Theory - Introduction
⚠️ The term response (or resp) is used here to denote the integral response (experimental or calculated). ⚠️
When a response value is calculated, e.g., a keff, there is a calculation bias Δresp between the simulated model's response value and the experimental observed value. This bias is primarily composed of bias due to nuclear data ΔrespND and bias due to approximations in calculation schemes ΔrespSC. The methods described here allow estimating the calculation bias and its associated uncertainty.
The estimation method Δresp used is the GLLSM (Generalized Linear Least Squares Method). It assumes that ΔrespND dominates the total bias (valid e.g. for Monte Carlo). It also assumes that the response variation due to small nuclear data changes is linear.
The uncertainty σrespND due to nuclear data is computed using the propagation matrix product formula (i.e. "sandwich formula"), which propagates sensitivities (of the response to ND) through a variances-covariances of ND matrix.
Note: here "nuclear data" refers to microscopic cross-section data, as an example. Other data such as angular distributions could be included. All formulas are valid for data expressed in relative values.