Transcript kaneko_Status_of_AIF_analysisx

Status of AIF analysis
Daisuke Kaneko
Reminder 1
Decided issues
BG PDF : Contains “true AIF” and “accidental AIF”
→ Use trigger 0 data directory
Signal, RD PDF : Only “accidental AIF”
→ Simulate accidental by shuffling γ and e+ combination.
PDF
value
entry
comment
Projection
3×1D
Binning
▲Cannot treat correlation between AIF
observables
3D Binning
1×3D
Binning
▲Much statistics needed for multi
dimension
3D Fit
1×3D
Fitting
▲It is difficult to model distribution is whole
analysis area
MVA
1×1D
Binning
● Correlation can be considered
● Less statistics needed because 1D
Reminder 2
From recent collab. meeting
- The shape of PDF for Signal & RMD agrees better
with MEG trigger e+ data than random trigger
- But Δφ has un-negligible discrepancy
- Some AIF are found very close to main e+ track
- Probably this is ghost of main e+
ΔθAIF
ΔφAIF
AccBG
Sig(trig00)
Sig(trig22)
Sig(trig31)
AccBG
Sig(trig00)
Sig(trig22)
Sig(trig31)
[deg
]
[deg
]
Remaining issues
How to make Sig+RMD PDF
1. Why Δφ distribution still have difference
2. How to set “No-AIF-candidate” ratio
Is AIF finding algorism correct ?
Why found candidates are so different by year
Inefficient candidates are found in small R
etc.
Δφ distribution
shuffled pair
XEC V γ
true pair
TC φ
↑Events that
do not exist in
real pair
↑ structure due
to select one
TIC hit
Trigger simulation
LXe γ
DCHAIF
Interconnection
VGamma
DCHAIF
Finder
AIF
candidate
Positron event list
before : vector <Int_t>
new
: vector [#bar] <Int_t>
Used list is randomly
selected according to
VGamma and table.
Used e+ event is randomly
selected from the list.
mix
Trigger table which was made
from true γ and e+ pair ↓
Result of trigger simulation
Base shape of Sig+RMD
PDF became more
similar to BG PDF.
Δθ & Δt PDF did not
changed by this.
How much is No AIF candidate event
In Signal or RMD event, the probability to find AIF candidate
is smaller than that of BG event, because BG event contain real-AIF event.
If e+ of trigger 0 (BG) event is used for Sig & RMD PDF, the ratio to find AIF
become higher than it really is.
Red : AccBG
Blue : Sig & RMD
No-AIF-events are stored in
exceptional bin
MVA Index
Classification of event type
signal
RMD
γ
BG
by AIF
BG not
by AIF
e+0 e +1 …
no AIF e+
γ
e+0 e+1 …
no AIF e+
γ
No AIF candidate
Category 1
γ
?
e+0 e+1 …
e+AIFa …
γ
?
e+0 e+ 1 …
e+AIFa
Category 3-1
e+0 e+1 …
e+AIFa …
AIF is found but not real
Category 2
γ
e+0 e+ 1 …
e+AIFa e+AIFb …
?
Category 3-2
Ways to calculate “no AIF ratio”
/ trigger
signal
RMD
Trig0
Remove most Scale
likely
Cat.1 & 2
candidate e+
from Cat.3
1
B3-2
A2
B3-1
C2
D2
C1
D2
B2
A1
0
B1
Check by example
2012 year, 2000 runs
γ : 48MeV & Acceptance
e+ : positron selection
TOTAL
AIF found
70499
54400
calculate “C-1”
Select Cat.3 (peak)
by small box
No AIF B1
16059
22.7
%
2852
B3-1
Only 1
candidate
B3-2
More than 2
candidates
54400
381
2417
C-1 = (B1 + B3-1) / (ALL) = 23.3%
calculate “D-1”
D-1 = (B2) / (B1) = 23.7%
remove most likely candidate e+
Black 1
Blue
2
Green 3
no AIF
region cut
Black
Blue
1
2
Peak area
cut
The 2nd AIF like e+ when the 1st e+ is in peak similar to base shape.
Treatment for few event bin
Example of AIF MVA PDF by
BDT method.
Basically the P(SIG)/P(BG)
has increasing trend, but
statistically fluctuate at end part.
We think to rebin or use variable
width histo to avoid strange
likelihood value.
Another way
: Cut far remote region, and
store it in OverFlow bin.
Summary to make Signal AIF PDF
・Use γ and e+ both from _open trigger 0 data
・If the same event is selected, select a e+ again
・Simulate γ-e+ direction matching
・Simulate also θ and t direction ?
・PDFs already match in its base part
・Ratio of “no candidate”
・C-1 or D-1
・Difference should be systematic ?
・Do include category 3-2 e+ data in shuffling ?
・It is already known that, whether
remove Cat3 doesn’t make no significant
difference in distribution shape
Remaining issues
Finalize Reconstruct Methods
→ Test Reconstruction Process
Decide MVA methods
→ Calculation MVA index
Prepare Signal & BG PDF
→ Start fitting
Estimate systematic error in AIF analysis
→ Start sensitivity calculation
Schedule
3
4
5
6
Finalize Reconstruct
Test Reprocess
Year/Period investigate
MVA selection
Main Reprocess
PDF preparation
Final check about AIF
7