Non-intrusive stochastic approach for nuclear cross-sections adjustment

Abstract

Bayesian adjustment of nuclear cross-section data within their prior uncertainties has been well-established in the neutronic community as the most mathematically-disciplined approach for improving the quality of neutronic calculations. The premise is that nuclear data are believed to contribute the most to the observed discrepancies between code calculations and measurements for key neutronic performance metrics such as critical eigenvalue, flux, power distribution, isotopic concentrations, etc. In its standard rendition, the adjustment procedure requires the solution of an optimization problem to calculate the optimum adjustments, with the optimality denoting the best possible agreement between measurements and predictions that constrains the adjustments to be statistically consistent with the prior uncertainties. To achieve that the optimization problem requires the derivatives of the responses of interest, e.g., keff, spectral indices, etc., with respect to all nuclear data. In the absence of an adjoint sensitivity capability, the calculation of these derivatives is computationally infeasible, thereby limiting applicability to adjoint-enabled codes only. This manuscript proposes a stochastic non-intrusive approach to preclude the need for derivatives, thereby allowing general neutronic codes with no adjoint capability to benefit from the nuclear data adjustment applications. Furthermore, the proposed approach serves as an independent tool for verifying the implementation of other adjustment capabilities, as it only requires non-intrusive access to the neutronic codes. This manuscript develops the theoretical basis for the new approach and compares its performance to an existing rendition, denoted by the generalized linear least-squares (GLLS) methodology, under the SCALE’s TSURFER module, for a criticality safety application.