0 - Electron Discrimination in Liquid Argon Time Projection Chamber

0 - Electron Discrimination in Liquid Argon Time Projection

Chamber

Liquid Argon TPC in T2K

disappearance, e appearance measurements

Liquid Argon TPC•100 tons of liquid argon in the intermediate detector (2km)•Measure the parameters of low momentum particles (below Cherenkov threshold)•Background from NC interactions and e contamination measurements

T2K experiment•1GeV accelerator neutrino beam from J-PARC•22.5 ktons water Cherenkov far detector (295 km from J-PARC)

Tomasz Wąchała, Epiphany, Cracow 2006

MC Data

MC Data

Important reactions

e + n e- + p

+ n + 0 + X

e CC reaction: (e appearance signal)

NC reaction:(background for e appearance signal)

electromagnetic shower

Decay: 0

electromagnetic shower

e-


Neural network as a classificator

•Simple Multilayer Perceptron network used: Input layer, 1 hidden layer and output layer•Learning with a supervisor on the MC events

Input layer

Hidden layer

Output layer


Neural network as a classificator

Signal(electrons

)

Background(0s)

NeuralNet

INPUT OUTPUT

Electron

Number of events

Network output


Classification quality

)(

)(%100

)()(

)(%100

InputN

OutputNEfficiency

OutputNOutputN

OutputNPurity

sig

sig

bgsig

sig

Nsig/bg - number of the signal/background events above the threshold

Purity [%]

Efficiency [%]

500

0

Number of events

Network output

Better qualityof classification


Events geometry

Wireplanes

Electron / 0

•Monoenergetic (1GeV)Monte-Carlo events

•Without noise


Using <dE/dx> informationNumber of events

Number of eventsdE/dx [MeV/cm]

dE/dx [MeV/cm]

Wire3 Wire4

Wire5 Wire6

i

N

i dx

dE

dx

dE

3

Purity [%]

Efficiency [%]

Best quality for N=3


Information from the second wire plane

Purity [%]

Efficiency [%]

Number of events

<signal> [ADC] <dE/dx> [MeV/cm]

Inductionplane

Collectionplane Improved quality


Results using 2-2-1 network

Number of events

Network output

Efficiency [%] Purity [%]94.11 88.00 +/- 0.6490.01 97.16 +/- 0.35


<dE/dx>

<signal>

Using event topology information

Number of events


Electron/Pi0

Using event topology information

•Average width of event•Length of the track with the largest number of hits•Total number of hits

Number of events

Number of events Number of events


Adding new parameters

Purity [%]

Efficiency [%]

Best qualityfor

7 parameters




Number of events

Network output


<signal>

<dE/dx>

Event topology

Primary vertex information

Number of events

Distance from the vertex to the first ionization signal

xion

Ionization

Ionization

e-

e-

0

e+

e-


Primary vertex information

7 parameters

+ xion

Purity [%]

Efficiency [%]

7 parameters(<dE/dx> + event topology)


Adding hidden neurons

Purity [%]

Efficiency [%]

Best qualityfor N = 3

hidden neurons



Network output

Number of events


<signal>

<dE/dx>

Event topology

xion


Purity [%]

Efficiency [%]

Network architecture Efficiency [%] Purity [%]2-2-1 94.11 88.00 +/- 0.64

90.01 97.16 +/- 0.357-2-1 94.11 93.21 +/- 0.51

89.97 97.73 +/- 0.328-3-1 94.06 99.34 +/- 0.17

89.80 99.77 +/- 0.10

Summary

Only <dE/dx>

<dE/dx> + topology

<dE/dx> + topology+ xion


Future plans

•Adding noise to the events - how does it affect the results?

• Influence of additional particles

• Applying this analysis to the ICARUS TPC (requires extra work on software)

•Testing algorithms on the real data in the ICARUS T600 liquid argon TPC detector


0 - Electron Discrimination in Liquid Argon Time Projection Chamber

Documents

Transcript of 0 - Electron Discrimination in Liquid Argon Time Projection Chamber