Back Ground Rejection Status

1
Back Ground Rejection Status
for DC2
a la Ins. Miner Analysis
2
First Look at glast-ts production
3-Oct-2005 Series 1 409160 in tuple / 123556
after RF (only FilterStatus _HI 0 used by TB)
(This series does not
have the single tile energy vars)
after Act.DistSSDVeto 82306
after GoodEnergyProb gt .1 81674 Series 2
408151 in tuple / 123120 after
RF after Act.DistSSDVeto
82440 after GoodEnergyProb gt .1
81795 Series 3 409098 in tuple /
123110 after RF after
Act.DistSSDVeto 82349
after GoodEnergyProb gt .1 81746 Each series
equals .01 days or 864 sec. on orbit with 744
per series live time due to the SAA
the 3 together represent
2232 sec. The following is from Toby's e-mail
dated 4-Oct-05 533 pm
Series incoming triggers downlink 1
38191926 6100518 409281 2 38053681
6078677 409415 3 38130276 6093686
409221 4 37845955 6054390
406189 total 152221838 24327271
1634106 Rates 8177 Hz
549 Hz
Note that the numbers I see are a little
different in the incoming ntuples Toby says
this is due to him taking numbers from the logs
rather then the tuples and failed jobs don't get
counted properly. Total live time 2976 sec.
for the 4 series
3
Vtx Topologies, THICK Radiators,
CalEnergyRaw lt 350 MeV (Note I'm trying breaking
the lowest energy bin into THICK THIN radiators
at Steve's suggestion)
Prefilter Exclude AcdUpperTileCount gt 0
AcdLowerTileCount gt 1 AcdRibbonActDist gt
-1999 where UpperTileCount Top 0 1 2 and
LowerTileCount 3
1000
AGs as indicated 1547 Bkg. Events
By Varying the of AGs used to train the CT,
the relative efficiency of the selection of Bkg
and AGs changes. These are for Prob(g) gt .50
700
500
400
300
200
100
The nominal rate for Bkg. in each channel
must be kept to lt .03 Hz
By changing the number of AGs - its equivalent
(? similiar at least) to weighting the events
except for the statistical fluctuations.
4
300 AGs
This is what the CT vars. and probs. look for 3
different mixes of AGs
BestEnergy gt 100 MeV
500 AGs
BestEnergy gt 100 MeV
1000 AGs
BestEnergy gt 100 MeV
5
Try Training CTs using BestEnergy gt 100 MeV
PreFilter
as before
BestEnergy gt 100 MeV
This is marginally worse... so I will inculded
all events to train CTs
The event mix changes with cos(q) as well.
We'll probably de-select on events near the edge
of the FoV (e.g. cos(q) lt -.3 rather going out to
cos(q) lt -.2)
No cut case
PreFilter
cos(q) lt -.3
as before
AG eff 129/235 .54 Bkg. eff 25/792 .03
Prob. gt .5
CT Variables and Prob for Vtx, Thick, lt 350 MeV
6
VTX, THIN, CalEnergyRaw lt 350 MeV, cos(q) lt -.3
This is much worse in terms of efficiencies
PreFilter
THICK eAG .87 eBkg .46
THIN eAG .73 eBkg .63
as before
PreFilter
THIN eAG .92 eBkg .84
Still its bad - but???? This PreFilter is
considerably worse for the THICK events then the
one dev. for it. THICK and THIN seem to be just
DIFFERENT
AcdActiveDist gt -199 AcdRibbonActDist gt -1000
THIN eAG .67 eBkg .02
Probgt.5
cos(q) lt -.3 BestEnergy gt 100 MeV
7
VTX, All Layers, 350 lt CalEnergyRaw lt 3500,
cos(q) lt -.3
PreFilter
All Layers eAG .97 eBkg .55
This g ineff. is consistent with conversions in
blanket and tiles
AcdActiveDist gt -199 AcdRibbionActDist gt -1990
All events now have BestEnergy gt 100 and cos(q) lt
-.3
8
VTX, All Layers, 3500 lt CalEnergyRaw, cos(q) lt -.3
The Remaining Bkg Event
PreFilter
e- (e- Splash)
Event Well Reconstructed
AcdActiveDist gt -10 CalTrackAngle gt .5
CalTrackDoca gt 40
This is NUTS! The event clearly has to be
pointing at an ACD Side tile and there is no
SSD Veto. The Doca Calc. says the event had a
track well within a tile so why is the
ActiveDistance set to its rogue value?
This is a glast-ts generated event.
Can Toby find it? (It comes from either series
2 or 3) We NEED a picture of it!
9
Vtx Topology Summary (All numbers for BestEnergy
gt 100 MeV BestZDir lt -.3)
Prob. gt .50
PreFilters
Background Rejection Path (BR.Path) not yet
determined
All Gamma
Loss .013 .120 .465
(g/Bkg)
Background
Background Source Breakdown
MIP Finder Failure
HeavyIons have gone away
10
1Tkr, THICK layers, CalEnergyRaw lt 350 MeV
Desperation This is the toughest subclass. In
DC1 we had to through all these away. This is
better but clearly not there yet.
cos(q) lt -.3
PreFilter
AcdActiveDist gt -199 AcdRibbonActDist gt -1999
AcdDoca lt 1999 CalTrackDoca gt 200
EvtECalTransRms gt 2.5 CalMaxXtalRatio gt .8
Tkr1FirstChisq gt 2.5 Tkr1ToTTrAve gt 2
Note that 50 of the events are below
BestEnergy lt 100 MeV
THIN eAG .68 eBkg .03
Probgt.5
cos(q) lt -.3 BestEnergy gt 100 MeV
11
1Tkr, THIN layers, CalEnergyRaw lt 350 MeV
cos(q) lt -.3
PreFilter
Desperation This is the 2nd toughest subclass.
AcdActiveDist gt -199 AcdRibbonActDist gt -1999
CalTrackDoca gt 200 EvtECalTransRms lt .8
Note that 40 of the events are below
BestEnergy lt 100 MeV
THIN eAG .72 eBkg .04
Probgt.5
cos(q) lt -.3 BestEnergy gt 100 MeV
12
1Tkr, All layers, 350 MeV lt CalEnergyRaw lt 3500
MeV cos(q) lt -.3
PreFilter
AcdActiveDist gt -199 AcdRibbonActDist gt -1900
CalTrackDoca gt 40 CalTrackAngle gt .5
CalXtalRatio gt .85
All Layers eAG .73 eBkg .04
Probgt.5
cos(q) lt -.3 BestEnergy gt 100 MeV
13
1Tkr, All layers, CalEnergyRaw gt 3500 MeV
cos(q) lt -.3
Re-Make of 1Tkr Cal_Hi
PreFilter
This PreFilter was not done properly first time
around. Here's a re-do
CalTrackDoca gt 30 CalTrackAngle gt .3
Leaving out all ACD vars. in PreFilter
All Layers eAG .98 eBkg .015
Probgt.5
Note however that the CTs insist on using the ACD.
cos(q) lt -.3 BestEnergy gt 100 MeV
14
2nd Time Around
1Tkr Topology Summary (All numbers for
BestEnergy gt 100 MeV BestZDir lt -.3)
Prob. gt .50
PreFilters
Background Rejection Path (BR.Path) not yet
determined
All Gamma
Loss .075 .394 .641
(g/Bkg)
Background
Note how much larger the Bkgs are w.r.t. Vtx
except for CalEnergyRaw gt 3500 MeV
Background Source Breakdown
MIP Finder Failure
15
v7r2 SLAC 2M All Gamma Run
Pr(GAM) gt .50
Number gt 1 GeV 1726 in 28 bins
62/bin x 50 bins 3100 3100/(2x106 / 100) x 6
m2 ltNBINgt 150 Aeff (cm2) or Ngt1GeV 5.36
Aeff(cm2) Aeff 9300 cm2
2469/bin
SRD Limit
Number gt 1 GeV 70883 in 28 bins
(2531) x 50 Bins 113500 113500/2x106 x 37.7
m2-str ltNBINgt .943x10-3 Aeff x DW(m2-str) or
Ngt1GeV 3.37x10-5 Aeff x DW(m2-str) Aeff x DW
2.39m2-str
Fall-off past 10 GeV improved however this
results from the CTs using AcdActiveDist to
kill the high energy cosmic electron component
in the background it may be a fact of life
for now...
Background 319 Events or .16 Hz
(SRD .03 Hz)
16
cos(q) lt -.9
Near On-Axis
These basically meet the SRD
cos(q) gt -.6
Edge of FoV
17
Direction Anti-Correlated events 416 Events
Residual Background Dissection
PSF Class
Prob.GAM gt .3 (relaxed cut) 1151 Events
These are dominated by upward moving protons.
This is what the MIP Finder was suppose to
identify
No Correlation 560 Events large component of
Limb gs
PSF Class
Direction Correlated events 175 Events
PSF Class
Location of First Hit on Found Track
Come in through the corners and there is a
definite handedness this suggest a ERROR IN THE
ACD CODE / GEOMETRY
18
Event Spectra both MC Truth and as Reconstructed
After PreFilter Stage No Prob.GAM Cut
61 Events Rate .03 Hz
Prob.GAM gt .7 (Meets Aeff and
Background rate req.)
62352 events gt 1 GeV Aeff x DW 2.1
m2-str