Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
GENIE models and global fits of neutrino scattering data
Marco Roda - [email protected] on behalf of GENIE collaboration
University of Liverpool
27 September 2017 NUFACT 2017 Uppsala
1/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Generators Neutrino MC Generators: A Theory/Experiment Interface
Access the flux distorsion due to oscillation Every observable is a convolution of flux, interaction physics and detector effects
Connect truth and observables Event topologies and kinematics Model dependencies
Good Generators ⇒ Support oscillation analyses uncertainty validation tune the physics models
⇒ Tuning proved to be difficult So far no results
Several MC Generators in use: GENIE, GiBUU , NuWro, NEUT
2/49 Global fits Generator is the ideal place for global fits We control the model implementation
Constraints on Cross Section for oscillation analysis Neutrino Cross sections priors Eventually based on data
What generators can do depends on the available datasets
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Role of generators Roles of MC generators in Oscillation Physics
Comparing data and models ⇒ You cannot study oscillations without fully understood models Validity region Highlight tensions
Feedback for experiments Drive the format of cross section releases Hint toward key measurements
3/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Role of generators Roles of MC generators in Oscillation Physics
Comparing data and models ⇒ You cannot study oscillations without fully understood models Validity region Highlight tensions
Feedback for experiments Drive the format of cross section releases Hint toward key measurements
Global fits Generator is the ideal place for global fits We control the model implementation
Constraints on Cross Section for oscillation analysis Neutrino Cross sections priors Eventually based on data
What generators can do depends on the available datasets
3/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Dataset Evolving datasets - Old datasets
2003-2017, GENIE - http://www.genie-mc.org
ANL_12FT,2 ANL_12FT,4 BEBC,0 BEBC,2
/ GeV] BEBC,5 BEBC,8 BNL_7FT,0 BNL_7FT,4
2 1 CCFR,2 CCFRR,0 CHARM,0 CHARM,4 FNAL_15FT,1 FNAL_15FT,2 Gargamelle,10 Gargamelle,12 cm IHEP_ITEP,0 IHEP_ITEP,2 IHEP_JINR,0 MINOS,0 SKAT,0 SciBooNE,0 -38 0.5 trunk:G00_00a:numu_freenuc
trunk:G16_02a:numu_freenuc CC [10 µ
ν 0 −1 2 10 1 10 10 Eν [GeV]
Functions of Eν Ignore nuclear effects Not flux-integrated Poor statistical interpretation “Only” statistical errors Poor model discrimination power
4/49 Sometimes incomplete Helped the development of new models 2p/2h
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Dataset Evolving datasets - Present datasets
Functions of experimental Cosθµ ∈[0.9; 1 ] observables flux-integrated
/GeV/n] Phys. Rev. D81, 092005 (2010) Usually differential cross-sections 2 1D, 2D cm -38 2
Organized by topology, not [10 µ
process T ∂ / µ
Higher statistics θ Cos
More statistically robust ∂ 1 )/ ⇒ Fermilab Neutrino seminar by π Mikael Kuusela - 2017/04/13 CC 0 µ ν ( σ 2 ∂ 0 0.5 1 1.5 2
miniboone_nuccqe_2010 Tµ [GeV]
5/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Dataset Evolving datasets - Present datasets
Functions of experimental Cosθµ ∈[0.9; 1 ] observables flux-integrated
/GeV/n] Phys. Rev. D81, 092005 (2010) Usually differential cross-sections 2 1D, 2D cm -38 2
Organized by topology, not [10 µ
process T ∂ / µ
Higher statistics θ Cos
More statistically robust ∂ 1 )/ ⇒ Fermilab Neutrino seminar by π Mikael Kuusela - 2017/04/13 CC 0 µ ν ( σ
Sometimes incomplete 2 ∂ 0 Helped the development of new 0.5 1 1.5 2 models 2p/2h miniboone_nuccqe_2010 Tµ [GeV]
5/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Dataset Future of datasets - a personal view
One big covariance matrix per experiment ? Correlation between datasets Differential cross sections, dim > 2
No data releases with this format in SBND we are thinking about a solution ? It is usually a big effort but ...
We finally have a way to use these datsets Statistically coherent Complete error analysis
6/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
GENIE - www.genie-mc.org
GENIE Collaboration Luis Alvarez Ruso8, Costas Andreopoulos2,5, Chris Barry2, Francis Bench2, Steve Dennis2, Steve Dytman3, Hugh Gallagher7, Tomasz Golan1,4, Robert Hatcher1, Libo Jiang3, Rhiannon Jones2, Anselmo Meregaglia6, Donna Naples3, Gabriel Perdue1, Marco Roda2, Jeremy Wolcott7, Julia Yarba1
[ Faculty, Postdocs, PhD students]
1 - Fermi National Accelerator Laboratory, 2 - University of Liverpool, 3 - University of Pittsburgh, 4 - University of Wroclaw, 5 - STFC Rutherford Appleton Laboratory, 6 - IPHC Strasbourg, 7 - Tufts University, 8 - Valencia University
Core GENIE mission 1 ... provide a state-of-the-art neutrino MC generator for the world experimental neutrino community
2 ... simulate all processes for all neutrino species and nuclear targets, from MeV to PeV energy scales
3 ... perform global fits to neutrino, charged-lepton and hadron scattering data and provide global neutrino interaction model tunes
7/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
GENIE status and prospects GENIE Version 2.12.6
RES CCQE models Rein-Sehgal Llewellyn Smith Berger-Sehgal Nieves, Amaro and Kuzmin-Lyubushkin- Valverde Naumov
MEC models COH Empirical Rein-Sehgal Nieves Simo Vacas Berger-Sehgal Alvarez Ruso Nuclear Models Relativistic Fermi Gas FSI - Intranuke Local Fermi Gas Single Kaon Full Intra-Nuclear Effective Spectral cascade Functions Λ production Schematic based on Hadron-nucleus data
Only one Comprehensive Model Configuration (CMC) Default tune has not changed
8/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
GENIE status and prospects GENIE Version 3
“Comprehensive Model Configurations” Self-consistent collections of primary process models Tune names are supposed to become commonly used Help cooperation between collaborations
single command-line flag --tune G16_02a
Complete characterization against public data Willing to host configuration provided by experiments
Tunes for each CMC will also be available
A step closer toward GENIE core mission
graphics by [email protected] 9/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
GENIE status and prospects Comprehensive Model Configurations
Dedicated web page
List of available configurations
10/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
GENIE status and prospects Comprehensive Model Configurations
Details and configuration
11/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Models Comprehensive Model Configurations
Configurations of interest for this talk G00_00a - Default No MEC CCQE process is Llewellyn Smith Model Dipole Axial Form Factor - Depending on MA = 0.99 GeV Nuclear model: Fermi Gas Model - Bodek, Ritchie
G16_01b - Default + MEC with Empirical MEC CCQE process is Llewellyn Smith Model Dipole Axial Form Factor - Depending on MA = 0.99 GeV Nuclear model: Fermi Gas Model - Bodek, Ritchie
G16_02a - Nieves, Simo, Vacas Model Theory motivated MEC CCQE process is Nieves Dipole Axial Form Factor - Depending on MA = 0.99 GeV Nuclear model: Local Fermi Gas Model
Small variations changing FSI models Variation including Spectral Functions
12/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Comparisons The Comparisons
The GENIE suite contains a package devoted to comparing GENIE predictions against publicly released datasets. Provides the opportunity to improve and develop GENIE models
Crucial database for new GENIE global fit to neutrino scattering data
All sorts of possible formats and dimensions
Can store correlations, even between different datasets
13/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Comparisons The database
Modern Neutrino Cross Section measurement nuclear targets typically flux-integrated differential cross-sections MiniBooNE, T2K, MINERvA
Historical Neutrino Cross Section Measurement Bubble chamber experiment
Measurements of neutrino-induced hadronic system characteristics Forward/backward hadronic multiplicity distributions Multiplicity correlations ... Measurements of hadron-nucleon and hadron-nucleus event characteristics (for FSI tuning) For pion, Kaons, nucleons and several nuclear targets Spanning hadron kinetic energies from few tens MeV to few GeV
Semi-inclusive electron scattering data electron-nucleus QE data electron-proton resonance data
14/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
CC 0π datasets MiniBooNE CCQE
2003-2017, GENIE - http://www.genie-mc.org Both ν and ν¯ 2 [GeV]
Phys. Rev. D81, 092005 (2010) µ T
Phys. Rev. D88, 032001 (2013) 2 1.5 Double differential cross section
flux integrated 1 1 No correlations Preferred model is Nieves Model 0.5 (G16_02a) 0 excellent agreement for ν −1 −0.5 0 0.5 1 2 χ = 101/137 DoF Cosθµ
2 -38 2 ∂ σ(νµ CC 0π)/∂ Cosθµ/∂ Tµ [10 cm /GeV/n] worse for ν¯ Data: miniboone_nuccqe_2010 χ2 = 176/78 DoF
15/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
CC 0π datasets MiniBooNE CCQE
Tµ ∈[0.4; 0.5 ] GeV
Both ν and ν¯ 2 /GeV/n] Phys. Rev. D81, 092005 (2010) 2 cm
Phys. Rev. D88, 032001 (2013) -38 1.5 [10 µ T ∂ /
Double differential cross section µ θ 1
flux integrated Cos ∂ )/ π 0.5
No correlations CC 0 µ ν ( σ 2
Preferred model is Nieves Model ∂ 0 (G16_02a) −1 −0.5 0 0.5 1 ν excellent agreement for Cosθ χ2 = 101/137 DoF µ miniboone_nuccqe_2010
worse for ν¯ G00_00a χ2 = 40.6/17 DoF 2 χ = / 2 176 78 DoF G16_01b χ = 64.8/17 DoF
G16_02a χ2 = 14.3/17 DoF
15/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
CC 0π datasets T2K ND280 0π
2003-2017, GENIE - http://www.genie-mc.org [GeV]
µ 6
Double differential cross section P 1 flux integrated Phys. Rev. D93 112012 (2016)
Fully correlated 4
0.5 Tensions between datasets Preferred model is G16_01b 2 χ2 = 135/67 DoF
0 0 all models look reasonable "By eye" −1 −0.5 0 0.5 1
estimation Cosθµ correlation is complicated 2 -38 2 ∂ σ/∂ Cosθµ/∂ Pµ [10 cm /GeV/n] We can’t ignore it! Data: t2k_nd280_numucc0pi_2015
16/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
CC 0π datasets T2K ND280 0π
Cosθµ ∈[0.9; 0.94 ]
1 /GeV/n] 2
Double differential cross section cm -38
flux integrated [10 µ P ∂ / µ Fully correlated θ 0.5 Cos ∂ / σ 2 Tensions between datasets ∂ Preferred model is G16_01b 2 χ = 135/67 DoF 0 0 1 2
all models look reasonable "By eye" Pµ [GeV] estimation t2k_nd280_numucc0pi_2015 correlation is complicated We can’t ignore it! G00_00a χ 2 = 17.2/8 DoF
G16_01b χ 2 = 7.75/8 DoF
G16_02a χ 2 = 25.8/8 DoF
16/49 Requirements granted by the comparisons: Data Metric
Minimizer ? Old problem in High Energy Physics CPU demanding Solution found in the Professor suite http://professor.hepforge.org Numerical assistant Developed for ATLAS experiment
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Tuning
Why tuning? Constraint parameters Provide specific tuning for experiments Liquid Argon tuning
Expected Output: Best parameters Parameter covariance matrix To be used for prior constructions
17/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Tuning
Why tuning? Constraint parameters Provide specific tuning for experiments Liquid Argon tuning
Expected Output: Best parameters Parameter covariance matrix To be used for prior constructions
Requirements granted by the comparisons: Data Metric
Minimizer ? Old problem in High Energy Physics CPU demanding Solution found in the Professor suite http://professor.hepforge.org Numerical assistant Developed for ATLAS experiment 17/49 1 Select points of param space 2 Evaluate bin’s behaviour with brute force 3 Parametrization I(p) Repeat for each bin
a parameterization Ij (p) for each bin N dimension polynomial Including all the correlation terms up to the order of the polynomial
Minimize according to~I(p)
∼ 15 parameters
Special thanks to H. Schulz
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Professor
Parametrization instead of a full MC
18/49 2 Evaluate bin’s behaviour with brute force 3 Parametrization I(p) Repeat for each bin
a parameterization Ij (p) for each bin N dimension polynomial Including all the correlation terms up to the order of the polynomial
Minimize according to~I(p)
∼ 15 parameters
Special thanks to H. Schulz
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Professor
Parametrization instead of a full MC 1 Select points of param space
p
18/49 3 Parametrization I(p) Repeat for each bin
a parameterization Ij (p) for each bin N dimension polynomial Including all the correlation terms up to the order of the polynomial
Minimize according to~I(p)
∼ 15 parameters
Special thanks to H. Schulz
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Professor
Parametrization instead of a full MC 1 Select points of param space 2 Evaluate bin’s behaviour with brute force
p
18/49 Repeat for each bin
a parameterization Ij (p) for each bin N dimension polynomial Including all the correlation terms up to the order of the polynomial
Minimize according to~I(p)
∼ 15 parameters
Special thanks to H. Schulz
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Professor
Parametrization instead of a full MC 1 Select points of param space 2 Evaluate bin’s behaviour with brute force 3 Parametrization I(p) interpolated value
p
18/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Professor
Parametrization instead of a full MC 1 Select points of param space 2 Evaluate bin’s behaviour with brute force 3 Parametrization I(p) Repeat for each bin interpolated value
a parameterization Ij (p) for each bin N dimension polynomial Including all the correlation terms up to the order of the polynomial
Minimize according to~I(p) p
∼ 15 parameters
Special thanks to H. Schulz
18/49 Advanced system Take into account correlations weights specific for each bin and/or dataset Proper treatment while handling multiple datasets Restrict the fit to particular subsets
Priors can be included Avoid unphysical result
Nuisance rescaling parameters can be inserted proper treatment for datasets without correlations (MiniBooNE)
Reliable minimization algorithm based on Minuit
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Advantages
Highly parallelizable independent from the minimization
All kind of parameters can be tuned Not only reweight-able
19/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Professor Advantages
Highly parallelizable independent from the minimization
All kind of parameters can be tuned Not only reweight-able
Advanced system Take into account correlations weights specific for each bin and/or dataset Proper treatment while handling multiple datasets Restrict the fit to particular subsets
Priors can be included Avoid unphysical result
Nuisance rescaling parameters can be inserted proper treatment for datasets without correlations (MiniBooNE)
Reliable minimization algorithm based on Minuit
19/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
The first tuning
Tuning against CC 0π datasets
20/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Inputs Datasets - 311 data points
MiniBooNE νµ CCQE 2D histogram 137 points Data No correlation matrix 2.2
2
MiniBooNE ν¯µ CCQE 1.8 2D histogram 1.6 1.4
78 points 1.2 No correlation matrix 1 0.8
T2K ND280 0π (2016) V2 0.6 2D histogram 0.4 0.2
80 points 0 full covariance matrix
MINERvA νµ CCQE 1D histogram Missing Covariance between 8 points Neutrino and antineutrino data full covariance matrix Minerva released this information! MINERvA ν¯µ CCQE 1D histogram 8 points full covariance matrix
21/49 Parameter Range Default value 2 QEL-MA (GeV/c ) [0.7 ; 1.8] 0.99 QEL-CC-XSecScale [0.8 ; 1.2] 1 RES-CC-XSecScale [0.5 ; 1.5] 1 FSI-PionMFP-Scale [0.6 ; 1.4] 1 FSI-PionAbs-Scale [0.4 ; 1.6] 1 MEC-FracCCQE - G16_01b only [0 ; 1] 0.45 MEC-CC-XSecScale - G16_02a only [0.7; 1.3] 1
Other inputs: Nuisance scaling parameters 30 % for MiniBooNE Dataset Priors on QEL-CC-XSecScale and RES-CC-XSecScale Gaussian with sigma 0.1
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Inputs Models and parameters
Default + Empirical MEC Full Nieves Model G16_01b in naming scheme G16_02a in naming scheme
22/49 Other inputs: Nuisance scaling parameters 30 % for MiniBooNE Dataset Priors on QEL-CC-XSecScale and RES-CC-XSecScale Gaussian with sigma 0.1
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Inputs Models and parameters
Default + Empirical MEC Full Nieves Model G16_01b in naming scheme G16_02a in naming scheme
Parameter Range Default value 2 QEL-MA (GeV/c ) [0.7 ; 1.8] 0.99 QEL-CC-XSecScale [0.8 ; 1.2] 1 RES-CC-XSecScale [0.5 ; 1.5] 1 FSI-PionMFP-Scale [0.6 ; 1.4] 1 FSI-PionAbs-Scale [0.4 ; 1.6] 1 MEC-FracCCQE - G16_01b only [0 ; 1] 0.45 MEC-CC-XSecScale - G16_02a only [0.7; 1.3] 1
22/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Inputs Models and parameters
Default + Empirical MEC Full Nieves Model G16_01b in naming scheme G16_02a in naming scheme
Parameter Range Default value 2 QEL-MA (GeV/c ) [0.7 ; 1.8] 0.99 QEL-CC-XSecScale [0.8 ; 1.2] 1 RES-CC-XSecScale [0.5 ; 1.5] 1 FSI-PionMFP-Scale [0.6 ; 1.4] 1 FSI-PionAbs-Scale [0.4 ; 1.6] 1 MEC-FracCCQE - G16_01b only [0 ; 1] 0.45 MEC-CC-XSecScale - G16_02a only [0.7; 1.3] 1
Other inputs: Nuisance scaling parameters 30 % for MiniBooNE Dataset Priors on QEL-CC-XSecScale and RES-CC-XSecScale Gaussian with sigma 0.1
22/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Outputs Sheer results
G16_01b - Default + MEC G16_02a - Full Nieves Model
Parameter Best fit Nominal Parameter Best fit Nominal 2 2 MA (GeV/c ) 1.17 ± 0.03 0.99 ± 0.01 MA (GeV/c ) 1.00 ± 0.03 0.99 ± 0.01 QEL-CC-XSecScale 0.92 ± 0.02 1 QEL-CC-XSecScale 0.91 ± 0.02 1 RES-CC-XSecScale 1.02 ± 0.07 1 RES-CC-XSecScale 1.01 ± 0.04 1 MEC-FracCCQE 0.55 ± 0.06 0.45 MEC-CC-XSecScale 1.18 ± 0.02 1 FSI-PionMFP-Scale 0.86 ± 0.04 1.0 ± 0.2 FSI-PionMFP-Scale 1.17 ± 0.04 1.0 ± 0.2 FSI-PionAbs-Scale 0.76 ± 0.09 1.0 ± 0.3 FSI-PionAbs-Scale 1.02 ± 0.09 1.0 ± 0.3
MA is reasonably low Nieve’s model is compatible with free nucleons fit Precision of MA reduced ⇒ Our choice not to add a strong prior
QEL reduced by ∼ 10% MEC increased by ∼ 20% FSI parameters strongly correlated They are better constrained than the GENIE prior
23/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Agreement Agreement with respect to datasets
G16_01b - Default + MEC G16_02a - Full Nieves Model
Dataset Best fit χ2 Nominal χ2 Dataset Best fit χ2 Nominal χ2 Miniboone νµ CC 0π 177 / 137 441 / 137 Miniboone νµ CC 0π 89.3 / 137 101 / 137 MiniBooNE ν¯µ CC 0π 66.2 / 78 50.4 / 78 MiniBooNE ν¯µ CC 0π 48.1 / 78 176 / 78 T2K ND 280 CC 0π 94 / 80 56.6 / 80 T2K ND 280 CC 0π 102 / 80 98.9 / 80 Total 337 / 289 548 / 295 Total 239 / 289 376 / 295
Improvement possible for both models ⇒ The fit is working
Fit driven by MiniBooNE datasets Lowest information ⇒ No correlations Room for improvement
These T2K and Minerva datasets cannot be fit on their own They cover a small phase space region ⇒ Parameters goes to the boundaries
24/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Best fit plots
Best fit - G16_01b - MiniBooNE νµ CCQE
θCosµ [0.8; 0.9 ]∈ Cosθµ ∈[0.9; 1 ] /GeV/n] /GeV/n] 2 2 cm cm
-38 2 -38 2 [10 [10 µ µ T ∂ / T ∂ / µ µ θ Cos ∂ )/ π CC 0 θ Cos
∂ 1
1 )/ π CC 0 µ µ ν ( σ ν ( σ 2 2 ∂ 0 ∂ 0 0.5 1 1.5 2 0.5 1 1.5 2
Tµ [GeV] Tµ [GeV] miniboone_nuccqe_2010 miniboone_nuccqe_2010
trunk:G16_01b:miniboone_fhc χ2 = 37.3/16 DoF trunk:G16_01b:miniboone_fhc χ2 = 95.8/18 DoF
trunk:Tuned:miniboone_fhc χ2 = 7.46/16 DoF trunk:Tuned:miniboone_fhc χ2 = 11.9/18 DoF
Fit has a big impact 25/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Best fit plots
Best fit - G16_01b - MiniBooNE ν¯µ CCQE
Cosθµ ∈[0.6; 0.7 ] Cosθµ ∈[0.7; 0.8 ]
0.4 0.6 /GeV/n] /GeV/n] 2 2 cm cm
-38 0.3 -38
[10 [10 0.4 µ µ T T ∂ ∂ / / µ µ
θ 0.2 θ Cos Cos ∂ ∂
)/ )/ 0.2 π π 0.1 CC 0 CC 0 µ µ ν ν ( ( σ σ 2 2 ∂ 0 ∂ 0 0.5 1 1.5 2 0.5 1 1.5 2
Tµ [GeV] Tµ [GeV] miniboone_nubarccqe_2013 miniboone_nubarccqe_2013
trunk:G16_01b:miniboone_rhc χ2 = 6.28/8 DoF trunk:G16_01b:miniboone_rhc χ2 = 9.62/9 DoF
trunk:Tuned:miniboone_rhc χ2 = 6.1/8 DoF trunk:Tuned:miniboone_rhc χ2 = 11.8/9 DoF
Improvement not really necessary in this case 26/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Best fit plots Best fit - G16_01b - MINERvA Neutrinos Antineutrinos
2003-2017, GENIE - http://www.genie-mc.org 2003-2017, GENIE - http://www.genie-mc.org 1.5 PRL 111, 022501 (2013) 1.5 PRL 111, 022502 (2013) /proton] 2 /neutron] 2 1 /GeV 2
/GeV 1 2 cm cm -38
-38 0.5 0.5 [10 [10 2 QE 2 QE /dQ σ /dQ d σ 0 0 d 0 0.5 1 1.5 2 0 0.5 1 1.5 2 Q2 [GeV2] Q2 [GeV2] MINERvAExDataCCQEQ2 QE MINERvAExDataCCQEQ2 QE
trunk:G16_01b:minerva_numu_2013 χ2 = 17.5/8 DoF trunk:G16_01b:numubar_2013 χ2 = 6.23/8 DoF
trunk:Tuned:minerva_numu_2013 χ2 = 10.9/8 DoF trunk:Tuned:numubar_2013 χ2 = 4.7/8 DoF
⇒ "Eye evaluation" wouldn’t prefer a model over the other
27/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Best fit plots Best fit - G16_01b - T2K ND280
Pµ ∈[0.8; 0.95 ] GeV
0.8 /GeV/n] 2 cm -38 0.6 [10 µ P ∂
agreement with T2K has worsened / µ θ 0.4
not surprising Cos ∂ / σ 2
⇒ Tensions already highlighted ∂ 0.2 χ2: 57 → 94 / 80 DoF 0 0.6 0.7 0.8 0.9 1
t2k_nd280_numucc0pi_2015_rps Cosθµ
trunk:G16_01b:t2k_nd280_numu_fhc χ2 = 6.9/8 DoF
trunk:Tuned:t2k_nd280_numu_fhc χ2 = 9.32/8 DoF
28/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Best fit plots Best fit - G16_02a
Cosθµ ∈[0.9; 1 ] /GeV/n] 2
cm 1 -38 [10 µ T ∂ /
Nieves’ model already works well µ θ Agreement is preserved Cos 0.5 ∂ )/ π Notable improvement only w.r.t. CC 0
MiniBooNE ν¯µ µ ν ( σ 2 ∂ 0 0.5 1 1.5 2
Tµ [GeV] miniboone_nubarccqe_2013
trunk:G16_02a:miniboone_rhc χ2 = 53.8/15 DoF
trunk:G16_02a_tuned:miniboone_rhc χ2 = 13.5/15 DoF
29/49 Data from Liquid argon experiments Part of GENIE collaboration is in SBND Plan for argon tunings
Look forward to more data
Release these results Paper is in preparation Implementation in GENIE v3
Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Tuning program Next steps
More tunings can be done hadronization retune Pythia 6 and 8 (implementation is ongoing) Tune of FSI Both hN and hA intranuke Free nucleon cross section model including dσ/dQ2 data
30/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Tuning program Next steps
More tunings can be done hadronization retune Pythia 6 and 8 (implementation is ongoing) Tune of FSI Both hN and hA intranuke Free nucleon cross section model including dσ/dQ2 data
Data from Liquid argon experiments Part of GENIE collaboration is in SBND Plan for argon tunings
Look forward to more data
Release these results Paper is in preparation Implementation in GENIE v3
30/49 Introduction GENIE Comparisons Tuning CC 0π tuning Conclusion
Conclusions Conclusion
We are renewing GENIE New models Systematic validation against Cross section data Maintained and rich database
We have a very powerful fitting machinery Validated A new branch of analyses Alternative tool to propagate systematic uncertainties
What we can do depends on data quality Look forward a promising collaboration between generators, experiments and theorists
31/49 Backup
Backup slides
32/49 Backup
Parametrization residuals
/global 350000
50000 300000
250000 40000
200000 30000
150000 20000 100000
10000 50000
0 0 0.4 0.2 0.0 0.2 0.4 0.5 0.0 0.5 Good Bad
33/49 Backup
Data covariance
Data Covariance
0.2
bin 300
250 0.15
200
0.1 150
100 0.05
50 0
0 0 50 100 150 200 250 300 bin
34/49 Backup
Model dependencies Model dependencies in Oscillation Analyses
A simple ratio between Near and Far spectra is not enough Detectors exposed to different flux “functionally identical” detectors do not exists
Near flux has to be fitted at the near detector and then propagated ⇒ Models required
35/49 Backup
Model dependencies Model dependencies in Oscillation Analyses (2)
2 2 2 CCQE is a 2-body reaction mp − (mn − Eb) − m` + 2(mn − Eb)E` Eν = Eν depends is just a function of 2(mn − Eb − E` + p`cosθ`) lepton momentum and angle
2p/2h is not a 2-body reaction low energy tails in reconstructed energy distributions
2p/2h also relevant for CP searches np-nh is different for ν/ν¯
⇒ 2p/2h modelling is important to achieve required precision Martini et al.
36/49 Backup
Model Comparisons Model comparison
37/49 Backup
Model Comparisons Model comparison
[M.Martini, FUNFACT J Lab workshop]
38/49 Backup
Importance of Covariance Importance of the covariance - an example
2003-2017, GENIE - http://www.genie-mc.org
Real dataset 1.5 8 points /neutron] 2 Which is the best agreeing curve?
/GeV 1 Black 2 Red cm
-38 Difference in terms of sigma? 0.5 [10 < 1 >
2 QE 1
2 /dQ Black χ = 17.5/8 DoF σ 0 d 0 0.5 1 1.5 2 Red χ2 = 10.9/8 DoF 2 2 ⇒ Almost 2 σ QQE [GeV ]
39/49 Not a full covariance matrix Neglecting the same flux Same detector/reconstruction
We can check agreement we can not fit these data without neglecting a correlation
Backup
Importance of Covariance Example of easy improvement
arXiv:1705.03791v1 Minerva experiment -39 - νµ C → µ p 16 ×10 Cross sections of CC 1-proton on 3.06e+20 Data POT Data 14 GENIE with FSI different targets GENIE No FSI 12 NuWro with FSI NuWro No FSI
C, Fe, Pb /nucleon) 2 10
/GeV 8 2 Wonderful dataset 6 ( cm 2 2p/2h and FSI tuning 4 /dQ C σ
d 2 0 Covariance matrices for each target 0 0.2 0.4 0.6 0.8 1 1.2 1.4 2 2 Qp(GeV )
Best format among present data releases -39 - νµ Pb → µ p 16 ×10 3.06e+20 Data POT Data 14 GENIE with FSI GENIE No FSI 12 NuWro with FSI
/nucleon) NuWro No FSI 2 10 /GeV
2 8
( cm 6 2 4 /dQ Pb
σ 2 d 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 2 2 Qp(GeV )
40/49 Backup
Importance of Covariance Example of easy improvement
arXiv:1705.03791v1 Minerva experiment -39 - νµ C → µ p 16 ×10 Cross sections of CC 1-proton on 3.06e+20 Data POT Data 14 GENIE with FSI different targets GENIE No FSI 12 NuWro with FSI NuWro No FSI
C, Fe, Pb /nucleon) 2 10
/GeV 8 2 Wonderful dataset 6 ( cm 2 2p/2h and FSI tuning 4 /dQ C σ
d 2 0 Covariance matrices for each target 0 0.2 0.4 0.6 0.8 1 1.2 1.4 2 2 Qp(GeV )
Best format among present data releases -39 - νµ Pb → µ p 16 ×10 3.06e+20 Data POT Data 14 GENIE with FSI Not a full covariance matrix GENIE No FSI 12 NuWro with FSI
/nucleon) NuWro No FSI Neglecting the same flux 2 10
Same detector/reconstruction /GeV
2 8
( cm 6 2 4 We can check agreement /dQ Pb
σ 2 we can not fit these data d 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 2 2 without neglecting a correlation Qp(GeV )
40/49 Backup
0π puzzle CC Quasi-Elastic - 0π on single nucleons
Theoretically well understood One diagram μ N’
A, B and C are form factors They have to be measured W+ B and C are known from e-N Ʋ scattering N A to be extracted from ν data
Axial Form factor Dipole standard g = 1.26 from neutron β decay parameterization A −2 fitted based on ∂σ/∂Q2 data 2 Q2 A(Q ) = gA 1 + 2 MA
41/49 Backup
0π puzzle CC Quasi-Elastic - Data
2003-2017, GENIE - http://www.genie-mc.org ] 2 2 cm -38 Hydrogen / Deuterium data 1.5 from 0.1 GeV to ∼ 100 GeV 1 For both Neutrinos and CCQE [10 0.5 Anti-neutrinos µ ν 0 Critical parameter: MA 10−1 1 10 M ∼ 1 GeV A ANL_12FT,1 ANL_12FT,3 Eν [GeV] BEBC,12 BNL_7FT,3 FNAL_15FT,3 Gargamelle,2 NOMAD,2 SERP_A1,0 SERP_A1,1 SKAT,8
42/49 Backup
0π puzzle CC Quasi-Elastic - Data ) 2 2.4 ν n→µ − p / GeV 2.2 µ 2
cm 2 − 38 1.8 Allasia et al., CERN WA25 (BEBC) 1990 (10 Target: D2 (converted to free neutron) >
Hydrogen / Deuterium data 2 1.6 Event number: 552 5
For both Neutrinos and < d σ / dQ 1.2 MA = 0.890 ± 0.044 GeV(best fit) Anti-neutrinos 1
0.8 Critical parameter: MA 0.6 MA ∼ 1 GeV 0.4
0.2
0 0 0.5 1 1.5 2 2.5 3 3.5 Q2 (GeV2)
42/49 Backup
0π puzzle 0π on heavy nuclei
MiniBooNE data
AIP Conf. Proc. 1189: 139-144 (2009); Phys. Rev. D 81, 092005 (2010)
On heavy nuclei things got complicated MiniBooNE ⇒ first evidence Carbon target
Possible explanation from enhanced MA ⇒ incompatibility with "historical" datasets
43/49 Backup
0π puzzle 0π on heavy nuclei - Solution
MoniBooNE is Cherenkov 2p-2h scheme detector Not able to see nucleons
miniBooNE dataset is a CCQE-like sample
genuine CCQE Multinucleon Emission np-nh Leading contribution is 2p-2h (2 particles - 2 holes)
44/49 Backup
0π puzzle 2 Particles - 2 Holes
M. Martini NN correlations MEC NN correlation-MEC interference
16 diagrams 49 diagrams 56 diagrams
Not easy to have a complete model Different approaches include different diagrams
45/49 Backup
Data from comparisons Hadronization example
46/49 Backup
Data from comparisons Hadronization example
47/49 Backup
Data from comparisons Hadronization example
> 4 0 π 2 FNAL 15’ Ne+H2ν BEBC Ne+H2ν BEBC H2ν p G00_00aν 1 n G00_00aν p G16_02aν n G16_02aν 0 1 10 102 103 W2(GeV2/c4) 48/49 Propagate the result to other observables Data Constraints for Oscillation analyses Propagate parameters uncertainty through the parameterization Muon Angle for 0π events 3 ×10 450 Events 400 350 300 Default 250 200 Tuned 150 100 50 0 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 Cos θ Backup Systematic propagation Tuning Output Parameters best fit Parameters covariance Prediction covariance due to the propagation of parameter covariance 49/49 Propagate the result to other observables Propagate parameters uncertainty through the parameterization Muon Angle for 0π events 3 ×10 450 Events 400 350 300 Default 250 200 Tuned 150 100 50 0 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 Cos θ Backup Systematic propagation Tuning Output Data Constraints for Oscillation Parameters best fit analyses Parameters covariance Prediction covariance due to the propagation of parameter covariance Muon Angle for 0π events 3 ×10 450 Events 400 350 300 Default 250 200 150 100 50 0 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 Cos θ 49/49 Propagate parameters uncertainty through the parameterization Backup Systematic propagation Tuning Output Data Constraints for Oscillation Parameters best fit analyses Parameters covariance Propagate the result to other observables Prediction covariance due to the propagation of parameter covariance Muon Angle for 0π events Muon Angle for 0π events 3 3 ×10 ×10 450 450 Events Events 400 400 350 350 300 Default 300 Default 250 250 Tuned 200 200 150 150 100 100 50 50 0 0 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 Cos θ Cos θ 49/49 Backup Systematic propagation Tuning Output Data Constraints for Oscillation Parameters best fit analyses Parameters covariance Propagate the result to other observables Prediction covariance due to the propagation of parameter Propagate parameters covariance uncertainty through the parameterization Muon Angle for 0π events Correlation 3 ×10 450 1 45 0.8 Events 400 bin 40 0.6 350 35 0.4 300 30 0.2 Default 250 25 0 200 Tuned 20 −0.2 150 15 −0.4 100 10 −0.6 50 5 −0.8 0 0 −1 0 5 10 15 20 25 30 35 40 45 −1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1 bin Cos θ 49/49