Analysis of thermal dielectron channels (status)
Dilepton meeting GSI, Germany
September 2020
Etienne Bechtel University of Frankfurt
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 2
Simulation overviewSoftware OCT19 release
• The TRD software was on the status of APR20 Event-based simulation and reconstruction GEANT3 was used
Collision system 5 Million 12 A GeV Au+Au (10% most central) collisions as calculated with UrQMD LMVM cocktail added as pluto input Thermal radiation added with a uniform mass distribution an rescaled to the expected yields
Cuts > 2 STS hits > 5 RICH hits > 2 TRD hits
< 3 Pre-pairing mass > 25 keV
Target 25 mu gold target
Setup Sis100_electron_setup
χ2 /NDF
90% e-eff in RICH 80% e-eff in TRD
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 3
Dielectron signal input
Full phase space pluto input for the dielectron analysis
•The y-axis is scaled to the expected yields per event
•The signals were calculated with a temperature of 0.13 GeV
•The thermal radiation is calculated with a fireball model and a coarse-graining approach
Galatyuk, T., Hohler, P., Rapp, R. et al., Eur. Phys. J. A 52, 131 (2016)
Ralf Rapp, Hendrik van Hees., Physics Letters B, 753:586, Feb 2016
For the expected yields see: https://cbm-wiki.gsi.de/foswiki/bin/view/PWG/CbmDileptonInfoFilesAuAu11000
*as shown by Tetyana in the last collab. meeting
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 4
Phase space of in-medium rho
Comparison of MC (top) and reconstructed (bottom) pt_y distribution
•We have sufficient mid-radpidity coverage
6−10
5−10
4−10
Yiel
d [1
/eve
nts]
2− 1− 0 1 2 3 4 5 6MClabY
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
(GeV
/c)
MC
Tp
Au+Au 12A GeV - 0-10% centrality
= 1.66cmy
6−10
5−10
4−10
Yiel
d [1
/eve
nts]
2− 1− 0 1 2 3 4 5 6labY
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
(GeV
/c)
Tp
Au+Au 12A GeV - 0-10% centrality
= 1.66cmy
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 5
Detection efficiency and correction matrix
Rapidity and transverse momentum dependence •The detection efficiency is in the order of 1-2%
• A 2D correction matrix was extracted and applied to the reconstructed spectra
• A 3D correction as a function of pt,y and invariant mass was not possible due to statistics
0 0.5 1 1.5 2 2.5 3 (GeV)
Tp
5
10
15
20
25
30
35
403−10×
det.
prob
.
Au+Au 12A GeV0-10% centrality
ρin-med.
QGP
1
10
210
310
Cor
rect
ion
fact
or
0 0.5 1 1.5 2 2.5 3 3.5 4labY
0
0.5
1
1.5
2
2.5
3
3.5
4
(GeV
/c)
Tp
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 6
Combinatorial background
Invariant mass spectrum of all uncorrelated unlike-sign pairs from the same event
• The background is dominated by electron-electron and electron-pion contributions
• The electron-pion contribution is dominant
from about 1.5 GeV/ • The pion-pion contribution is suppressed but still significant (especially at larger masses)
• Pion suppression seems to be a bit low
c2
0 0.5 1 1.5 2 2.5)2 (GeV/ceeM
10−10
9−10
8−10
7−10
6−10
5−10
4−10
3−10
2−10
1−10
1
10)2 (1
/GeV
/cee
dN/d
M
SE+-ee (comb.)
(comb.)πe (comb.)ππ
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 7
Signal-to-background ratio
Impact of the electron-pion dominance at larger invariant masses
• The signal-to-background ratio is in the order
of 1:10000 for masses larger than 1.5 GeV/ • This is a result of the remaining pions
• Additionally, the ratio is further decreasing towards larger masses
c2
0 0.5 1 1.5 2 2.5)2 (GeV/ceeM
5−10
4−10
3−10
2−10
1−10
1
10
S/B
12 GeV
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 8
Corrected invariant mass spectrum of embedded signalsThermal radiation • The thermal component has very good statistics and shows the expected yields per event
• The intermediate mass range looks very promising for the extraction of the inverse slope parameter
Vectormesons The vectormeson cocktail is also well reconstructed and suffers less from statistical limitations
Unlike-sign spectrum (SE+-) Very low background statistics in the IMR
• Background subtraction is not reasonable with these fluctuations
0 0.5 1 1.5 2 2.5)2 (GeV/ceeM
8−10
7−10
6−10
5−10
4−10
3−10
2−10
1−10
1
10
210)2 (c
orr.)
[1/e
v.] (
1/G
eV/c
eedN
/dM
SE+--e+ eγ → 0π
-e+ eγ → η-e+ e0π → ω
-e+ e→ ω-e+ e→ φ
-e+ eγ → 'η-e+ e→ 0ρ
-e+ p e→ Δρin-medium
thermal radiation (Rapp)
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 9
Inverse slope parameter
Background subtraction was done with MC information due to statistics
The original information seems to be simulated and reconstructed very well
0 0.5 1 1.5 2 2.5 3)2 (GeV/ceeM
9−10
7−10
5−10
3−10
1−10
10
310
510
710
910
1110) [
corre
cted
.]2
(1/G
eV/c
eedN
/dM
/ ndf 2χ 05 / 1198− 1.319ep0 9.03e+06± 9.07e+04 p1 2.1067± 0.1787
6Theory spectra x 10
3MC Signal x 10
+ QGPρRec in-med
Au+Au 12 A GeV 0-10%CBM simulation
0.5 MeV±Theory T: 178.6 1 MeV±MC T: 176.6 7 MeV±REC T: 174.9
f(Mee) = c ⋅ M3/2ee ⋅ exp−Mee/T
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 10
Improvement of the pion suppression with machine learning
Idea Using an ANN to enhance pion suppression via the combination of information
Architectures I tried different network architectures of fully connected and convolutional networks
Input As input I combined detector information from multiple sub-systems:
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 11
Combinatorial background with ANN
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2)2 (GeV/ceeM
5−10
4−10
3−10
2−10
1−10
1
10
210
S/B regular cuts
linear convolutional
softmax fully connected
regular cuts + softmax fully connected
0 0.5 1 1.5 2 2.5)2 (GeV/ceeM
8−10
7−10
6−10
5−10
4−10
3−10
2−10
1−10
1
10
210)2 (c
orr.)
[1/e
v.] (
1/G
eV/c
eedN
/dM
SE+-ee (comb.)
(comb.)πe (comb.)ππ
Signal-to-background comparison (top) •The usage of machine learning seems to improve the pion-suppresion in general
•However, the best performance was achieved via a combination of regular PID cuts and machine learning (blue line)
Background contributions •The pion-pion contributions are far stronger suppressed and the electron-electron contributions is dominant over the whole invariant mass range
•The statistics from the pairing process reduce event further, which increases fluctuations even more
Etienne Bechtel - Dilepton meeting - GSI, Germany4.9.20 12
ConclusionAnalysis procedure • The general dielectron analysis chain works very well • Event-mixing and corrections are possible • Analysis is limited by background statistics
Thermal signal • The embedding with a uniform mass distribution provides sufficient statistics for the
IMR • Invariant mass spectra scaled to the yield per event look as expected
Machine learning • The usage of machine learning looks promising • Further investigations could be made but are time intensive
Discussion: How can the fluctuations in the IMR background be reduced?
-> Realistic estimation of the errors?