EXAMINATION OF VARIOUS ROLES FOR COVARIANCE MATRICES IN THE DEVELOPMENT, EVALUATION, AND APPLICATION OF NUCLEAR DATA
Argonne National Laboratory
9700 South Cass Avenue
Argonne, Illinois 60439, USA
The last decade has been a period of rapid development in the implementation of covariance-matrix methodology in nuclear data research. This paper offers some perspective on the progress which has been made, on some of the unresolved problems, and on the potential yet to be realized. These discussions address a variety of issues related to the development of nuclear data, the evaluation of nuclear data, and the applications for nuclear data. Topics examined are: the importance of designing and conducting experiments so that error information is conveniently generated; the procedures for identifying error sources and quantifying their magnitudes and correlations; the combination of errors; the importance of consistent and well-characterized measurement standards; the role of covariances in data parameterization (fitting); the estimation of covariances for values calculated from mathematical models; the identification of abnormalities in covariance matrices and the analysis of their consequences; the problems encountered in representing covariance information in evaluated files; the role of covariances in the weighting of diverse data sets; the comparison of various evaluation procedures involving covariance matrices; the role of covariances in updating existing evaluations; the influence of primary-data covariances in the analysis of covariances for derived quantities (sensitivity); and the role of covariances in the merging of diverse nuclear data information.
KEYWORDS: Statistics, covariance matrices, error propagation, sensitivity, fitting, evaluation