1. A First DC2 Analysis with New Recons
DC1 Recap
Recon Re-write Review
TkrRecon
CalRecon
Good Energy Analysis
PSF Analysis
THIS IS A PROGRESS
REPORT
ALL ANALYSIS ARE
PRELIMINARY

2. DC1 Recap Eff. = 82% Bad-Cal = 4.5%
Cuts: 1/1
Ratio 95/68 > 3
Meets SR
Events Eff.: 94.5%
Cuts: 3/2
Events Eff.: 52.3%
Cuts: 3/4
Events Eff.: 19.1%
Cuts: 2/1
An extensive analysis was performed
to optimize PSF and energy resolution
while maintaining the maximum Aeff.
This was the first complete analysis
done using the Geant 4 Simulation
(previous Glast Analysis used the
Gismo simulation)
Bad-Cal = 4.5%
Eff. = 82%

3. Background Rejection in DC1
3 Classes of Background after selection
for good energy and direction
reconstruction
Prob. g > .5
Upward moving events
Earth Limb events
Downward moving events
Prob. g > .9
A Classification Tree analysis was constructed
allowing cutting on the probability that an event
was a g (vs. Background)
However – to meet the Science Req. for Background
contamination – the remaining Aeff was lower then
the goal (it did meet the Sci. Req.)

4. Conclusions from DC1 Major Effort was Launched to completely re-write
The Status of the Reconstruction in GLAST was marginal
1) The code had been patch, re-patched, etc. many times
2) The code had become needlessly complex and
essentially un-maintainable
3) The algorithms (particularly for the Calorimeter) were
undeveloped
Major Effort was Launched to completely re-write
the Tracker and Calorimeter Recon
1) Emphasis on uniformity and simplicity
2) Migrate related task to one area
3) Re-vamp and re-assess poorly functioning
algorithms (esp. Vertexing and Calorimeter)

5. TDS Classes Condensed and Simplified:
Tracker Recon Re-write
TDS Classes Condensed and Simplified:
External Controls
Control Name
Internal Variable
Def. Value
Description
MinEnergy
m_minEnergy
30.
Min. energy to use for setting search regions
SigmaCut
m_sigmaCut
9.0
Sigma cut for picking up points
FirstTrkEnergyFrac
m_1stTkrEFrac .8
First track energy fraction
MinTermHitCount
m_termHitCnt 16
Min. no. of hits on best track to terminate search
MaxNoCandidates
m_maxCandidates 10
Max. allowed number of candidate tracks
MaxChisq
m_maxChiSqCut
40.
Max allow Combo Pat. Rec. Chisq. (1st fit)
NumSharedFirstClusters
m_hitShares 6
Number of first clusters which can be shared
MaxNumberTrials
m_maxTrials 50
Max. number of trial candidates to test
FoVLimit
m_PatRecFoV
.19
Minimum cos(theta) for track trials
MinCosKink
m_minCosKink .7
Minimum cos(theta) for a track kink
MaxTripletRes
m_maxTripRes
Max. un-normalized residual for first 3 TkrPoints
TkrFitHit
TkrFitMatrix
TkrFitPar
TkrFitPlane
TkrFitTrackBase
TkrKalFitTrackBase
TkrKalFitTrack
TkrPatCand
TkrPatCandHit
TkrRecInfo
TkrTrackTab
TkrTrackParams
TkrTrackHit
TkrTrack
Program architecture simplified

6. New Vertex Algorithm Vertex Location Combining Tracks
Multivariate Averaging:
Put Vertex
at Radiator
Mid-Point
Preferred Solution
If 2 Tracks share the same first hit
W RADIATOR
SSD PLANE
SUPPORT PANEL
Found
Tracks
Put Vertex
DOCA Point
SUPPORT PANEL
where Pi are the parameter vectors of the combination (Pair) and tracks (P1 and P2)
and Ci are the covariance matrices
where
Next Best Solution
If DOCA location of 2 tracks
lies before 1st Hit
W RADIATOR
SSD PLANE
All Other Case
Put Vertex at Z location
of start of the 1st Track
Found
Tracks
The parameter vectors are (x, Sx, y, Sy)

7. CalRecon Re-write 2) Better Background Rejection Cuts
1) Put GLAST’s Fracture Energy Back Together
1 GeV g
Thin Radiator
Hits
Gap Between
Tracker Towers
Thick Radiator
Hits
Blank Radiator
Hits
Gap Between
CAL. Towers
Calorimeter
Xtals
Leakage out
CAL. Back
2) Better Background Rejection Cuts
Shower Shape parameters
3) Move code to where it belongs!
AnalysisNtuple CalRecon

8. CalRecon Re-write (cont.)
A Major Change: Replace Layer-by-Layer Analysis with Energy-Moments
Energy Moments: Same as Classical Mech. Mass moments with energy
replacing mass (see Goldstein) (first used by S. Ritz in GLAST)
Long. Moment
Trans. Moment
LSQ Fit to Layers
Moments

9. Good Energy Analysis But First: The New Data Sets Initial Cuts:
- Major effort by many people
- Took a couple of shots to get right
- 1.5x106 All Gamma*
run 23-June-2005
- 6 x106 Orbit-Average Background
run 24-June-2005
- All data run on SLAC Batch Farm
available at GLAST ftp site
- These are the first large data sets
using new Recons
Initial Cuts:
CalEnergyRaw > 5 MeV and
CalCsIRLn > 4 rad. len.
2) Tkr1ZDir < -.2 (FoV)
color range:
Pruning Selections:
1) TkrNumTracks > 0 standard mode
2) TkrNumTracks = 0 Cal-Only Events!
*All Gamma: 2p str. , 18 MeV – 180 GeV,
1/E Spectrum

10. Energy Assessment EvtDeltaEoE = Top plots show dep. on MC parameters
was EvtMcEnergySigma
Top plots show dep. on
MC parameters
Bottom plots show dep. on
Recon parameters

11. Good Energy CT Analysis
Data divide into Energy Ranges using CalEnergyRaw
LowCal
MedCal
HiCal
CalEnergyRaw < 350 MeV
MeV
CalEnergyRaw > 3500 MeV

12. Energy CT Details Low CAL
Produce a set of variables that are ~independent of cos(q) and E
Parametric function of Tkr1ZDir and EvtLogEnergy
Define Class "GoodEnergy" & "BadEnergy"
GoodEnergy: (will explore dependence on this)
2) Derive a CT for each of the EvtEnergyRaw Bins
Low CAL

13. more Energy CT Details
Med CAL
High CAL

14. Energy CT Dependencies
Since the energy resolution is improved – try more restrictive
definition of "Good Energy"
100 80 60 20 40
100 80 60 20 40
100 80 60 20 40
Eff.
DC1 Choice
Select this one
NOISE
Expected Behavior
Anomaly
Use of
Tkr1TotTrAve
in Cal Low CT
This is part of a problem at
High Energies .... more later
What's happening here?

15. Preliminary DC2 Energy Resolution
Integrated Over FoV
On Axis
DC1 Resolution
DC1 Resolution
DC2 Resolution
DC2 Resolution
High Energy
Deficit caused
by Pat. Rec.
Confusion!
Aeff x DW = 4.02 m2-str
Aeff x DW = 2.76 m2-str

16. Root Cause of Hi E Deficit
Pictures and Words....
Incoming g
Event #6
Bad Cal Dir
These events come from Old CalRecon
but effect remains.
First attempt to fix problem:
USE CAL DIRECTION
100 GeV g
Event #9
Bad Cal Dir
Incoming g
Missed Track
100 GeV g
Event #6
Bad Cal Dir
Clearly there's more
work to do here!
Missed Track

17. A First PSF Analysis for DC2
Strategy: Divide and Conquer!
Split Events in Thin and Thick
Conversion locations
For Events with > 1 Tkr. Vertices
decide whether or not to use
vertexed solution
Example: Thin Radiator – Vertex CT

18. More PSF CT Details Strategy: Divide and Conquer!
3) Find and tag (a la probability)
Events with large PSFs
4) Finally – add a knob to describe
quality of reconstruction
- Using a REGRESSION TREE

19. Preliminary DC2 PSFs DC1 DC2 Integrated over FoV On Axis 2.6 46.2
Sel. 1
Sel. 2
Sel. 3
Sel. 4
PSF68(100MeV)
4.7
4.1
3.6
3.0
4.2
3.8
3.3
2.6
PSF95/PSF68
4.4
2.3
2.2
3.2
Rel. Eff. (%)
100
94.5
52.3
19.1
95.2
74.0
46.2

20. DC2 Background Rejection
Nothing has been done with the recently run data.
(only became available late last Friday)
But – Here's is a peek –
Preemptory cuts:
1) GoodEnergy.Prob > .25
2) AcdDOCA > 250
3) Only Thin Radiator Events Analyzed so far
Data: 5.1x106 Orbit Ave. Bkg.
.75x106 All Gamma
15642 Events
942 Events (Good News)

21. more DC2 Background Rejection

22. more DC2 Background Rejection
Energy & Angle Compensate CalTransRms
Note: In AnalysisNtuple all variables beginning with
EvtE..... are previously defined variables which
have been compensated for log(E) and cos(q)
dependencies.

23. Conclusions The 1st DC2 Workshop has already accomplished 90+%
its goals! - look at all the work it stimulated!
The Recon re-writes show good improvements in
- Energy Reconstruction
- PSF
- New Variables to Reject Background
The (first) large data samples are providing the next
level of insight in the Recons
- TkrRecon need more work in the area of
very High Energy reconstructions
- CalRecon needs MIP finding and more work
to refine moments analysis (emphasis on directions)