DC1 Instrument Response: DC1 Instrument Response Agenda
Where are we?
V3R3P7 Classification Trees
Covariance Scaled PSF
Slide2: A Brief History of Resolution andamp; Rejection Preparing for DC1 is a LARGE TASK
- Not likely to get right the 1st, or the 2nd, or the 3rd, or.... time!
1st Time: April-May
Discover Mult-scattering in G4 'too good to believe!'
Took till end of June to fix!
2nd Time: July (SAS Workshop)
OOPS! The ACD geometry!
3rd Time: July-August
Where did all the Run Numbers go?
4th Time: August
Will Bill never stop changing variables - well at least
he shouldn't make so many coding errors! Steve's variables added.
5th Time: August-September
Data of the day! But it's certainly not 'The rest of the story!'
6th Time: .... IS A CHARM! Slide3: A Brief History Continues!
6th Time: NOT CHARMED! September:
ToT's found to be effective at removing range-outs! Code added to
explore this handle on Backgrounds
7th Time: October- November
ACD ribbons added to seal up ACD cracks. Code added to analyze Ribbons.
5M All-Gammas produced over [18 MeV, 180 GeV] andamp; 2p str.
8th Time: November-December
Ribbon andamp; Tile Geometry discoveries!
9th Time: December 3
Background Data delivery: 160M+ BGEs. Note: just the BGEs
have been run. All-Gammas awaiting. Credit goes to Heather
and Berrie. THANK YOU!
Slide4: NoCal: andlt; 2 r.l. or andlt; 5 MeV
LowCal: andlt; 350 MeV
MedCal: andlt; 3500 MeV
HighCal: andgt; 3500 MeV CAL Energy Def's and
'Good': Energy CTs Slide5: Energy CT's Probability Evaluation Sci. Req. Cut for V3R3P7 (.2)
All-Gamma Eff.: 92%
Energy Tails: andlt; 10% All-Gamma Eff.: 78%
Energy Tails andlt; 3%
Slide6: DC1 Energy Post Good-Energy Cut: IMgoodCalProb andgt; .2 Energy s
Energy Energy s
cos(q) Meas. Energy
M.C. Energy Meas. Phase Space
cos(q) Slide7: Rome: Thin PSF's - Integrated over FoV
4 Combinations of Cuts (CORE/Pred) Cuts: 1/1 Ratio 95/68 andgt; 3 Meets SR
Events Eff.: 94.5% Cuts: 2/1 Cuts: 3/2 Events Eff.: 52.3% Cuts: 3/4 Events Eff.: 19.1% Slide8: DC1: Thin PSF's - Integrated over FoV
4 Combinations of Cuts (1-CORE/4-Pred) Cuts: 2/1 Cuts: 2/2 Cuts: 2/3 Cuts: 2/4 Slide9: Definitions: Where all the variables come from the Merit-ntuple.
(See my covariance ppt for details on Tkr1ThetaErr and
Tkr1PhiErr - these are derived from the covariance
matrix elements, event-by-event) Covariance Scaled PSF's A bit of math then shows that: and (from Covariance.ppt presentation to Analysis Group, July, 2003) Slide10: 18 andlt; E andlt; 56 56 andlt; E andlt; 180 180 andlt; E andlt; 560 560 andlt; E andlt; 1800 1800 andlt; E andlt; 5600 Comment:
Works well except in regions
where energies fed to Kalman
Filter are inaccurate. Specifically
below 50 MeV and above 10 GeV 5600 andlt; E andlt; 18000 18000 andlt; E andlt; 56000 56000 andlt; E andlt; 180000 Scaled PSFs: Energy Dependence Slide11: -1andlt;cos(q)andlt;-.8 -.8andlt;cos(q)andlt;-.6 -.6andlt;cos(q)andlt;-.4 -.4andlt;cos(q)andlt;-.2 McEnergy andlt; 10000 MeV Scaled PSFs: Angle Dependence On Axis Edge of FoV Slide12: 1) Scale Factors adj. to 2.38 andamp; 3.36
Thin /Thick respectively
2) IMcoreProb andgt; .2 andamp;
IMpsfErrPred andlt; 3.
Universal PSF Curve??? 3) Energy cut: .5 andlt;Tkr1ConEne/EvtEnergySumOpt andlt; 1.
Note: This cuts out almost 1/2 the data !!!! (44.4%)
Slide13: Pair Energies: The Missing Half Only Valid Region: [.5, 1.) Optimization done
using: Slide14: Pair Energies: The Missing Half (2) Optimization done
Consraint to 'QED' Slide15: Backgrounds: A First Look Input: - 27 BGE Events Files
- 168 MB/File 4.5 GB Total
- 161.5 M BGE Generated
Prune Step: AcdActiveDist andlt; -20 andamp; AcdRibbonActDist andlt; -20
Tkr1SSDVeto andgt; 2
Reduces BGE sample by 3.7X.
This used SSD Tracking layers as 'back-up' Vetos.
Tkr1SSDVeto Definition: # of live SSD back along trajectory from
start of First Track to ACD.
Adding Trk1SSDVeto saves 1/2 the killed g's allowing ~ +15% BGEs. Slide16: Yellow/Orange Bins: Cos(Mc-Theta) -1 to 1. All Gamma Sample
only over cos(q) andlt; 0. This is close
to the irreducible
limit from conversions
in the ACD andamp; Blanket
AND it's ~ flat in energy. Events Lost Events lost due to Global Veto Cut
7.6% AG's Lost andamp; 73% BGE's 4.8% 6.1% ~ 78% BGE's get killed on axis Slide17: CT Pruner Step Event Sample much too large (410 Hz Orbit Average Rate)
First allow only !NoCal Events: 3.7x BGE Reduction (107 Hz)
Apply a CT based secondary pruner build on Reconstruction Primatives.
55 Hz 97% Slide18: Good Cal Energy andamp; Minimal PSF Cut Good Cal Energy ( Prob.GoodCal andgt; .2): Remaining Rate: 23 Hz
All-Gamma Efficiency: 91.5% (Total so far: 82%) Efficiency vs Energy Cutting on PSF(CORE) (Kills PSF Tails - Prob. andgt; .2): Remaining 13.2 Hz
All-Gamma Efficiency: 88% (Total so far: 73%) - 2.2 m2 - str. Efficiency vs Energy Slide19: Final CT Background Tree Processing Good Cal Energy ( Prob.GoodGam andgt; .5): Remaining Rate: 1.3 Hz
All-Gamma Efficiency: 87.4% (Total so far: 63.4%) - 1.93 m2 - str. Now the 3 Problem
classes are clear! Slide20: Conclusions and Future Work 1) We're NOT THERE YET! Stay tuned - for the Rest of the Story.
2) V3R3P7 CTs are in production version of GLAST Sim
3) An event-by-event PSF analysis seems to be achievable
4) We straighten out the energies used in the Kalman Fit.
5) Backgrounds - First look at statistically useable event samples.
- Need to back the filtering up-stream to better
manage local resources!
GOAL: To have a Background Analysis in hand by DC1 Close-Out