Chapter 6.23. xafsX: a program to process, analyse and reduce X-ray absorption fine-structure spectra

All input and output data are column-oriented text files (ASCII code). xafsX contains a convenient, versatile data-import tool which facilitates user-defined column and header formats, including the importing and averaging of multiple data sets and the averaging of data from multiple fluorescence detectors. Available column data types are energy (eV or keV) or any abcissa, reference (I₀), transmission (I₁), fluorescence (I₁), single channel, multiple averaged fluorescence (I₁), calibration standard (I₂), absolute error, xafs [for example ln(I₀/I₁)] or any ordinate. Corresponding file formats are saved for further use. If original signals are imported, errors are estimated according to Lee et al. (1981). Additionally, simulated FEFF data (Ankudinov et al., 1998; Kas et al., 2024) can be imported either as phase and amplitude (from feff####.dat files) or EXAFS signal (from chi.dat files).

Data-preparation (preprocessing) tools contain deglitching, data rebinning and pre-edge analysis using a Victoreen (transmission data) or polynomial fit (fluorescence data). Data rebinning improves the signal-to-noise ratio at the cost of resolution.

The post-edge background subtraction is also performed empirically using either a spline function after subtraction of the edge (determined in the XANES fit) including white lines or alternatively (default) by a combination of an error function (for the absorption edge up to a chosen k value, the default value is 3 Å⁻¹) and a cubic spline. This improves the spline fit considerably as the steep absorption edge is already fitted and taken care of in the prior XANES fit.

Arithmetic operations on spectra are included to perform, for example, Δμ XANES analysis or extraction of partial signals such as atomic XAFS (Ramaker & Koningsberger, 2010). This tool uses mapping of a second data-set abscissa to a data set chosen first by cubic spline interpolation; it allows a constant shift in abscissa and adjustment of the amplitude of the second data set.

XANES data analysis in xafsX is based on empirical line-shape functions. Besides spectroscopic information about pre-edge and edge features, this analysis provides reliable and reproducible values for the step height and position of the absorption edge. This is important for subsequent EXAFS data-analysis steps. Line-shape functions implemented for the absorption edge are arctangent, error function, a general step function (linear combination of arctangent and error function) and those for peak features are Lorentzian, Gaussian or pseudo-Voigt functions, as well as a Breit–Wigner–Fano function (Bianconi, 1988).

In Fig. 2 the fit of the Co K-edge X-ray absorption spectrum of nanocrystalline ZnO doped with 20 atom% Co (Djenadic et al., 2010) is displayed with a corresponding fit of pre-edge peak, step and XANES features. The `deconvolution' by line-shape analysis provides unambiguous information about the position and height of the absorption edge as well as pre-edge and XANES features (Table 1).

EXAFS data analysis in xafsX can be performed using different empirical models: `standard' using the classical EXAFS formula (Sayers & Bunker, 1988), `general' (including the third and fourth cumulant), `analytical distribution' (Crozier et al., 1988), `numerical distribution' (histogram), cumulant analysis of phase and amplitude signals (Bunker, 1983; Dalba et al., 1993) and distribution reconstruction (splice method) (Crozier et al., 1988). EXAFS data analysis is performed in k-space. Multiple scattering is not considered.

The analysis of the Co K-edge EXAFS signal of nanocrystalline ZnO doped with 20 atom% Co (Djenadic et al., 2010) up to the second coordination shell (Table 2, Figs. 3 and 4) provides local structure information about the Co dopant which is identical within the estimated errors with analysis results obtained using ARTEMIS (Ravel & Newville, 2024) or RMCXAS (Djenadic et al., 2010) for the Co–O and Co–Zn contribution. In case of the (minor) Co–Co contribution to the second shell, this type of data analysis is at the limit since Co and Zn are difficult to distinguish as backscatterers. The large estimated errors in the Debye–Waller factor for the Co–Co contributions and the deviation in the Co–Zn coordination number are probably an artefact owing to the parameter correlation between N_Co–Co and N_Co–Zn used in the fit.

Figure 3 xafsX fit of EXAFS data of the Co K-edge EXAFS spectrum for nanocrystalline Co0.2Zn0.8O using xafsX after filtering the first and second shell in R space (0.5–3.4 Å, compare Fig. 4). The red points are filtered data, the blue line is the fit and the green line is the residual (for parameter values see Table 2).

Figure 4 Magnitude of the Fourier transform of the xafsX fit of the Co K-edge EXAFS spectrum (Fig. 3) for nanocrystalline Co0.2Zn0.8O after filtering the first and second shell in R space (0.5–3.4 Å, compare Fig. 3). The red points are original, unfiltered data, the blue line is the fit and the green line is the residual. Since filtered data were fitted, the residual and fit signals are flat above 3.4 Å.

Detailed analysis of the first shell using different methods (Table 3) shows that higher than second moments do not contribute significantly to the distribution of the Co–O coordination distances in this sample. This Gaussian distribution is consistent with the moment analysis of the partial pair distribution function obtained by RMC analysis of the same data (Djenadic et al., 2010). Deviations regarding the Co–O distance observed for the cumulant expansion and the general fit are likely to be due to probing different parts of the bond-distance distribution compared with other methods, since cumulant expansion and general fit include the third and fourth cumulants as fitting variables. This is likely to enlarge the flexibility of the fitting routine, although the higher cumulants remain insignificant. The rather small errors in the Co–O distance are a consequence of the single shell fits and the narrow Co–O distribution of Co substituting Zn in wurtzite-type ZnO.

Together with results at the Zn K edge, the local structure shows that Co replaces Zn substitutionally in the wurtzite lattice, creating a solid solution of Co in ZnO consistent with X-ray diffraction and RMC data analysis (Djenadic et al., 2010).

The postprocessing tools in xafsX are forward Fourier transformation including optical, phase-corrected transformation as well as backward Fourier transformation for filtering.