Supplemental information for “Analysis of Fe K-edge XAFS spectra of phyllosilicates of varying crystallinity”
Leslie L. Baker and Daniel G. Strawn, Department of Soil and Land Resources, University of Idaho, Moscow, ID 83844-2339;
Purification of Garfield nontronite sample
The nontronite was purified by size fractionation, as follows. Granules of clay were hand-picked and hand-crushed in an agate mortar. This material was Na-saturated, washed in DI water, and centrifuged to separate the <1-µm size fraction. The very fine clay fraction was further refined by resuspending and centrifuging the <1-µm size fraction at 10,800 x g for 1 hour at 5°C. The supernatant contained a very fine clay gel with pure nontronite particles approximately less than 0.05 µm in diameter. This supernatant was decanted and freeze-dried, leaving flakes of pale green material. X-ray diffractometry (XRD) and Fourier-transform infrared spectroscopy (FTIR) confirmed that the clay was pure nontronite.
Detailed fitting methods
Initially, the Garfield nontronite spectrum was modeled out to 6 Å from the core Fe atom using 92 single and multiple scattering EXAFS paths that had contributions to the EXAFS spectra > 6%. Nontronite crystal structure data for the Feff input file were taken from Manceau et al. (1998). Amplitude reduction factor was set to 0.85 (O'Day et al. 1994), which was verified by fitting 6 O backscatterers to the first peak in the windowed Fourier transform of the experimental Garfield nontronite EXAFS spectrum. Multiple scattering was evaluated for up to 4 successive legs. Debye-Waller parameters for the Fe-O1, Fe-Fe1, and Fe-Si1 (Figure 1a) were initially optimized, and found to be near 0.005 Å2, which they were fixed at for the simulation. Debye-Waller factors for all other single and multiple scattering shells were constrained to be equal and fit; although this does not represent a structural reality, in fitting it is reasonable to include such a constraint to minimize degrees of freedom. This approach produced a global Debye-Waller value of 0.0089 Å2 for the single- and multiple-scattering shells at distances greater than 3.4 Å, which is in agreement with values reported in other fitting papers (Manceau et al. 1998; O'Day et al. 1994; Vantelon et al. 2003). The theoretical nontronite EXAFS structure (i.e., no fitting) was in close agreement with experimental, as also observed by Manceau et al. (1998). However, there was a slight mismatch of the Fe-Fe1 shell distance features in the EXAFS spectrum. Thus, fitting of the Fe-O1, Fe-Fe1, and Fe-Si1 single-scattering shell bond distances was done to refine the 92-shell optimization. Except for Fe-Fe1, the fit single-scattering path distances were in very good agreement with the theoretical distances (additional details are provided in results). Analysis of multiple scattering below 3.4 Å in the Fourier transform showed that contributions are small, which is in agreement with the findings of Manceau et al. (1998) and Vantelon et al. (2003).
Unlike the nontronite, the simulated 88-shell full-scattering spectrum of Fe-substituted montmorillonite did not agree well with the experimental EXAFS data. However, fitting of the single-scattering bond distances in the 88-shell model resulted in a model that was in good agreement with the experimental EXAFS spectrum, and was in good general agreement with Fe positions in the montmorillonite structure as proposed by Vantelon et al. (2003). A 6-shell model with only single scattering contributions was used to fit the montmorillonite EXAFS spectral features out to ~3.4 Å in the Fourier transform magnitude data. The montmorillonite 6-shell model paths were analogous to the 5-shell model used for Garfield nontronite, plus an octahedral Fe-Al1 backscattering path (Table 1).
SBId-1 beidellite was fit using the 6-shell montmorillonite model. Coordination number and Debye-Waller factor were fixed in the fitting of the clay spectra to values similar to those of the SWy-2 montmorillonite full shell fit. In a few instances Debye-Waller factors were optimized (Table 2) after the initial coordination number fit to account for a small amount of amplitude fit residual. However, because coordination number and Debye-Waller factor are directly correlated, simultaneously varying both parameters makes interpretation more uncertain. Fe-Fe1 and Fe-Al1 coordination numbers were constrained to sum to three. Path lengths were optimized for all shells. However, to stabilize the fit optimization, in some cases the Fe-O2 and Fe-O3 path lengths were constrained after an initial fit.
Results
Simplifying the full nontronite fitting model to a 5-shell or 6-shell model
Fitting phyllosilicate EXAFS data beyond 3-4 Å in R-space is difficult without additional information, such as polarized EXAFS, because of the closeness of Si and O scattering atoms in the structure (Manceau et al. 1998). This range includes second shell Fe-Me1 (where Me1 is an Fe, Al, or Mg atom) and Fe-Si1 shells. Windowing this range in exclusion of the Fe-O2 and O3 shells (Table 1), which occur at 3.4-3.8 Å in the structure, is difficult because of the overlap of the peaks in the FT (Figure 3b). Vantelon et al. (2003) showed that fitting of Fe EXAFS data on smectites in the range 1-3.4 Å in FT is reliable for interpreting local molecular structure. Manceau et al. (1998) showed that MS in the nontronite model has a small contribution. Vantelon et al. (2003) also showed that for a C2 smectite model (cis-vacant) theoretical MS paths have minimal contributions, and thus can be ignored. Our modeling confirmed that contributions by MS to the EXAFS are relatively small, and benefits of including them, particularly on poorly crystalline structures such as allophane, are negated by over parameterization of the fit. Thus, based on the minimal contributions of MS and the limitations of fitting, using a single-scattering model to fit Fe-substituted aluminosilicate EXAFS data is a feasible approach.
Ignoring the MS shells and analyzing only the region of the FT between 1 and 3.5 Å (the region of the two major peaks) reduces the number of possible shells to 9 (see Table 1 and Figure 1 for theoretical paths). However, the first three O-scattering paths (Fe-O1) are similar, as are the last three O-scattering paths (Fe-O3) that occur between 3.74-3.82 Å in the structure. Thus the number of theoretical paths needed for fitting the two main peaks in the FT of nontronite reduces to 5. If Al were present in the octahedral sites, then the number of paths is 6. Figure 4 and Table 2 show the best fit results of the Garfield nontronite EXAFS spectrum using the 5-shell model. The 5-shell model fit the data very well, and resulting fit distances (Table 2) are the same as the 92-shell model.
Montmorillonite fitting using the full 88-shell model
Unlike the simulation of the nontronite model, the Feff multi-shell simulation of Fe-substituted montmorillonite using all shells > 6% (88 shells) is dissimilar to the EXAFS spectrum of the SWy-2 montmorillonite sample. The discrepancy in the Feff model of montmorillonite may arise because the unit cell structure used to generate the atomic coordinates for the Feff input model does not account for short-range structural distortion caused by isomorphic substitution. Since Fe is a larger atom than Al (0.785 Å vs. 0.675 Å), an Fe-centered octahedral site should be larger than an Al-centered site, and thus have longer interatomic bond distances. To account for expansion of the structure to accommodate the Fe, fitting of the multi-shell Feff model was done.
Allophane nanoball model
In the proposed proto-imogolite structure, the nanomineral is constructed of a rolled octahedral sheet with unpolymerized Si coordinated to the interior of the sphere. Each Si atom is coordinated to three Al atoms via Al-O-Si bonds, and the fourth O atom (hydroxyl) (Figure 1b) points towards the center of the sphere. Many of the backscattering shells around the octahedral atom in this model are similar to those in a clay mineral; i.e., the octahedral cation is surrounded by six oxygen atoms in ordinary octahedral configuration, and three neighboring octahedral cations are present at distances comparable to those in montmorillonite (Table 2). The most significant difference in the theoretical allophane structure is the distance of the Si atom; three Si atoms are present at a modeled reffective distance of 3.14-3.16 Å, rather than the four Si atoms at 3.18 to 3.24 Å found in 2:1 phyllosilicates (Figure 1, Table 1). In addition, because there is no Si polymerization in the proto-imogolite model, no tetrahedral O2 atoms are present. Thus the second backscattering O shell in this structure consists of six octahedral O atoms corresponding to the O3 atoms in the nontronite structure (also labeled O3 in Figure 1b), and three backscattering Oi atoms at 3.98 Å from the central Fe atom (the hydroxyl on the Si-OH apices pointing toward the interior of the nanosphere). Additional differences between the models occur at greater distances.
Wavelet analysis
Wavelet analysis can be used to extract additional information from XAFS spectra that may be of use in interpreting the data. In particular, it is useful for discriminating between different backscattering atoms at similar distances (Funke et al. 2005). Here, we used wavelet analysis to help discriminate the contributions of close Fe and Al backscattering shells in the hisingerite and allophane spectra.
Morlet wavelet analysis was carried out on spectra using the FORTRAN program HAMA (Funke et al. 2005) using a kappa value of 6 (to examine the Fe-Fe1 backscattering peak near 3 Å) and a sigma value of 1. To enhance the signal from more distant backscatterers, the analysis was carried out on k3-weighted chi spectra. To examine the effect of varying Fe content on an otherwise unchanging structure, several synthetic spectra were generated using the Feff 92-shell model for the Garfield nontronite structure. Parameters for generating these model spectra were taken from Manceau et al. (1998) (amp = 0.85; DW = 0.0047 for Fe-O1 and Fe-Me1, 0.0055 for Fe-Si1, 0.01 for all additional SS shells, and 0.121 for MS shells). The model was weighted to contain different numbers of Fe atoms (out of a possible total of 3) with the remainder being Al, assuming a random distribution of these atoms in the octahedral sheet. The spectra modeled using this approach were saved and were processed using wavelet analysis as were the spectra of the natural samples. The Garfield nontronite sample wavelet graph (not shown) was very similar to the model with 3 Fe, as expected.
The wavelet graphs for the allophane and hisingerite samples can be compared to the nontronite model at different Fe contents to assess the Fe coordination number fit to the spectrum. We use the nontronite model here rather than the montmorillonite because, as discussed in the text, the spectrum of Garfield nontronite can be reproduced almost perfectly by the model without fitting, whereas the montmorillonite model required parameter fits. This therefore makes the nontronite model a better choice for generation of entirely theoretical spectra. The wavelet graph for the beidellite SBId-1 is also shown.
Shown here (Figure S-2) are the wavelet graphs for model spectra containing 1, 1.5, 2, and 3 Fe atoms. These graphs illustrate the effect of changing Fe-Al ratio. Each graph is scaled to its maximum value for the Fe-Me1 peak, so different graphs are scaled differently, but the shape and positions of the Fe-Me1 peaks can be compared. These peaks are observed at R = 2.5 - 3.5 Å and k = 6 - 10 Å-1, with maxima generally near 8.5 Å-1. It can be observed that the shape of these maxima varies with Fe-Al ratio, extending to lower k for higher-Fe samples. The beidellite graph also illustrates this behavior, with the maximum only at higher k for this Al-rich sample.
The wavelet graph of hisingerite also follows this trend, with a very strong shift of the maximum to lower k values as would be expected for this almost Al-free sample. Some differences are observable between the hisingerite graph and that of the 3-Fe nontronite model, particularly at R values >4, supporting the argument that hisingerite is not in fact a poorly crystalline nontronite.
In the allophane wavelet graph, the maximum corresponding to the Fe-Fe1 and Fe-Al1 peaks in the FT most closely resembles the wavelet graph of the model nontronite containing 1.5 Fe atoms, in terms of both shape of the feature and intensity distribution. This observation is supportive of the shell fitting model result suggesting a CN of Fe = 1.85. A 20% error applied to this CN would correspond to a range in CN of approximately 1.5 – 2.25 Fe. Thus the lower bound is consistent with the wavelet graph, based on the wavelet graph, the upper bound appears to be too high.
References cited
Funke H, Scheinost AC, Chukalina M (2005) Wavelet analysis of extended X-ray absorption fine structure data. Physical Review B 71:94-110
Manceau A, Chateigner D, Gates WP (1998) Polarized EXAFS, distance-valence least-squares modeling (DVLS), and quantitative texture analysis approaches to the structural refinement of Garfield nontronite. Physics and Chemistry of Minerals 25 (5):347-365. doi:10.1007/s002690050125
O'Day PA, Rehr JJ, Zabinsky SI, Brown GE, Jr. (1994) Extended X-ray absorption fine structure (EXAFS) analysis of disorder and multiple-scattering in complex crystalline solids. Journal of the American Chemical Society 116 (7):2938-2949. doi:10.1021/ja00086a026
Vantelon D, Montarges-Pelletier E, Michot LJ, Pelletier M, Thomas F, Briois V (2003) Iron distribution in the octahedral sheet of dioctahedral smectites. An Fe K-edge X-ray absorption spectroscopy study. Physics and Chemistry of Minerals 30 (1):44-53