Total-reflection X-ray fluorescence determination of thorium and uranium in the presence of interfering elements in solid geological objects of natural and technogenic origin

Abstract

The insufficient spectral resolution of the energy-dispersive total-reflection X-ray fluorescence spectrometer (TXRF) makes the quantitative determination of U and Th with the use of Lα lines a challenging task, especially in the case of analysis of natural solid samples with low concentrations of these elements. In this study, we used the PLS regression and the spectrum deconvolution to overcome the peak overlap problem (Th Lα overlaps with Fe Kα + Fe Kα and Rb Kα; U Lα overlaps with Rb Kα and Sr Kα). Samples of rocks, ashes, and ores with different amounts of interfering elements (Rb, Sr, and Fe) were prepared as suspensions for the direct TXRF determination of U and Th. We applied the PLS regression coupled to the special design of experiment (DoE) based on the low-correlated concentrations of U, Th, and Rb in the calibration set of samples. To validate our combination of the DoE, preprocessing procedure and PLS models, we used the independent test sets of the Rb-rich and Rb-depleted samples. Our study revealed that in the presence of considerable line overlapping the deconvolution leads to a significant systematic error (30–60%) within the low concentration range (10–30 ppm), while the PLS regression reduces the systematic error by 2–3 times. Additionally, we demonstrated that the borderline of the semi-quantitative and quantitative determination for the PLS regression is 3 and 10 times lower, respectively. We found that the high RSD of the Th concentrations found by the deconvolution reflects not a poor repeatability of the TXRF measurements, but a poor stability of the deconvolution. The application of the Rb-depleted test set helped us to demonstrate that the PLS regression does not improve the accuracy in the absence of strong line overlapping. In this study, we conclude that the best strategy for the determination of U in the Rb-depleted samples is the implementation of the deconvolution, while the determination of Th and U in the Rb-rich samples is best carried out with the PLS regression.