[Paper Review] Combining Total Monte Carlo and Benchmarks for nuclear data uncertainty propagation on an LFRs safety parameters

E. Alhassan, Henrik Sjöstrand+6 more|arXiv (Cornell University)|Mar 26, 2013

Nuclear reactor physics and engineering3 citations

TL;DR

This paper proposes a novel method to reduce nuclear data uncertainty in reactor safety parameters by integrating integral benchmarks into the Total Monte Carlo (TMC) framework. By applying a chi-squared-based accept/reject criterion using the Pu-239 Jezebel benchmark, the study reduces $k_{\text{eff}}$ uncertainty from 748 pcm to 443 pcm, demonstrating that benchmark-informed filtering significantly improves TMC accuracy and reproducibility in uncertainty propagation for lead-cooled reactor designs.

ABSTRACT

Analyses are carried out to assess the impact of nuclear data uncertainties on keff for the European Lead Cooled Training Reactor (ELECTRA) using the Total Monte Carlo method. A large number of Pu-239 random ENDF-formated libraries generated using the TALYS based system were processed into ACE format with NJOY99.336 code and used as input into the Serpent Monte Carlo neutron transport code to obtain distribution in keff. The keff distribution obtained was compared with the latest major nuclear data libraries - JEFF-3.1.2, ENDF/B-VII.1 and JENDL-4.0. A method is proposed for the selection of benchmarks for specific applications using the Total Monte Carlo approach. Finally, an accept/reject criterion was investigated based on chi square values obtained using the Pu-239 Jezebel criticality benchmark. It was observed that nuclear data uncertainties in keff were reduced considerably from 748 to 443 pcm by applying a more rigid acceptance criteria for accepting random files.

Motivation & Objective

• To reduce nuclear data uncertainty in criticality parameters for the European Lead Cooled Training Reactor (ELECTRA) using the Total Monte Carlo (TMC) method.
• To develop an automated, reproducible method for selecting relevant benchmarks for TMC applications, replacing subjective 'by-eye' selection.
• To investigate the impact of limiting $\chi^2$ values on nuclear data uncertainty in reactor parameters.
• To quantify the reduction in $k_{\text{eff}}$ uncertainty achievable through benchmark-based filtering of random nuclear data files.

Proposed method

• Generated 740 random Pu-239 ENDF-formatted nuclear data files using the TALYS-based TENDL system.
• Processed all random files into ACE format using NJOY99.336 at 1200 K for use in the Serpent Monte Carlo code.
• Performed criticality calculations on the full-core 3D ELECTRA model with 50,000 neutrons per cycle and 500 active cycles to obtain $k_{\text{eff}}$ distributions.
• Compared the TMC-derived $k_{\text{eff}}$ distribution with results from major libraries (JEFF-3.1.2, ENDF/B-VII.1, JENDL-4.0) to validate the method.
• Applied a chi-squared accept/reject criterion using the Pu-239 Jezebel benchmark to filter random files, with $\chi^2$ thresholds used to select only the most representative files.
• Used correlation analysis between $k_{\text{eff}}$ values in ELECTRA and the Jezebel benchmark to assess benchmark representativeness, with Pearson correlation coefficient R = 0.84 indicating strong agreement.

Experimental results

Research questions

• RQ1 Can integral benchmarks be systematically used to improve the accuracy and reduce uncertainty in TMC-based nuclear data uncertainty propagation?
• RQ2 How does applying a $\chi^2$-based accept/reject criterion on random nuclear data files affect the uncertainty in $k_{\text{eff}}$ for a lead-cooled reactor?
• RQ3 To what extent does the Pu-239 Jezebel benchmark represent the behavior of the ELECTRA reactor core in terms of $k_{\text{eff}}$?
• RQ4 Can a correlation-based method for benchmark selection be automated and generalized across different nuclides and reactor types?

Key findings

• The mean $k_{\text{eff}}$ from the TMC distribution was 1.001540 ± 0.00021 pcm, in good agreement with reference libraries such as JEFF-3.1.2 (1.00023 ± 0.00022 pcm).
• The nuclear data uncertainty in $k_{\text{eff}}$ was quantified as 748 ± 19 pcm using the standard TMC approach without filtering.
• By applying a limiting $\chi^2$ threshold of 5.31 × 10⁻⁶, the nuclear data uncertainty was reduced to 443 pcm, representing a 25% reduction in uncertainty.
• A strong correlation (Pearson R = 0.84) was observed between $k_{\text{eff}}$ values computed for ELECTRA and the Pu-239 Jezebel benchmark, indicating that the benchmark is a reliable representative for the reactor system.
• The study proposes a weighted benchmark selection method where the weight is proportional to the absolute value of the correlation coefficient, enabling automated and reproducible benchmark prioritization in TMC.
• The method demonstrates that filtering random nuclear data files based on integral benchmark performance significantly enhances the precision of uncertainty quantification in reactor safety parameters.