Electromagnetic calorimeter reconstruction in Belle II.€¦ · Electromagnetic calorimeter...

Belle

Electromagnetic calorimeter reconstruction in Belle II.

2019-03–12, ACAT, Saas-Fee Torben Ferber ([email protected]) for the Belle II ECL group

Electromagnetic calorimeter reconstruction in Belle II (Torben Ferber) �2

SuperKEKB

7/34Torben Ferber, DESY

Nano beam scheme.

KEKB Super-KEKB

L=γ±

2er e

(1+σ y

*

σ x

*)I± ξ y±

βy±

RL

Rξy

vertical beta function at IP

beam current

factor 2-3factor 20

83mrad

KEKB e+/e-E (GeV): 3.5/8.0I (A): ~ 1.6/1.2β*y (mm): ~5.9/5.9Crossing angle (mrad): 22

7/34Torben Ferber, DESY

Nano beam scheme.

KEKB Super-KEKB

L=γ±

2er e

(1+σ y

*

σ x

*)I± ξ y±

βy±

RL

Rξy

vertical beta function at IP

beam current

factor 2-3factor 20

83mrad

SuperKEKB e+/e-E (GeV): 4.0/7.0I (A): ~ 3.6/2.6β*y (mm): ~0.27/0.3Crossing angle (mrad): 83→ Luminosity increase x40

4.0 GeV

7.0 GeV


Belle II at SuperKEKB

http://www-superkekb.kek.jp/img/ProjectedLuminosity_v20190128.png

Belle (1ab-1)

Belle II goal(50ab-1)

Belle II up to today:0.0005 ab-1


Belle II at SuperKEKB

positrons e+

electrons e-

KL and muon detector (KLM): Resistive Plate Counters (RPC) (outer barrel) Scintillator + WLSF + MPPC (endcaps, inner barrel)

Particle Identification (PID):Time-Of-Propagation counter (TOP) (barrel) Aerogel Ring-Imaging Cherenkov Counter (ARICH) (FWD)

Electromagnetic calorimeter (ECL):CsI(Tl) crystals, waveform sampling

Vertex detectors (VXD): 2 layer DEPFET pixel detectors (PXD) 4 layer double-sided silicon strip detectors (SVD)

Central drift chamber (CDC):He(50%):C2H6 (50%), small cells, fast electronics

Magnet:1.5 T superconducting

Trigger:Hardware: < 30 kHz Software: < 10 kHz

DEPFET: depleted p-channel field-effect transistorWLSF: wavelength-shifting fiberMPPC: multi-pixel photon counter


Belle II Calorimeter (ECL)

CrystalPhotodiode

Preamplifier

Sum

offline analysis

L1 trigger

Digitizer1.76 MHz

18 bitShaper

FPGA

• 8736 crystals

• ~5×5×30 cm3 CsI(Tl) →16.1 X0

• Crystals, photodiodes and preamplifiers reused from Belle experiment

• New shapers and digitizers/FPGAs

6

7

5 8

1

23

4

HdD

HaAaA

c C

bB

dD

_2

_1

_3

_4

dg68

dg24

dg57

dg13

1

2

9

3

4

5

6

10

7

8

_9

_2

_1 _4

_3

a

A

dD c C

b2

B2

b1

B1

HdD

HaA

dg68 dg57

dg13 dg24


Signal processing: Belle and Belle II

t

Shaper output (τ=0.5μs)

Waveform Digitizer, 1.76 MHz, 18 bit

FPGA fits to extract Amplitude and Time

Shaper output (τ=1.0μs)

Gate width (Δt=100ns)

t

Signal charge

Charge-to-time converter (QTC)

Digitizer (TDC)

Amplitude


Calorimeter tasks at Belle II

Calorimeter basics:Photon energy and position (30 MeV - 7 GeV)

Neutral trigger (L1 and HLT) Online luminosity

Multiple detectors:Track-cluster matching

Electron ID

Advanced: “Extra energy” (hermeticity at an e+e- collider)

Hadron IDPulse-shape discrimination Low pt muon identification


Calorimeter challenges at Belle II

Higher beam backgrounds

Higher collision rates (= tighter trigger conditions)

More demanding analyses


Calorimeter challenges at Belle II: L1 and HLT Trigger

Higher collision rates (= tighter trigger conditions)

0.5kHz, σ≈(5+450) nb(no HLT)20

18

10kHz, σ≈(5+8) nb (after HLT, compare to BaBar: (5+70) nb)20

25

Calibration, low multiplicity events:Dominated by ee→ee(γ) debris


Calorimeter challenges at Belle II: Beam background

Higher beam backgrounds

Belle II Simulation(2018)

Belle II Simulation(2025)

~20 cluster candidates per event

~100 cluster candidates per event


Offline reconstruction flow

Crystal calibration(time, energy)

User analysis

Track-Cluster Matching

Charged PID

Tracking

Cluster shapes

Energy Position Relation to track(s) and PID information Cluster shapes (similar to image moments) Crystal information is not available!

Seed Finder

Connected Region

Shower splitting

Clu

ster

ing

Energy calibration

Tail calibration

Leakage Correction

Cal

ibra

tion

Electromagnetic calorimeter reconstruction in Belle II (Torben Ferber)

x (cm)100 150 200 250 300 350

y (c

m)

250−

200−

150−

100−

50−

0

(E /

MeV

)10

Log

012

3

Crystal energy

x (cm)100 150 200 250 300 350

y (c

m)

250−

200−

150−

100−

50−

0Shower (no timing selection)

x (cm)100 150 200 250 300 350

y (c

m)

250−

200−

150−

100−

50−

0 > 20 MeV)

ClusterCluster (timing selection, E

θΦ

θΦ

θΦ

�12

Offline reconstruction in pictures

Clustering

Belle II Simulation Belle II Simulation Belle II Simulation

User analysis


Offline reconstruction For the case of the large pile-up noise the Belle algorithm for cluster reconstruction is not optimal

Belle 1: sum of the Ei in 5x5 matrix E

i>0.5 MeV

Other approach: Sum of N most energetic hits, N depends on energy and background

Cluster reconstruction

To get the photon energy: cluster energy is corrected by function depending on E, angles

and the background level

Background level is estimated from multiplicity off time events

Measured per event

#crystals in energy sums

[GeV]trueE1−10 1

/ E

[%]

Eσ

0

5

10

15BGx0.0 (no material) FWDBGx0.0 FWDBGx0.1 FWDBGx1.0 FWD

[GeV]trueE1−10 1

/ E

[%]

Eσ

0

5

10

15 BGx0.0 (no material) BarrelBGx0.0 BarrelBGx0.1 BarrelBGx1.0 BarrelBGx1.0 Barrel (MC5)

[GeV]trueE1−10 1

/ E

[%]

Eσ

0

5

10

15

20

25 BGx0.0 (no material) BWDBGx0.0 BWDBGx0.1 BWDBGx1.0 BWD

Fig. 16: Photon energy resolution as function of true photon energy for FWD endcap (a),

barrel (b), and BWD endcap (c). Note the di↵erent y–axis ranges of the plots. A smooth

curve has been fitted to the points to guide the eye. An older implementation of ECL

reconstruction (used in MC5) is also plotted in (b).

the cost of requiring analysis specific optimisation. The uncertainty on the selection e�ciency

cannot be pre-determined using this method.

The likelihood selectors rely on likelihood ratios constructed in the following way. First,

the PID log likelihoods from each detector are summed to create a combined PID likelihood

for each of six long-lived charged particle hypotheses: electron, muon, pion, kaon, proton

and deuteron. Next, the di↵erence in log likelihood between two particle hypotheses is used

to construct a PID value L(↵ : �) according to

L(↵ : �) =1

1 + eln L↵�ln L�

=

Qdet L(↵)Q

det L↵ +Q

det L�, (5)

where ↵ and � represent two di↵erent particle types and the product is over the active

detectors for the PID type of interest. The value L(↵ : �) is greater than 0.5 for a charged

track that more closely resembles a particle of type ↵ than one of type � and is less than

0.5 otherwise. More details on the PID types are given in the following sections.

The performance plots included in this section were generated from inclusive samples

of 106 cc̄ events generated during the fifth and sixth MC campaigns. These samples

were reconstructed with release-00-05-03 and release-00-07-00 of the Belle II software,

respectively.

68/689

Belle algorithm

arXiv:1808.10567 [hep-ex]

Belle II Simulation


Performance: π0

)2 (GeV/cγγm0.40 0.45 0.50 0.55 0.60 0.65 0.70

)2

Entri

es /

(0.0

03 G

eV/c

0

100

200

300 Data

> 0.30 GeVγE

Belle II 2018 (Preliminary)

-1 L dt = ~5 pb∫

FIG. 2: m�� distribution in Exp 3, runs 112-1355. A � peak at about 542 MeV/c2 is visible.

)2 (GeV/cγγm0.08 0.10 0.12 0.14 0.16 0.18

)2

Entri

es /

(0.0

01 G

eV/c

0.0

0.5

1.0

1.5

310×

Data

> 0.15 GeVγE

Belle II2018 (Preliminary)

-1 L dt = ~5 pb∫

FIG. 1: m�� distribution in Exp 3, runs 112-1355. A ⇡0 peak at about 132 MeV/c2 is visible.

6

Figure 1: This figure shows the invariant mass distribution of ⇡0 ! �� in 5 pb�1 of collisiondata. Events are required to contain at least three good tracks to purity the sample withprocesses of the type e+e� ! hadrons, while rejecting beam induced background, Bhabhascattering, and other low multiplicity background sources. The photon daughters of the⇡0 candidates are required to have an energy of greater than 150 MeV, and to be withinthe acceptance of the Central Drift Chamber (CDC). The internal document reference isBELLE2-NOTE-PH-2018-002.

1

BELLE2-NOTE-PL-2018-009

Few days after the start of data taking.


Offline reconstruction flow

Crystal calibration(time, energy)

User analysis

Track-Cluster Matching

Charged Particle Identification

Tracking

Cluster shapes

Energy Position Relation to track(s) Cluster shapes (similar to image moments) Crystal information is not available!

Seed Finder

Connected Region

Shower splitting

Clu

ster

ing

Energy calibration

Tail calibration

Leakage Correction

Cal

ibra

tion


Pulse-Shape Discrimination (PSD)• Online FPGA waveform fits use photon

templates only and provide time and amplitude fit results (2 variables)

• New: Exploit the fact that hadronic and electromagnetic scintillation components are different

• If crystal energy E > 30 MeV: Store waveform data (31 variables) and repeat fit offline with different templates.

• Third information from a crystal: PSDSavino Longo ([email protected]) ICHEP 2018

CsI(Tl) Pulse Shape Discrimination at Belle II

13

• Offline waveforms are characterized with photon+hadron component fit.

• Hadron response for all crystals is calibrated using Fourier analysis to compute the impulse response for each calorimeter crystals signal chain.

CsI(Tl) Diodes Preamp ShaperDSP

RHadron(t) = LHadron(t) ⇤ IShaperDSP(t)IShaperDSP(t) = IFT�FT (R�(t))FT (L�(t))

�FT = Fourier Transform IFT = Inverse Fourier Transform

Sample Fit of Hadron Pulse in Collision Data

Belle II Phase 2 data – preliminary

S. Longo, ICHEP2018


Pulse-Shape Discrimination (PSD)

• First time pulse shape discrimination (PSD) is used in an e+e- collider experiment

• New variable based on a BDT trained (on MC) to separate photons and K0L using all pulse shapes in a cluster

• Will be included in charged particle identification to improve muon vs. pion separation

Savino Longo ([email protected]) ICHEP 2018

First Beam = First Hadrons Observed!

• When SuperKEKB beams started circulating (April 2018) hadronic backgrounds in ECL allowed for hadronic pulse shapes to be observed at Belle II.

14

HER = 0.15 A LER < 0.001 A

ECL Data all barrel crystals

HER = 0A LER = 0A

(beams OFF)


Observing Hadronic Pulse Shapes with Belle II Calorimeter

electromagnetic pulse shapes

proton bands

alphas Belle II Phase 2 data – 1st glimpse-preliminary

Belle II Phase 2 data – 1st glimpse-preliminary

SuperKEKB info: http://www-linac.kek.jp/skekb/snapshot/ring/dailysnap-20180404-0001.png

protons

alphas

Two component total energy [GeV]

Had

ron

Com

pone

nt In

tens

ity

Savino Longo ([email protected]) ICHEP 2018

First Beam = First Hadrons Observed!

• When SuperKEKB beams started circulating (April 2018) hadronic backgrounds in ECL allowed for hadronic pulse shapes to be observed at Belle II.

14

HER = 0.15 A LER < 0.001 A


HER = 0A LER = 0A

(beams OFF)


Observing Hadronic Pulse Shapes with Belle II Calorimeter

electromagnetic pulse shapes

proton bands

alphas Belle II Phase 2 data – 1st glimpse-preliminary

Belle II Phase 2 data – 1st glimpse-preliminary

SuperKEKB info: http://www-linac.kek.jp/skekb/snapshot/ring/dailysnap-20180404-0001.png

Electromagnetic pulse shapes

Had

ron

Com

pone

nt In

tens

ity

S. Longo, ICHEP2018


Low pt Particle Identification plans: Muon vs Pion

• Particles with low transverse momentum (pt < 0.5 GeV/c) do not reach our muon system:→ Baseline particle identification depends on E/p and is very poor→ Clustering itself difficult, since these particles leave long, charge dependent, trails in the calorimeter

ECL

High ptLow pt

muon system


Low pt Particle Identification plans• Approach under study:

• No clustering

• Extrapolate tracks to calorimeter

• Analyse 5×5 pixel calorimeter images around impact crystal using convolutional networks

Baseline identificationCalorimeter images + track extrapolation

Baseline identificationCalorimeter images + track extrapolation

Belle II Simulation (work in progress)

Belle II Simulation (work in progress)

Pions Electrons Muons


Summary

• Belle II is starting physics runs in ~7 days

• Calorimeter reconstruction priorities so far: Robust reconstruction, calibration

• Calorimeter reconstruction priorities now: Getting ready for high and varying beam backgrounds, and complex physics analyses

• Bonus: The knowledge of the initial state at an e+e- collider and the huge event rate makes this a perfect playground to develop data-driven ML applications

Deutsches Elektronen Synchrotron www.desy.de

Torben Ferber [email protected] ORCID: 0000-0002-6849-0427

Contact

http://www.desy.de

mailto:[email protected]

Electromagnetic calorimeter reconstruction in Belle II.€¦ · Electromagnetic calorimeter...

Documents

Transcript of Electromagnetic calorimeter reconstruction in Belle II.€¦ · Electromagnetic calorimeter...