Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
All-electron analysis
Group A meeting
Update on signal selection
Valerio Vagelli
Status at Jan2014 GM
Binning● Redefinition of binning
– Same binning up to 100 GeV (3 sigma)
– 5 sigma binning up to 220 GeV
– Bins ad-hoc above
● Results in a more smooth spectrum in [100-200] GeV
Double_t AMSBinning[58] = { 0.50, 0.72, 0.98, 1.29, 1.65, 2.05, 2.51, 3.01, 3.57, 4.19, 4.87, 5.60, 6.40, 7.26, 8.20, 9.20, 10.28, 11.44, 12.68, 14.01, 15.43, 16.95, 18.57, 20.30, 22.14, 24.10, 26.18, 28.40, 30.76, 33.26, 35.93, 38.76, 41.76, 44.96, 48.35, 51.94, 55.76, 59.81, 64.11, 68.67, 73.51, 78.64 ,84.08, 89.86, 95.98, 102.47,113.95, 126.57, 140.47, 155.76, 172.59, 191.10, 220.00, 260.00, 350.00, 500.00, 700.00, 1000.00 };
Energy Corrections● EcalShower::GetCorrectedEnergy(2,2) instead of EnergyE
– Templates and efficiencies calculated in the same energy definition
– We need to re-evaluate the systematic due to energy migration
– The very last bin [700-1000] GeV is now more in line with the others ( probably less contamination? )
Statistics added● Added ISS.B620/pass4 5 months of new data
– Preselection routine run. Files available in CASTOR/CNAF● We tried to use the really same selection as before, some
differences may come from bug-fixes to the patch / different DB
Statistics added● Added ISS.B620/pass4 5 months of new data
DOE Review
TOF scans
● Seems there is a slightly higher amount of events/time in the last preselected events
Difference in trigger efficiency at low energies
More evident in the ECAL triggers
BDT v6● BDT v6
– trained in CorrEne bins
– More training statistic
– “Same” cuts to define training sample as in the main analysis
BDT v6
*with a “temporary” choiche of templates
Errors comparison
● Zhili○ Us
● Zhili○ Us
● Zhili○ Us
*BDTv6 analysis
Roofit Error
Toy: check how the RooFit error changes with different fits scenario.
GENERATE AND FIT WITH SAME TEMPLATES → TEMPLATES ARE 100% CORRECT1) One Gaussian + One Gaussian with tail (Cball,n=1,alpha=4), б=1, different separations, S/N=0.2
GENERATE AND FIT WITH DIFFERENT TEMPLATES → 1 TEMPLATE IS NOT CORRECT2) Generate two gaussians, б=1, different separations, S/N=0.2 Fit with a wrong gaussian (gaussian with tail (Cball, n=1.5, alpha=8) ) Simulates electron tails in proton templates
3) Generate two gaussians, б=1, different separations, S/N=0.2 Fit with a gaussian with less shoulder (splitted gaussian, бleft=0.85, бright=0.85) Simulates a bias induced by ECAL selection on the shoulder on the opposite side from the signal in the proton template
Roofit Error
μ=±5 μ=±2
μ=±1 μ=±0.5
1) One Gaussian + One Gaussian with tail (Cball), б=1, different separations, S/N=0.2
Fiterror/sqrt(fitted)
Roofit Error2) Generate two gaussians, б=1, different separations, S/N=0.2 Fit with a wrong gaussian (gaussian with tail (Cball, n=1.5, alpha=8) ) Simulates electron tails in proton templates
μ=±5 μ=±2
μ=±1 μ=±0.5
Roofit Error3) Generate two gaussians, б=1, different separations, S/N=0.2 Fit with a gaussian with less shoulder (splitted gaussian, бleft=0.85, бright=0.85) Simulates a bias induced by ECAL selection on the shoulder on the opposite side from the signal in the proton template
μ=±5 μ=±2
μ=±1 μ=±0.5
Selection Systematics● Mainly given by the ambiguity to determine the shape of
the proton templates from data– R>0 protons, R<0 protons, AllR protons
– No cut on BDT, BDT<0
– Strenght of E/P cut
● All these cuts bias the shape is some direction
Selection Systematics● Different template definition gives different efficiency curves
R<0, no BDT cut→ too much electrons in tail
● The most visible effect is the rigidity sign, but also the BDT cut and E/P have an effect
*BDTv6 analysis
Selection Systematics● Effect on the final result
Template used for the signal extraction
Tem
plat
e u
sed
for
the
effic
ienc
y de
term
inat
ion
Fixed BDT cut (noBDT in this case)Fixed R cut (R>0 in this case)Change the strenght of the E/P cut induces a trend
Selection Systematics● Effect on the final result
Template used for the signal extraction
Tem
plat
e u
sed
for
the
effic
ienc
y de
term
inat
ion
Fixed R cut (R>0 in this case)For a given E/P cut, applying or not a BDT cut on the proton template gives a different result
No BDT cut
BDT<0
Selection Systematics● Effect on the final result
Template used for the signal extraction
Tem
plat
e u
sed
for
the
effic
ienc
y de
term
inat
ion
Adding also negative protons, again changes the result
No BDT cut
BDT<0
R>0 No cut on R
Selection Systematics● Effect on the final result
Template used for the signal extraction
Tem
plat
e u
sed
for
the
effic
ienc
y de
term
inat
ion
The same is true for fixed signal extraction and changing the template used to evaluate the efficiency
Selection Systematics● Effect on the final result
Template used for the signal extraction
Tem
plat
e u
sed
for
the
effic
ienc
y de
term
inat
ion
We can think to add also the templates with R<0.This has some sense only when fitting efficiencies (because we fit the negative sample)Also, we cannot use these templates without a cut on BDT, otherwise the signal contamination in the electron tail fakes the normalization, which is critical for the efficiency evaluation.Therefore, the only meaningful addition is the highlighted one.
But: it is correct to add these to the final result? The systematic also will increase by a lot!
Same square
Asbefore
R<0R<0
R<0
Selection Systematics● A reasonable choice of templates
– Signal extraction:
● R>0 + allR, noBDTcut, EoP [0.5-0.7]
● R>0 + allR, BDT<0, EoP [0.7-0.9]
– Efficiency evaluation
● R>0 + allR, noBDTcut, EoP [0.5-0.7]
● R>0 + allR, BDT<0, EoP [0.7-0.9]
● R<0, BDT<0, EoP [0.7-0.9] AllR
NoBDTcutEoP [0.5-0.7]
AllRBDT<0
EoP [0.7-0.9]
AllRBDT<0
EoP [0.7-0.9]
AllRNoBDTcut
EoP [0.5-0.7]AllR
BDT<0EoP [0.7-0.9]
R>0NoBDTcut
EoP [0.5-0.7] R>0BDT<0
EoP [0.7-0.9]
*BDTv6 analysis
Selection Systematics● A reasonable choice of templates
– Signal extraction:
● R>0 + allR, noBDTcut, EoP [0.5-0.7]
● R>0 + allR, BDT<0, EoP [0.7-0.9]
– Efficiency evaluation
● R>0 + allR, noBDTcut, EoP [0.5-0.7]
● R>0 + allR, BDT<0, EoP [0.7-0.9]
● R<0, BDT<0, EoP [0.7-0.9] AllR
NoBDTcutEoP [0.5-0.7]
AllRBDT<0
EoP [0.7-0.9]
AllRBDT<0
EoP [0.7-0.9]
AllRNoBDTcut
EoP [0.5-0.7]AllR
BDT<0EoP [0.7-0.9]
R>0NoBDTcut
EoP [0.5-0.7] R>0BDT<0
EoP [0.7-0.9]
*BDTv5 analysis
(Old) ECAL Simple Cuts efficiency● Cut on the energy deposit in the first 3 superlayers (normalized on data)
– F1SL in [9%-91%] efficiency
– F2SL in [9%-100%] efficiency
– F3SL in [9%-100%] efficiency
● Cut efficiency estimated from MC
Is this a real trend? Need more MC
(new) Simple Cuts
Mainly transformationsNeed to gaussianize variablesMore stable than box-cox transf.