Bayesian Monte Carlo Method for Nuclear Data Evaluation

Bayesian Monte Carlo applied to Ni isotopes Arjan Koning Head of Nuclear Data Section International Atomic Energy Agency, Vienna CSEWG Covariance session, Nuclear Data Week BNL, Brookhaven November 5-8 2018

Contents Introduction Goodness-of-fit and Bayesian Monte Carlo Some examples for Ni Towards TALYS-2.0 Conclusions 2

Goodness-of-fit estimators Utsunomiya et al, to be published PRC (2018)

F factors for TALYS vs EXFOR Allows to determine global predictive power Also available for each world library Used to clean up EXFOR (SG30) Used to create pseudo-experimental database to keep optimization and Bayesian inference under control

Ethr

Pseudo-experimental data: Global TALYS calculations + uncertainties Prior available, for all projectiles, target nuclides and reaction channels

Bayesian Monte Carlo Random TALYS samples k for parameters p and exp data x give likelihoods and weights All methods, UMC, BFMC, BMC etc are based on such weighting. The real trouble is in the definition of chi2: Not available in practice Too simple Four main problems which prohibit taking chi2 exactly 1. There are wrong data in EXFOR 2. Models can be really wrong 3. There is, in general, no covariance matrix or distinction between systematic and experimental uncertainties available 4. The inference process only depends on cross sections which happen to have been measured, potentially leading to unphysical results for unmeasured channels Issues 1-3 lead to extremely large chi2 values and make weights useless

BMC goodness-of-fit estimator until something better comes along 1. Wrong data 2. Model defect 3. Strong correlation assumption 4. Pseudo data A.J. Koning, Bayesian Monte Carlo method for nuclear data evaluation Eur. Phys. J.A51, 184 (2015)

Zooming in Prior: Global TALYS uncertainties from all EXFOR data Use weights based on EXFOR for 90Zr Final 10

Uncertainty Quantification everywhere, smoothness above 20 MeV

Example: capture (groupwise)

Example: capture (groupwise) To be done: Establish current status of resonance evaluations for all Ni isotopes

Detailed structure Challenge (also for TENDL): automate inclusion of partial and total inelastic cross section measurements (JRC Geel)

Adopt dosimetry evaluations in general purpose library

Cover all applications Gas production

2018-2019: TALYS-2.0 No new physics (for these two years only) Complete modernization: Fortran-90/95/03/08 Statistical variation and uncertainty quantification (Unpublished) TASMAN code integrated in TALYS-2.0 Uncertainties, covariances, sensitivity profiles, Bayesian Monte Carlo, Total Monte Carlo Automatic fitting (Simulated Annealing) of cross sections, and model parameters, to EXFOR (also included) Complete ENDF formatting (Unpublished) TEFAL code integrated in TALYS-2.0 Allows TALYS users to create ENDF data libraries New website, discussion forum etc.

2018-2019: TALYS-2.0 (This time) A real tutorial of about 1000 pages All physics of the Old TALYS manual Sample cases throughout the tutorial Guide to fitting experimental data from EXFOR or your own data Guide to making large databases for cross sections and astrophysical reaction rates All ENDF-6 formatting aspects Guide to TENDL production and Total Monte Carlo uncertainty propagation Acknowledge ALL TALYS use (i.e. by you) throughout the tutorial (very challenging) After 2019 (preliminary): Youtube video for installation, basic use, possibilities etc. Courses: Ready-to-use material for e.g. 4-hour, 16-hour and 40 hour course

Thank you!