Fe-based metallic glasses as efficient oxygen scavengers

Abstract

Ubiquitous oxygen drives degradation, spoilage, and side reactions, making oxygen scavenging essential for materials preservation and reaction stabilization. However, conventional oxygen scavengers exhibit limited oxygen removal rates and capacities far below their theoretical maximums, wasting resources while lagging behind industry needs. Here, we report Fe-based metallic glasses as efficient oxygen scavengers, achieving oxygen removal rates 1–4 orders of magnitude higher than conventional systems. In FeSiB metallic glass, Fe oxidation synergistically activates Si, delivering a 24-hour oxygen removal capacity of 1.439 L g^-1—reaching the Fe-based theoretical limit—and a 48-hour capacity of 1.596 L g^-1, surpassing it. Density functional theory calculations reveal that the amorphous structure significantly lowers the oxygen adsorption energy barrier and facilitates O–O bond cleavage. Moreover, the generated self-reinforcing microdomains mediate O₂/H₂O transport via robust autocatalytic cycling. These results highlight a promising strategy for oxygen potential control and suggest a possible paradigm for catalytic applications.

Fe-based metallic glasses are reported as efficient oxygen scavengers, achieving removal rates 1–4 orders of magnitude higher than conventional systems through rapid adsorption and autocatalytic effect, while approaching theoretical capacity limits.

Introduction

Though essential to life, the ubiquity of oxygen drives unwanted reactions, making effective scavenging crucial for food and pharmaceutical preservation, cultural relics conservation, and chemical synthesis^1–5. Fe-based scavengers, the most prevalent type, can remove oxygen through a redox reaction that proceeds in electrolyte-containing environments^1,6. Theoretically, 1 g of iron can remove oxygen from approximately 1566 mL of air at 298 K; however, the actual performance falls far short⁷, reflecting decades of stagnant efficiency and significant resource waste. Moreover, the performance of conventional scavengers remains fundamentally limited by three key challenges: intrinsically low oxygen adsorption activity, surface passivation due to oxidation product accumulation, and a sharp decline in reaction kinetics under low-oxygen conditions—a limitation widely deemed unavoidable^6–11. These constraints hinder their use in demanding applications like gas purification and catalyst protection.

Metallic glasses (MGs) are known for their exceptional mechanical and functional properties^12–16, with Fe-based MGs being particularly promising for large-scale applications due to their good glass-forming ability and low cost¹⁷. While they have been widely used as soft-magnetic materials, their potential in chemical reactions remains largely unexplored.

In this study, we report the use of Fe-based MGs as oxygen scavengers, achieving unmatched efficiency that far exceeds conventional materials. Fe₇₈Si₉B₁₃ (FeSiB) MG exhibits excellent oxygen scavenging performance even at a low-temperature of 278 K, whereas conventional Fe-based oxygen scavengers are nearly ineffective. Kinetic studies reveal a self-accelerating oxygen scavenging process, while density functional theory (DFT) calculations show that FeSiB MG possesses significantly lower oxygen adsorption energy barrier than crystalline iron. Moreover, the formation of a unique coupled structure—comprising moisture-retaining and adsorption zones—endows the system with strong autocatalytic behavior. These features overcome long-standing limitations of conventional systems and enable exceptional performance under low-oxygen conditions. Their marked superiority and low cost over existing benchmarks position them as enabling materials for next-generation oxygen-sensitive applications (Fig. 1a).

Fig. 1. Schematic structure, application, and oxygen removal performance of Fe-based metallic glass (MG) scavenger.

a Schematic of the structural and functional characteristics of Fe-based MG oxygen scavenger. b, c Oxygen concentration curves (B-spline fitted) for FeSiB MG, pure iron, and reduced iron powders under non-saline (b) and saline (c) conditions. The error bars represent the standard deviations from three parallel measurements, showing good repeatability with variabilities of about 2%–5%. d, e Pseudo-first-order kinetic fittings corresponding to b and c, the value of -ln(C/C₀) represents the time-dependent evolution of the apparent first-order rate constant k. f, g First derivatives of the fitted curves in d and e, respectively, showing the evolution of instantaneous reaction rate over time. All measurements for each material were obtained from the same sample; each experimental group includes 3 samples. Source data are provided as a Source Data file.

Results

Accelerating oxygen removal kinetics

To evaluate oxygen removal performance, we compared FeSiB MG, pure iron, and reduced iron (commonly used in industry) powders in 500 mL sealed air under saline and non-saline conditions. Using 2 g samples with matched volumetric particle sizes (Supplementary Fig. 1), we added either 1 g ultrapure water (non-saline) or 1 wt% NaCl solution (saline), plus 0.8 g diatomite as a moisture-retaining agent. This controlled setup enabled direct performance comparisons under identical conditions.

Figure 1b, c demonstrates the superior oxygen removal kinetics of FeSiB MG. Even under non-saline conditions, it achieved complete oxygen removal in just 11.9 h, showing high reactivity (Fig. 1b). The complete oxygen removal time here is defined as the period required for the oxygen concentration <0.05%, corresponding to >99.7% oxygen removal. In contrast, pure iron and reduced iron showed much slower kinetics and activity decay after 4 h due to surface passivation. Reduced iron outperformed pure iron owing to its porous structure with much larger surface area. Adding 1 wt% NaCl significantly enhanced the oxygen removal performance of all three materials (Fig. 1c), confirming the role of electrolyte in promoting electron transfer. FeSiB MG achieved a shorter removal time of 5.6 h, while pure and reduced iron powders still failed to reach complete removal within 24 h, even under optimized conditions.

In this reaction system, given the ample metal supply and continuous oxygen depletion, oxygen removal kinetics can be described using the pseudo-first-order kinetic model^18–20, the apparent rate constant k can be obtained by Eq. (1):

$$k=-\mathrm{In}\left(C/C_0\right)/t$$ 1

where C represents the real-time oxygen concentration, C₀ is the initial oxygen concentration, and t is the reaction time. Using this equation, the data from Fig. 1b, c were replotted in Fig. 1d, e, where the curve slope at each point reflects the instantaneous reaction rate.

As shown in Fig. 1d, the -ln(C/C₀) curve for FeSiB MG displays exponential-type growth, indicating a continuously accelerating reaction rate that deviates from traditional kinetics. In contrast, pure iron and reduced iron show curves that level off, reflecting declining activity. Under saline conditions, FeSiB MG still follows an accelerating trend with a steeper slope (Fig. 1e), confirming electrolyte-enhanced kinetics. Meanwhile, the curves for pure iron and reduced iron become nearly linear, aligning with pseudo-first-order behavior.

Differentiating the curves in Fig. 1d, e yields Fig. 1f, g, where the vertical coordinate represents the instantaneous oxygen removal rate. FeSiB MG shows a much higher initial rate than the two crystalline iron powders, confirming its intrinsically high reactivity. All materials exhibit a slight rate dip in the first 1–2 h, likely due to initial oxide formation hindering internal oxidation. Uniquely, FeSiB MG displays accelerating kinetics—starting after 3 h without electrolyte, and as early as 1 h with electrolyte. This suggests a strong autocatalytic effect during oxygen removal. In such reactions, a product also acts as a catalyst, causing the rate to increase as the reaction proceeds—as seen in acid-catalyzed ester hydrolysis and liquid-phase crystallization^21,22.

For comparison, pure iron shows a continuous rate decline to near zero under both conditions. Reduced iron behaves similarly under non-saline conditions, but under saline conditions, it displays a V-shaped curve: the rate drops sharply in the first 2 h, then gradually rises. The X-ray diffraction (XRD) results in Supplementary Fig. 2 show that Fe₃O₄ and FeO(OH) are simultaneously formed during the reaction. At the early stage (2 h), Fe₃O₄ dominates, leading to a relatively low oxygen removal rate. With continued reaction, the FeO(OH) phase progressively increases and becomes pronounced at 6 and 8 h, giving rise to the V-shaped variation in Fig. 1g. The scanning electron microscopy (SEM) images shown in Supplementary Fig. 3, together with the energy-dispersive spectroscopy (EDS) maps in Supplementary Fig. 4, further corroborate the result.

Record-high oxygen removal capacity

Figure 2a compares the oxygen removal time and percentage for the three materials, underscoring the superior performance of FeSiB MG. To quantify this, we integrated the instantaneous rate curves from Fig. 1f, g and divided by the reaction time to obtain the average rate kₐ. As shown in Fig. 2b, under non-saline conditions, the kₐ of FeSiB MG is 60 times that of pure iron and 41 times that of reduced iron powder; under saline conditions, it remains 40 and 9 times higher, respectively.

Fig. 2. Comparative oxygen scavenging performance.

a Oxygen removal time and final removal percentage for FeSiB metallic glass (MG), pure iron, and reduced iron powders. b k_a comparison under non-saline and saline conditions; FeSiB MG demonstrates markedly higher k_a in both cases. c Equivalent 12-h and 24-h oxygen removal capacities, relative to the theoretical maximum for each material, the error bars represent the standard deviations from three parallel measurements and the green circles display the distributed data points. d Radar plot comparing FeSiB MG (red line and shading) and reduced iron (blue line and shading) across four normalized metrics: k_a, weight-based, cost-based, and surface area-based oxygen removal capacities. BET surface areas: FeSiB MG, 0.0340 m² g⁻¹; reduced iron, 0.412 m² g⁻¹. e Oxygen removal rate comparison between Fe-based MGs and various reported scavengers^8,11,23–30. Source data are provided as a Source Data file.

Besides reaction rate, oxygen removal capacity is another key metric for evaluating scavenger performance. To assess it more accurately, we used 0.5 g of sample and increased the air volume to 1200 mL to ensure sufficient moisture and oxygen. Based on their iron mass fraction (FeSiB MG: 91.7%, pure iron: 100%, reduced iron: 98.8%), the theoretical maximum oxygen removal capacities at 298 K are 1.436, 1.566, and 1.547 L, respectively.

Figure 2c compares the equivalent oxygen removal capacities—defined as the volume of air from which oxygen is removed—at 12, 24, and 48 h, relative to theoretical limits. In practical use, performance within 12 h is most critical. In 12 h, 1 g of FeSiB can remove oxygen from 1.045 L of air, approaching its theoretical maximum and far exceeding pure iron (0.149 L) and reduced iron (0.440 L). After 24 h, FeSiB reaches 1.439 L g⁻¹, attaining its Fe-based theoretical limit, and further increases to 1.596 L g⁻¹ at 48 h—surpassing even the theoretical limit of pure iron (1.566 L g⁻¹). The highly reactive Fe atoms in FeSiB promote the synergistic involvement of Si in the reaction, a phenomenon first reported here (Supplementary Fig. 5).

To comprehensively compare material performance, a radar chart (Fig. 2d) was constructed using k_a, and weight-, cost-, and area-normalized equivalent oxygen removal capacities (12- and 24-h). Given the porous nature of reduced iron powder, specific surface areas were measured via the BET method. FeSiB MG outperforms reduced iron powder across all metrics, delivering much higher oxygen removal rates and capacities despite having less than one-tenth of the surface area. Owing to its low material and processing costs, FeSiB MG also shows superior cost-effectiveness—approximately 2.3× higher than reduced iron based on 12-h performance (Supplementary Fig. 6). Moreover, as Fe-based MGs do not require specific formability, thickness, or magnetic properties for use as oxygen scavengers, processing costs are expected to decrease further.

The oxygen removal performance of FeSiB MG can be further improved. Pre-reacting 2.1 g of FeSiB powder with oxygen to consume 0.1 g of FeSiB significantly enhances its activity. This modified material is referred to as pre-reacted FeSiB (see “Methods” for details). Additionally, atomized amorphous powders Fe₇₈Si₉B₁₃ (denoted as FeSiB-2) and Fe_77.3Si_6.3B_13.9C_2.6 (FeSiBC) with smaller particle sizes show markedly improved performance, removing >99.7% of oxygen from 500 mL of air within just 1.6 and 2.0 h, respectively. Such rapid kinetics demonstrate its potential for oxygen potential regulation.

Figure 2e compares the oxygen removal rates of Fe-based MG scavengers from this study with various reported systems^8,11,23–30. For reliable comparison, all rates here were calculated as volume of oxygen removed per unit time and weight. The results show that Fe-based MGs outperform all others by 1–4 orders of magnitude in oxygen removal rate.

Low-temperature and long-term storage performance

In practical applications, Fe-based oxygen scavengers are typically operated within a temperature range of 288–303 K. For conventional reduced iron powders, decreasing temperature suppresses the reaction kinetics of oxygen consumption, rendering them largely ineffective below approximately 283 K. To systematically evaluate the low-temperature adaptability of FeSiB MG and reduced iron powder, comparative oxygen scavenging experiments were therefore conducted at 278 K.

As shown in Fig. 3a, the oxygen absorption performance of conventional reduced iron powder deteriorates substantially under low-temperature conditions: only about half of the oxygen is consumed within 24 h, which is insufficient to meet the preservation requirements in food and pharmaceutical packaging. In contrast, the FeSiB oxygen scavenger maintains a high oxygen removal rate at 278 K and is capable of completely depleting the oxygen in a sealed container within 15 h. These results demonstrate that the FeSiB MG scavenger effectively overcomes the low-temperature performance limitations inherent to conventional iron-based systems.

Fig. 3. Low-temperature and long-term storage performance of oxygen scavengers.

a Oxygen removal curves of FeSiB metallic glass (MG) and reduced iron powder at 278 K. The samples were prepared at 298 K and subsequently transferred to a constant-temperature chamber at 278 K for testing. Reduced iron releases physically adsorbed oxygen upon cooling, causing a fluctuation in the blue curve between 1 and 2 h. b Pseudo-first-order kinetic fittings corresponding to a. c Oxygen removal curve of the MG after 30 days of sealed storage. The error bars in a and c represent the standard deviations from three parallel measurements. Source data are provided as a Source Data file.

The kinetic fitting results in Fig. 3b further reveal that the FeSiB MG retains an autocatalytic reaction behavior even at low-temperature, whereas no such characteristic is observed for reduced iron powder. This sustained autocatalytic feature under low-temperature conditions accounts for the superior oxygen removal performance and substantially broadens the applicable temperature window of Fe-based oxygen scavengers.

In addition to operational performance, long-term storage stability is a critical consideration for practical deployment, as oxygen scavengers are commonly stored in sealed environments for extended periods prior to use. To assess storage stability, the as-prepared FeSiB scavengers were sealed in 10 mL centrifuge tubes and stored at room temperature for 30 days before oxygen scavenging measurements were conducted. As shown in Fig. 3c, no noticeable degradation in oxygen removal performance is observed after storage, indicating that prolonged sealing does not induce surface passivation of the MG. This long-term retention of reactivity highlights the robustness of the FeSiB scavenger and underscores its suitability for applications requiring both low-temperature operability and extended shelf life.

Atomic-scale insights from DFT

To better understand the exceptional oxygen removal performance of FeSiB MG, we simulated the O₂ molecule adsorption process on its surface in the presence of H₂O molecules and compared it with that on the body-centered cubic (BCC) Fe(110) surface—the most stable and densely packed surface of BCC Fe³¹. Using climbing image-nudged elastic band (CI-NEB) calculations based on DFT, we estimated the transition states and energy barriers for O₂ adsorption. The energy barrier in the pure Fe case is 3.39 eV, while in the FeSiB MG case it is only 1.29 eV (Fig. 4a; transition states in Fig. 4b; full structures in Supplementary Fig. 7), indicating a much easier O₂ adsorption process on FeSiB MG and explaining its superior oxygen removal activity.

Fig. 4. Density functional theory simulations of O₂ adsorption on FeSiB metallic glass (MG) and crystalline Fe surfaces.

a Energy barriers of O₂ adsorption on Fe(110) and FeSiB MG surfaces calculated by the climbing image-nudged elastic band (CI-NEB) method. IS, TS, FS represent the initial state, transition state, and final state, respectively. b Simulated O₂ adsorption pathways. The numbers denote reaction coordinates. c Projected density of states (PDOS) and d-band centers of the surface Fe atoms before O adsorption. The arrows indicate the d-band center positions. d Charge density difference of 0.01e Å⁻³ isosurfaces around the O₂ molecules. Yellow and blue denote electron accumulation and depletion, respectively. Source data are provided as a Source Data file.

We have also calculated the energy barriers of the O₂ adsorption process by using the implicit solvent model to include the effect of the NaCl electrolyte. It should be noted that the possible specific effects of chloride ions were not considered in this model. The FeSiB MG still shows a lower energy barrier and therefore easier O₂ adsorption (Supplementary Fig. 8a). Besides, the O₂ adsorption process on the diamond-structured pure Si(111) surface and Si sites on the FeSiB MG surface were also calculated, with and without considering implicit solvation effects. The results show lower energy barriers of the O₂ adsorption on both the pure Si and Si atom in FeSiB MG (see Supplementary Figs. 9–12 for more details), supporting that the Si can actively react with O₂. However, in bulk crystalline Si, the formation of a surface Si oxide layer prevents the inner Si atoms from reacting with oxygen, whereas in FeSiB MG, the preferential oxidation of Fe promotes the oxidation of Si and maintains the exposure of active sites. The role of Si will be discussed in detail in the following section.

According to energy band theory, the adsorption process involves the hybridization of Fe 4s and 3d orbitals with O orbitals, forming bonding and antibonding states. This interaction leads to electron rearrangement, changes in charge density, bond breaking (O–O), and bond formation (Fe–O)^32,33. Since the Fe 4s orbitals are broad in the energy and their projected density of states (PDOS) is much smaller than Fe 3d orbitals (see Supplementary Fig. 13), the primary effects arise from the hybridization between O and Fe 3d orbitals³⁴. The overall energy change in this process can be effectively described by the Fe d-band center (see “Methods” section). An upward shift of the Fe d-band center strengthens the Fe–O interaction by reducing the occupation of Fe–O antibonding states, which facilitates electron transfer into the O₂ antibonding orbitals, weakens the O–O bond, and ultimately promotes O₂ dissociation^35–38, as shown in Supplementary Fig. 14. This d-band center theory has been successful in describing transition metal surfaces^39–42.

In our results, the PDOS for the 3d orbitals of surface Fe atoms before O adsorption was obtained, and the corresponding d-band centers were calculated (Fig. 4c). FeSiB MG shows a higher d-band center at −1.55 eV compared to −1.85 eV for pure Fe, which explains its stronger O adsorption. The pure Fe value closely matches the reported value of −1.80 eV⁴³, confirming the result of our calculation. The higher d-band center of FeSiB MG is also confirmed in the calculation with implicit solvent model (Supplementary Fig. 8b). To further understand the stronger adsorption, the charge density difference around the O₂ molecule in the initial states (Fig. 4d) is presented. For FeSiB MG, there is a decrease in charge density between the O atoms and an increase around them, indicating a stronger tendency for O–O bond breaking. In contrast, O₂ on pure Fe shows a decrease in overall charge density with no significant O–O bond breaking, consistent with its lower d-band center and higher adsorption energy barrier. Overall, DFT calculations reveal a low energy barrier for O adsorption on FeSiB MG, explaining its exceptional oxygen removal activity on the atomic scale.

Origin of robust autocatalytic effect

Based on the experimental data shown in Fig. 1b, c, the correlation curves between the oxygen removal kinetics and the mass of reaction products under the two conditions were constructed (see Supplementary Fig. 15a, b). The evolution trends of these curves indicate that the oxygen removal process exhibits a typical autocatalytic behavior, characterized by a rapid acceleration of the reaction rate with increasing product formation. To probe the mechanism of autocatalytic behavior, we analyzed the reacted composite powder under saline conditions. The reacted samples were washed with ethanol to remove loosely bound iron oxides and part of residual diatomite, then dried and characterized by XRD, as shown in Fig. 5a. Due to interference from diatomite and moisture, reliable and systematic information could not be readily obtained from operando XRD measurements; therefore, the reaction products were subjected to post-treatment.

Fig. 5. Origin of the pronounced autocatalytic behavior in FeSiB metallic glass (MG) oxygen scavenger.

a X-ray diffraction (XRD) patterns of reaction products from FeSiB MG, pure iron, and reduced iron under saline conditions. b–d Scanning electron microscopy (SEM) and energy-dispersive spectroscopy (EDS) mapping images of the reacted FeSiB MG, revealing a coupled structure of iron oxides (red) and diatomite (cyan). The observations were repeated three times, yielding similar results. e Schematic illustration of the proposed autocatalytic oxygen removal mechanism in FeSiB scavenger. Source data are provided as a Source Data file.

XRD analysis reveals distinct oxidation behavior of FeSiB MG. The reaction products consist mainly of diatomite and residual amorphous phase. γ-FeO(OH) is identified as the primary oxidation product, though its weak diffraction intensity suggests poor adhesion to the MG matrix, with most oxides removed during washing. This facilitates continuous exposure of active sites, which is essential for sustained oxygen scavenging. Poor adhesion and easy spallation of FeO(OH)-type oxides are widely recognized and well documented in corrosion science and oxidation literature^20,44. XRD pattern (Supplementary Fig. 16) and high-resolution transmission electron microscopy (HRTEM) image (Supplementary Fig. 17) of the separated orange-red products further confirm γ-FeO(OH) as the dominant oxidation species. In contrast, pure iron and reduced iron exhibit markedly different oxidation behavior. Post-reaction XRD pattern shows dominant Fe₃O₄ along with minor γ-FeO(OH) and α-FeO(OH), indicating sluggish kinetics in crystalline iron that lead to incomplete oxidation and passivating Fe₃O₄ formation. These results underscore the fundamental contrast between FeSiB MG and crystalline iron. DFT calculations provide mechanistic insight: the strong oxygen adsorption and dissociation abilities of FeSiB enhance surface reactivity, enabling complete oxidation and suppressing Fe₃O₄ formation.

To elucidate the autocatalytic mechanism of FeSiB MG, we conducted microscopic characterization of the reacted powder. SEM and EDS mapping images (Fig. 5b–d) reveal a coupled microstructure of diatomite (cyan) and nano iron oxides (red) on the particle surface. This interfacial structure is key to enhanced oxygen removal. Such regions are characteristic of the post-reaction particles, as illustrated in Supplementary Fig. 18. BET analysis shows the oxidation products (including adhered diatomite) possess a high specific surface area of 121 m² g⁻¹, far exceeding that of pristine diatomite (1.65 m² g⁻¹; Supplementary Fig. 19). These findings, along with oxygen removal data, clarify the autocatalytic behavior of FeSiB MG.

As illustrated in Fig. 5e, the initial oxygen removal by FeSiB MG is driven by surface adsorption of O₂ and H₂O, with diatomite providing moisture to sustain the reaction. Over time, a nanoporous γ-FeO(OH) layer rapidly forms, coupling with diatomite to create a hierarchical interfacial structure. This architecture promotes reactant transport, prevents surface passivation to maintain active site exposure, and provides stable interfaces for electron transfer. As the reaction progresses, the evolving porous nano-oxide layer further refines the interface, underpinning the observed autocatalytic behavior. This stands in stark contrast to that of crystalline iron. Autocatalytic reactions commonly show slow initial rates because of low product concentrations^21,22. In contrast, FeSiB MG deviates fundamentally from this conventional pattern: its amorphous structure enables high intrinsic reactivity from the onset.

Prior studies show that Fe²⁺ and Fe³⁺ species in Fe(OH)₂ and FeO(OH) can reorganize into Fe₃O₄⁴⁵, but FeSiB MG—with its superior oxygen adsorption and dissociation capabilities—undergoes complete oxidation, effectively avoiding Fe₃O₄ formation. Its highly adsorptive surface layer further promotes oxygen transport, ensuring full utilization of Fe. These findings align with reports linking Fe₃O₄ formation to Fe(OH)₂ oxidation⁴⁶ and its possible transformation into γ-FeO(OH) and α-FeO(OH) under suitable conditions⁴⁷, offering a mechanistic basis for the distinct reactivity of FeSiB MG.

A simple experiment confirms the strong link between FeSiB MG’s autocatalytic effect and its surface structure. Pre-reacting the powder to form a dual-zone interface—comprising a water-retention zone and a strong adsorption zone—significantly enhanced oxygen removal, reducing the complete oxygen removal time from 5.6 to 3.5 h (Supplementary Fig. 20).

Analysis of oxidation behavior

To elucidate the oxidation mechanism of the FeSiB MG oxygen scavenger, a series of controlled experiments was designed. Figure 6a compares the Fe 2p_3/2 X-ray photoelectron spectroscopy (XPS) spectra of the native oxide layers formed on FeSiB MG and pure Fe. The FeSiB sample exhibits a higher initial oxidation state of Fe, indicating a stronger intrinsic affinity for oxygen.

Fig. 6. Oxidation behavior of oxygen scavengers.

a Fe 2p_3/2 X-ray photoelectron spectroscopy (XPS) spectra of primary oxide layers on FeSiB metallic glass (MG) (red line) and pure Fe (black line), indicating a higher initial Fe oxidation state in FeSiB. b Relative fractions of Si and Fe oxides in FeSiB surface oxide layers before and after reaction, with the nominal composition shown for comparison. c X-ray diffraction (XRD) patterns of FeSiB after 2, 5, and 20 cycles. d XRD patterns of pure Fe after 2 and 5 cycles; the inset shows a magnified view. e Fe K-edge X-ray absorption near edge structure (XANES) spectra of FeSiB before and after reaction, and of pure Fe after reaction. f R-space extended X-ray absorption fine structure (EXAFS) spectra (phase shift not corrected) of FeSiB before and after reaction, and of pure Fe after reaction. Source data are provided as a Source Data file.

To isolate the contribution of Si, additional experiments were conducted in the absence of diatomite. Specifically, 2 g of FeSiB powder was mixed with 30 μL of 1 wt% NaCl solution and allowed to react under ambient air. After 1 h, an additional 30 μL of solution was introduced, and this procedure was repeated multiple times. Samples subjected to n cycles were denoted as FeSiB-Rn. The small volume of electrolyte was deliberately chosen to achieve micro-wetting without full immersion, ensuring that oxygen uptake occurred at the solid–gas–liquid interface rather than via bulk solution reactions. This condition is critical for effective oxygen scavenging.

XPS spectra were first detected for samples subjected to 10, 20, and 30 reaction cycles. The relative fractions of Si oxides and Fe oxides in the resulting surface oxide layers were quantified and compared with those of the native oxide on the as-prepared alloy, as well as with the nominal bulk composition. The XPS spectra of Si and Fe oxides are shown in Supplementary Figs. 21 and 22, respectively.

The results are summarized in Fig. 6b. Notably, the native oxide layer on FeSiB MG is already strongly enriched in Si oxides relative to the nominal composition, indicating a pronounced oxygen affinity of Si, in good agreement with the DFT predictions. More importantly, the Si-to-Fe oxide ratio increases monotonically with progressive oxidation. As shown in the XPS survey spectra in Supplementary Fig. 23, the signal corresponding to oxidized Si is clearly pronounced. This trend provides compelling evidence for the active participation of Si in the oxygen scavenging process; in the absence of a significant Si contribution, continued oxidation of Fe would be expected to yield a surface dominated by Fe oxides, thereby precluding any further increase in the relative Si content.

Figure 6c, d show the XRD patterns of reacted FeSiB and pure Fe samples from the diatomite-free experiments. To exclude thermal effects, all samples were dried naturally at room temperature. In the FeSiB series, the diffraction features are weak due to the poor crystallinity of the oxidation products; nevertheless, the characteristic reflections of γ-FeO(OH) are discernible in the FeSiB-R20 sample. In contrast, for pure Fe, the formation of Fe₃O₄ is detected after only two reaction cycles. These observations are fully consistent with the above XRD results.

To further quantify the oxidation degree and Fe–O coordination environment in the FeSiB MG and pure Fe, synchrotron-based X-ray absorption measurements were performed on samples reacted for the same duration (5.5 h). Figure 6e presents the Fe K-edge X-ray absorption near edge structure (XANES) spectra of FeSiB before reaction, FeSiB after reaction, and reacted pure Fe. As the absorption edge position reflects the Fe oxidation state, the edge of reacted pure Fe (blue curve) is clearly shifted to lower energy relative to that of reacted FeSiB (red curve), confirming a lower average oxidation state. Moreover, the pronounced intensity differences highlighted by the red arrow indicate a higher oxidation degree in FeSiB MG.

Figure 6f shows the corresponding extended X-ray absorption fine structure (EXAFS) spectra (phase shift not corrected) in R space, which provide insight into the local Fe–O coordination. Comparison of the pristine alloy (black) and reacted FeSiB (red) confirms that the post-reaction spectra are not significantly influenced by residual Fe coordination in the MG. The FeSiB scavenger exhibits a broadened first-shell Fe–O contribution (R ≈ 1.3–1.5 Å), accompanied by a weak, continuous second-shell feature (R ≈ 2.6–2.9 Å), while higher-shell contributions are largely suppressed. This coordination signature is characteristic of low-crystallinity Fe oxyhydroxide species closely resembling γ-FeO(OH).

In contrast, the crystalline Fe-based scavenger displays a different EXAFS profile, with a strong first-shell Fe–O peak at 1.1–1.5 Å and a pronounced second-shell feature at 2.6–2.9 Å, together with retained medium-intensity signals in the 3.5–4.5 Å range. These features are indicative of a mixed coordination environment dominated by Fe₃O₄, with additional contributions from FeO(OH). A weak feature at ~2.3 Å suggests a minor fraction of residual metallic Fe. Notably, the intensity observed around 1.1–1.2 Å does not correspond to an independent coordination shell but arises from the low-R extension of the first-shell Fe–O contribution, a common characteristic of Fe–O-based systems. This effect is intrinsic to both iron oxides and oxyhydroxides and is further influenced by phase shifts, window functions, and multiple-scattering contributions.

Taken together, the XANES and EXAFS results consistently demonstrate that Fe in the FeSiB MG is oxidized more readily and to a higher degree than in crystalline Fe, corroborating its superior oxygen scavenging capability.

Analysis of the contribution of Si

To evaluate the contribution of Si in the FeSiB MG, SEM observation was performed on the FeSiB-R20 sample prepared without diatomite to eliminate interference. As shown in Fig. 7a, the reacted FeSiB-R20 exhibits a highly porous surface morphology, with a large fraction of oxidation products appearing as delaminated or exfoliated features. The corresponding EDS elemental maps (Fig. 7b–e) acquired from the region highlighted in Fig. 7a show pronounced enrichment of Fe and O within the oxide layer, accompanied by a clear Si signal, confirming the involvement of Si in the oxidation process. The EDS spectrum (Fig. 7f) yields an average composition of Fe_31.77Si_2.74O_65.49 (at%), corresponding to a Si-to-Fe atomic ratio of 8.62%, slightly lower than the nominal value of 11.54%. Assuming the oxidized Si is present as SiO₂, the estimated contribution of Si to oxygen uptake reaches ~11.49% for controlled experiments without diatomite. Considering the presence of intermediate Si oxidation states, the actual contribution is expected to be close to but slightly lower than this value. It should be noted that the removal of diatomite in this experimental series may lead to minor variations in the apparent Si contribution compared with diatomite-containing systems. Nevertheless, these results unambiguously demonstrate the active role of Si in oxygen scavenging and provide a reasonable quantitative estimate of its contribution.

Fig. 7. Analysis of the role of Si.

a Scanning electron microscopy (SEM) image of the FeSiB-R20 sample surface, showing spontaneous delamination of the oxide layer after reaction, the SEM observations were repeated three times, yielding similar results. b–e Energy-dispersive spectroscopy (EDS) elemental maps of the region highlighted in a, indicating pronounced Fe, Si, and O signals, with only weak B signal. f Corresponding EDS spectrum showing the measured composition, revealing that the oxide layer is primarily composed of Fe and Si oxides. Source data are provided as a Source Data file.

Raman and infrared spectroscopy were also employed to probe the oxidation behavior of Si (see Supplementary Figs. 24 and 25); however, their phase discriminability was limited. On one hand, the presence of Fe–Si and Fe–B bonding in the FeSiB alloy gives rise to multiple intrinsic vibrational features that complicate spectral assignment. On the other hand, the poor crystallinity of the oxidation products leads to significant peak broadening, further reducing spectral resolution. Nevertheless, both Raman and infrared spectra clearly reveal the formation of γ-FeO(OH) and α-FeO(OH), with γ-FeO(OH) being the dominant phase, in good agreement with the XRD results.

Discussion

Fe-based MGs exhibit a broad universality in both oxygen removal activity and autocatalytic behavior. As summarized in Supplementary Fig. 26, melt-spun Fe₈₀P₁₃C₇ (FePC) and Fe_73.5Si₉B_13.5Nb₃Cu₁ (FeSiBNbCu), together with atomized FeSiB-2 and FeSiBC powders, all demonstrate high oxygen scavenging efficiency despite substantial differences in composition and processing routes. These results indicate that the observed oxygen removal behavior is not limited to a specific alloy chemistry or fabrication method, but rather represents a general feature of Fe-based MG systems. Among them, FeSiB-2 achieves a k_a as high as 3.34 h⁻¹, highlighting the strong kinetic advantage of this material class. When combined with low-cost raw materials and minimal processing requirements, Fe-based MGs emerge as a potentially economical and scalable platform for oxygen scavenging.

Beyond individual material performance, the present results reveal a broadly applicable reaction mechanism. The composite architecture, which integrates oxygen adsorption sites with reactant and water storage within a confined micro–nano environment, provides a flexible framework for regulating redox reactions. At the atomic scale, the disordered bonding environment of the MG promotes rapid oxygen adsorption and dissociation, while at the micro–nano scale, the formation of multifunctional nanoporous iron oxyhydroxide layers facilitates efficient O₂/H₂O transport and sustains active reaction interfaces. This hierarchical coupling of structure and chemistry explains the simultaneous achievement of high reaction rates, complete oxidation, and autocatalytic behavior, and suggests that the underlying mechanism can be extended to the rational design of other high-performance oxygen scavengers and related functional materials.

Looking ahead, further cost reduction and performance optimization of Fe-based MG oxygen scavengers will require closer integration between MG manufacturing technologies and the oxygen scavenger industry to enable scalable and application-oriented fabrication. Although this study demonstrates clear performance advantages under laboratory conditions, extended field testing under realistic operating environments remains essential, particularly to assess long-term stability and system compatibility in sensitive application domains such as advanced packaging, gas purification, and semiconductor-related processes. These aspects warrant further investigation beyond the scope of the present work.

In summary, our Fe-based MG scavengers achieve oxygen removal rates that are one to four orders of magnitude higher than conventional systems, while approaching the theoretical oxygen removal capacity limit. The combination of rapid kinetics, high efficiency, and mechanistic versatility establishes Fe-based MGs as a promising materials platform for next-generation oxygen scavenging and broader redox-related applications.

Methods

Materials

The flaky FeSiB and FeSiBNbCu amorphous powders used in the experiments were purchased from Qingdao Yunlu Advanced Materials Technology Co., Ltd. (China) and prepared using the airflow crushing method. The subglobular Fe₇₈Si₉B₁₃ (prepared by gas-water atomization) and spherical Fe_77.3Si_6.3B_13.9C_2.6 (prepared by gas atomization) amorphous powders were purchased from Chengdu Ketailong Alloy Co., Ltd. Pure iron powder was purchased from Hebei Chuangying Metal Materials Co., Ltd. (China) and prepared by the gas atomization method, with Fe content > 99.9 wt%. The reduced iron powder (composition: Fe_98.8Mn_0.4Cr_0.2Si_0.2S_0.2Ca_0.1Al_0.1, wt%) was secondary reduced iron powder, purchased from Tianjin Beichen District Fangzheng Reagent Factory (China). The diatomite (300 mesh, composition: (SiO₂)_92.5(Al₂O₃)_3.6(Na₂O)_2.7(Fe₂O₃)_1.2, wt%) used for the oxygen removal experiments was sourced from Henan Boxu Environmental Protection Technology Co., Ltd. (China). The water used in the experiments was obtained from an ultrapure water system, and the 1% NaCl saline solution was prepared using analytical-grade NaCl (China National Pharmaceutical Group Chemical Reagent Co., Ltd.) and ultrapure water.

In the preparation of FePC ribbons, pure iron (>99.95 wt%), ferrophosphorus (22 wt% P), and carbon particles (>99.9 wt%) were weighed according to the nominal composition and melted five times into button ingots using a non-consumable vacuum arc furnace under argon protection. The ingots were then cut and rapidly quenched into ribbons using a vacuum melt-spinner (copper roller speed: 40 m s⁻¹). The amorphous ribbons were vacuum-annealed (pressure <5 × 10⁻³ Pa) at 623 K for 1 h, then ground for 3 min using a commercial grinder and sieved through a 100-mesh screen.

Sample characterization

A Co-source diffractometer (D8 Advance X, Bruker, Germany) and a Cu-source diffractometer (SmartLab, Rigaku, Japan) were used to analyze the phase information of samples. The scanning rates were 10° min⁻¹ and 5° min⁻¹, respectively. The XRD results for the starting metal powders are shown in Supplementary Fig. 27a. HRTEM and selected-area electron diffraction of the FeSiB powder were performed using a Talos F200x TEM (FEI, USA), with representative HRTEM images shown in Supplementary Fig. 27b. The reaction products were added into 20 mL of anhydrous ethanol, and after filtration through a 0.22 μm filter membrane, the samples containing nano-iron oxide particles were subjected to TEM observation. SEM (GeminiSEM 300, ZEISS, Germany) equipped with an energy-dispersive spectrometer was used for sample observation and analysis. The poorly conductive powders were priorly sputter-coated with platinum. All the SEM observations were repeated three times, yielding similar results. Particle size distributions of the Fe-based MGs, pure iron, and reduced iron powders were measured using a laser particle size analyzer (Mastersizer 3000, Malvern, UK). The BET specific surface area of the samples was tested using a fully automated surface area and porosity analyzer (ASAP 2460, Micromeritics, USA). A Raman spectrometer (LabRAM HR Evolution, Horiba, Japan) and a Fourier transform infrared spectroscopy (FT-IR, RTracer-100, SHIMADZU, Japan) were used to analyze the reaction products. Prior to the FT-IR measurement, the samples were dried at 393 K for 5 min to remove moisture. Compositions on the oxide films of initial and reacted FeSiB samples were measured by an XPS (Scientific K-Alpha, Thermo Fisher, USA), with charge correction referenced to the C 1s peak at 284.8 eV. The XAFS spectra of reacted FeSiB and Fe samples were measured at the Energy Materials Beamline (hard branch) of the Shanghai Synchrotron Radiation Facility to analyze the oxidation degree and Fe–O coordination environment. For sample preparation, 5 g of the reacted scavenger was vacuum-dried and then milled in a planetary ball mill for 2 h using a ball-to-powder weight ratio of 6:1. ZrO₂ milling balls with a diameter of 5 mm were employed, and 8 mL of ethanol was used as the milling medium. After milling, the powder was dried at 393 K and subsequently sieved through a 400-mesh screen; the collected fine fraction was used for XAFS measurements.

Oxygen removal performance testing

Each oxygen removal experiment was conducted as follows: 0.8 g of diatomite was weighed and added to a 50 mL polypropylene tube, followed by 1 mL of ultrapure water or 1 wt% NaCl solution. The mixture was stirred for 1 min using a polystyrene rod to ensure uniform wetting. Fe-based powder (2 g for rate testing or 0.5 g for capacity testing) was then added and stirred for another minute to ensure thorough mixing. The diatomite dosage of 0.8 g was optimized through preliminary experiments. Routine experiments were conducted at 298 K, while low-temperature experiments were performed at 278 K in a temperature-controlled refrigerator. All measurements for each material were obtained from the same sample, with three samples included in each experimental group, and all experiments exhibited good reproducibility. In the long-term storage experiment, each prepared mixed sample consisting of 2 g of FeSiB powder, 0.8 g of diatomite, and 1 g of saline solution was sealed in a 10 mL centrifuge tube, which was subsequently stored in a vacuum bag. After 30 days, the samples were taken out and subjected to oxygen removal experiments.

The mixed powder was placed in a 70 mm glass Petri dish and immediately transferred to a transparent sealed chamber equipped with an oxygen detector (accuracy: 0.1% VOL, Nanjing Yice Technology Co., Ltd., China). A camera recorded the real-time change in oxygen concentration. When the reading reached 0, the actual oxygen concentration was < 0.05%, corresponding to an oxygen removal efficiency > 99.7%. The sealed chamber had a volume of 560 ± 10 mL for rate testing and 1260 ± 15 mL for capacity testing, measured by water displacement. The Petri dish and detector occupied ~60 ± 5 mL. The recorded videos were analyzed to extract oxygen concentration data over time, enabling evaluation of oxygen removal rate and capacity.

After rate testing, the reacted powders were cleaned by dispersing in 20 mL anhydrous ethanol in a 50 mL beaker, shaking for 20 s, and separating using a magnet. The cleaning process was repeated three times. The powders were then dried on filter paper on a 373 K hot plate for subsequent characterization. In one experiment, the ethanol suspension from washing the FeSiB reaction products was transferred to a 200 mL beaker and left to settle for 1 min. The supernatant was decanted, and the remaining powder was dried at 373 K, yielding iron oxide composite powders with residual diatomite. In the pre-reaction experiment of the FeSiB MG oxygen scavenger, the powder mass was increased from 2.0 to 2.1 g, with all other conditions identical to the standard saline tests. When the oxygen concentration dropped to 15.4%, approximately 0.1 g of FeSiB was fully oxidized, at which point the reaction vessel was opened to restore oxygen levels. To compensate for slight moisture loss during pre-reaction, water was replenished based on the measured weight reduction in the Petri dish before conducting subsequent oxygen removal tests.

DFT calculations

Our spin-polarized first-principles DFT calculations for pure Fe and FeSiB MG were carried out by the Vienna ab initio simulation package (VASP) version 5.4.4^48–51. The Perdew–Burke–Ernzerhof functional in the generalized gradient approximation was employed to describe the exchange-correlation functional⁵², and the projector augmented-wave method with a plane wave basis set was used to model the core-valence electron interactions with an energy cutoff of 500 eV for both calculations. Due to the relatively large simulation cells, the k-points sampling of the first Brillouin zone was carried out using a 2 × 2 × 1 Gamma-centered mesh. To accurately treat the 3d orbitals of Fe atoms, we applied the empirical simplified (rotationally invariant) DFT + U approach with an effective U value (U_eff) of 4.3 eV^53–55. The DFT-D3 corrections were used to account for van der Waals interaction⁵⁶. Electronic self-consistency was achieved when the energy change was less than 10⁻⁵ eV. For geometry optimization, the systems were relaxed until the norms of all the forces were smaller than 0.05 eV·Å⁻¹. The BCC Fe slab with the (110) surface exposed was modeled as a BCC structure with 4 layers, and each layer contains 3 × 4 supercells with 2 atoms per supercell (24 atoms per layer). While the Fe-based metallic glass slab with 75 Fe atoms, 12 B atoms, and 9 Si atoms was generated using ab initio molecular dynamics method with NVT ensembles of Nose-Hoover thermostats under 1600 K for 2 ps with a timestep of 2 fs. A vacuum layer of 12 Å in the z direction was added to the slabs to prevent interactions between the neighboring slabs, and two H₂O molecules were attached to the surfaces of the slabs in both calculations. Then, one O₂ molecule or two O atoms were attached to the surface as the initial state model or final state model in both calculations. The structures of these models were then optimized by the conjugate gradient method. The path and the energy barriers of the O₂ adsorption were then calculated using the CI-NEB method, which had been implemented in the VASP transition state tool^57,58.

The Fe d-band center (ε_d) was defined as Eq. (2):

$${\epsilon }_d=\frac{{\int }_{-\infty }^{\infty }PDOS\left(\epsilon \right)\cdot \epsilon \cdot d\epsilon }{{\int }_{-\infty }^{\infty }PDOS\left(\epsilon \right)\cdot d\epsilon }$$ 2

where PDOS(ε) is the PDOS of Fe d orbitals and ε is the energy.

Author contributions

J.S. conceived the idea. J.S., G.X., J. Lu, and K.Y. supervised the research and established the research route. J.S. and H. Liu. carried out the main the experiments. H. Luan designed the DFT calculations. Y.L. optimized the calculations. J.S., H. Luan, Y.S., G.X., and J. Lv analyzed the data. J.S., H. Luan, and Y.S. prepared the paper. All authors discussed the results. J.S. and H. Luan contributed equally to this work.

Peer review

Peer review information

Nature Communications thanks Baran Sarac and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Source data are provided with this paper (ref. ⁵⁹). All the raw data relevant to the study are available from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71713-4.