T89 Spectroscopy for the Geosciences in the 21st Century

Monday, 10 October 2011
8:00 AM - 12:00 PM
Room 102D-F

MULTIVARIATE ANALYSIS OF XANES SPECTRA FOR MEASUREMENT OF FERRIC IRON IN GARNETS, AMPHIBOLES, MICAS, AND GLASSES

Paper number: T89-3
Program: Technical Sessions
Time: 8:30 AM – 8:45 AM

Authors

Abstract

Microfocused x-ray absorption near-edge spectroscopy (XANES) presents arguably the most promising technology for making Fe3+/ΣFe measurements at ca. 10 µm scales on standard thin sections. It utilizes the energy of the Fe Kα absorption edge and its pre-edge, which are known to be very sensitive to the oxidation state of Fe. Previous work has concentrated on the pre-edge region, and studies of isotropic glasses and garnets with restricted compositional ranges have yielded excellent results. However, broader application of micro-XANES has been hampered by differential interactions of the plane-polarized x-ray beam with anisotropic crystals that result in ±15-20% errors for Fe3+/ΣFe measurements based on pre-edge spectra of randomly-oriented crystals. Moreover, different mineral groups require separate calibration lines because the geometries of their Fe polyhedra are so variable. Finally, use of solely the pre-edge region for modeling Fe3+ contents ignores valuable information contained in the main absorption edge. We demonstrate here that these problems can be overcome by use of oriented standards and multivariate analysis.

For anisotropic mineral groups, we have acquired spectra with the crystals oriented with the polarization direction of the synchrotron beam is parallel to the X, Y, and Z optical orientation directions in the crystals, providing end-members of spectra for samples with extremes of Fe3+/ΣFe and orientation. We extracted the pre-edge region of the XANES spectra to fit component peaks representing contributions from both Fe2+ and Fe3+ in this region, and created a calibration curve that regresses the area-normalized centroid of those pre-edges to known Fe3+/ΣFe from bulk methods. These results were compared with those from multivariate analysis models such as partial least-squares analysis (PLS), which predict Fe3+/ΣFe using spectra from the entire XANES region. Results shows that multivariate analysis yields equivalent or more accurate predictions of Fe3+ in garnets, amphiboles, micas, and glasses to techniques based on the pre-edge region only, with improved robustness and generalizability.