On random sampling of correlated resonance parameters with large uncertainties

Three different methods for multivariate random sampling of correlated resonance parameters are proposed: the diagonalization method, the Metropolis method, and the correlated sampling method. For small relative uncertainties (typical for s-wave resonances) and weak correlations all methods are equivalent. Differences arise under difficult conditions: large relative uncertainties of inherently positive parameters (typical for widths of higher-l-wave resonances) and/or strong correlations between a large number of parameters. The methods are tested on realistic examples; advantages and disadvantages of each method are pointed out. The correlated sampling method is the only method which produces consistent samples under any conditions. In the field of reactor physics, these methods are mostly used for the sampling of nuclear data, however, they may be used for any data with given uncertainties and correlations.