ppt - glast

Tracker Reconstruction Software
Performance Review, Oct 16, 2002
Summary of
Core “Performance Review”
for TkrRecon
How do we know the Tracking is working?
GLAST Science Analysis Software
Wednesday, Oct 17, 2002
The TkrRecon Group
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
Is the Tracking working?
The Simple Answer
YES!
2
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
How well is the Tracking is working?
The real question


We know the tracking is working to a certain level:
–
–
–
–
The event display tells us we are not that far off most of the time
Basic distributions of TkrRecon output quantities look reasonable
Output from merit is not complete garbage
Bill Atwood’s presentation from last week outlined some studies he has done:
–
Gut feeling is that base performance already better than PDR
•
•
•
Studies of track fits with muons
Studies of 100 MeV gammas
Some initial PSF plots
Detailed studies of TkrRecon performance are in the early stages
–
Yesterday’s report outlined the following:
–
Other tasks in progress
•
•
•
•
•
Event Display
Initial studies of Simulation/Digitization/Clustering
Some Basic TkrRecon output distributions
Some very preliminary comparisons of MC predictions versus TkrRecon output
Algorithm timing
•
•
•
TkrReconTestSuite – help to compare the performance of different algorithms
Vertexing algorithm studies
Etc.
3
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
The Event Display
What does it tell us?

Pattern Recognition
–
–

Event Display
–
–

Problem: associate clusters to
form tracks
Single biggest problem for
TkrRecon
Human eye very good at solving
pattern recognition problem
Can very quickly “see” major
flaws in the pattern recon
Example
–
–
–
1 GeV muon
Point source (z = 1000)
-1.0 < cos(theta) < -0.8
4
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
Event Display
Gamma Example

Example
–
–
–
100 MeV gamma
Point source (z = 1000)
-1.0 < cos(theta) < -0.8
5
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
McHits, Fit to Landau Distribution
“expect” 0.155
MeV
We need to investigate the “landau” distribution in thin layers
6
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
McHits, Energy Deposit
“long” hits
“long” hits
!!!
“short” hits
Energy from photons comes from Compton scatters below Geant range
cutoff, and from stopping photons. We need to fix digitization to dump all
this energy into one or two strips.
7
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
Summary of Simulation and Digitization

Energy loss of charged particles
–
For both electrons and muons, the energy loss in the silicon is less that predicted
by the naïve wallet card formula, even without considering the relativistic rise:
•
–

<eloss> = 0.140 MeV (peak at 0.166 MeV) compared to expectation of 0.155 MeV
Need to understand effect of thin radiator, perhaps do a better count of
secondary energy deposit.
Energy loss of neutral particles!
–
–
–
Because of the way energy below the cutoff is counted, photons appear to lose
energy in the silicon when they emit a low-energy compton, or such-like. With a
range cutoff of 0.7 mm, there are photon energy losses up to about 0.15 MeV
These losses are not handled correctly in digitization. The loss is spread evenly
among all of the strips crossed by the photon, whereas it almost certainly occurs in
on place. This leads to an underestimation of this source of noise.
The fix:
•
•
If the photon stops in the silicon, add all the energy to the strip in which it stops.
If it exits, add the energy to one of the strips that it crosses, randomly.
8
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
TkrRecon – Basic Recon Plots
100 MeV Gammas
-Back
-Front
-Front
-Back
-Back
-Front
9
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
TkrRecon / Monte Carlo Comparisons
Monte Carlo prediction (from McPositionHits, etc.) vs Recon output
100 MeV Gammas generated into the cone –0.8 < cos(theta) < -1.0
-Recon
-MC
-Recon
-MC
-Recon
-MC
-Recon
-MC
10
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
TkrRecon / Monte Carlo Comparisons
Monte Carlo prediction (from McPositionHits, etc.) vs Recon output
100 MeV Gammas generated into the cone –0.8 < cos(theta) < -1.0
Reconstructed Gamma
Pointing Resolution
Front
Blue is MC
Red is Recon
Back
11
Tracker Reconstruction Software
Performance Review, Oct 16, 2002
TkrRecon Algorithm Timing
Clustering
Track Finding
~ 2 ms
~ 400 ms
Total TkrRecon
Track Fit
~ 400 ms
~ 7 ms
~10 ms resolution
Vertexing
Reconstruction time dominated
by Track Finding (No surprise)
~ 200 us
12