0 from
  System Tests
• Also cut on CalTransRms gt 0

8
Cal PSF Revisited1 GeV Normal Gammas
Two effects 1st No events past 1.7 radians,
compared to AllGamma run. So, problem we were
looking for has gone away? See next slide 2nd
Now have a pile up of events with a Cal axis that
is wrong by about 90º from incident photon
direction. What is happening here? (see slides
following the next slide)
9
Cal PSF vs MC Cos(?)AllGamma
Moments Analysis applies an axis signing
convention with cos(?) positive. At large
incoming photon angles it is easy to flip this
axis (since cos(?) 0), in some cases the cal
axis will be 180º to the incoming photon (so, it
points in exactly the wrong direction!) This
cannot happen for photons incident normally on
the LAT, here the worst you can do is get the
axis wrong by 90º Final Cal axis signing needs
to be thought about some more
AllGamma Run
Flipped axis events
1 GeV Normal Gammas on previous slide populate
this region
Cal PSF (radians)
Monte Carlo cos(theta)
The Good Events
10
Bad Axis LocalizationIts a gap problem only
on-axis problem?
1 GeV Normal Gammas
Cal Centroid X vs Y all events
Cal Centroid X vs Y PSF gt 1.4
Cal Centroid Y PSF gt 1.4
Cal Centroid X PSF gt 1.4
11
Illustration of Problem w/ Event Display1 GeV
Normal Gammas

• Gap problem easy to see in Event Display
• Restrict launch point to x400, y400, z1000
• Not hard to find example events with this source
  (Event 3 in this example)
• The problem with this event is that most of the
  gamma energy is missed in the gap, the outliers
  have disproportionate energy for their position
  and the axis is pulled by them.
• Once a bad axis calculated, not much hope for
  recovery in the current iterative analysis as it
  throws out outliers based on distance from axis

Incoming Gamma
Conversion Point
Cluster Axis
Cluster Centroid
Blue boxes are MC Integrating hits Red boxes are
hit crystals Blue crosses are cluster layers
12
Possible to identify bad events?1 GeV Normal
Gammas
Plot Transverse RMS vs Longitudinal RMS
Distribution of all events
Distribution of events with a Cal PSF gt 1.4 Just
to see what happens, apply a cut with Transverse
RMS gt 30 Longitudinal RMS gt 35
13
On Axis Cal PSF with Selection1 GeV Normal Gammas
Not complete solution but does look like both the
transverse and longitudinal RMS can be used as
measures of the quality of the Cal axis in the
current ntuple. Need to work on First,
Fixing, to the extent possible, the Cal axis
(clustering algorithms?)Second Finding better
variables for determining the quality of the axis
determination
1 GeV Normal Gammas
Blue All Events Red Cut Events
MC-Cal Angle (radians)