Repository for running SINBAD benchmarks with OpenMC and nuclear data uncertainty propagation. Nuclear data is propagated with Monte Carlo simulation, à la Total Monte Carlo (TMC).
Fusion Science and Technology article
Required:
Post processing:
Nuclear Data sampling:
- Download relevant endf and hdf5 data with openmc/data scripts (NNDC, TENDL and JEFF used)
- Set relevant environmental variables in environment.sh
- (If TENDL) Set names of TENDL endf files with random_data/rename_tendl_lib.py
- Sample benchmark data with scripts in random_data
- SANDY used for sampling
- Slurm submission scripts provided, appropriate changes required
- Default number of samples: 500
- Some covariance files missing for some nuclides, see random_data/README.MD
- Variance reduction cannot be used with these cross sections due to issue #1699
Benchmarks and Uncertainty Propagation:
- Set appropriate submission script for TMC_tools/openmc_sub.sh
- For Ni and FNG, build source with build_source.sh
- Various uncertain libraries may be propagated with provided scripts, ie oktav_fe/run_endf.sh
- One job is submitted per random evaluation, i.e if 500 random samples, 500 jobs
- Multiple random nuclides are ran simultaneously. Inter-nuclide dependence is assumed to be independent.
Sensitivity Analysis (todo):
- Not yet supplied, but quite simple.
- If you would like to know the sensitivity of eg Fe56, run the simulation will all other nuclides uncertain except this one
- Then measure the drop in uncertainty with either difference in variances (Saltelli), entropy (information based), area metric (pinching), ...
Post processing and plotting (todo):
- For tally arithmetic, we use ProbabilityBoundsAnalysis.jl
- Allows you to perform arithmetic considering both ND and statsitical uncertainties.
- Arithmetic with independence, dependent (with correlation), or unknown correlation.
- Plotting script available in TMC_tools
In TMC random evaluations of the nuclear data quantities are created either by using the endf covaraiance information (produces gaussian random evalutations) or by using a bayesian method with a nuclear reaction model code (TENDL library produced this way). The particle transport simulation is then repeated with each random evaluation. Each transport simulation has statistical uncertainty assosiated with it, which also must be considered. If the individual statistical uncertainty is negliable compared to the ND uncertainty, then standard probability theory may be applied. If this is not the case and both need to be considered, we use imprecise probabilities to perform further analysis. This is especially relevent in "Fast-TMC", where larger statistical uncertainty is exchanged for compute time.
