QUANTIFICATION OF FERROUS/FERRIC RATIOS IN MINERALS: NEW EVALUATION SCHEMES OF FE L23 ELECTRON ENERGY-LOSS NEAR-EDGE SPECTRA

Abstract

Determination of Fe3+/ΣFe in minerals at submicrometre scale has been a long-standing objective in analytical mineralogy. Detailed analysis of energy-loss near-edge structures (ELNES) of the Fe L23 core-loss edges recorded in a transmission electron microscope (TEM) provides chemical information about the iron oxidation state. The valence-specific multiplet structures are used as valence fingerprints. Systematic investigations on the Fe L23 ELNES of mono and mixed-valence Fe-bearing natural minerals and synthetic solid solutions of garnets (almandine-skiagite and andradite-skiagite), pyroxenes (acmite-hedenbergite) and spinels (magnetite-hercynite) are presented where the presence of multiple valence states is distinguished by a splitting of the Fe L3 edge. We demonstrate the feasibility of quantification of the ferrous/ferric ratio in minerals by analyzing the Fe L23 ELNES as a function of the ferric iron concentration resulting in three independent methods: (1) The method of the modified integral intensity ratio of the Fe L23 white lines employs two 2-eV-wide integration windows centring around both the Fe L3 maximum for Fe3+ and the Fe L2 maximum for Fe2+. This refined routine, compared to the previously published quantification method of the ferrous/ferric ratio in minerals, leads to an improved universal curve with acceptable absolute errors of about ±0.03 to ±0.04 for Fe3+/ΣFe ratios. (2) The second method uses a simple mathematical description of the valence-dependent splitting of Fe L3 ELNES by fitting several Gaussian functions and an arctan function. The systematic analysis of the integral portions of the individual Gaussian curves for different mineral groups provides a further Fe3+/ΣFe quantification method with an absolute error of about ±0.02 to ±0.03. (3) The Fe L3 ELNES can also be modelled with the help of reference spectra, whereby the Fe3+/ΣFe ratio can be determined with an absolute error of ca. ±0.02.