New approach to improving the reliability of X-ray fluorescence analysis

It is shown that the probability of a mistake in the recognition of compositions close to each other in X-ray fluorescence analysis can be decreased by several tens of orders of magnitude if sets of signals from the X-ray spectrometer are selected to be specially correlated to each other. It is proved that the reliability of control can be sharply improved only with the use of sets of signals for which the longer axes of ellipsoids of equal probability densities are nearly parallel to each other and the ellipoids themselves exhibit a high eccentricity. Informativeness tests are analytically deduced for sets of two, three and n signals with different covariance matrices. The root-mean-square deviation of the measurement error is determined from the informativeness test calculated for the given set of closely correlated signals