Upload
najila
View
27
Download
0
Tags:
Embed Size (px)
DESCRIPTION
AA->HH Study ~About the Neural Network analysis and other issues~. Shin-ichi Kawada (Advanced Science of Matter, Hiroshima University). About PLC. Electrons. Backward Compton scattering. Photon-photon collision. Reaction in photon-photon collision. Signal: gamma ・ gamma->HH - PowerPoint PPT Presentation
Citation preview
16th ILC general meeting (2010/7/17)
1
AA->HH Study~About the Neural Network analysis and other issues~Shin-ichi Kawada
(Advanced Science of Matter, Hiroshima University)
16th ILC general meeting (2010/7/17)
2
About PLCElectrons
Backward Compton scattering
Photon-photon collision
16th ILC general meeting (2010/7/17)
3
Reaction in photon-photon collision Signal: gamma ・ gamma->HH
Final goal: measurement of higgs self-coupling constant
Many backgrounds gamma ・ gamma->WW, ZZ, 4b, etc...
Analysis of background reduction is necessary. It was assumed that higgs mass is 120GeV in this
study.
16th ILC general meeting (2010/7/17)
4
Main backgrounds
Optimized CM energy: 270GeV
There are 2 main backgrounds.
gamma・ gamma->WW (11.6pb)gamma・ gamma->ZZ (9.42fb)
(studied by N.Maeda)
16th ILC general meeting (2010/7/17)
5
Previous results (studied by N.Maeda) The neural network (NN) analysis can reduce
WW backgrounds. However, ZZ backgrounds remain because of
inaccuracy from the jet clustering and the b-tagging.
We will write the paper about the analysis of the background reduction using NN.
16th ILC general meeting (2010/7/17)
6
Current issues
Toward the preparation of the paper, there are 3 issues to be solved. Consideration of other backgrounds Treatment of the ZZ events Check of the NN training
16th ILC general meeting (2010/7/17)
7
1. Other backgrounds
We have to consider other intermediate states in gamma ・ gamma->4b. We have asked Kurihara-san to calculate the cros
s-section of gamma ・ gamma->4b.
16th ILC general meeting (2010/7/17)
8
2. Treatment of ZZ events
γγ ZZ
Present method
1st step 2nd stepZZ 4b
ON-shell mode is helicity state.
More realistic method
γγ intermediate state including ZZ 4b
OFF-shell mode is not helicity state.
reaction
16th ILC general meeting (2010/7/17)
9
2. Treatment of ZZ events
We have to compare the helicity amplitude using ON-shell mode with OFF-shell mode.
16th ILC general meeting (2010/7/17)
10
Results
CM energyCos(theta)
Phi
Helicity
ON-shell mode
OFF-shell mode
16th ILC general meeting (2010/7/17)
11
Results
ON-shell mode results and OFF-shell mode results are almost same.
ON-shell mode calculation is effective. “Treatment of ZZ events” issue is solved!
16th ILC general meeting (2010/7/17)
12
3. Check of the NN training
We have to make an estimation of the systematic error from NN analysis.
16th ILC general meeting (2010/7/17)
13
3. Check of the NN trainingNN analysis (conventional)
NN weight file No.1signal file No.1
BG file No.1
Training step
Analysis step
some events
some events
weight file No.1
signal file No.1
BG file No.1
ALL events
ALL eventsNN
output
Weight file No.1 depends on the signal file No.1 and BG file No.1.
16th ILC general meeting (2010/7/17)
14
3. Check of the NN trainingWe have to check of statistical independence.
NN weight file No.2signal file No.2
BG file No.2
Training step
Analysis step
some events
some events
weight file No.2 (1)
signal file No.1 (2)
BG file No.1 (2)
ALL events
ALL eventsNN output
No.2 files are independent from No.1 files.
16th ILC general meeting (2010/7/17)
15
Events file & weight file
There are 4 MC events files & 4 weight files. Maeda MC events file & weight file Kawada MC ON-shell events file & weight file Kawada MC OFF-shell events file No.1 & weight
file Kawada MC OFF-shell events file No.2 & weight
file Events files: HH 50k events (signal)
ZZ 1M events (BG)
16th ILC general meeting (2010/7/17)
16
Pre-selection condition
99.0cos
05.0
# of b-flavor jets (nsig=3.0 & noffv=1) > 2
# of b-flavor jets (nsig=3.0 & noffv=2) > 1
These condition come from gamgamZZ-code.
b-tagging
Condition of reconstructed particles
16th ILC general meeting (2010/7/17)
17
Pre-selection# of generate
# of remained events after pre-selection
Maeda
Kawada-ON
Kawada-OFF1
Kawada-OFF2
HH=50000
ZZ=1000000
HH=50000
ZZ=1000000
HH=50000
ZZ=1000000
HH=50000
ZZ=1000000
HH=29897
ZZ=86538
HH=29958
ZZ=87056
HH=29958
ZZ=87486
HH=29823
ZZ=87143
These difference is caused by statistical fluctuation.
16th ILC general meeting (2010/7/17)
18
How to check statistical independence
NNweight file
Kawada OFF1
Kawada OFF1 signal file
29958
Kawada OFF1 BG file 87056
3000 (fixed)
5000 (fixed)
Example
(4 events file) * (4 weight file) = (16 combinations)
2. We calculate the maximum significance for each combination.
1. Preparation of weight files
3. We compare the values of significance to check the statistical independence among the combination.
16th ILC general meeting (2010/7/17)
19
Combination results
MaedaKawada
OFF1
Kawada
OFF2
Kawada
ON
weights
events
Maeda29897 86538
Kawada-OFF129958 87056
Kawada-ON29958 87486
Kawada-OFF229823 87143
14086 43721.34
16051 66621.25
15520 57831.29
15078 55831.28
11704 21971.53
13575 30571.52
13168 28961.52
13738 30571.54
12501 26031.51
13880 32931.51
14524 34881.50
13738 31441.52
12167 23891.53
12538 25751.53
12830 27231.52
13291 28461.54
# of remained HH, # of remained ZZ, Maximum significance
16th ILC general meeting (2010/7/17)
20
Result & Discussion
Significance ~1.3 was obtained with Maeda events and there is large fluctuation.
Significance ~1.52 was obtained with Kawada events and there is small fluctuation. These fluctuation is caused by the statistical fluctuation among Kawada events.
16th ILC general meeting (2010/7/17)
21
Result & Discussion
In Kawada events, maximum significance does not depend on the weight file.
It is still unclear what causes large fluctuation among the Maeda events.
16th ILC general meeting (2010/7/17)
22
Summary
The status of 3 issues are presented as following. Consideration of other backgrounds -> Kurihara-s
an Treatment of ZZ events -> ON-shell mode calculat
ion is effective. Check of the NN training -> The fluctuation of sign
ificance is caused by the statistical fluctuation among Kawada events.
16th ILC general meeting (2010/7/17)
23
Talk in JPS meeting
I will talk about AA->HH study topics in JPS. English title: Feasibility study of the measurement
of higgs pair creation in the photon-photon linear collider
16th ILC general meeting (2010/7/17)
24
Backup slides
16th ILC general meeting (2010/7/17)
25
ON-shell mode program
ZZSpring.cxx
HELVector z1im(fP[0]+fP[1], kM_z, -1, -1);
HELVector z1i0(fP[0]+fP[1], kM_z, 0, -1);
HELVector z1ip(fP[0]+fP[1], kM_z, -1, -1);
This “kM_z” means ON-shell mode (91.19GeV).
16th ILC general meeting (2010/7/17)
26
OFF-shell mode programZZSpring.cxx
Double_t forinvE1 = TMath::Power((fP[0].E() + fP[1].E()),2);
Double_t forinvPx1 = TMath::Power((fP[0].Px() + fP[1].Px()),2);
Double_t forinvPy1 = TMath::Power((fP[0].Py() + fP[1].Py()),2);
Double_t forinvPz1 = TMath::Power((fP[0].Pz() + fP[1].Pz()),2);
Double_t invmassZ1 = TMath::Sqrt(forinvE1 – forinvPx1 – forinvPy1 – forinvPz1);
HELVector z1im(fP[0]+fP[1], invmassZ1, -1, -1);HELVector z1i0(fP[0]+fP[1], invmassZ1, 0, -1);HELVector z1ip(fP[0]+fP[1], invmassZ1, -1, -1);
This means invariant mass.
These parts show the calculation of invariant mass.
22 pEmZ