Cross-section systematics / inputs for VALOR DUNE ND analysis Costas Andreopoulos, Steve Dennis, Lorena Escudero, March 2016
Cross-section systematics in VALOR General approach for the 2nd pass-through implementation: 1. Cover all modelling aspects • For the 1st pass-through, only 6 simple normalizations cross-section systematics were used • For the 2nd pass-through ~40 systematics are included to cover neutrino-nucleon/ neutrino-nucleus cross-section, hadronization and intranuclear rescattering (next slides) 2. Use predominantly linear parameters • Some of them are not linear thought (e.g. pion mean free path systematic), and prior response functions are necessary (next slides) 3. Maintain relative independence from DUNE-specific tools (e.g weighting schemes) • Medium-term goal is to reuse this analysis for SBN by allowing multiple detectors in the fit, swapping the NuMI with the Booster flux covariance matrix and plugging-in oscillation probabilities. Aiming for all components of this analysis to be easily re-usable in SBN. Lorena Escudero, March 2016 2
Cross-section systematics in VALOR General approach for the 2nd pass-through implementation: Use predominantly model-independent parameters 4. • Model-independent parameters (e.g. `normalisation of CCQE events in a Q2 (or other kinematic variable range’ instead of MA, MV, Eb, kF,…) 1. Easier to explain and defend 2. Simplify treatment of different models choices within a single GENIE release 3. Simplify migration from one GENIE release/tune to another (where default models/parameters will change) • VALOR in sync with GENIE development, will use v3.0.0 tune and error analysis as soon as it becomes available • A covariance matrix with the input prefit errors will be computed (next slides) Lorena Escudero, March 2016 3
Cross-section systematics in VALOR 1. Cover all modelling aspects Studies to define binning ongoing • 6 correlated CCQE parameters, 3 kinematic bins each for Nu/NuBar. • 12 correlated CC1pi parameters 3 kinematic bins each for Nu/NuBar and charged/neutral pions. • 1 MEC normalisation systematic • 1 Other CC resonance (eg 1-gamma) systematic • 6 CC DIS (>1 pion) systematics, 3 kinematic bins each for Nu/NuBar • 1 NuE/NuMu normalisation systematic • 1 Nu/NuBar normalisation systematic Total: • 1 NC normalisation 37 parameters • 1 Coherent normalisation • 1 Pion and Nucleon mean free path systematic • 1 Pion charge exchange fraction systematic • 1 Pion absorption & multi-nucleon knockout systematic • 1 Pion inelastic fraction systematic • 1 Hadronization systematic for events containing Etas etc Not final but a comprehensive list for Even for the 2nd pass-through, still possible to the 2nd pass-through, even more tweak this list, should be finalised in the complexity can be added later upcoming weeks Lorena Escudero, March 2016 4
Cross-section systematics in VALOR 1. Cover all modelling aspects Why ~40 interaction modelling systematics? There are many modelling aspects to be constrained by the ND. • The VALOR DUNE/ND analysis is already substantially complex, currently constraining ~100 • flux + ~40 interaction modelling systematics. The chosen number of interaction modelling parameters (~40) is a trade-of between the • desire to start covering all modelling aspects and technical limitations (fit should be fast enough to allow numerous early studies / ~140-parameter fit using the standard 18 VALOR DUNE/ND samples takes several minutes) The goal during the second pass-through is to validate and understand the fit performance • Informed by the fit performance studies, the e ff ect of individual cross-section systematics • on CP sensitivity, and by parallel studies comparing new GENIE comprehensive model configurations with external data, new cross-section systematics parameters will be added (or, possibly, existing ones will be removed) early in the summer. At the same time we will manage the complexities arising from constructing the marginal • likelihood (integrating out detector systematics via toy-MC generation) at each minimisation step. Goal is to have a nearly-final version of the VALOR DUNE/ND fit by the Sept. collaboration • meeting Lorena Escudero, March 2016 5
Prefit cross-section errors in VALOR Considering 40 interaction modelling systematics, we plan to feed into the VALOR • DUNE/ND analysis a 40 x 40 interaction modelling error matrix We develop a procedure to construct this 40 x 40 error matrix using the to large extent • (but not exclusively) the standard GENIE reweighing tools A covariance matrix is created by tweaking GENIE systematic parameters of the • di ff erent interaction models. For each tweak, a weight of the number of the events in each model-independent parameters is calculated and the ensemble of those weights allows us to construct the covariance matrix (more details in next slides). The procedure itself is largely model-agnostic, so can be easily adapted to the new • GENIE tune (v3.0.0). This error matrix encapsulates a systematic error analysis on the default model . • Will be supported by detailed data/MC comparisons performed internally by GENIE. • Lorena Escudero, March 2016 6
Prefit cross-section errors in VALOR The error matrix can be extended to incorporate the result of other • independent studies, so it will serve as interface with other (non VALOR/ GENIE) groups interested in interaction modelling If you think there is an important source of systematic uncertainty • which is neglected in our studies, please provide a covariance matrix in the appropriate format and we will add it to the one we use. This systematic error analysis on the default model used in the VALOR • DUNE/ND fit is not critically important. The only requirement is that error assignments are amply conservative. • We will validate using GENIE data/MC comparisons. We will be presenting studies and constructed covariance matrix in following • meetings Lorena Escudero, March 2016 7
Prefit cross-section errors in VALOR Considering 40 interaction modelling systematics, we plan to feed into the VALOR DUNE/ND analysis a 40 x 40 matrix of interaction modelling pre-fit errors Translate into e ff ect in the Tweak internal model- correlated model-independent dependent parameters parameters in GENIE used in the VALOR/ND fit • 6 correlated CCQE parameters, 3 Using simple GENIE • MA reweight tools kinematic bins each for Nu/NuBar. • MV • 12 correlated CC1pi parameters 3 • EB kinematic bins each for Nu/NuBar and • kF charged/neutral pions. • etc • etc Construct covariance matrix by tweaking multiple internal parameters at the same time (using a normal distribution with their 1 σ error inside GENIE) , reweighting the number of events to obtain the e ff ect in terms of the model-independent parameters Lorena Escudero, March 2016 8
Prefit cross-section errors in VALOR FRACTIONAL COVARIANCE Quick test! • With simple MC events (25k ν + 25k anti- ν ) with 0-120 GeV • Just hundred tweaks of only MA CC QE and MA CC RES • In a preliminary binning in y_reco (smeared with a 10% resolution from true values) CORRELATION 0: ν CCQE y reco (bin1) 9: anti ν CC1piC y reco (bin1) 1: ν CCQE y reco (bin2) 10: anti ν CC1piC y reco (bin2) 2: ν CCQE y reco (bin3) 11: anti ν CC1piC y reco (bin3) 3: anti ν CCQE y reco (bin1) 12: ν CC1pi0 y reco (bin1) 4: anti ν CCQE y reco (bin2) 13: ν CC1pi0 y reco (bin2) 5: anti ν CCQE y reco (bin3) 14: ν CC1pi0 y reco (bin3) 6: ν CC1piC y reco (bin1) 15: anti ν CC1pi0 y reco (bin1) 7: ν CC1piC y reco (bin2) 16: anti ν CC1pi0 y reco (bin2) 8: ν CC1piC y reco (bin3) 17: anti ν CC1pi0 y reco (bin3) y reco binning different for each category and flavour Lorena Escudero, March 2016 9
Input response functions We will be using predominantly linear parameters (normalizations), but some parameters are not linear (e.g. pion mean free path systematic) For those non linear parameters we will construct a response function: • Precomputed functions storing averaged weights as a function of the tweak of the parameter • They allow to account for variations in the interaction models without re-generating the MC • One response function is to be calculated for each bin of Dummy the 3D(2D) templates used to build our prediction, i.e. for example each neutrino flavour, interaction mode, true and reconstructed variables bin They are computed as well using the reweighting functionality of GENIE Methods to use response functions are already in place in the VALOR DUNE ND fit, only necessary to compute them now Lorena Escudero, March 2016 10
Summary We will be using a ~40x40 covariance matrix with the pre-fit interaction modelling errors, which will be the input to our VALOR ND fit • Currently performing studies to define binning and construct it • It will be validated with GENIE data/MC comparisons • It will be redone using a more realistic approach, with more statistics, more throws of the systematic parameters, and reweighting multiple GENIE parameters at the same time This matrix will serve as interface with other groups interested in interaction modelling, as it can be easily extended to incorporate the e ff ect of other systematic uncertainties if necessary For non linear systematics, response functions are necessary • Machinery is ready to use response functions inside VALOR DUNE ND fit • Constructing them Lorena Escudero, March 2016 11
BACKUP
Recommend
More recommend