[PPT] - Sensitivity study of at the Belle II experiment Outline Michel PowerPoint Presentation

SLIDE 1

Outline

B-factories and 𝜐

physics.

Second class currents
𝜐→𝜃𝜌𝜉 decay
Outlook.

Sensitivity study of 𝜐→𝜃𝜌𝜉 at the Belle II experiment

Michel Hernández Villanueva,   Cinvestav Group  Mexico City

28 Sep 2017

SLIDE 2

B Factories

B-Factory 
𝝊 factory too!

𝝉(e+e- —> 𝜱(4s)) = 1.05 nb  𝝉(e+e- —> 𝝊 𝝊) = 0.92 nb

10.58 GeV

2 Michel H. Villanueva 2

BR(Υ(4S) → B ¯ B) > 96%

SLIDE 3

Integrated Luminosity of B factories

3 Michel H. Villanueva 3 High-luminosity experiments.

6.54x108 𝝊’s 3.98x108 𝝊’s

SLIDE 4

4 Michel H. Villanueva

SuperKEKB

Super B-Factory

(And 𝝊 factory too!)

Integrated

luminosity expected: 50 ab-1  (4.6x1010 𝝊 pairs)

Full physics

program starts:   late 2018 @KEK  Tsukuba, Japan

SLIDE 5

5 Michel H. Villanueva

Belle II Detector

SLIDE 6

6 Michel H. Villanueva

Belle II MC samples

MC Sample:  ~ 2 ab-1  (1 ab-1 for training,   1 ab-1 for analysis).

SLIDE 7

Mexican Contribution

7

~1.4% CPU usage 

f the grid
504 cores

3.7 KHS06    70 TB storage

SLIDE 8

In this work, we are studying

the feasibility to measure the decay  

𝜐→𝜃𝜌𝜉 ,  

in order to get information related at:

Second class currents.
Scalar and tensorial

currents.

8 Michel H. Villanueva

The 𝜐→𝜃𝜌𝜉 decay

Disadvantage: We cannot detect 𝜉

SLIDE 9

Mechanisms in the SM: isospin violation1

9

The corresponding suppression of the SM contribution can make new

physics visible.

Charged Higgs exchange

Michel H. Villanueva

The 𝜐→𝜃𝜌𝜉 decay

1 R. Escribano, S. Gonzalez, P. Roig; Phys.Rev. D94 (2016) no.3, 034008

Leptoquark  exchange

+

SLIDE 10

10

BR(𝜐→𝜃𝜌𝜉) ~ 10-5

Accesible at Belle II luminosity.

[8] S. Nussinov + A. Soffer, PRD78, (2008) [9] N. Paver + Riazuddin, PRD82, (2010) [10] M. Volkov D. Kostunin, PRD82, (2012) [11] S. Descotes-Genon+B. Moussallam, EJPC74, (2014) [12] R. Escribano, S. Gonzalez, P. Roig; Phys.Rev. D94 (2016) no.3, 034008

Ref BRV (x105) BRS (x105) BRV+S (x105) Model [8] 0.36 1.0 1.36 MDM, 1 resonance [9] [0.2, 0.6] [0.2, 2.3] [0.4, 2.9] MDM, 1 and 2 resonances [10] 0.44 0.04 0.48 Nambu-Jona-Lasinio [11] 0.13 0.20 0.33 Analiticity, Unitarity [12] 0.26 1.41 1.67 3 coupled channels

Michel H. Villanueva

Some recent theoretical predictions

Largest difference comes  from scalar form factor.

SLIDE 11

NP contributions (scalar and tensorial currents) can be studied in the

framework of an effective field theory 1

11 Michel H. Villanueva

The 𝜐→𝜃𝜌𝜉 decay

1 E. A. Garcés, MHV, G. López Castro, P. Roig; arXiv:1708.07802

SM Belle BaBar CLEO

Constraints on scalar and tensor

couplings can be obtained from experimental upper limits on branching fractions.

Belle BaBar CLEO SM

SLIDE 12

This decay mode should have already been discovered

if there were no strong background.

Control of the background is essential.

12

470 fb-1 670 fb-1

Michel H. Villanueva

Previous Results

SLIDE 13

13 Michel H. Villanueva

Thrust axis

Thrust axis: such that

is maximum.

ˆ nthrust Vthrust

Vthrust = P

i |~

pi

cm · ˆ

nthrust| P

i |~

pi cm|

ˆ nthrust

The thrust axis define a plane which splits the space in two.

tag side signal side

SLIDE 14

14

τ τ π η γ γ ντ ντ ` ¯ ν` τ τ π π π η π0 γ γ ντ ντ ` ¯ ν`

Tag side

Signal side

Michel H. Villanueva

e+ e−

1-prong 3-prong

2 ways to reconstruct 𝜃

BR(𝜃 —> 𝛿𝛿) = 39.41% BR(𝜃 —> 𝜌𝜌𝜌0) = 22.92%

Thrust axis: such that

is maximum.

ˆ nthrust Vthrust

Vthrust = P

i |~

pi

cm · ˆ

nthrust| P

i |~

pi cm|

e+ e−

SLIDE 15

Selection criteria :tag + 1 or 3 charged + 2 or 3 𝛿.
Signal events generated: 4M.

(2M for training and 2M for sensitivity study).

1-prong 3-prong

η → γγ

η → π+π−π0

] 2 [GeV/c γ γ Invariant Mass 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2 Events / ( 0.002 ) 20 40 60 80 100 120 140 160 180 200 3 10 ×

π0 → γγ

Mis-reconstructed 𝜌0

15

𝛿 from other sources

Michel H. Villanueva

𝜐→𝜃𝜌𝜉 signal events

Eff: 13.56% Eff: 3.70%

BR(𝜃 —> 𝛿𝛿) = 39.41% BR(𝜃 —> 𝜌𝜌𝜌0) = 22.92%

SLIDE 16

16 Michel H. Villanueva

𝜐→𝜃𝜌𝜉 bkg events

Background sources: 
𝝊𝝊 pair 
bb pair 
qq pair

1-prong

bb pair qq pair tau pair

1 ab-1 MC 1 ab-1 MC

1 ab-1 MC

Eff: 0.002% Eff: 0.34% Eff: 0.006%

𝜌0 veto  applied.

SLIDE 17

17 Michel H. Villanueva

BDT variables (1-prong)

∠(𝜃,𝜌)
∠(pmiss, Vthrust)
Mmiss
Pt(𝜌)
𝜃(𝜃)
∠(𝛿, 𝛿)𝜃

TMVA used for this test.

cos(𝜄miss)
PIDe(𝜌)
PIDµ(𝜌)
PIDK(𝜌)
E(𝛿)

Corte en BDT 0.1 − 0.05 − 0.05 0.1 0.15 0.2 0.25 0.3

No. de Background

200 400 600 800 1000 1200

3

10 ×

Eficiencia 0.02 0.04 0.06 0.08 0.1 0.12

Punzi

Optimal   cut

100 − 80 − 60 − 40 − 20 − 20 40 60 80 100

η ) γ , γ ( ∠ π P t η η miss M ) π , η ( ∠ ) 2 γ ) + E ( 1 γ E ( ) π ( K # P I D ) π ( µ # P I D ) π ( e # P I D ) miss θ c

s

( ) thrust , V miss ( p ∠ η ) γ , γ ( ∠ π Pt η η miss M ) π , η ( ∠ ) 2 γ ) + E( 1 γ E( ) π ( K #PID ) π ( µ #PID ) π ( e #PID ) miss θ cos( ) thrust ,V miss (p

Correlation Matrix (background)

100 1 10

3
3

5

2

4 6 1 100

1
19
6
8

3

6

1 1 4 10

1

100 10 15

5
10

3

55

4

3
19

10 100 26

50
4

6 3

6
3
6

15 26 100

9
2

12

4
19

5

8
5
50
9

100

6

6

3

33 3

10
4
2
6

100

3

8 1

2
6

6 12 6

3

100

7
2

4 1 3 3

7

100

2

6 1

55
6
4
3

8

2

100

18

4 4

19

33 1

2
18

100

Linear correlation coefficients in %

100 − 80 − 60 − 40 − 20 − 20 40 60 80 100

η ) γ , γ ( ∠ π Pt η η miss M ) π , η ( ∠ ) 2 γ ) + E( 1 γ E( ) π ( K #PID ) π ( µ #PID ) π ( e #PID ) miss θ cos( ) thrust ,V miss (p ∠ η ) γ , γ ( ∠ π Pt η η miss M ) π , η ( ∠ ) 2 γ ) + E( 1 γ E( ) π ( K #PID ) π ( µ #PID ) π ( e #PID ) miss θ cos( ) thrust ,V miss (p

Correlation Matrix (signal)

100 12 19

7
15

4

1
4
2
3
1

12 100

37
60
24

7

24
9

25 19 100 1

20
3
2
54
7
37

100 41

54
5

8 2

2
15
60

41 100

8
12

18 9

32

4

24

1

54
8

100

5

6 2

1

37

1

7

20
5
12
5

100

4

14 4

4
24
3

8 18 6

4

100

2

2

6
2
9
2

2 9 2

2

100

1
3
54
1

14 2 100

1

25

2
32

37 4

6
1

100

Linear correlation coefficients in %

BDT response

0.6 − 0.4 − 0.2 − 0.2 0.4

dx / (1/N) dN

1 2 3 4 5

Signal (test sample) Background (test sample) Signal (training sample) Background (training sample)

Kolmogorov-Smirnov test: signal (background) probability = 0.81 (0.063)

U/O-flow (S,B): (0.0, 0.0)% / (0.0, 0.0)%

TMVA overtraining check for classifier: BDT

Optimization proposed by Punzi, G.  at arXiv preprint physics/0308063

✏ a/2 + √ B

SLIDE 18

]

2

[GeV/c γ γ Invariant Mass 0.4 0.45 0.5 0.55 0.6 0.65 Events / ( 0.0025 ) 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000

]

2

) [GeV/c γ γ ( η Invariant Mass 0.4 0.42 0.44 0.46 0.48 0.5 0.52 0.54 0.56 0.58 0.6 Events / ( 0.004 ) 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000

0.010 ± = 1.039 α 0.000051 ± = 0.540854 µ 0.000038 ± = 0.010962 σ 0.039 ± n = 2.309

18 Michel H. Villanueva

Optimal BDT cut

Nsig=157,680  Eff: 7.88%

Nsig = 271,258 Eff: 13.56%

Effcut = 41.87% Signal

1-prong

SLIDE 19

]

2

[GeV/c γ γ Invariant Mass 0.4 0.45 0.5 0.55 0.6 0.65 Events / ( 0.0025 ) 1000 2000 3000 4000 5000 6000 7000

MC events ν ) π π → ρ ( ν ) π π π →

1

(a ν γ π π b b q q

]

2

[GeV/c γ γ Invariant Mass 0.4 0.45 0.5 0.55 0.6 0.65 Events / ( 0.0025 ) 10000 20000 30000 40000 50000 MC events ν ) π π → ρ ( ν ) π π π →

1

(a ν γ π π b b q q

19 Michel H. Villanueva

Optimal BDT cut

Nbkg=417,217  Eff: 5.26x10-4

Nbkg = 2,694,408 Eff: 0.34%

Effcut = 84.51% Background

1-prong

MC events ν ) π π → ρ ( ν ) π π π →

1

(a ν γ π π b b q q

𝜈 ± 3σ

98,146 events

SLIDE 20

20 Michel H. Villanueva

𝜐→𝜃𝜌𝜉 bkg events

Background sources: 
𝝊𝝊 pair 
bb pair 
qq pair

3-prong

1 ab-1 MC 1 ab-1 MC

1 ab-1 MC

Eff: 5.6x10-6 Eff: 0.028% Eff: 7.6x10-6

bb pair qq pair tau pair

3𝛒𝛒0 is the mayor issue. (This depends of the hadronic input in the generation of MC)

SLIDE 21

BDT cut 0.1 − 0.05 − 0.05 0.1 0.15 0.2 0.25 0.3 Background number 20 40 60 80 100 120

3

10 ×

Efficiency 0.005 0.01 0.015 0.02 0.025 0.03 0.035

Punzi

21 Michel H. Villanueva

BDT variables (3-prong)

∠(𝜃,𝜌)
∠(pmiss, Vthrust)
∠(𝜌, 𝜌0)
∠(𝛿, 𝛿)𝜌0
Mmiss
Pt(𝜌)
Pt(𝜃)
Pt(𝜌0)
𝜃(𝜌0)
E(𝛿)

Optimal   cut

100 − 80 − 60 − 40 − 20 − 20 40 60 80 100

π ) γ , γ ( ∠ π P t η P t π P t π η miss M ) π , η ( ∠ ) π ^ , π ( ∠ ) 2 γ ) + E ( 1 γ E ( ) miss θ c

s

( ) thrust , V miss ( p ∠ π ) γ , γ ( ∠ π Pt η Pt π Pt π η miss M ) π , η ( ∠ ) π ^0 , π ( ∠ ) 2 γ ) + E( 1 γ E( ) miss θ cos( ) thrust ,V miss (p

Correlation Matrix (signal)

100 9

54
2

2 33 7 34

76
1
24

9 100

17

1

1
38
57

10

15

25

54
17

100 8

4
52
8
61

61 1 34

2

1 8 100

6
2
5

7 4 2

1
4

100 2 2

1
56
2

33

38
52
6

2 100 41 38

37
2
3

7

57
8
2

41 100 5

6
31

34 10

61
5

2 38 5 100

36
1
27
76
15

61 7

1
37
6
36

100 24

1

1

56
2
1

100

1
24

25 34 4

2
3
31
27

24

1

100

Linear correlation coefficients in %

100 − 80 − 60 − 40 − 20 − 20 40 60 80 100

π ) γ , γ ( ∠ π P t η P t π P t π η miss M ) π , η ( ∠ ) π ^ , π ( ∠ ) 2 γ ) + E ( 1 γ E ( ) miss θ c

s

( ) thrust , V miss ( p ∠ π ) γ , γ ( ∠ π Pt η Pt π Pt π η miss M ) π , η ( ∠ ) π ^0 , π ( ∠ ) 2 γ ) + E( 1 γ E( ) miss θ cos( ) thrust ,V miss (p

Correlation Matrix (background)

100 16

56
8

10 24 2 29

76
15

16 100

19
1
8
45
41

11

19

7 21

56
19

100 5

29
46
2
58

61 12 23

8
1

5 100

3
5

12 1 10

8
29
3

100 17 10 10

10
54

1 24

45
46

17 100 34 31

25
10

2

41
2

10 34 100 4

4
18

29 11

58
5

10 31 4 100

29
3
17
76
19

61 12

10
25
29

100 1 15 7 12 1

54
10
4
3

1 100

14
15

21 23 1

18
17

15

14

100

Linear correlation coefficients in %

BDT response

0.6 − 0.4 − 0.2 − 0.2 0.4

dx / (1/N) dN

1 2 3 4 5

Signal (test sample) Background (test sample) Signal (training sample) Background (training sample)

Kolmogorov-Smirnov test: signal (background) probability = 0.005 (0.149)

U/O-flow (S,B): (0.0, 0.0)% / (0.0, 0.0)%

TMVA overtraining check for classifier: BDT

✏ a/2 + √ B

SLIDE 22

]

2

) [GeV/c π π π ( η Invariant Mass 0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 0.6 Events / ( 0.002 ) 1000 2000 3000 4000 5000 6000

0.017 ± = 1.167 α 0.000034 ± = 0.545065 µ 0.0000051 ± = 0.0050000 σ 123 ± Nb = 9408 206 ± Ns = 36420 35 ± = 102 a 18 ± n = 98

22 Michel H. Villanueva

Optimal BDT cut

Nsig=36,420  Eff: 1.82%

Nsig = 74,088 Eff: 3.70%

Effcut = 38.14% Signal

3-prong

SLIDE 23

]

2

[GeV/c π

π

+

π Invariant Mass 0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 0.6 Events / ( 0.001 ) 100 200 300 400 500 600

MC events ν ) π π π π ( ν ) π π π →

1

(a ν π ω π b b q q

]

2

[GeV/c γ γ Invariant Mass 0.5 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 0.6 Events / ( 0.001 ) 500 1000 1500 2000 2500 3000 3500 4000

MC events ν ) π π π π ( ν ) π π π →

1

(a ν π ω π b b q q

23 Michel H. Villanueva

Optimal BDT cut

Nbkg=39,634  Eff: 5.0x10-5

Nbkg = 240,438 Eff: 3.03x10-4

Effcut = 83.52% Background

3-prong

𝜈 ± 3σ

12,120 events

SLIDE 24

24 Michel H. Villanueva

Estimation @ 1 ab-1

BR(𝜐→𝜃𝜌𝜉) ~ 10-5

In the mass window of 𝜃:

Nbkg = 98,146

Nsig = ✏ · ττ · BR(⌧ → `⌫¯ ⌫) · L · BR(⌧ → ⌘⇡⌫) Nsig p Nbkg ' 0.786 In the mass window of 𝜃:

Nbkg = 12,120

Nsig p Nbkg ' 0.516

3-prong 1-prong

SLIDE 25

25 Michel H. Villanueva

Estimation @ 50 ab-1

BR(𝜐→𝜃𝜌𝜉) ~ 10-5

In the mass window of 𝜃:

Nbkg ≃ 4.9x106

Nsig = ✏ · ττ · BR(⌧ → `⌫¯ ⌫) · L · BR(⌧ → ⌘⇡⌫) In the mass window of 𝜃:

Nbkg ≃ 6.06x105

3-prong 1-prong

Nsig p Nbkg ' 5.56 Nsig p Nbkg ' 3.65

SLIDE 26

)

1

Luminosity (ab

1 −

10 × 2 1 2 3 4 5 6 7 8 10 20

5

10 × ) ν π η → τ Upper Limit of BR(

1 2 3 4 5 6 7 8 9 10 11 12 13

26 Michel H. Villanueva

Estimated Upper Limits

3-prong

BaBar  Belle  3 channels model  Other SM models

SLIDE 27

)

1

Luminosity (ab

1 −

10 × 2 1 2 3 4 5 6 78 10 20

5

10 × ) ν π η → τ Upper Limit of BR( 1 2 3 4 5 6 7 8

27 Michel H. Villanueva

Estimated Upper Limits

1-prong

BaBar  Belle  3 channels model  Other SM models

SLIDE 28

28

SuperKEKB will produce a sample of 𝜐 pairs 50 times larger than

previous B-factories. 𝜐 physics is now considered “precision physics”.

BR measurement (or upper limit), invariant mass of 𝜃𝜌, and Dalitz

plots will be very important to disentangle models.

Better selection of variables or more MVA techniques have to be

tested.

Some extra contributions to the background has to be studied.
The comparison of channels generated, with the data obtained in

the beginning of the experiment, is important to control the bkg.

Michel H. Villanueva

Summary

SLIDE 29

29

Thank you

Michel H. Villanueva

SLIDE 30

30

Backup

Michel H. Villanueva

SLIDE 31

The 𝜐 lepton is the only lepton massive

enough to decay into hadrons.

Semileptonic decay channels

𝜐 → H 𝜉𝜐 allow a clean theoretical analysis of the hadronization, determination of SM parameters and properties of weak currents1: 𝛽s CKM parameters CPV LNV and LFV SM and NP interactions etc.

B-factories provide a large dataset of 𝜐

leptons to precision studies.

31 Michel H. Villanueva

Semileptonic decays of 𝜐 lepton

τ ντ

Hadronization

h1 h2 h3 W q q > 200 hadronic channels

1Pich, A. Progress in Particle and Nuclear Physics, 75, 41-85 (2014).

Disadvantage: We cannot detect 𝜉

SLIDE 32

V-A currents can be classified by their

transformation proprieties under G-parity 1.

First-class currents:
Second-class currents (SCC):

JP G = 0+−, 0−+, 1++, 1−−, ... JP G = 0++, 0−−, 1−+, 1+−, ...

32

GVµG−1 = +Vµ GA0

µG1 = +A0 µ

GAµG−1 = −Aµ GSG−1 = −S

Hadronic Currents

Michel H. Villanueva

Standard Model New physics

1S. Weinberg. Physical Review, 112(4), 1375 (1958).

G = CeiπI2

SLIDE 33

SCC are isospin violating processes, suppressed by isospin

symmetry.

Unsuccessful searches of SCC in nuclear Physics.

33

/ G ∼ / SU(2) ∼ md − mu Λ

Second-class currents

1Leroy, C., & Pestieau, J. (1978). Physics Letters B, 72(3), 398-399.

Another possibility:

Search in tau decays1

G|πi = |πi G|ηi = +|ηi

τ − → ηπ−ντ

Michel H. Villanueva

G| ¯ dγµui = +| ¯ dγµui

SLIDE 34

G-parity is defined by1
Is a good symmetry of the strong interactions

Convenient to analyze process where the initial or final state contains only mesons

However, G-Parity is not exact.

G = CeiπI2

34

[Hstr, Ii] = 0; [Htot, Ii] 6= 0;

G-parity

Michel H. Villanueva

[Hstr, C] = 0

1T. D. Lee, and Chen Ning Yang. Il Nuovo Cimento 3.4 (1956): 749-753.

SLIDE 35

NP contributions (scalar and tensorial currents) can be studied in the

framework of an effective field theory 1

35 Michel H. Villanueva

The 𝜐→𝜃𝜌𝜉 decay

1 E. A. Garcés, MHV, G. López Castro, P. Roig; arXiv:1708.07802

New Physics effects can appear in the

distribution of Dalitz plots, with a large enhancement expected towards large values of the hadronic invariant mass1.

Ratio between the squared amplitude of EFT  with 𝜗T = 0.3 and squared amplitude of SM.

SLIDE 36

36

.
BR ~ 10-5 !

(Not suppressed by G-parity, unlike the   channel without photon.) 

Veto of photons with Eγ > 100 MeV should get rid of this background.

τ − → ηπ−ντγ

1A. Guevara, G. López-Castro, P. Roig (2016). Phys.Rev. D95 no.5, 054015 (2017)

Michel H. Villanueva

A new bkg source1

SLIDE 37

Michel H. Villanueva

B-Factories

PEP-II KEKB SuperKEKB Detector BaBar Belle Belle II Año de inicio 1999 1999 2016 Fin de

peraciones

2008 2010

Energía del haz

(GeV) e-: 9.0 e+: 3.1 e-: 8.0 e+: 3.5 e-: 7.0 e+: 4.0 Luminosidad max 550 fb-1 1 ab-1 50 ab-1

SLIDE 38

SLIDE 39

Hadronic matrix element: 2 form factors
Invariant mass distribution

39 Michel H. Villanueva

SLIDE 40

Form factors
MDM: Meson dominance models.
Sum of Breit-Wigner formulae
Chiral theory, etc.

40 Michel H. Villanueva

SLIDE 41

For tau physics study, roughly TinyDST (tdst) is designed1.
Events having:

Less than 6 charged tracks with |dr|<0.5 cm, |dz|<3.0 cm,   pt>0.1 GeV/c and -0.8660<cos 𝜄<0.9535. Less than 10 photons with E𝛿>50 MeV and -0.8660<cos 𝜄<0.9535.

Thrust vector information contained.
To squeeze the size, one lepton is required.

In SM precise measurement, to avoid qq BG, usually, leptonic decay is required for tag tau (tau with non-signal decay ).

50MBytes for 200k events. (In original mdst, 50MBytes for 20k events.)

Michel H. Villanueva

TinyDST

SLIDE 42

42 Michel H. Villanueva

Boosted Decision Trees

What is a Decision Tree? 
Consecutive set of questions

(nodes). 

Two possible answers per

node. 

Final verdict (leaf) is reached

after a defined maximum of   nodes.

Advantages 
Easy to understand. 
Fast training.
Disadvantages 
Single tree not strong (that’s

why we use Random   Forests).

SLIDE 43

43 Michel H. Villanueva

Boosted Decision Trees

Random Forest is an

ensemble method that combines different trees.

Final output is determined by

the majority vote of all the trees.

Boosting: 
Misclassified events are

weighted higher so that future learners concentrate

n these.

The score of an event is a weighted average

f the scores the event receives from each

tree in the forest.

bdt = P

i wiNi

P

i wi

; Ni = -1 or 1

SLIDE 44

Michel H. Villanueva

BDT tests

Signal efficiency

0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

Background rejection

0.3 0.4 0.5 0.6 0.7 0.8 0.9

MVA Method: BDT BDTMitFisher BDTG BDTD BDTB

Background rejection versus Signal efficiency