[PPT] - AHCAL Energy Resolution Katja Seidel MPI for Physics & PowerPoint Presentation

SLIDE 1

AHCAL Energy Resolution

Katja Seidel

MPI for Physics & Excellence Cluster ’Universe’ Munich, Germany for the CALICE Collaboration

International Linear Collider Workshop 2010 Beijing, China 27 March 2010

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 0 / 15

SLIDE 2

Outline

1 CALICE calorimeter prototypes 2 Calibration of the AHCAL 3 Electromagnetic Showers 4 Hadronic Showers - Software Compensation

Global Method

Cluster Energy Density Weighting Neural Network

Local Method

Single Cell Energy Weighting

5 Conclusions

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 1 / 15

SLIDE 3

CALICE Calorimeter Prototype Program

Sc1 Sc2 Sc4 Sc3 Muon Trigger Drift Chambers Cherenkov Detector Scintillators ECAL HCAL TCMT Beam

Extensive Test Beam Program DESY: 2006 CERN: 2006, 2007 FNAL: 2008, 2009 Particle Types: µ, e±, π±, p Particle Energies: 1 GeV to 80 GeV

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 2 / 15

SLIDE 4

CALICE Calorimeter Prototype Program

Sc1 Sc2 Sc4 Sc3 Muon Trigger Drift Chambers Cherenkov Detector Scintillators ECAL HCAL TCMT Beam

CERN 2007 ECAL: Silicon-Tungsten Calorimeter 30 Layers; 1×1 cm2 readout pads, 1.4, 2.8, 4.2 cm thick absorber plates; 30 X0, 1 λ0 HCAL: Scintillator-Steel Calorimeter 38 Layers; 1.8 cm thick absorber plates, 47 X0 4.5 λ0 TCMT: Scintillator-Steel Calorimeter 8 layers: 2 cm thick absorber plates, 8 layers: 10 cm thick absorber plates, 5×100 cm scintillator bars; 5.8 λ0

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 2 / 15

SLIDE 5

CALICE Analog HCAL

Iron absorber structure Active layers: scintillator tiles Tile sizes: 3×3 cm2, 6×6 cm2, 12×12 cm2 Light collection via wavelength shifting fiber Readout via SiPM High granularity in AHCAL center → in the shower core

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 3 / 15

SLIDE 6

Calibration of the AHCAL

Signal Saturation SiPM pixel number limited → only limited number of photons can be counted Auto-calibration of SiPM gain: Low-intensity LED light coupled into each detector cell Gain measurement MIP-Calibration with Muons Complete detector illuminated with high energy muons Equalization of cell response by matching the MPV position Temperature effect correction SiPM gain SiPM amplitude ⇒ All effects included into event reconstruction and Monte Carlo digitization.

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 4 / 15

SLIDE 7

Calibration of the AHCAL

Signal Saturation SiPM pixel number limited → only limited number of photons can be counted Auto-calibration of SiPM gain: Low-intensity LED light coupled into each detector cell Gain measurement MIP-Calibration with Muons Complete detector illuminated with high energy muons Equalization of cell response by matching the MPV position Temperature effect correction SiPM gain SiPM amplitude ⇒ All effects included into event reconstruction and Monte Carlo digitization.

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 4 / 15

SLIDE 8

Calibration of the AHCAL

Signal Saturation SiPM pixel number limited → only limited number of photons can be counted Auto-calibration of SiPM gain: Low-intensity LED light coupled into each detector cell Gain measurement MIP-Calibration with Muons Complete detector illuminated with high energy muons Equalization of cell response by matching the MPV position Temperature effect correction SiPM gain SiPM amplitude ⇒ All effects included into event reconstruction and Monte Carlo digitization.

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 4 / 15

SLIDE 9

Electromagnetic Showers in the AHCAL

Beam energy [GeV]

10 20 30 40 50

Residual to linearity [%]

8
6
4
2

2 4

CALICE preliminary

Beam energy [GeV]

10 20 30 40 50

Residual to linearity [%]

8
6
4
2

2 4

Digitized MC Positron data systematic uncertainty Beam energy [GeV]

10 20 30 40 50

relative reconstructed width

0.02 0.03 0.04 0.05 0.06 0.07 0.08

CALICE preliminary

Positron data Digitized MC true MC systematic uncertainty

Positron test beam data from 10 GeV to 50 GeV Comparison to Monte Carlo data Data taking without ECAL in front of HCAL Linearity of detector response of 1.5 % up to 30 GeV Non-Linearity at higher energies not yet reproduced in MC → Saturation handling Energy Resolution Data Fit in the range from 10 GeV to 30 GeV with:

σ E[GeV ] = a

√

E[GeV ] ⊕ b

a = 22.5 ± 0.1(stat) ± 0.4(syst) % b = 0.0 ± 0.1(stat) ± 0.1(syst) %

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 5 / 15

SLIDE 10

Hadronic Showers

Detector Response ◮ CALICE: non-compensating sampling calorimeter ◮ Calorimeter response to hadrons is smaller than to electrons of the same energy ◮ CALICE AHCAL e

π ∼ 1.2

Software Compensation ⇒ Identification of electromagnetic and hadronic shower component fractions ⇒ Improve energy resolution ⇒ Improve linearity of detector response Method: Electromagnetic showers tend to be denser than purely hadronic ones Correlations between reconstructed energy and energy in high density shower regions ⇒ Test Local and Global Techniques

Total HCAL Energy [MIP]

200 400 600 800 1000 1200

Total HCAL Energy in Cells > 4.5 MIP/cell [MIP]

200 400 600 800 1000 1200 E ( > t h r e s h

l

d ) / E ( t

t

a l ) = 1 E ( > t h r e s h

l

d ) / E ( t

t

a l ) = . 3

(a)

CALICE Preliminary

20 GeV pions, no weighting

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 6 / 15

SLIDE 11

Cluster-Based Software Compensation

Two global methods based on cluster as a whole - no subcluster analysis Look at global cluster properties

1 Shower reconstruction in AHCAL and TCMT

Showers are required to start in the AHCAL

2 Determination of shower variables

from test beam and simulated data

3 Analyses developed on Monte Carlo data

FTF BIC Energy density weighting technique Neural Network from TMVA

4 Application of weight or trained neural network

n test beam data

Sc2 Muon Trigger ECAL HCAL TCMT

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 7 / 15

SLIDE 12

Cluster Energy Density Weighting Technique

Hadronic Showers with high energy density ρ ⇒ Higher electromagnetic content ⇒ Higher reconstructed energy Erec[GeV ] = Erec[MIP] · ω(ρ, E)

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 8 / 15

SLIDE 13

Cluster Energy Density Weighting Technique

Hadronic Showers with high energy density ρ ⇒ Higher electromagnetic content ⇒ Higher reconstructed energy Erec[GeV ] = Erec[MIP] · ω(ρ, E) Individual weights with minimization of function χ2 = Erec · ω − Ebeam

Cluster Energy Density [MIP/volume] 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 entries 1 10

2

10

3

10 Cluster Energy Density [MIP/volume] 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 entries 1 10

2

10

3

10

CALICE Preliminary

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 8 / 15

SLIDE 14

Cluster Energy Density Weighting Technique

Hadronic Showers with high energy density ρ ⇒ Higher electromagnetic content ⇒ Higher reconstructed energy Erec[GeV ] = Erec[MIP] · ω(ρ, E) Individual weights with minimization of function χ2 = Erec · ω − Ebeam

Cluster Energy Density [MIP/volume] 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 entries 1 10

2

10

3

10 Cluster Energy Density [MIP/volume] 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 entries 1 10

2

10

3

10

CALICE Preliminary

Parameterization of the individual weights via ω = a(E) · ρ + b(E)

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 8 / 15

SLIDE 15

Cluster Energy Density Weighting Technique

Hadronic Showers with high energy density ρ ⇒ Higher electromagnetic content ⇒ Higher reconstructed energy Erec[GeV ] = Erec[MIP] · ω(ρ, E) Individual weights with minimization of function χ2 = Erec · ω − Ebeam Parameterization of the individual weights via ω = a(E) · ρ + b(E) Parameterization of energy dependence with function for a(E) und b(E), E = Erec

beam energy [GeV] 10 20 30 40 50 60 70 80 fit parameter a

0.07
0.06
0.05
0.04
0.03
0.02

beam energy [GeV] 10 20 30 40 50 60 70 80 fit parameter b 0.03 0.0305 0.031 0.0315 0.032 0.0325 0.033 0.0335 0.034 0.0345

⇒ Determination of weights independent of beam energy!

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 8 / 15

SLIDE 16

Cluster Energy Density Weighting - Results

Energy Resolution:

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

E/E ∆

0.05 0.1 0.15 0.2 0.25

c GeV/E ⊕ b ⊕ E Fit: a/ 0.208 [GeV] ± 0.80% c = 0.000 ± 0.2% b = 0.00 ± a = 64.8 0.118 [GeV] ± 0.37% c = 0.488 ± 0.9% b = 2.21 ± a = 53.5 test beam data: constant cluster weight energy dependent parametrization Energy resolution - FTF_BIC weights

CALICE Preliminary

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

single

σ /

weight

σ

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05

CALICE Preliminary

test beam data energy dependent parametrization / constant cluster weight Ratio of energy resolutions - FTF_BIC weights

Weight parametrization from Monte Carlo derived Weights applied on test beam data Energy resolution improvement: → approx. 15 %

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 9 / 15

SLIDE 17

Cluster Energy Density Weighting - Results

Linearity of detector response:

10 20 30 40 50 60 70 80 90

reconstructed Energy [GeV]

10 20 30 40 50 60 70 80 90

CALICE Preliminary

test beam data constant cluster weight energy dependent Linearity of detector response - FTF_BIC weights

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

beam

)/E

beam

E

rec

(E

0.1

0.1

Significant improvement of linearity of detector response → better than 4 % Test beam data: π+ and π− runs. Elimination of proton content of π+ runs should improve the linearity even further.

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 10 / 15

SLIDE 18

Neural Network of TMVA

TMVA - Toolkit for Multivariate Data Analysis Training with Monte Carlo events with continuous energy

f hadronic model FTF BIC

6 input variables Reconstructed energy Cluster volume Cluster length Mean cluster width Cluster energy in last 5 AHCAL layers Cluster energy in Tail Catcher Target variable: beam energy Output value: reconstructed energy ⇒ Trained Neural Network applied on test beam data

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 11 / 15

SLIDE 19

Neural Network Technique - Results

Energy Resolution:

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

E/E ∆

0.05 0.1 0.15 0.2 0.25

c GeV/E ⊕ b ⊕ E Fit: a/ 0.208 [GeV] ± 0.80% c = 0.000 ± 0.2% b = 0.00 ± a = 64.8 0.048 [GeV] ± 0.24% c = 1.070 ± 1.0% b = 2.84 ± a = 43.1 test beam data: constant cluster weight neural network Energy resolution - FTF_BIC training

CALICE Preliminary

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

single

σ /

weight

σ

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05

CALICE Preliminary

test beam data neural network / constant cluster weight Ratio of energy resolutions - FTF_BIC training

Energy resolution improvement: → approx. 23 %

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 12 / 15

SLIDE 20

Neural Network Technique - Results

Linearity:

10 20 30 40 50 60 70 80 90

reconstructed Energy [GeV]

10 20 30 40 50 60 70 80 90

CALICE Preliminary

test beam data constant cluster weight neural network Linearity of detector response - FTF_BIC training

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

beam

)/E

beam

E

rec

(E

0.1

0.1

Significant improvement of linearity of detector response → better than 3 %

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 13 / 15

SLIDE 21

Single Cell Weighting - Technique

Simple reconstruction without weighting One calibration factor (MIP to GeV) per subdetector (ECAL, AHCAL, TCMT) Noise rejection applied Weight calorimeter cells according to their energy content ⇒ Apply higher weights to cells with low energy density Weights are energy dependent No knowledge of beam energy needed to apply weights Results: Energy resolution improvement: → approx. 18 % Linearity of detector response better than 4 %

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

E/E Δ

0.05 0.1 0.15 0.2 0.25

c GeV/E ⊕ b ⊕ E Fit: a/ 0.041 [GeV] ± 0.10% c = 0.000 ± 0.1% b = 2.54 ± a = 61.3 0.042 [GeV] ± 0.12% c = 0.504 ± 0.4% b = 2.34 ± a = 49.2 Energy resolution single weight energy dependent parametrization

CALICE Preliminary

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

beam

)/E

beam

E

rec

(E

0.1

0.1

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 14 / 15

SLIDE 22

Conclusions

Calibration of analog AHCAL with SiPM redout Energy Reconstruction of electromagnetic data ⇒Test beam and simulated data in the AHCAL Energy Reconstruction of hadron data ⇒High granularity can be used for software compensation Local and global software compensation methods Cluster energy density weighting technique: ⇒ 15 % energy resolution improvement for AHCAL and TCMT Neural Network: ⇒ 23 % energy resolution improvement for AHCAL and TCMT Tile energy density weighting technique: ⇒ 18 % energy resolution improvement for the complete CALICE setup Optimum maybe in between both methods Outlook Application of method on full ILD simulations and ILD reconstruction software PandoraPFA

K. Seidel (MPI for Physics)

energy reconstruction 27 March 2010 15 / 15

SLIDE 23

Backup slides

SLIDE 24

Energy Resolution and Linearity for all energy in AHCAL and TCMT

No clustering No software compensation MIP to GeV factor 0.028

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

E/E ∆

0.05 0.1 0.15 0.2 0.25

c GeV/E ⊕ b ⊕ E Fit: a/ 0.103 [GeV] ± 0.17% c = 0.000 ± 0.3% b = 3.44 ± a = 56.8 0.098 [GeV] ± 0.14% c = 0.000 ± 0.4% b = 4.66 ± a = 52.6 0.144 [GeV] ± 0.16% c = 0.000 ± 0.4% b = 3.21 ± a = 53.4 Energy resolution - Hcal & Tcmt Data FTF_BIC QGSP_BERT

CALICE Preliminary

reconstructed Energy [GeV]

10 20 30 40 50 60 70 80 90

CALICE Preliminary

Linearity of detector response - Hcal & Tcmt Data FTF_BIC QGSP_BERT

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

beam

)/E

beam

E

rec

(E

0.1

0.1

SLIDE 25

Energy Resolution and Linearity for test beam data and QGSP BERT Monte Carlo with Neural Network

Neural Network trained with FTF BIC simulated data

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

E/E ∆

0.05 0.1 0.15 0.2 0.25 c GeV/E ⊕ b ⊕ E Fit: a/ 0.048 [GeV] ± 0.24% c = 1.070 ± 1.0% b = 2.84 ± a = 43.1 0.042 [GeV] ± 0.16% c = 1.319 ± 1.3% b = 3.84 ± a = 31.3 FTF_BIC neural network training: Data with NN QGSP_BERT with NN Energy resolution

CALICE Preliminary

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

single

σ /

weight

σ

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05

CALICE Preliminary

FTF_BIC neural network training Data: neural network / constant cluster weight QGSP_BERT: neural network / constant cluster weight Ratio of energy resolutions

reconstructed Energy [GeV]

10 20 30 40 50 60 70 80 90

CALICE Preliminary

FTF_BIC neural network training: Data with NN QGSP_BERT with NN Linearity of detector response

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

beam

)/E

beam

E

rec

(E

0.1

0.1

SLIDE 26

Energy Resolution and Linearity for test beam data and QGSP BERT Monte Carlo with Cluster Energy Density Weighting

Weights extracted from FTF BIC simulated data

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

E/E ∆

0.05 0.1 0.15 0.2 0.25 c GeV/E ⊕ b ⊕ E Fit: a/ 0.118 [GeV] ± 0.37% c = 0.488 ± 0.9% b = 2.21 ± a = 53.5 0.058 [GeV] ± 0.20% c = 0.860 ± 1.0% b = 3.08 ± a = 41.0 FTF_BIC weights: Data with weighting QGSP_BERT with weighting Energy resolution

CALICE Preliminary

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

single

σ /

weight

σ

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05

CALICE Preliminary

FTF_BIC weights Data: energy dependent para. / constant cluster weight QGSP_BERT: energy dependent para. / constant cluster weight Ratio of energy resolutions

reconstructed Energy [GeV]

10 20 30 40 50 60 70 80 90

CALICE Preliminary

FTF_BIC weights: Data with weighting QGSP_BERT with weighting Linearity of detector response

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

beam

)/E

beam

E

rec

(E

0.1

0.1

SLIDE 27

Energy Resolution Improvement for complete CALICE setup

Single Tile Energy Weighting Technique Weights extracted from data No clustering

beam Energy [GeV]

10 20 30 40 50 60 70 80 90

single

σ /

weight

σ

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05

CALICE Preliminary

Ratio of energy resolutions energy dependent parametrization / single weights

SLIDE 28

Monte Carlo Energy Correction

beam energy [GeV]

10 20 30 40 50 60 70 80

beam

/E

beam

E

rec

E

0.1
0.08
0.06
0.04
0.02

0.02 0.04 0.06 0.08

beam energy [GeV]

10 20 30 40 50 60 70 80

beam

/E

beam

E

rec

E

0.1
0.08
0.06
0.04
0.02

0.02 0.04 0.06 0.08

beam energy [GeV]

10 20 30 40 50 60 70 80

beam

/E

beam

E

rec

E

0.1
0.08
0.06
0.04
0.02

0.02 0.04 0.06 0.08

Data FTF_BIC corrected MC energy