Shan
Li
abcd,
Ke
Wen
abc,
Yiping
Yang
abc,
Xiaoju
Lin
abc,
Yonghua
Cao
a,
Yao
Xiao
abcd,
Haiyang
Xian
*abcd,
Jianxi
Zhu
abcd and
Hongping
He
abcd
aState Key Laboratory of Deep Earth Processes and Resources, Guangzhou Institute of Geochemistry, Chinese Academy of Sciences, Guangzhou 510640, P. R. China. E-mail: xianhaiyang@gig.ac.cn
bGuangdong Provincial Key Laboratory of Mineral Physics and Materials, Guangzhou Institute of Geochemistry, Chinese Academy of Sciences, Guangzhou 510640, P. R. China
cGuangdong Research Center for Strategic Metals and Green Utilization, Guangzhou 510640, P. R. China
dUniversity of Chinese Academy of Sciences, Beijing 100049, P. R. China
First published on 10th June 2025
The oxidation state of iron (Fe) in minerals is crucial for Earth and planetary sciences. Accurate determination of the oxidation state of Fe aids in the estimation of redox conditions during mineral formation and thus the reconstruction of the evolutionary history of Earth and other solid celestial bodies. Compared to conventional methods that provide the average bulk Fe3+/ΣFe ratio in minerals, the recently developed scanning transmission electron microscopy electron energy loss spectroscopy (STEM-EELS) technique enables the accurate determination of Fe oxidation states at sub-micrometer scales. However, the focused electron beam in STEM mode can induce the generation of Fe3+ and consequently introduces uncertainty in the quantification of Fe3+ in minerals. In this study, a calibration relationship was established between the relative ratios of the Fe-L3 and Fe-L2 peak areas in the EELS spectra and the Fe3+/ΣFe ratio based on the characterization of a series of pyroxene reference materials with varying Fe3+ contents. The EELS spectra were collected in the STEM-EELS-mapping mode using focused ion beam (FIB)-derived sections. The impacts of sample preparation, data acquisition, and data processing procedures on the determination of the Fe3+ contents in minerals were examined and discussed to gain insights into the constraints of parameters for accurate quantification of Fe3+ in minerals. The obtained calibration relationship was further validated for the accurate determination of Fe3+ content in other Fe-bearing minerals, suggesting its promising applicability in the quantification of Fe3+ in precious nanoscale terrestrial and extraterrestrial materials.
The conventional methods for the quantification of Fe3+ in minerals primarily include wet chemical analysis, Mössbauer spectroscopy, and electron probe microanalysis (EPMA).12–14 These techniques are limited by their large sample consumption and low spatial resolution, making them unsuitable for in situ characterization of small or microsamples. X-ray photoelectron spectroscopy (XPS) and X-ray absorption spectroscopy (XAS) have enabled the measurement of Fe3+ content at the micrometre scale in individual mineral grains.15–17 However, these micro-scale (with beam spot sizes generally ranging from tens to hundreds of micrometres) in situ examination methods still fall short of meeting the demands at the nanoscale.18
Electron energy loss spectroscopy (EELS) coupled with transmission electron microscopy (TEM) allows for the quantification of Fe3+ in tiny samples with exceptionally high spatial resolution.19–25 The Fe-L2,3 electron energy loss near-edge structure provides chemical information about the specific multiple structures of iron valence states, which can be used as valence state fingerprints. The L2,3 edges of Fe exhibit sharp energy loss peaks that appear as two white lines on the recording camera and are thus referred to as the “white line” peaks. The Fe-L2,3 white lines correspond to the transitions from 2p to unoccupied 3d orbitals,26 and the distinct features of Fe-L3 and Fe-L2 edges are due to the spin–orbit splitting of the Fe-2p hole, with an energy separation of approximately 13 eV. The EELS spectra of Fe2+ and Fe3+ in minerals demonstrate distinct Fe-L2,3 edge features. Therefore, the accurate quantification of Fe3+ in minerals can be achieved using TEM-based Fe-L2,3 EELS.22
The quantitative determination of Fe3+ content in minerals using EELS was proposed by van Aken et al. in 1998,21 and a modified method with a universal curve was set up in 2002.22 The universal curve was derived by synthesizing a series of minerals with varying Fe3+ contents and further analysing the correlation between the EELS-Fe-L2,3 edge “white line” ratios and the Fe3+/ΣFe ratios measured using Mössbauer spectroscopy. The measurements were operated with the electron beam in a parallel beam condition (i.e., TEM mode), in which only the EELS spectrum is obtained without simultaneously capturing an image of the examined area, thus making it impossible to determine the exact location of the analyzed region.
For conventional materials requiring solely EELS spectral acquisition, particularly homogeneous single-mineral specimens, parallel-beam illumination with single-spectrum acquisition may suffice. However, with recent advancements in analytical techniques, geological research has progressively focused on nanoscale investigations, particularly for ultra-rare samples such as those from deep mantle environments and extraterrestrial sources. These precious specimens necessitate in situ characterization at sub-micron scales to resolve complex mineralogical compositions and structural heterogeneities. Consequently, conventional TEM specimen preparation protocols and parallel-beam spectral mapping approaches fall short in addressing these advanced analytical requirements.
In contrast to the TEM mode, the scanning transmission electron microscopy (STEM) mode allows for the collection of EELS spectra along with imaging information of the test location, enabling precise positioning of the analyzed area. The STEM mode in transmission electron microscopy, utilizing a focused electron beam, offers distinct advantages over conventional TEM mode with parallel illumination. Specifically, STEM enables simultaneous acquisition of both imaging data and EELS spectra, allowing precise localization of the analyzed area. This integrated capability contrasts with TEM-EELS, which only provides spectral information without spatial context for the measurement position. Spectral mapping enables direct visualization of redox heterogeneity, which is essential for deciphering paragenetic sequences.
However, the highly focused electron beam in STEM mode (with a nanometre-scale beam spot) inevitably induces electron beam damage to the sample, particularly in electron-beam-sensitive materials, causing significant deviations in the measured Fe3+ content. Acquiring high-quality EELS spectra requires the samples to be sufficiently thin, often 30–50 nm or even thinner. Consequently, the spectra collection mode and parameters must be carefully selected to accurately measure the Fe3+ content. EELS data collection can be performed in three modes: point, line, and mapping scan. The point and line scan modes involve continuous data collection at a single point. Shorter dwell times result in inferior data quality; however, extending the dwell time may cause damage to electron-beam-sensitive samples, consequently introducing bias in the data. For instance, using the point scan mode, Gu et al. observed an increase in the Fe3+/ΣFe ratio with the increase of the dwell time at a single point, when measuring the Fe3+ content in the Chang'e 5 impact glass.27 In contrast, the mapping scan mode can ensure data quality even with a short dwell time at each point. The operation principle of the mapping scan mode involves using a scanning probe to collect parallel spectral data at (x, y) positions and sequentially acquiring image pixels by changing the (x, y) positions. This enables the simultaneous collection of EELS and annular dark field (ADF) signals and performs spatial drift correction, thus obtaining the EELS spectra and spatial image of the sample. The mapping scan mode distributes the number of electrons across multiple pixels within a given time, allowing for pixel stacking in data processing, which can yield higher-quality signals even with a shorter dwell time at each point. Consequently, the mapping scan mode is expected to effectively reduce the potential electron beam damage to the sample.
Sample preparation is another factor that may affect the measurements of Fe3+ content using EELS. Techniques including powder ultrasonic dispersion, ultra-thin sectioning, and ion milling are commonly used for preparing TEM specimens. Ultrasonic dispersion requires samples to be ground to particles that are finer than 200 mesh. Ultra-thin sectioning necessitates complex pre-treatments such as resin embedding and thus is unable to achieve precise micro-cutting at target sites. In ion milling, the sample must be pre-processed to a disc with a diameter of 3 mm and thickness of a few micrometres. These techniques are inadequate for samples in the Earth and planetary science community, which preferentially involve natural minerals, rocks, meteorites, interstellar dust, and space mission returns. These samples are characterized by their varied, heterogeneous, and complex nature, and can be irreproducible due to age or scarcity. Alternatively, the focused ion beam-scanning electron microscope (FIB-SEM) dual-beam system has emerged as a promising alternative, allowing SEM-guided selection of regions of interest, with FIB then used to extract precise sections for further TEM observations.28
Although van Aken et al. (2002) established a “universal curve” for the quantification of Fe3+ in minerals through the analysis of powder samples in TEM mode,22 the applicability of this curve to FIB specimen preparation in conjunction with STEM-EELS mapping remains to be confirmed. In the present study, we conducted a methodological investigation using FIB-STEM-EELS-mapping to determine the Fe3+ content in a series of pyroxene reference materials. Additionally, we investigated the influence of FIB specimen preparation and experimental settings on the accurate quantification of Fe3+ in minerals.
The prepared FIB thin sections were subjected to STEM-EELS measurements in a Thermo Scientific FEI Talos F200S TEM. The EELS data acquisition was performed in STEM mode using the Gatan Continuum 1077 EELS spectrometer, with a TEM camera length of 98 mm and EELS entrance aperture of 5 mm. The zero-loss and core-loss peaks were simultaneously acquired using the STEM SI Dual mode, with a core-loss range of 580–980 eV. To explore the impacts of the experimental conditions on the EELS data, spectra were acquired using various channel energy dispersion (0.05, 0.15, and 0.3 eV) and pixel time (0.01, 0.02, 0.04, 0.06, 0.08, and 0.1 s) values. The acquisition area size was set to 300 × 15 pixels for single-point dwell times of ≤0.02 s, but to 200 × 15 pixels for dwell time >0.02 s. As the ionization edges of the inner shells typically occur at energy losses higher than 100 eV, the electron excitation process can overlay a monotonically decreasing background signal onto the ionization edge signal of the inner shell. Consequently, it is necessary to subtract this background signal from the intensity of interest at the energy loss for quantitative elemental analysis. In this study, the following steps were taken in the Digital Micrograph software for accurate processing of the EELS spectra: (1) a background was fitted to the original spectrum using the least squares method at the 15–30 eV front of the Fe-L3 edge at 708 eV, which yielded the subtracted spectrum (Fig. 2a). (2) Due to the multiple scatting effects, a broad hump may exist at the inner shell ionization edge due to the merging of higher-order satellite peaks and secondary scattering peaks, which can significantly influence the shape of the loss peak and eliminate fine structure. This multiple scatting impact could be neglected in the case of extremely thin samples (th/λ ≤ 0.3), but in thick samples, a Fourier-ratio deconvolution method was applied to eliminate the multiple scattering from the loss spectrum, which generated the deconvolved spectrum (Fig. 2a). (3) All spectrum intensities were normalized using a double arctan function.22 To establish the numerical proportionality between the white-line intensity ratio I(L3)/I(L2) and the ferric iron concentration Fe3+/ΣFe, the white lines must be isolated from the background intensity, which is due to transitions to unoccupied states in the continuum, using a double arctan function (Fig. 2b). Instead of calculating the total integral intensities of the L3- and the L2-white lines, Van Per et al. (2002)22 developed a method based on changes in the I(L3)/I(L2) intensity ratio, chemical shift, and near-edge structure, depending on the mean iron valence state. Two integrating windows of 2 eV width were applied to the L2,3-edges from 708.85 eV to 710.85 eV and from 719.85 eV to 721.85 eV centered around the maximum at the L3-edge for Fe3+ and at the L2-edge for Fe2+, respectively. In our work, the background-function step sizes were scaled to the minima behind the L3- and the L2-edge with fixed inflection points at 709.85 eV and 720.95 eV (Fig. S1†).
The height of the two arctan functions (h1 and h2) were scaled to the minimum behind the Fe-L3 and the Fe-L2 edge, respectively, with fixed inflection centres at E1 = 709.85 eV and E2 = 720.95 eV and fixed widths of w1 = w2 = 1 eV. Two integration windows of 2 eV in width were applied to the Fe-L3 and the Fe-L2 edges, respectively, to calculate the integral areas and I(L3)/I(L2) ratios (Fig. 2b). The rationale for selecting a 2 eV integration window lies in the chemical shift of ∼2 eV between Fe2+ and Fe3+. While integration windows narrower than 2 eV could theoretically be applied, excessively small widths amplify measurement errors due to reduced signal intensity. To balance data integrity and measurement reliability, we adopted a 2 eV integration window as the optimal threshold (Fig. S2†).
Sample | Hed1904 | HYP-1 | GZF-A | CL-R | XWT-1 | Age1903 |
---|---|---|---|---|---|---|
a bdl denotes data below detection limit of EPMA. b All the valence states of iron are calculated as FeO. Based on EPMA analytical results and reference material certifications, total compositions within the range of 98.5% to 101.5% are considered analytically acceptable. | ||||||
Data points | 12 | 5 | 5 | 4 | 5 | 12 |
SiO2 | 48.93 | 51.96 | 48.74 | 46.93 | 49.31 | 51.31 |
TiO2 | bdl | 0.47 | 1.16 | 1.64 | 0.83 | 1.83 |
Al2O3 | 0.01 | 3.59 | 8.89 | 9.12 | 4.15 | 0.15 |
FeO | 21.84 | 6.5 | 7.43 | 6.31 | 23.16 | 29.16 |
MnO | 5.14 | 0.07 | 0.13 | 0.11 | 0.51 | 0.51 |
MgO | 1.61 | 14.5 | 13.67 | 12.44 | 0.03 | 0.03 |
CaO | 23.64 | 21.17 | 17.24 | 20.82 | 13.54 | 0.54 |
Na2O | 0.02 | 1.61 | 1.73 | 1.31 | 6.99 | 12.99 |
K2O | bdl | bdl | 0.01 | 0.01 | 0.01 | 0.01 |
Cr2O3 | bdl | 1.04 | bdl | 0.01 | bdl | bdl |
Totalb | 101.19 | 100.91 | 99.01 | 98.69 | 98.53 | 96.52 |
Fe3+/ΣFe Mössbauer | 0.00 | 0.16 | 0.21 | 0.36 | 0.54 | 1.00 |
Sample | Chemical formula |
---|---|
Hed1904 | (Ca1.025Fe2+0.675Mn0.176Mg0.097Fe3+0.040Na0.002)Σ2.015(Si1.980Fe3+0.020)Σ2.000O6 |
HYP-1 | (Ca0.872Mg0.788Na0.114Fe2+0.17Fe3+0.06Cr0.030Ti0.0132Mn0.002)Σ2.005(Si1.894Al0.106)Σ2.000O6 |
GZF-A | (Ca0.685Fe2+0.144Mn0.004Mg0.756Ti0.032Al0.196Fe3+0.084Na0.124)Σ2.025(Si1.807Al0.193)Σ2.000O6 |
CL-R | (Ca0.820Fe3+0.228Mg0.696Al0.122Na0.093Fe2+0.076Ti0.045Mn0.003)Σ2.070(Si1.726Al0.274)Σ2.000O6 |
XWT-1 | (Ca0.573Na0.523Fe3+0.480Fe2+0.300Al0.090Ti0.025Mn0.017Mg0.002)Σ2.010(Si1.903Al0.189)Σ2.092O6 |
Age1903 | (Fe3+0.959Na0.948Ca0.022Mn0.016Mg0.002Ti0.052)Σ1.999(Si1.932Fe3+0.061Al0.007)Σ2.000O6 |
Most of the Mössbauer spectra of the pyroxene samples exhibited tall and thin doublets, suggesting similar occupation status of Fe in the mineral structures (Fig. 3). The occurrence of hyperfine magnetic doublets indicates poor magnetic ordering of these pyroxene samples at room temperature, which is consistent with the typical features of Fe-bearing silicates. The quadrupole doublet derived from the spectral fitting mainly originates from the presence of Fe2+, particularly at the M1 (weakly distorted octahedral) positions in the pyroxene structures. Two quadrupole doublets were observed in the Hed1904, suggesting that Fe2+ occupies both M1 and M2 positions in the mineral structure (Fig. 3a), which is consistent with the absence of Fe3+ based on the spectrum fitting (Table 1). In contrast, only one quadrupole doublet was observed in samples with low Fe content (Fig. 3b–d), suggesting that the small amount of Fe preferentially exists as Fe2+ at the M1 positions in the mineral structures. The spectrum of XWT-1 was different from those of the other samples (Fig. 3e), likely because the Fe3+ occupies not only the octahedral M1 positions but also tetrahedral positions in the pyroxene structure. Only a narrow quadrupole splitting was identified in Age1903, which is because of the presence of Fe3+ at the M1 position in the pyroxene structure (Fig. 3f). The Fe3+/ΣFe reference values of the pyroxene samples were calculated based on the Mössbauer spectra fitting and are provided in Table 1.
Reduced energy dispersion enhances the spectral resolution of the EELS-Fe-L-edges within narrower energy windows, although this improvement comes at the expense of increased filtration of inelastic scattered electrons through the spectrometer slit and concomitant SNR degradation. Systematic comparisons of spectral outcomes across dispersion settings (0.05–0.3 eV per channel) are being conducted to establish optimal acquisition parameters balancing edge definition and counting statistics. Increasing the dwell time from t = 0.01 to 0.02 s can improve the signal-to-noise ratio in the spectrum of HYP-1 while minimizing the electron beam damage on the sample (Fig. 4b). The electron beam damage on HYP-1 induced by the prolonged dwell time was also observed in the data acquired at d = 0.05 eV (Fig. 4c) but not that at d = 0.3 eV (Fig. 4d). This is because the low energy resolution of d = 0.3 eV is limited to revealing the changes in the fine electronic structure.
d (eV) | t (s) | T (s) | R 2 | a | b | c |
---|---|---|---|---|---|---|
0.05 | 0.02 | 45 | 0.97 | 0.19 ± 0.06 | −0.40 ± 0.08 | 0.30 ± 0.02 |
0.15 | 0.02 | 45 | 0.99 | 0.25 ± 0.05 | −0.48 ± 0.07 | 0.33 ± 0.02 |
0.30 | 0.02 | 45 | 0.98 | 0.18 ± 0.05 | −0.38 ± 0.06 | 0.30 ± 0.02 |
Several Fe-bearing minerals, including two natural hornblende (labelled BD #4 and SC #1), one natural magnetite (Fe3O4_tw), one synthetic nano magnetite (Fe3O4_sy), and one synthetic hematite (Fe2O3_sy) sample, were selected to assess the reliability and applicability of the calibration curve established under the experimental condition of d = 0.15 eV. The EELS spectra were collected with parameters of d = 0.15 eV, t = 0.02 s, T = 50 s, and a sample thickness of th = 0.6 (Fig. 8a). The Fe3+ content in these samples, as determined by EPMA, was found to be 82% for SC #1, 32% for BD #4, 67% for the magnetite, and 100% for the hematite. The EELS spectra correspondingly displayed features that correlated with the Fe3+ contents in the samples (Fig. 8a). For instance, the peak maximum for Fe-L3 was observed at 708.5 eV for BD #4, and at 710 eV for Fe2O3_sy, which is in line with the 32% Fe3+ content in BD #4 and the 100% Fe3+ content in hematite. The Fe3+ contents derived from the EELS analyses for these minerals aligned well with those obtained from the EPMA analyses (Fig. 8b), as indicated by the data points clustering around the 1:
1 line. This alignment validates the high reliability and broad applicability of the calibration curves obtained under the data acquisition conditions of t = 0.02 s, T = 45 s, and d = 0.15 eV (as well as d = 0.05 and 0.3 eV).
The total integration time has a pronounced impact on the accurate determination of Fe-EELS I(L3)/I(L2) ratios (Fig. 9a). Insufficient integration time can lead to spectra with inadequate signal-to-noise ratios, particularly in minerals with low Fe3+ contents. For example, the EELS I(L3)/I(L2) ratio error for HYP-1, which does not contain Fe3+, can be as high as 0.35 when the total integration time is less than 45 s. In contrast, the error for Age1903, which contains 100% Fe3+/ΣFe, is relatively modest at 0.3. Prolonging the total integration time to T ≥ 45 s can effectively reduce the error to a maximum of 0.14. Notably, the Fe-EELS I(L3)/I(L2) ratio does not vary linearly with Fe3+ content in the calibration curve (Fig. 7). Consequently, the error associated with quantifying Fe3+ content using the Fe-EELS I(L3)/I(L2) ratio can vary across samples with various Fe3+ contents. The error in Fe3+ content quantification initially decreases and then increases with the increasing Fe3+/ΣFe ratio, reaching a minimum of 0.012 at a Fe3+/ΣFe ratio of 0.7 (Fig. 9b). The error values for samples with 0% and 100% Fe3+ are 0.032 and 0.025, respectively, which are slightly lower than the 0.04 reported in the previous study by van Aken et al.22
![]() | ||
Fig. 9 Experimental error analysis considering the impacts of (a) the total integration time and (b) the nonlinear relationship between the EELS I(L3)/I(L2) ratio and the mineral Fe3+ contents. |
Furthermore, in the microanalysis (EPMA) data and EELS-derived I(L3)/I(L2) ratios in Fig. 8b, we attribute these discrepancies primarily to two dominant sources of uncertainty: (1) systematic measurement errors inherent to EPMA instrumentation and (2) computational uncertainties propagated through the charge-balance oxygen modelling framework. We employed the conventional ZAF + standards routine of the JEOL JXA-8230 EPMA to determine the compositions of Fe-bearing minerals. Based on the analysis of an in-house monitor standard (chrome-diopside), the error was less than 3% for the major elements. The Fe3+/ΣFe ratios were determined through charge-balance oxygen calculation, a methodology intrinsically associated with computational uncertainties. When implementing the standard calibration curve established in this study for retrospective quantification, the methodological error was constrained to 4%. However, the combined uncertainty—accounting for both the inherent limitations of the charge-balance approach and the propagated uncertainties from the calibration protocol—may reach approximately 10%.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5ja00002e |
This journal is © The Royal Society of Chemistry 2025 |