Lambda Weak Decay Constant – how to assess different results
D. G. Ireland Bayesian Inference in Subatomic Physics Gothenburg, 17-20 September, 2019
1 Objectives
Statistical Challenge:
Two independent measurements of same physical quantity - two results are (apparently) not compatible.
Discussion questions for talk today: Are there any obvious mistakes? • How should results be judged as compatible? • What are possible strategies for identifying sources of uncertainty? •
2 The Recent BESIII Result Nature Physics paper1
1M. Ablikim et al. “Polarization and entanglement in baryon-antibaryon pair production in electron-positron annihilation”. In: Nature Physics (May 2019), p. 1.
3 ΛΛ¯ Production
ΛΛ¯ production process.
4 ΛΛ¯ Production
BES Λ Λ¯ event Polarization signal... −
5 ΛΛ¯ Parameter Estimation
Intensity distribution of ΛΛ¯ events
Observations are counts in kinematic (ξ) space (total N 400000) • ∼ αψ, ∆Φ, α− and α+ are parameters to estimate • Maximise likelihood: • N N Y i i data = ξ ; αψ, ∆Φ, α−, α+ ξ L C W i=1
6 Λ Weak Decay Parameter- BES Results
7 Λ Weak Decay Parameter- BES Results
7 Λ Weak Decay Parameter- BES Results
> 5σ difference between new result and PDG2.
2(Particle Data Group) Tanabashi, M. et al. “Review of Particle Physics”. In: Phys. Rev. D 98.3 (Aug. 2018), p. 030001.
7 So What? pp¯ ΛΛ¯ Polarization3 →
3E. Klempt et al. “Antinucleon nucleon interaction at low energy: Scattering and protonium”. In: Phys. Rept. 368 (2002), pp. 119–316.
8 Global Λ hyperon polarization in nuclear collisions4
STAR Au-Au collision Average Λ (Λ)¯ polarization in collisions...
4The STAR Collaboration. “Global Λ hyperon polarization in nuclear collisions”. In: Nature 548.7665 (2017), pp. 62–65.
9 Λ(Λ)¯ Transverse Polarization with ATLAS5
5ATLAS Collaboration, G. Aad, et al. “Measurement of the transverse polarization of Λ and Λ¯ hyperons produced in proton-proton...”. In: Phys. Rev. D 91 (2015), 10 p. 032004. Decay Parameters of Higher Mass States6
6(Particle Data Group) Tanabashi, M. et al. “Review of Particle Physics”. In: Phys. Rev. D 98.3 (Aug. 2018), p. 030001.
11 Light Baryon Spectrum - PDG 20187 List of N∗ Resonances
N Resonances 3000
2500 ) 2 c / V e
M 2000 (
s s a M 1500
**** 1996 **** 2018 *** *** 1000 ** ** * *
+ + + + + + 1 + 3 + 5 7 9 11 13 15 1− 3− 5− 7− 9− 11− 13− 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Spin-Parity
7(Particle Data Group) Tanabashi, M. et al. “Review of Particle Physics”. In: Phys. Rev. D 98.3 (Aug. 2018), p. 030001. 12 Kaon Photoproduction Kaon Photoproduction - KΛ example
13 Fierz Identity for LUY plus CUY
Intensities for Experimental Polarization Configurations: LUY: Linear photon beam; unpolarized target; measured recoil
γ 1 + α− cos θy P Σ + α− cos θy T P cos 2(α φ) − { } L − γ + α− cos θx Ox + α− cos θz Oz P sin 2(α φ) { } L − CUY: Circularly photon beam; unpolarized target; measured recoil
γ 1 + α− cos θy P + (α− cos θx Cx + α− cos θz Cz) PC
Fierz identities connecting two experiments:
2 2 2 2 2 2 2 Ox + Oz + Cx + Cz + Σ T + P = 1 − ΣP CxOz + CzOx T = 0 . − −
14 Fierz Identities as a Statistical Ensemble
Generate pseudodata:
Pick random point in amplitude space • Calculate observables from amplitudes • Treat observables as independent and adjust value by a random • number sampled from (0, σ), N Single polarization observables have σ [0.01, 0.05] • ∈ Double polarization observables have 3 σ • × Writing 2 2 2 2 2 2 2 = Ox + Oz + Cx + Cz + Σ T + P F − Also define pull as 1 F − σF
15 Key Result: Histogram of Fierz Identities
3.0 Pseudodata
2.5
2.0
1.5
1.0
0.5
0.0 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2 2 2 2 2 2 2 Ox + Oz + Cx + Cz + T + P
16 Key Result: Histogram of Fierz Identities
3.0 Pseudodata Original 2.5
2.0
1.5
1.0
0.5
0.0 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2 2 2 2 2 2 2 Ox + Oz + Cx + Cz + T + P
16 Key Result: Histogram of Fierz Identities
3.0 Pseudodata Original Scaled 2.5
2.0
1.5
1.0
0.5
0.0 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2 2 2 2 2 2 2 Ox + Oz + Cx + Cz + T + P
16 Estimate α from KΛ data? − Parameter Estimation
Infer α− from KΛ data with: M. D¨oring,D. I. Glazier, J. Haidenbauer, R. Murray-Smith, M. Mai and D. R¨onchen8.
Fierz Identities as Random Variables
(1) =a2l 2 2 + 2 2 + a2c2 2 + 2 + l 2Σ 2 + a2 2 , Fi Ox,i Oz,i − Ti Cx,i Cz,i i Pi (2) =al [Σi i i ac( x,i z,i z,i x,i )] , Fi P − T − C O − C O PDG where a = α− /α− and c, l are relative calibrations (i.e. systematics) for circular and linear photon polarization, resp. Instances of (1) and (2) for given a, c and l are f (1) and f (2) Fi Fi i i
8D. G. Ireland et al. In: arXiv:1904.07616 (2019).
17 Parameter Estimation
Likelihood Function
(1) !2 (2) !2 fi 1 fi pi (Oi a, l, c) exp − , | ∝ − σ (1) − σ (2) Fi Fi assuming observables are independent and identically distributed 7 (gaussian). Oi = j,i represents observables at point i ∪j=1O For observables from N bins:
n Y (O a, l, c) pi (Oi a, l, c) P | ∝ | i=1
n where O = Oi represents entire data set. ∪i=1 i.e. the probability that the observables obey Fierz identity
18 Parameter Estimation
Include knowledge of calibrations Calculate posterior PDF from likelihood and prior:
(a, l, c O) (O a, l, c) (l, c) . P | ∝ P | P
Impose α− 0 • ≥ Quoted systematic uncertainties in PL are 3-6% (use 5%) • γ Quoted systematic uncertainties in PC are 2% (use 2%) • γ Which PDF to use? Gaussian (1, σ)? Uniform (1 σ, 1 + σ)? • N U − Posterior PDF Function of a, c and l, (a, l, c O), calculations done by P | 1. Markov Chain Monte Carlo (MCMC) 2. Direct calculation on a c l grid − − 19 Gaussian Prior
�� �(α-) �� �� �� Posterior looks �� PDG • �� III BES nice! � � Most probable • �(α-� �) �(�) value of l ��� ��
��� questionable!
� ��
��� �
��� �
���� �(α-� �) �(�� �) �� �(�) ���� �� �� � ���� �� ���� � ���� � ���� ���� ���� ���� ��� ��� ��� ��� ���� ���� ���� ���� ���� α- � � 20 Uniform Prior
(α ) �� � - �� Posterior in l • ��
PDG and c does not E III BES �� look nice! � Most probable ���� �(α-� �) �(�) �� • ���� values are
� ���� �� reasonable! ���� � ���� �
�(α-� �) �(�� �) �� �(�) ���� �� ��
� �� �� �� ���� � � ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� ���� α- � � 21 Posterior PDFs for α− - Summary of results
120
100
80
60
40
Gaussian prior 20 Fixed l and c Uniform prior 0 0.62 0.64 0.66 0.68 0.70 0.72 0.74 0.76 0.78
22 Posterior PDFs for α− - Summary of results
120 PDG Value 100 BES III Value
80
60
40
Gaussian prior 20 Fixed l and c Uniform prior 0 0.62 0.64 0.66 0.68 0.70 0.72 0.74 0.76 0.78
22 Posterior PDFs for α− - Summary of results
120 PDG Value 100 BES III Value
80
60
40
Gaussian prior 20 Fixed l and c Uniform prior 0 0.62 0.64 0.66 0.68 0.70 0.72 0.74 0.76 0.78
α− = 0.721 0.006 (stat.) 0.005( sys.) ± ± 22 Estimates of α− - Summary of results
L C Source Value Prior Assumption Pγ , Pγ PDG 0.642 0.013 ± BES III 0.750 0.009 0.004 (new PDG value) ± ± 0.719 0.013 (1.0, 0.052), (1.0, 0.022) ± N N 0.721 0.006 (?) (0.95, 1.05), (0.98, 1.02) CLAS ± U U 0.727 0.007 (0.90, 1.10), (0.96, 1.04) ± U U 0.717 0.004 Both fixed at 1.0 ± 0.721 0.006 0.005 Summary of our result ± ±
23 Conclusions Kaon photoproduction data can • independently determine α−
Our result: α− = 0.721 0.006 (stat.) • ± 0.005( sys.) ± Other analyses will have to be • reviewed!
Summary
New BES III result for α− is 17% • higher than PDG value
24 Our result: α− = 0.721 0.006 (stat.) • ± 0.005( sys.) ± Other analyses will have to be • reviewed!
Summary
New BES III result for α− is 17% • higher than PDG value Kaon photoproduction data can • independently determine α−
24 Other analyses will have to be • reviewed!
Summary
New BES III result for α− is 17% • 120 higher than PDG value PDG Value 100 BES III Value Kaon photoproduction data can • 80 independently determine α− 60
40
Our result: α = 0.721 0.006 (stat.) Gaussian prior − 20 Fixed l and c • ± Uniform prior 0.005( sys.) 0 ± 0.62 0.64 0.66 0.68 0.70 0.72 0.74 0.76 0.78
24 Summary
New BES III result for α− is 17% • 120 higher than PDG value PDG Value 100 BES III Value Kaon photoproduction data can • 80 independently determine α− 60
40
Our result: α = 0.721 0.006 (stat.) Gaussian prior − 20 Fixed l and c • ± Uniform prior 0.005( sys.) 0 ± 0.62 0.64 0.66 0.68 0.70 0.72 0.74 0.76 0.78 Other analyses will have to be • reviewed!
Details posted on arXiv9
9D. G. Ireland et al. In: arXiv:1904.07616 (2019).
24 Objectives
Statistical Challenge:
Two independent measurements of same physical quantity - two results are (apparently) not compatible.
Discussion questions for talk today: Are there any obvious mistakes? • How should results be judged as compatible? • What are possible strategies for identifying sources of uncertainty? •
25 Backup Slides
26 Light Baryon Spectrum - PDG 199610 List of ∆ Resonances
∆∗ Resonances: PDG 1996 3000
2500 ) 2
2000 Mass (MeV/c 1500
**** *** 1000 ** *
+ + + + + + + + 1 3 5 7 9 11 13 15 1− 3− 5− 7− 9− 11− 13− 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Spin-Parity
10R.M. Barnett et al. In: Phys. Rev. D 54 (1996), p. 1.
26 Light Baryon Spectrum - PDG 201811 List of ∆ Resonances
∆∗ Resonances 3000
2500 ) 2
2000 Mass (MeV/c 1500
**** 1996 **** 2018 *** *** 1000 ** ** * *
+ + + + + + + + 1 3 5 7 9 11 13 15 1− 3− 5− 7− 9− 11− 13− 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Spin-Parity
11(Particle Data Group) Tanabashi, M. et al. “Review of Particle Physics”. In: Phys. Rev. D 98.3 (Aug. 2018), p. 030001. 26 g8 K Λ Coverage −
1.00
0.75
0.50
0.25
s 0.00 o c
0.25
0.50
0.75 g8 g1c 1.00 1.7 1.8 1.9 2.0 2.1 2.2 W (GeV)
26 Gaussian Process Interpolation
26 Using Gaussian Process to Interpolate g1c to g8 Bins: Cx
1
0
W = 1.72 GeV W = 1.74 GeV W = 1.76 GeV W = 1.78 GeV W = 1.8 GeV W = 1.82 GeV 1 1
0
W = 1.84 GeV W = 1.86 GeV W = 1.88 GeV W = 1.9 GeV W = 1.92 GeV W = 1.94 GeV 1 x
C 1
0
W = 1.96 GeV W = 1.98 GeV W = 2.0 GeV W = 2.02 GeV W = 2.04 GeV W = 2.06 GeV 1 1
0
W = 2.08 GeV W = 2.1 GeV W = 2.12 GeV W = 2.14 GeV W = 2.16 GeV W = 2.18 GeV 1
1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 cos
26 Using Gaussian Process to Interpolate g1c to g8 Bins: Cz
1
0
W = 1.72 GeV W = 1.74 GeV W = 1.76 GeV W = 1.78 GeV W = 1.8 GeV W = 1.82 GeV 1 1
0
W = 1.84 GeV W = 1.86 GeV W = 1.88 GeV W = 1.9 GeV W = 1.92 GeV W = 1.94 GeV 1 z
C 1
0
W = 1.96 GeV W = 1.98 GeV W = 2.0 GeV W = 2.02 GeV W = 2.04 GeV W = 2.06 GeV 1 1
0
W = 2.08 GeV W = 2.1 GeV W = 2.12 GeV W = 2.14 GeV W = 2.16 GeV W = 2.18 GeV 1
1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 cos
26 Key Result: Pulls of Fierz Identities
0.9 Pseudodata 0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0 3 2 1 0 1 2 3 Pull
26 Key Result: Pulls of Fierz Identities
0.9 Pseudodata 0.8 Original 0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0 3 2 1 0 1 2 3 Pull
26 Key Result: Pulls of Fierz Identities
0.9 Pseudodata 0.8 Original Scaled 0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0 3 2 1 0 1 2 3 Pull
26 MCMC Chain - no prior
1.05
0.90
a 0.75
0.60
1.20
1.05
0.90 l
0.75
0.60
1.75
1.50 c 1.25
1.00
0.75 0 500 1000 1500 2000 MCMC Time Step 26 MCMC Chain - gaussian prior
1.04
0.96 a 0.88
0.80
1.05
0.90 l
0.75
0.60
1.20
1.12
c 1.04
0.96
0.88 0 500 1000 1500 2000 MCMC Time Step 26 MCMC Chain - uniform prior
1.00
0.95 a 0.90
0.85
1.050
1.025
l 1.000
0.975
0.950
1.02
1.01
c 1.00
0.99
0.98 0 500 1000 1500 2000 MCMC Time Step 26 Light Baryon Spectrum - PDG 199612 List of N∗ Resonances
N∗ Resonances: PDG 1996 3000
2500 ) 2
2000 Mass (MeV/c 1500
**** *** 1000 ** *
+ + + + + + + + 1 3 5 7 9 11 13 15 1− 3− 5− 7− 9− 11− 13− 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Spin-Parity
12R.M. Barnett et al. In: Phys. Rev. D 54 (1996), p. 1.
26 Resonance Hunting
Resonances found from πN scattering...
Coupling to other channels13?
13Simon Capstick and W. Roberts. “Strange decays of nonstrange baryons”. In: Phys. Rev. D 58 (1998), p. 074011.
26 Kaon Photoproduction - KΛ Experimental Database
Accumulated KΛ data points
26 Recent Experimental Data
Data also from ELSA, GRAAL, LEPS, MAMI,...
26 Fierz Identity Values for Original Data
2 Raw 2 P 1 + 2 T 0 W = 1.72 GeV W = 1.74 GeV W = 1.76 GeV W = 1.78 GeV W = 1.80 GeV W = 1.82 GeV 2 2 + 2 z
C 1 + 2 x C 0 W = 1.84 GeV W = 1.86 GeV W = 1.88 GeV W = 1.90 GeV W = 1.92 GeV W = 1.94 GeV + 2 z 2 O + 2 x 1 O
: y t
i W = 1.96 GeV W = 1.98 GeV W = 2.00 GeV W = 2.02 GeV W = 2.04 GeV W = 2.06 GeV
t 0 n
e 2 d I
z r 1 e i F 0 W = 2.08 GeV W = 2.10 GeV W = 2.12 GeV W = 2.14 GeV W = 2.16 GeV W = 2.18 GeV 1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 cos
26 Fierz Identity Values for Scaled Data
2 Raw
2 Scaled P 1 + 2 T 0 W = 1.72 GeV W = 1.74 GeV W = 1.76 GeV W = 1.78 GeV W = 1.80 GeV W = 1.82 GeV 2 2 + 2 z
C 1 + 2 x C 0 W = 1.84 GeV W = 1.86 GeV W = 1.88 GeV W = 1.90 GeV W = 1.92 GeV W = 1.94 GeV + 2 z 2 O + 2 x 1 O
: y t
i W = 1.96 GeV W = 1.98 GeV W = 2.00 GeV W = 2.02 GeV W = 2.04 GeV W = 2.06 GeV
t 0 n
e 2 d I
z r 1 e i F 0 W = 2.08 GeV W = 2.10 GeV W = 2.12 GeV W = 2.14 GeV W = 2.16 GeV W = 2.18 GeV 1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 1 0 1 cos
26 Refit of J¨uBoCoupled Channels Model
Observable χ2/n (Refits)
(# data points) α− = 0.642 0.75 0.721 dσ/dΩ (421) 1.11 1.03 0.95 Σ (314) 2.55 2.61 2.56 T (314) 1.75 1.74 1.69 P (410) 1.84 1.66 1.62
Cx (82) 2.15 1.72 1.34 Cz (85) 1.58 1.83 1.62 Ox (314) 1.44 1.53 1.51 Oz (314) 1.34 1.58 1.49 all (2254) 1.67 1.66 1.59
Improved fit!
26