Comparison and evaluation of resolved resonance region covariance representations

The covariance matrices for the evaluated resolved resonance region parameters are typically processed into a grouped covariance representation for use in uncertainty propagation studies. An alternative method for performing uncertainty propagation would be to use the resonance parameters covariance matrix to sample many realizations of resonance parameters, and then perform a simulation with each sample, employing the total Monte Carlo method. In this paper, we demonstrate that these two techniques can yield drastically different estimations of the uncertainty on k_eff. Furthermore, we provide reasoning as to why the grouped covariance approximation is a nonphysical approximation for the resolved resonance region, and how sampling resonance parameters is able to capture variations in the structure of the point-wise cross sections. Our recommendation is that the community move away from the grouped covariance approximation and towards a continuous energy representation, which could be generated by sampling resonance parameters, and ensure that the resonance parameter covariance matrices in evaluated libraries yield realistic grouped cross section uncertainties, regardless of group structure. (author)