Many thanks to Martin Heck Belle & Neural Networks To Physics - - PowerPoint PPT Presentation

▶
many thanks to martin heck
SMART_READER_LITE
LIVE PREVIEW

Many thanks to Martin Heck Belle & Neural Networks To Physics - - PowerPoint PPT Presentation

Dec 27, 2022 212 likes •660 views

Many thanks to Martin Heck Belle & Neural Networks To Physics From Tracks Florian Bernlochner florian.bernlochner@cern.ch Setting the Scene The Precision Frontier of Particle Physics ` New Physics New Physics p . . . p New

slide-1
SLIDE 1

From Tracks & Neural Networks To Physics Florian Bernlochner florian.bernlochner@cern.ch Many thanks to Martin Heck

Belle

slide-2
SLIDE 2

Setting the Scene

slide-3
SLIDE 3

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

The Precision Frontier of Particle Physics

3

γ γ

New Physics

New Physics

p p

New Physics . . . `

New Physics

Energy Frontier Ansatz Intensity Frontier Ansatz e.g. LHC, Tevatron e.g. BaBar, CLEO, Belle

slide-4
SLIDE 4

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

p p

New Physics . . . `

The Precision Frontier of Particle Physics

4

γ γ

New Physics

New Physics New Physics

Intensity Frontier Ansatz e.g. BaBar, CLEO, Belle

Direct searches

Flavour observables

sures their decays into light flavours

Talks of Uli, Teppei and Wouter

Plot: Andreas Crivellin

Large degree of complementarity At B-Factories with B-Mesons

Belle

Energy Frontier Ansatz e.g. LHC, Tevatron

slide-5
SLIDE 5

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

The Experiment

5

‘→ el le ←’

‘B’ breaks the symmetry In elle, hence Belle :-)

slide-6
SLIDE 6

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

Υ(4S)

e+ e−

√s = 10.58 GeV

hb¯ bi

hb¯ qi h¯ bqi

Fragmentation into two bound states: B-Mesons

¯ q q

quark-antiquark-pair produced from vacuum

q q

6

Υ(1S) = hb¯ bi Υ(4S) = hb¯ bi

B!threshold

Plot: CLEO

The Experiment: Collision energy

slide-7
SLIDE 7

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 7

The Experiment: B-Meson Decays

inclusive charged particle multiplicity: ~5.4 per B-Meson

  • r ~ 11 per B-Meson pair

b

q

light Quark heavy anti-b-Quark

B-Meson

7

τB ≈ 1.5 × 10−12 s

Stolen from Martin Heck

Murphy’s Law of Flavour Physics

Was einfach zu messen ist, kann nicht ausgerechnet werden, was einfach zu rechnen ist, ist schwierig zu messen.

4

What you can measure without a problem, you cannot calculate. What you can calculate easily, you cannot measure

h¯ bqi

r m

…

slide-8
SLIDE 8

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

The Experiment: Asymmetric Beam Energies

8

Asymmetric Beam energies: allow to directly observe CPV in B-system

!

PEP-II KEKB Υ(4S)

e+ e−

hb¯ bi

√s = 10.58 GeV

Υ(4S)

BaBar

E(e−) = 9 GeV E(e+) = 3.1 GeV E(e−) = 8 GeV E(e+) = 3.5 GeV

βγ = 0.56

βγ = 0.42

Belle Belle II

B-meson lifetime

B0 B0

hb ¯ di h¯ bdi

hb¯ bi

E(e−) = 7 GeV E(e+) = 4 GeV

βγ = 0.28

∆z ≈ c β γ τB ≈ 126 µm

slide-9
SLIDE 9

50:1

What is the difference between Belle and Belle II?

slide-10
SLIDE 10

50:1

What is the difference between Belle and Belle II?

Expected data set increase and ~ increase in inst. Luminosity

50 1 = LHCb Upgrade LHCb today = Belle II BaBar and Belle = BaBar and Belle CLEO

As significant for us as the energy increase from 7/8 TeV to 13 TeV at the LHC

slide-11
SLIDE 11

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 11

12

present KEKB SuperKEKB

5mm 1m 100m

(without crab)

L

Hourglass condition: y*>~ L=x/

Half crossing angle:

1m 5mm 100m

~50nm 83 mrad crossing angle

22 mrad crossing angle

13 15

KEKB → SuperKEKB

Nano-beam scheme: Squeeze vertical beam spot to 50 nm

slide-12
SLIDE 12

April 11, 2017

final focussing magnets

slide-13
SLIDE 13

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 13

Belle → Belle II

Electrons (7 GeV) Positrons (4 GeV)

Increased luminosity comes at a price: much larger beam backgrounds

slide-14
SLIDE 14

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 14

Belle → Belle II

KL and Muon detection system RPC based

Electromagnetic Calorimeter:

Thallium activated Caesium Iodide scintillation crystals

Central drift chamber:

Gas mixture of Helium and Ethan (C2H6)

Particle identification

Time-of-propagation counter, Aerogel Cherenkov ring detector

Vertex detectors

2 layers of Pixel (DEPFET) + 4 layers of strips (DSSD)

Vertex Drift chamber Cherenkov / TOP Calorimeter KL & Muon

slide-15
SLIDE 15

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 15

Belle → Belle II

KL and Muon detection system RPC based

Electromagnetic Calorimeter:

Thallium activated Caesium Iodide scintillation crystals

Central drift chamber:

Gas mixture of Helium and Ethan (C2H6)

Particle identification

Time-of-propagation counter, Aerogel Cherenkov ring detector

Vertex detectors

2 layers of Pixel (DEPFET) + 4 layers of strips (DSSD)

slide-16
SLIDE 16

Tracks

slide-17
SLIDE 17

14336 sense wires 56 layers

Central drift chamber:

Gas mixture of Helium and Ethan (C2H6)

rdrift ∝ tdrift

Cells of sense wires and field wires

Aim: Convert unmarked Hits …

Illustrations: Oliver Frost, Sarah Neuhaus

slide-18
SLIDE 18

14336 sense wires 56 layers … into charged particle trajectories

Central drift chamber:

Gas mixture of Helium and Ethan (C2H6) Illustrations: Oliver Frost, Sarah Neuhaus

e+ e−

Physics collision

Beam background
slide-19
SLIDE 19

Original question by E. Paoloni

slide-20
SLIDE 20

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

VXD Online and Offline Tracking

20

Sector-Map

Einteilung des SVD in Sektoren; Sektoren durch die “vernünftige” Spur in großer Simulation geht, sind “Freunde”;

37

To reduce combinatorics: Group hits into sectors that will contain a likely neighbouring hit Done in 3D Used in track reconstruction Vertex detector (VXD)

slide-21
SLIDE 21

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

z-Vertex Trigger

21

1 2 > 2 1 2 > 2 20 % 40 % N forward N backward

e−e+ − → τ −τ +

  • Typical B-Meson trigger requires 3 tracks (at least one in each hemisphere)
  • A lot of interesting low-multiplicity events are missed

3 tracks or more 2 track requirement

slide-22
SLIDE 22

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

z-Vertex Trigger

22

1 2 > 2 1 2 > 2 20 % 40 % N forward N backward

e−e+ − → τ −τ +

  • Typical B-Meson trigger requires 3 tracks (at least one in each hemisphere)
  • A lot of interesting low-multiplicity events are missed

3 tracks or more 2 track requirement

slide-23
SLIDE 23 z (cm)
  • 40
  • 30
  • 20
  • 10
10 20 30 40 # of events / 5 mm 200 400 600 800 1000

Z distribution

Belle

p h y s i c s background e− e+

Use FPGA based L1 trigger with neural network to “learn” z direction from drift chamber input

Input:

Track Segment

≈ 15 mm

  • position and drift time
  • f TS priority wires
  • 2D track estimates

(pT, ϕ)

Output:

ut z estimate

Plots: Sebastian Skambraks

Interesting physics collisions z

(TS)

slide-24
SLIDE 24

Neural Networks

slide-25
SLIDE 25

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

Neural Networks and Lepton colliders

25

  • Fairly clean environment (even with beam background) and no pile-up

Simuliertes Beispiel- Ereignis Elektron- Positron- Beschleuniger

http://www.nikhef.nl/~i93/img/Event6_top.png 6
  • Allows use of multivariate methods to implement a “Full Event Interpretation”
slide-26
SLIDE 26

Vollständige Ereignisinterpretation

  • Rekombinations-Effizienz O(1%) nach …

○ Rekonstruktions-Effizienz ○ “angemessener” Reinheit … erfordert bereits Betrachtung von > 10,000 Zerfallsketten

52

~ pBtag = −~ pBsig

Allows one to reconstruct the missing Four-momentum on the signal side

slide-27
SLIDE 27

Vollständige Ereignisinterpretation

  • Rekombinations-Effizienz O(1%) nach …

○ Rekonstruktions-Effizienz ○ “angemessener” Reinheit … erfordert bereits Betrachtung von > 10,000 Zerfallsketten

52

Reconstruct O(1000-10000) of hadronic and semileptonic modes, achieves an efficiency of about O(1%)

B0, B+

slide-28
SLIDE 28

Physics

slide-29
SLIDE 29

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

The big flavour questions and one anomaly

29

B-Factory measurement candy bowl

ρ η

0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5

| cb V / ub Exclusive |V | cb V / ub Inclusive |V γ Direct | cb V / ub and |V γ Future σ , 5 σ , 3 σ 1 at s m ∆ / d m ∆ ) β sin(2 Current CKM fit contours hold 68.3%
slide-30
SLIDE 30

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 30

B-Factory measurement candy bowl

ρ η

0.1 0.2 0.3 0.4 0.5 0.1 0.2 0.3 0.4 0.5

| cb V / ub Exclusive |V | cb V / ub Inclusive |V γ Direct | cb V / ub and |V γ Future σ , 5 σ , 3 σ 1 at s m ∆ / d m ∆ ) β sin(2 Current CKM fit contours hold 68.3%

Vqb

W −

−

¯ ν b q u u

Vcb

c

τ − ¯ ντ

R(D)

0.2 0.3 0.4 0.5 0.6

R(D*)

0.2 0.25 0.3 0.35 0.4 0.45 0.5 BaBar, PRL109,101802(2012) Belle, PRD92,072014(2015) LHCb, PRL115,111803(2015) Belle, PRD94,072007(2016) Belle, PRL118,211801(2017) LHCb, FPCP2017 Average SM Predictions = 1.0 contours 2 χ ∆ R(D)=0.300(8) HPQCD (2015) R(D)=0.299(11) FNAL/MILC (2015) R(D*)=0.252(3) S. Fajfer et al. (2012) ) = 71.6% 2 χ P( σ 4 σ 2 HFLAV FPCP 2017

The big flavour questions and one anomaly

slide-31
SLIDE 31

Decay in the Standard Model

Vqb

W −

−

¯ ν b q u u

Vcb

c

τ −

¯ ντ

Decay with New Physics e.g. with charged Higgs boson

Vqb

W −

−

¯ ν b q u u c

¯ ντ H−

τ −

v mτ

R(D(∗)) = B( ¯ B → D(∗)⌧ ¯ ⌫⌧) B( ¯ B → D(∗)`¯ ⌫`)

R(D/D*)

Electron or Muon

slide-32
SLIDE 32

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

How does one measure R(D) or R(D*)?

32

b

W −

¯ ν

Υ(4S)

e+ e−

hb¯ qi h¯ bqi

h¯ bbi

b u

  • W −

u

¯ d

  • π+

c u

¯ D0

Bsig Btag

K+

π−

E.g.

Vqb

W −

−

¯ ν b q u u

Vcb

c

τ − ¯ ντ

Xq = D(∗)

¯ ν`

`−

ντ

slide-33
SLIDE 33

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

How does one measure R(D) or R(D*)?

33

b

W −

¯ ν

Υ(4S)

e+ e−

hb¯ qi h¯ bqi

h¯ bbi

b u

  • W −

u

¯ d

  • π+

c u

¯ D0

Bsig Btag

K+

π−

E.g.

Vqb

W −

−

¯ ν b q u u

Vcb

c

τ − ¯ ντ

Xq = D(∗)

  • pBsig

⌫ =

  • pe+ e− − ptag − pXq − p`
  • ¯

ν`

`−

ντ

pmiss

slide-34
SLIDE 34

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics

How does one measure R(D) or R(D*)?

34

b

W −

¯ ν

Υ(4S)

e+ e−

hb¯ qi h¯ bqi

h¯ bbi

b u

  • W −

u

¯ d

  • π+

c u

¯ D0

Bsig Btag

K+

π−

E.g.

Vqb

W −

−

¯ ν b q u u

Vcb

c

τ − ¯ ντ

Xq = D(∗)

pBsig

¯ ν`

`−

ντ Separation of Signal and Background:

p2

miss = m2 miss

(pν)2 = m2

miss Number of Events semileptonic decays with light leptons

  • ther B-Meson Decays with missing

particles or other background semileptonic decays with tau leptons

⌫ =

  • pe+ e− − ptag − pXq − p`
  • pmiss

Light leptons:

pν =

slide-35
SLIDE 35

20 40 60 80 20 40 60 80

  • 2

2 4 6 8 20 40 60

  • 2

2 4 6 8 20 40 60

m2

miss (GeV2)

Dτν D∗τν Dℓν D∗ℓν D∗∗(ℓ/τ)ν Bkg.

D+ℓ D∗+ℓ

BaBar: Phys.Rev.D 88, 072012 (2013)

slide-36
SLIDE 36

R(D)

0.2 0.3 0.4 0.5 0.6

R(D*)

0.2 0.25 0.3 0.35 0.4 0.45 0.5

BaBar, PRL109,101802(2012) Belle, PRD92,072014(2015) LHCb, PRL115,111803(2015) Belle, PRD94,072007(2016) Belle, PRL118,211801(2017) LHCb, FPCP2017 Average

SM Predictions

= 1.0 contours

2

χ ∆

R(D)=0.300(8) HPQCD (2015) R(D)=0.299(11) FNAL/MILC (2015) R(D*)=0.252(3) S. Fajfer et al. (2012)

) = 71.6%

2

χ P( σ 4 σ 2

HFLAV

FPCP 2017

Plot: HFLAV 2017 For SM prediction see also: FB et al, Phys. Rev. D 95, 115008 (2017)

SM

R(D)SM = 0.299 ± 0.003 R(D∗)SM = 0.257 ± 0.003

slide-37
SLIDE 37

R(D) R(D*)

0.3 0.35 0.4 0.45 0.24 0.26 0.28 0.3 0.32 0.34

LHCb Belle II Future WA SM prediction

SM σ 1 σ 3 σ 5 σ 7 σ 9

  • 1

8fb

  • 1

22fb

  • 1

50fb

  • 1

5ab

  • 1

50ab

Plot: FB, J. Albrecht M. Kenzie S. Reichert D. M. Straub A. Tully (To appear)

slide-38
SLIDE 38

LHC Shutdown LHC Shutdown

~ 22 fb-1

LHC Shutdown

2017

Q1 Q2 Q3 Q4

2018

Q1 Q2 Q3 Q4

2019

Q1 Q2 Q3 Q4

2020

Q1 Q2 Q3 Q4

2021

Q1 Q2 Q3 Q4

2022

Q1 Q2 Q3 Q4

2023

Q1 Q2 Q3 Q4

2024

Q1 Q2 Q3 Q4

2025

Q1 Q2 Q3 Q4

2026

Q1 Q2 Q3 Q4

2027

Q1 Q2 Q3 Q4

2028

Q1 Q2 Q3 Q4

2029

Q1 Q2 Q3 Q4

2030

Q1 Q2 Q3 Q4

Belle II LHCb

Start of Data taking period ~ 50 ab-1 ~ 8 fb-1 ~ 50 fb-1

Belle II LHCb LHCb

~ 5 ab-1

Milestone I Milestone II Milestone III End of Data taking period End of Data taking period

Run 1 Run 2 Run 3

slide-39
SLIDE 39

Additional Material & Slides

slide-40
SLIDE 40

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 40

emittance

⇣±y ∼ p ∗/✏

𝛄 function

L = γ± 2ere ✓ 1 + σ∗

y

σ∗

x

◆ ✓I± ζ±y β∗

y

◆ ✓RL Ry ◆

vertical 𝛄 function beam current beam-beam parameter geometric factors beam size aspect ratio Lorentz factor

LER / HER KEKB SuperKEKB Energy [GeV] 3.5 / 8 4.0 / 7.0 𝛄y* [mm] 5.9 / 5.9 0.27 / 0.30 𝛄x* [mm] 1200 32 / 25 I± [A] 1.64 / 1.19 3.6 / 2.6 𝛈±y 0.129 / 0.09 0.09 / 0.09 𝜻 [nm] 18 / 24 3.2 / 4.6 # of bunches 1584 2500 Luminosity [1034 cm-2 s-1] 2.1 80

KEKB → SuperKEKB

slide-41
SLIDE 41

Florian Bernlochner Belle II @ KIT : From Tracks to Neural Networks to Physics 41

Upgrade positron capture section

Replaced old beam pipes with TiN coated beam pipes with antechambers

New superconducting final focusing magnets near the IP Reinforced RF (radio frequency) system for higher beam currents, improved monitoring & control system

Damping ring

Low emittance positrons to inject Low emittance electrons to inject

Low emittance gun

Redesign the lattices of HER & LER to squeeze the emittance. Replace short dipoles with longer ones (LER)

KEKB → SuperKEKB

slide-42
SLIDE 42

Original question by E. Paoloni

If you have 721 hits in the Belle II detector and you want to reconstruct 12 physical trajectories (11 from B- Meson decays, 1 from beam background), how many unique combinations do you need to consider?

⇢721 12

  • ≈ 2.57 × 10769

Number of ways to partition a set of n elements into k non-empty sets

slide-43
SLIDE 43 / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 20 40 60 80 100 3 10 × hadronic + B / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 5000 10000 15000 20000 25000 hadronic (Classifier output > 0.01) + B / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 2000 4000 6000 8000 10000 hadronic (Classifier output > 0.1) + B / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 500 1000 1500 2000 2500 3000 3500 4000 hadronic (Classifier output > 0.5) + B + / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 10000 20000 30000 40000 50000 hadronic B / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 2000 4000 6000 8000 10000 12000 14000 hadronic (Classifier output > 0.01) B / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 1000 2000 3000 4000 5000 6000 7000 hadronic (Classifier output > 0.1) B / GeV bc m 5.24 5.245 5.25 5.255 5.26 5.265 5.27 5.275 5.28 5.285 5.29 Entries per 0.0005 GeV 500 1000 1500 2000 2500 3000 hadronic (Classifier output > 0.5) B

Stronger cut On classifier

mbc = q E2

beam, CMS − ~

p2

CMS,

Three momentum of tag-side B-Meson

slide-44
SLIDE 44