Improving precision of X-ray fluorescence analysis of lanthanide mixtures using partial least squares regression

This study addresses the problem of simultaneous quantitative analysis of six lanthanides (Ce, Pr, Nd, Sm, Eu, Gd) in mixed solutions by two different X-ray fluorescence techniques: energy-dispersive (EDX) and total reflection (TXRF). Concentration of each lanthanide was varied in the range 10(-6)-10(-3) mol/L, low values being around the detection limit of the method. This resulted in XRF spectra with very poor signal to noise ratio and overlapping bands in case of EDX, while only the latter problem was observed for TXRF. It was shown that ordinary least squares approach in numerical calibration fails to provide for reasonable precision in quantification of individual lanthanides. Partial least squares (PLS) regression was able to circumvent spectral inferiorities and yielded adequate calibration models for both techniques with RMSEP (root mean squared error of prediction) values around 10(-5) mol/L. It was demonstrated that comparatively simple and inexpensive EDX method is capable of ensuring the similar precision to more sophisticated TXRF, when the spectra are treated by PLS. (C) 2015 Elsevier B.V. All rights reserved.