Introduction

The presence of anthropogenic organic pollutants in water resources has become a global concern, posing significant threats to the natural environment and human health1,2. Advanced oxidation processes (AOPs) are considered a viable water treatment option, as they target transforming organic pollutants to benign products by employing highly oxidative radicals such as hydroxyl radical (•OH)3,4,5. Among various catalytic schemes to achieve AOPs, membranes loaded with catalysts that activate AOP precursor such as hydrogen peroxide (H2O2) have been extensively pursued as a practical approach at the device level. These catalytic membranes are designed to function both as a membrane–rejecting larger organics and colloids via size exclusion–and as a catalyst–catalytically destroying smaller pollutants that pass through membrane pores6.

Catalysts with various morphologies, such as single-atom, metal-organic framework, and covalent organic framework, have been explored as active components of catalytic membranes7,8. While some catalytic membranes show great promise for high-efficiency water treatment, one widely acknowledged limitation is the challenge of balancing initial high catalytic activity with long-term catalyst stability. Catalyst deactivation becomes more pronounced when the catalytic scheme primarily targets producing a large amount of •OH for rapid pollutant oxidation. In addition to reaching target micropollutants, which typically occur at ppt-ppb levels9,10,11, •OH ( < 10 μs) can react adversely with the catalysts themseleves12,13, compromising system longevity and producing unintended byproducts.

Developing AOP catalysts that are both highly reactive and stable over long period presents a dilemma. Constructing protective composite structures by introducing additional chemically robust materials14,15,16 (e.g., Fe@Fe2O3 with a stable Fe2O3 outer layer17) would inevitably lower the catalysis kinetics. The same challenge emerges when milder reactive oxygen species (ROS) substitutes3, such as singlet oxygen (1O2), are pursued as the main oxidant. Circumventing the problem related to oxidants and employing a completely different mechanism, such as non-radical pathway18 (e.g., phenol polymerization on FeOCl19), can be applied to niche applications, but deviates from the original objective of AOP in non-selectively removing a wide range of pollutants20.

In this study, we propose spatial confinement as an innovative strategy to enhance the long-term efficacy of AOP catalysts in catalytic membranes. We employ single-layer graphene oxide as a flexible matrix to confine iron oxyfluoride (FeOF)—one of the most efficient heterogeneous Fenton catalysts reported to date—and form a catalytic membrane with an aligned layer structure. We demonstrate the catalytic reaction in angstrom-scale confined spaces ( < 1 nm) within the membrane channels significantly improve both the durability and activity of the confined FeOF. We discuss the key reasons for FeOF deactivation during AOP and the mechanism by which spatial confinement enhances the catalyst stability while preserving catalytic efficiency.

Results

Synthesis of representative iron oxyhalides

The selection of catalysts plays a critical role in the performance of catalytic membranes. Among readily available options of AOP catalysts, iron oxyhalides have gathered significant attention due to their exceptional catalytic efficacy. Studies have shown that FeOCl and FeOF outperform the other conventional catalysts in both powder suspensions and catalytic membranes for pollutant degradation21,22. However, they have been criticized for offering remarkable activation efficiency only in the short term but suffering from relatively poor material durability21,23,24,25,26.

To verify the (in)stability of iron oxyhalides, we fabricated FeOCl by pyrolyzing FeCl3·6H2O at 220 °C for 2 h in a muffle furnace21, and FeOF by heating FeF3·3H2O in methanol medium at 220 °C for 24 h in an autoclave (Supplementary Fig. 1). The X-ray diffraction (XRD) patterns confirmed alignments of both FeOCl and FeOF with their PDF cards retrieved from JCPDS database (Fig. 1a). The inverse Fourier transformation of the selected transmission electron microscopy (TEM) region revealed that the primary exposed crystalline plane of the synthesized FeOF was assigned to its (110) plane (Fig. 1b), consistent with the result reported in previous report27. Apart from the primary (010) plane obtained from XRD pattern, the fabricated FeOCl exhibited hybrid crystalline planes on the surface (Supplementary Fig. 2), including (110), (120), (121), etc. Both materials displayed a layered morphology which was corroborated by scanning electron microscopy (SEM) and TEM images (Supplementary Figs. 3-4). The original FeOCl and FeOF were fully digested (Supplementary Table 1 and Supplementary Text 1) to determine their accurate chemical compositions using inductively coupled plasma atomic emission spectroscopy (ICP-OES) for Fe and ion chromatography (IC) for halogens, which were Fe1.14OCl1.17 and Fe1.75OF3.45, respectively. It is important to determine whether additional ions were adsorbed on the material surface during synthesis, as these could potentially influence the subsequent catalytic processes. To unravel this, we also performed an acid-washing treatment to the freshly synthesized materials (Supplementary Table 1). The close elemental ratios before and after acid treatment indicate that the discrepancy between the theoretical atomic ratios and the measured values (from total digestion) is primarily attributable to the intrinsic material defects, rather than the surface-adsorbed species.

Fig. 1: Catalytic performance of iron oxyhalides in heterogenous Fenton systems and key factor in material deactivation.

a XRD patterns of iron oxyhalides. b SEM, SEM-EDS mapping, TEM, HAADF-TEM images, and inverse Fourier transformation of the selected region (6.252 nm2) in TEM for the FeOF. c Comparison of H2O2 activation efficiency for different catalysts from EPR ([catalyst] = 1 g L–1, [H2O2] = 50 mM, [DMPO] = 200 mM, pH = 6.2 ± 0.1). d DMPO–OH signal generated by different catalysts ([catalyst] = 1 g L–1, [H2O2] = 50 mM, [DMPO] = 200 mM, pH = 6.2 ± 0.1). e Removal efficiency for representative neonicotinoid ([catalyst] = 1 g L–1, [H2O2] = 10 mM, [THI] = 10 μM, pH = 6.2 ± 0.1, reaction time = 60 min, standard deviations (n = 3) were presented). f F 1 s XPS spectra of FeOF before and after H2O2 activation. g H2O2 activation efficiency of spent FeOF after reacting with H2O2 for a certain period ([catalyst] = 1 g L–1, [H2O2] = 10 mM, contact time = 1, 2, 3 or 4 h, [DMPO] = 200 mM, pH = 6.2 ± 0.1, reaction time = 1 min). h Element leaching over time in suspension systems ([catalyst] = 1 g L–1, [H2O2] = 10 mM, pH = 6.2 ± 0.1).

The H2O2 activation efficiency of the catalysts was evaluated using electron paramagnetic resonance (EPR) spectroscopy with 5,5-dimethyl-1-pyrroline N-oxide (DMPO) as the spin trapping agent (Supplementary Fig. 5 and Supplementary Text 2). The synthesized layered FeOF demonstrated the highest radical generation efficiency compared to benchmark iron-based catalysts, as well as FeOCl (Fig. 1c). The activation efficiency can be quantitatively compared by spin concentration of the generated DMPO–OH28. In accordance with the results in the literature22,29, both FeOCl and FeOF showed significantly superior catalytic performance over the benchmark catalysts, with FeOF surpassing FeOCl by 4.7 times in DMPO–OH signal intensity (Fig. 1d).

However, after recovering the catalysts by filtration and vacuum drying, the catalytic performance of second-run FeOCl and FeOF weakened markedly. The signal intensity of spent FeOCl and FeOF decreased by 67.1% and 70.7%, respectively, compared to their fresh forms. This trend was also observed in the removal rate of thiamethoxam (THI), one of neonicotinoids which are the most common insecticide pollution in the global groundwater30,31,32. The potency of FeOCl and FeOF showed a severe reduction of 77.2% and 75.3%, respectively, in the second run (Fig. 1e). The results imply that the powder-form iron oxyhalides would undergo deactivation during the catalytic oxidation in bulk suspension23. Although the FeOF led in •OH generation efficiency, its THI degradation did not transcend that of the FeOCl with a same level, possibly ascribed to the rapid quenching of •OH in bulk reaction.

Significant halide leaching during catalytic oxidation

To further investigate this deactivation process, the changes in material surface before and after catalytic oxidation were examined using X-ray photoelectron spectroscopy (XPS). Results indicate that FeOF lost a significant fraction of both F (40.2 at.%) and Fe (33.0 at.%) during H2O2 activation (Fig. 1f and Supplementary Fig. 6). SEM and TEM images revealed that the leaching of elements led to a corroded morphology (Supplementary Fig. 7). FeOCl showed even more pronounced leaching degree (Cl 76.1 at.%, Fe 43.2 at.%), possibly due to the lower electronegativity of Cl (3.16 on the Pauling scale), which decreased its coordination strength to the Fe core23 (Supplementary Fig. 8). After the reaction, XPS peaks assigned to halogens shifted to higher binding energies (FeOF +0.43 eV, FeOCl +0.18 eV), possibly indicating the crucial role of halogens as critical species in H2O2 activation. The efficiency of •OH generation also strongly correlated with the remaining surface halogen content in the iron oxyhalides (R² = 0.97–0.99), suggesting that halogen loss was probably the decisive factor in catalyst deactivation (Fig. 1g and Supplementary Fig. 9).

Heterogenous catalysts applied in catalytic oxidation are commonly reported to proceed through peroxide adsorption as the first reaction step3,33. However, there is limited research that comprehensively studied how the peroxide adsorption behaviors correlate with element leaching. Therefore, we monitored the elements leaching of Fe and halides over time using ICP-OES and IC. We calculated that the halide contents continuously leached into the solution over time, resulting in a loss of 93.5% Cl for FeOCl and 40.7% F for FeOF after a 12-hour reaction with H2O2 (Fig. 1h). Under the severe halogen loss, compositions of the spent FeOCl and FeOF changed to Fe0.47OCl0 and Fe0.85OF1.18, respectively. In contrast, the leached Fe amount for both iron oxyhalides increased in the initial 2 h but plateaued during the later phase of reaction. The overall Fe leaching amounts for both FeOF and FeOCl were not significant during reaction (Supplementary Table 1), and the leaching of both Fe and halides was negligible without the presence of H2O2. The results indicate the important role of halogens in catalytic reaction, and challenges the conventional understanding that the deactivation of catalysts is mainly due to leaching or overoxidation of metal elements34. The H2O2 consumption was also measured during leaching experiments (Supplementary Fig. 10 and Supplementary Text 3), and the results indicate that the FeOF exhibited a 2.3 times higher H2O2 activation rate compared to FeOCl. A higher DMPO–OH generation in initial stage (1 min) and a faster reaction rate over the long run (6 h) both suggest the superior capability of FeOF in catalytic oxidation.

Irreversible halide loss during bulk catalysis

The leaching-to-performance correlation was different for FeOCl and FeOF during reaction, where FeOF showed a lower leaching but a better H2O2 activation capability than FeOCl. To verify whether halogen-rich environment would impact the catalytic process, 0.5 M F− or Cl− was added during the H2O2 activation to intentionally increase the chance of halogen recombination with the Fe core. In the second run, the F-treated spent FeOF exhibited significantly better catalytic performance in THI removal (76.9% in 1 h) compared to the untreated spent FeOF (18.6%), approaching the performance of the fresh FeOF (83.5%) (Fig. 2a). Similarly, the Cl-treated spent FeOCl was also improved with a higher catalytic performance (26.8%) than the untreated spent FeOCl (14.6%), although less close to the fresh FeOCl (64.1%) compared with its counterpart. Different types of halogens failed to recover the catalytic performance of iron oxyhalides. Furthermore, the spent FeOF was placed in different levels of F− ion (0.1–0.5 M F−) to activate H2O2, and the resultant EPR signals showed a positive correlation with the ambient F− concentration (Fig. 2b). Similarly, the performance of spent FeOF in THI degradation increased under high concentration of F− (68.2% with 0.5 M F−). These results suggest that either mitigating halide diffusion or ensuring timely recovery of halide to the catalyst surface is the key to maintaining the activity of iron oxyhalides. Probe sonication was performed to exfoliate the fresh FeOF, which artificially reduced its F−-holding capacity by disrupting the intrinsic F−-bridged layered structure and introducing additional structural defects. The markedly decreased catalytic performance ( < 10%) and the reduced fluorine content (from 55.6% to 12.5%, Supplementary Table 1) further supported the correlation between fluorine retention and catalytic efficacy.

Fig. 2: Mechanisms of conventional iron oxyhalide Fenton systems and their structural properties.

a THI degradation by iron oxyhalides, their spent forms, and their halide-treated spent forms ([catalyst] = 1 g L–1, [H2O2] = 10 mM, reaction time = 1 h, [Cl– or F–] = 0.5 M, [THI] = 10 μM, pH = 6.2 ± 0.1, standard deviations (n = 3) were presented). b H2O2 activation efficiency and THI degradation of spent FeOF under F-rich environment ([catalyst] = 1 g L–1, [H2O2] = 10 mM, [F–] = 0.1, 0.2 or 0.5 M, [THI] = 10 μM for reaction time of 60 min, [DMPO] = 200 mM for reaction time of 1 min, pH = 6.2 ± 0.1, standard deviations (n = 3) were presented). c Theoretical bond length and Bader charge in iron oxyhalides, where the higher electronegativity of fluoride reflects its stronger nucleophilic nature to complete the redox cycle. d Fourier transform of k**2-weighted EXAFS spectra of iron oxyhalides. The structures of iron oxyhalides match with that of Fe2O3, and the spectra of other reference materials could refer to Supplementary Fig. 14. e Schematic illustration of the reaction mechanism in FeOF-based Fenton system.

To evaluate the affinity of iron oxyhalides for uncoordinated halide ions, we pre-saturated fresh FeOCl and FeOF in a halide-rich solution (0.5 M) overnight to allow for potential adsorption of free halide species. The halide-saturated materials were then filtered and immediately subjected to catalytic testing. As shown in Supplementary Fig. 11, no significant change in catalytic performance was observed. This suggests a low adsorptive affinity of FeOCl and FeOF toward uncoordinated halide ions, likely due to their highly crystalline nature and limited surface area (15.46 m2 g–1 for FeOCl and 3.19 m2 g–1 for FeOF; Supplementary Table 2), which restricts the availability of physisorption sites capable of retaining free halide ions.

We employed density functional theory (DFT) calculations, based on the VASP package, to evaluate structures of the iron oxyhalides with the surfaces in lowest energy (Fig. 2c and Supplementary Fig. 12). Both materials exhibit an octahedral structure35,36, with an Fe core coordinated with halide and oxygen atoms. The structures were confirmed by X-ray absorption spectroscopy (XAS, Supplementary Table 3). Both X-ray absorption near edge structure (XANES, Supplementary Fig. 13) and extended X-ray absorption fine structure (EXAFS, Supplementary Fig. 14) data also indicate that Fe sites in the FeOF and FeOCl resemble those in Fe2O3, which are octahedrally coordinated. Fourier transform of k2-weighted EXAFS spectra of iron oxyhalides clearly suggested a higher affinity of fluorine toward Fe core than chlorine due to a stronger signal intensity of Fe–F bond (Fig. 2d and Supplementary Fig. 15).

While both FeOF and FeOCl have bulk octahedral Fe cores coordinated with halide and oxygen atoms, the shorter Fe–F bonds in FeOF (2.1 Å) relative to the Fe–Cl bonds in FeOCl (2.4 Å) indicate that F atoms are more strongly held, and thus less susceptible to displacement under chemical action. Bader charge analysis (Supplementary Text 4) also reveals that the F atoms in FeOF have a higher charge (0.66 e−) at octahedral Fe sites than those in FeOCl (0.54 e−), corresponding to the greater electronegativity of fluorine than chlorine.

The high efficiency of iron oxyhalides in activating H2O2 originates, in theory, from the high electronegativity of halides to attract and deliver electrons from Fe cores. The catalytic process begins with the substitution of halides by H2O2, followed by a cascade of reactions that generate •OH, with the expectation that the detached halides reoccupy the active sites to complete the redox cycle22 (Fig. 2e). Fluorine, due to its highest electronegativity among all elements (3.98 on the Pauling scale), shows a stronger affinity to reclaim the occupied sites than chlorine. This strong electroactive nature has been applied in other fields, such as making high-performance electrodes or high-energy density batteries37,38.

To support our hypothesis, we conducted simulations involving H2O2 reacting with FeOCl or FeOF: (a) H2O2 substituting halogen atoms, (b) H2O2 filling halide vacancies, and (c) regeneration of halogen sites. The results showed that FeOF exhibits more negative reaction energies in both the halogen substitution (−1.59 eV) and vacancy filling (−1.04 eV) processes, indicating a stronger affinity toward H2O2 and a greater tendency to initiate catalytic reactions compared to FeOCl. Additionally, the lower energy barrier for halogen substitution than vacancy filling suggests that the F sites coordinated with Fe are the dominant active sites in the catalytic process. Moreover, the regeneration of halogen sites is essential for maintaining catalytic continuity and material stability. Our calculation revealed that the energy for F reoccupying a vacant site in FeOF is highly exergonic (−2.74 eV), further demonstrating that the released F ions have a strong tendency to return and restore the catalytic structure. This energetically favorable regeneration explains why the F atoms in FeOF are more preferably retained during reaction (Supplementary Fig. 16).

To validate the proposed reaction pathway involving fluorine release during the interaction of FeOF with H2O2, we calculated the reaction energies for three distinct mechanistic routes: (a) direct conversion of H2O2 to •OH on FeOF without fluorine release, (b) conversion of H2O2 to •OH on FeOF involving F− regeneration and formation of an intermediate •OOH species, and (c) conversion of H2O2 to •OH on FeOF involving F− regeneration without the formation of a •OOH intermediate. As shown in the reaction energy diagram (Supplementary Fig. 17), pathway b exhibits the most favorable energetics, with a significantly exothermic profile across all reaction steps. This result supports our hypothesis that the interim release of F− is critical for the catalysis.

While for catalytic reactions occurring in bulk solution with a high concentration of H2O2 (typically 1–10 mM) involved, the critical F species does not necessarily reclaim the Fe sites before diffusing away or participating in a side reaction. The above results collectively inspired us to implement a promising solution—leveraging the spatial confinement enabled by a membrane structure to facilitate the longevity of the catalytic system.

Spatial confinement with graphene oxide layers

Increasing the concentration of F− in real-world water treatment is impractical, particularly with respect to drinking water treatment39. Furthermore, the low affinity of pristine FeOF for uncoordinated fluorine ions limits its potential for functionalization. Alternatively, our approach focused on confining the two-dimensional FeOF within a suitable matrix to limit fluorine diffusion away from the Fe centers40,41,42,43. Graphene oxide layers, with a theoretical interlayer spacing less than 1 nm44, emerged as a suitable structure to spatially confine alien materials within their highly flexible matrix45,46. The graphene oxide sheets were expected to serve as a structurally flexible confinement matrix, effectively intercalating with the layered FeOF to form confined regions.

Single layer graphene oxide was produced by exfoliating commercial graphene oxide paste under high-energy sonication (Supplementary Text 5). SEM images showed a layered morphology of the exfoliated graphene oxide (Supplementary Fig. 18). The thickness of the exfoliated graphene oxide plus interlayer spacing was measured to be approximately 1.3 nm by atomic force microscopy (AFM), suggesting a typical single layer feature (Supplementary Fig. 19). The thickness of a single graphene oxide layer (excluding the space between graphene oxide and substrate during AFM test) ranges from 0.5 to 0.8 nm47,48, indicating that the interlayer spacing between our exfoliated graphene oxide layers would fall within angstrom-scale range ( < 1 nm) when forming a membrane.

By adjusting addition order and ratio of the FeOF and the single layer graphene oxide, we manufactured three different membranes via vacuum filtration, termed as GO (without FeOF), FeOF/GO (FeOF embedded within the substrate), and FeOF topping (FeOF deposited on top of the substrate) (Fig. 3a). The FeOF content in the FeOF/GO and the FeOF topping membranes was approximately 45.1 wt.% (or 2.5 g FeOF per m2 graphene oxide), as was calculated by weighing mass of the freeze-dried samples and material total digestion. The FeOF content was also determined to be 46.0 wt.% using thermogravimetric analysis (TGA, Supplementary Fig. 20).

Fig. 3: Construction and characterization of the FeOF membrane enhanced by spatial confinement.

a Schematic illustration of the membrane synthetic protocol. b Cross-sectional FIB-SEM images of the membranes. After the ion beaming, the FeOF was exhausted to leave voids while graphene oxide layers were slightly melted to form a compact surface, featured as a clean cut. c Cross-sectional SEM and SEM-EDS mapping images of the membranes. After the liquid nitrogen cracking, the FeOF between graphene oxide layers could be preserved and the uneven surface favored the observation of layer-by-layer structure, featured as an uneven cracking. d XRD patterns of the membranes. e 2D/3D AFM images of the membranes, where mica was adopted as the substrate (inset: top-view SEM image showing confined FeOF within graphene oxide matrix).

The inclusion of FeOF resulted in an increased membrane roughness (Rmax = 1394–1420 nm compared to 719 nm, Supplementary Fig. 21). High-energy ion beam imaging via focused ion beam-scanning electron microscopy (FIB-SEM) was used to profile the space where the FeOF intercalated with the graphene oxide (Fig. 3b and Supplementary Fig. 22). The cross-section of the GO membrane showed a smooth and compact morphology, while the FeOF/GO membrane possessed a high density of horizontal voids. When the FeOF dosages were reduced gradually (0.8, 0.5, 0.2, and 0.1 of the original dosage, termed as 0.1FeOF/GO–0.8FeOF/GO), the resultant FeOF/GO membranes exhibited gradually reduced voids density (Supplementary Fig. 23). The results corroborated the horizontal laminate structure of the membrane formed by FeOF and graphene oxide layers. It was implied that water would flow through these horizontal channels in between layers, which promises good contact between reactants and catalysts. Given that the tilting angle was fixed at 52° in all FIB-SEM tests (Supplementary Text 6), the thickness of GO and FeOF/GO membranes could be accurately calculated at 2.597 and 2.662 μm (Supplementary Fig. 24), respectively.

The cross-sectional surface created by the ion beam in FIB-SEM would be a “clean cut” which could revealed the laminate channels. The intercalating morphology was further exposed through the “uneven cracking” in SEM imaging performed after liquid nitrogen-induced cracking of the membranes (Fig. 3c and Supplementary Text 7). The addition of FeOF did not significantly alter the structural integrity, and the two-dimensional FeOF with approximately 45 nm in average thickness was compatibly confined within the graphene oxide matrix.

XRD patterns revealed an angstrom-scale interlayer spacing of 7.89 Å from the GO membrane49 (Fig. 3d), based on Bragg’s law. The original exfoliated graphene oxide layers displayed a weak intensity at this position (zoomed in by 100 times in Supplementary Fig. 25), further corroborating the single layer nature of the exfoliated graphene oxide. After introducing the FeOF, a new characteristic peak at 9.04° emerged while the peak associated with interlayer spacing of graphene oxide weakened. This new peak probably corresponded to the interlayer spacing between the FeOF and graphene oxide layers, calculated to be 9.77 Å. The XRD patterns of the FeOF/GO membranes with reduced FeOF dosages (0.1FeOF/GO–0.8FeOF/GO) exhibited gradually reduced intensity of this peak (Supplementary Fig. 26). For the FeOF topping membrane, where the least contact between the FeOF and graphene oxide layers was involved, the peak at this position still existed but showed the weakest intensity. 3D AFM images further confirmed the angstrom-scale spacing between FeOF and graphene oxide (49.3 nm vs. 50.4 nm, Fig. 3e). In past research, 20 nm was identified as the critical threshold to trigger the spatial confinement effect in H2O2 activation6. The angstrom-scale interlayer spacing ( < 1 nm) between the FeOF and graphene oxide layers favorably meets this requirement, confirming the successful construction of an angstrom-scale reaction vessel for the following heterogeneous Fenton reactions.

Long-lasting catalytic oxidation under angstrom-scale confinement

A flow-through membrane cell was employed to evaluate the applicability of the catalytic membranes (Fig. 4a). The FeOF/GO membrane exhibited greater hydrophobicity compared to the GO membrane in sessile drop experiments, corresponding to an improved water permeability (Fig. 4b). This enhancement is attributed to the inclusion of hydrophobic FeOF surface, which offers less resistance to water molecules when passing through the channels (Supplementary Fig. 27). This effect has been previously reported that the water molecules incline to permeate through a tortuous path primarily along the introduced hydrophobic surfaces rather than the hydrophilic surfaces of graphene oxide50.

Fig. 4: Set-up, properties, and catalytic performance of membrane systems in neonicotinoid degradation.

a Schematic diagram of the membrane setup. b Hydrophobicity test conducted via contact angle measurement. The first frame after the water droplet reached membranes was used for analysis. c Membrane specific surface area and porosity determined by BET. d Raman spectra for the membranes. e Permeability as a function of pressure. f THI purification and DMPO–OH generation in membrane systems during the first 8 h run ([H2O2] = 10 mM, [THI] = 10 μM, pH = 6.2 ± 0.1, pressure = 2 bar, standard deviations (n = 3) were presented). g Rate constants for THI purification in different membrane systems where standard deviations (n = 3) were presented.

To further understand the angstrom-scale reaction spaces, textural properties of the membranes were determined via N2 adsorption-desorption isotherms (Supplementary Table 2 and Supplementary Fig. 28). The inclusion of FeOF slightly influenced the specific surface area of membrane, which was reduced from 7.10 to 6.41 m2 g–1, while it generated a significantly higher density of micropores ( < 2 nm) which was 4.2 times greater than that of original GO membrane (Fig. 4c). The increased micropores were attributed to the higher density of defects generated by the introduced FeOF bending the graphene oxide layers. The defective levels were quantified by Raman spectroscopy, where the FeOF/GO membrane possessed a higher D band (1348 cm–1) to G band (1593 cm–1) ratio (ID/IG) at 1.71, compared with that of the GO membrane at 0.77 (Fig. 4d). The 0.1FeOF/GO–0.8FeOF/GO membranes also exhibited gradually reducing ID/IG values from 1.37 to 0.93, corroborating that the generation of micro-sized defects was caused by the FeOF intercalation (Supplementary Fig. 29). Notably, previous membrane studies often introduced alien materials into membrane substrates to generate defects as a convenient method to increase water flux51,52. This effect was also reflected in the increased water permeability of FeOF/GO membrane. The water flux could be increased to 5.8 LMH bar–1 (Fig. 4e), classifying it as a nanofiltration membrane which typically exhibits a water flux range of 4–10 LMH bar–1. In comparison, the GO membrane displayed a lower water flux of 2.3 LMH bar–1. The permeability decreased to 1.8 LMH bar–1 for FeOF topping membrane due to the blockage of surface channels.

The swelling of graphene oxide layers—primarily induced by water intercalation—can increase the interlayer spacing from approximately 0.8 nm in the dry state to ~1.2–1.4 nm when hydrated53. To investigate the swelling behavior of graphene oxide, we further measured the XRD patterns of pre-wetted GO and FeOF/GO membranes (Supplementary Fig. 30). The results confirm that the peak corresponding to the interlayer spacing of graphene oxide shifted to a lower 2θ angle upon wetting, indicating an increase in spacing from 7.98 Å to 12.1 Å. In contrast, the peak corresponding to the interlayer spacing between FeOF and graphene oxide exhibited only a slight shift, with the spacing increasing from 9.77 Å to 10.1 Å. This smaller change may be attributed to the hydrophobic nature of the FeOF surface, which likely retains fewer water molecules and thereby limits the extent of swelling in the confined channels. Additionally, the applied pressure ( > 2 bar in this study) can effectively mitigate this swelling to some extent54, helping to maintain narrower interlayer spacings and enhance membrane performance in long-term applications.

Volumes of water channels in the membranes were measured by water saturation test (Supplementary Text 8 and Supplementary Fig. 31). The results indicate that the water flow indeed preferentially passed through spaces or channels with lower resistance—those containing hydrophobic FeOF surfaces—despite