Bilayer electrified-membrane with pair-atom tin catalysts for near-complete conversion of low concentration nitrate to dinitrogen

Abstract

Discharge of wastewater containing nitrate (NO₃⁻) disrupts aquatic ecosystems even at low concentrations. However, selective and rapid reduction of NO₃⁻ at low concentration to dinitrogen (N₂) is technically challenging. Here, we present an electrified membrane (EM) loaded with Sn pair-atom catalysts for highly efficient NO₃⁻ reduction to N₂ in a single-pass electrofiltration. The pair-atom design facilitates coupling of adsorbed N intermediates on adjacent Sn atoms to enhance N₂ selectivity, which is challenging with conventional fully-isolated single-atom catalyst design. The EM ensures sufficient exposure of the catalysts and intensifies the catalyst interaction with NO₃⁻ through mass transfer enhancement to provide more N intermediates for N₂ coupling. We further develop a reduced titanium dioxide EM as the anode to generate free chlorines for fully oxidizing the residual ammonia (<1 mg-N L⁻¹) to N₂. The sequential cathode-to-anode electrofiltration realizes near-complete removal of 10 mg-N L⁻¹ NO₃⁻ and ~100% N₂ selectivity with a water resident time on the order of seconds. Our findings advance the single-atom catalyst design for NO₃⁻ reduction and provide a practical solution for NO₃⁻ contamination at low concentrations.

Introduction

Nitrate (NO₃⁻) contamination resulting from the discharge of inadequately treated wastewater disrupts aquatic ecosystems through eutrophication followed by harmful algal blooms¹. The algal blooms can be triggered by NO₃⁻ at a concentration as low as 1 mg-N L^−1 2, therefore, underscoring the need for water treatment systems to remove NO₃⁻ with low concentrations³. However, near-complete removal of NO₃⁻ at low concentrations in complex water matrices is difficult and costly. Electrocatalytic NO₃⁻ reduction reaction (NO₃RR) to dinitrogen (N₂) or ammonia (NH₃) offers a promising alternative to be applied for NO₃⁻ removal from water. Electrochemical processes offer advantages including rapid reaction kinetics, adaptability to various treatment scenarios, the potential to be modularized for decentralized application, and minimal generation of secondary wastes such as sludges or concentrates^4,5.

Numerous electrocatalysts have been developed for NO₃RR towards NH₃ instead of N₂ with the goal of resource recovery^6,7,8,9. However, the practicability of NH₃ recovery from low concentration NO₃⁻ remains questionable, considering the complexity and high cost of separating dilute NH₃ from water¹⁰. It is noteworthy that NH₃ can be even more toxic compared to NO₃⁻¹¹, posing higher environmental risks. Therefore, developing NO₃RR systems that can completely convert low-concentration NO₃⁻ to N₂ with close to 100% selectivity is yet to be developed. Conventionally, NO₃⁻ reduction to N₂ requires a precious noble metal^12,13 in combination with a promoter metal (e.g., Cu) to improve the adsorption and charge transfer of the reactants and intermediates. However, the limited availability and high cost of the noble metals restrict their application for NO₃RR¹⁴.

Single-atom catalysts (SACs) have emerged as a transformative alternative in catalyst material design. With close to 100% atomic efficiency, SACs expose more active surface sites than nanoparticle counterparts, offering unparalleled efficiency and selectivity in various catalytic schemes¹⁵. The strong interaction of SACs with the support via surface defects and N-atomic anchors also offers enhanced chemical stability^16,17,18. Despite these advantages, the spatial isolation of catalytic sites in SAC architecture is not considered ideal for NO₃RR toward N₂. With fully isolated SACs, the *N intermediate generated on each SAC site cannot easily interact with another *N to undergo the *N-*N coupling step, which is critical for N₂ generation^8,19,20. This is the reason that most reported SACs are prone to promote the NH₃ selectivity through *N-*H coupling^{21,22,23,24,25}. (Fig. 1, left panel).

Fig. 1: Design principles of this work.

Schematic illustrating the design principles of incorporating single-atom catalysts in traditional plate electrode (left panel) or pair-atom catalysts in an electrified membrane for realizing near-complete conversion of low concentration nitrate to dinitrogen (right panel).

In this study, we propose a viable SAC architecture, pairs of tin (Sn) single atoms, to enhance the NO₃RR selectivity toward N₂ production (Fig. 1, right panel). Closely-spaced dual active sites intend to facilitate *N-*N coupling that are favored in multi-atom catalytic sites²⁶. Sn is selected because of its rapid NO₃RR kinetics and high N₂ selectivity^27,28. We incorporate Sn pair-atoms in an innovative electrified membrane (EM) to ensure their exposure in the membrane nanopores. The EM intensifies the catalyst interaction with NO₃⁻ during flow-through operation^29,30,31,32, overcoming the mass transfer limitation of the conventional plate electrodes^33,34 and providing more *N sources for N₂ coupling^35,36,37. We demonstrate highly efficient conversion of low concentration NO₃⁻ (10 mg-N L⁻¹) to N₂ in a single-pass filtration through the Sn pair-atom-loaded EM. We further incorporate a reduced titanium dioxide (TiO_2-x) EM anodic bottom layer to generate free chlorines from chloride ion (Cl⁻) in water to fully oxidize the residual ammonia (NH₃) downstream (<1 mg-N L⁻¹) into N₂.

Results

Synthesis and characterization of Sn single-atom catalysts

We loaded Sn single atoms on carbon black (CB) support using a ligand-mediated method (Fig. 2a)³⁸. In brief, the Sn⁴⁺ complex with 1,10-phenanthroline (Phen) at a molar ratio of Sn⁴⁺:Phen = 1:3 was adsorbed onto CB at target Sn loading of 1 and 5 wt%. Subsequent annealing in Ar atmosphere at 600 °C for 3 h produced N-doped carbon black (NCB)-supported Sn SACs. Transmission electron microscopy (TEM) images showed the porous structure of NCB and confirmed the absence of nanoparticles (Fig. S1). Aberration-corrected high-angle annular dark field-scanning transmission electron microscopy (AC HAADF-STEM) images suggested the atomic dispersion of Sn atoms on NCB. At the target Sn loading of 5%, we predominantly observed pair-atom sites (i.e., two single atoms placed close to each other, denoted as Sn₂/NCB, Fig. 2b). In contrast, fully-isolated Sn single-atom sites became more dominant over pair-atom sites at 1 wt% Sn loading (denoted as Sn₁/NCB) (Fig. 2c). The formation of Sn pair-atoms may attribute to the generation of Phen-Sn⁴⁺ precursor binuclear complexes in a distorted trigonal bipyramidal configuration in the solution^39,40,41. In addition, increasing the density of single atoms on the substrate to tune their site-to-site distance can create correlating atomic interactions between neighboring single atom sites^42,43. When the dispersed single atoms are close enough in catalysts with a M₁–N–C structure (M₁ represents metal single atom), a strong interaction could occur between the two adjacent single atoms to form a double active center to improve the catalytic performance⁴³. However, increasing only the density of single atoms to create pair-atom sites cannot precisely control the coordination configuration and distribution uniformity compared to methods such as the metal binuclear precursor deposition^44,45. Therefore, both Sn pair-atom and single-atom sites were observed, yet the composition of the former was dominant with the increase of Sn loading. The energy dispersion spectroscopy (EDS) mapping of Sn₂/NCB confirmed the uniform dispersion of Sn on the NCB support (Fig. 2d). Statistical image analysis determined the averaged distance between Sn atoms in Sn₂/NCB at d = 4.8 ± 0.7 Å (Fig. 2e–g). We also confirmed the mass loading to be within 4% error from the target, using inductively coupled plasma-mass spectrometry (ICP-MS) analysis.

Fig. 2: Characterization of Sn₂/NCB.

a Schematics of synthesis procedures of Sn₂/NCB. b AC HAADF-STEM image of Sn₂/NCB. c AC HAADF-STEM image of Sn₁/NCB. d TEM EDS mapping images of Sn₂/NCB. e Histogram of the distribution of distances between Sn pair-atoms. Three representative Sn pair-atoms sites: f enlarged AC HAADF-STEM images and surface intensity plots; g the variation in brightness along the Sn pair-atoms. Source data are provided as a Source Data file.

X-ray diffraction (XRD) patterns of Sn₂/NCB and NCB showed similar broad peaks at 25° and 44° after pyrolysis (Fig. S2), indicating poor crystallinity of carbon. XRD analysis also confirmed the absence of metal nanoparticles. In Raman analysis, the relative intensity ratio between the D and G bands (I_D/I_G) of Sn₂/NCB and NCB were 1.01 and 1.02 respectively, higher than the 0.99 of pure NCB (Fig. S3), consisting of the increased surface defects due to nitrogen doping⁴⁶. X-ray photoelectron spectroscopy (XPS) of Sn₂/NCB also confirmed the absence of characteristic Sn⁰ peak at ~485.0 eV⁴⁷. XPS analysis further suggested that the binding energies of two Sn 3d peaks at 487.3 eV and 497.3 eV were close to the values reported for Sn⁴⁺ (Fig. S4a)⁴⁷. The N 1s spectrum showed the presence of pyridinic N (399.0 eV), Sn–N (400.0 eV), pyrrolic N (400.8 eV), and graphitic N (401.5 eV) (Fig. S4b)⁴⁸.

We further examined the oxidation state and coordination environment of Sn in Sn₂/NCB using X-ray adsorption (XAS) spectroscopy. The X-ray absorption near edge structure (XANES) spectrum at Sn K-edge shows that the edge energy of Sn₂/NCB was in between Sn foil and Sn(IV) phthalocyanine (SnPc) standards (Fig. 3a), suggesting that the oxidation state of Sn in Sn₂/NCB (δ₂⁺) was 0 < δ₂⁺ < 4⁺. Note that there was a shift towards higher edge energy (eV) from Sn₂/NCB to Sn₁/NCB (Fig. 3b). The lower edge energy suggests that Sn pair-atoms would be in a more reduced state compared to single atoms (i.e., δ₂⁺ < δ₁⁺). A similar shift of electron distribution towards clustered single atoms due to the electronic interactions between short-distanced atoms has been observed with neighboring or ensemble single atoms^49,50. The linear correlation analysis between the edge energy and oxidation state of Sn indicated that the oxidation states of Sn would be around δ₂⁺ = 2.8 for Sn₂/NCB and δ₁⁺ = 3.6 for Sn₁/NCB (Fig. 3c).

Fig. 3: Characterization of Sn₂/NCB.

a, b Normalized XANES spectra of Sn K-edge. The arrow indicates the rise of oxidation state. c The linear correlation between the edge energy (eV) and oxidation state of Sn. d FT-EXAFS spectra. e FT-EXAFS fitting of Sn₂/NCB. Inset: schematic diagram of the Sn-N₄ structure. WT-XAFS spectra of (f) Sn foil, (g) SnPc, (h) Sn₂/NCB, and (i) Sn₁/NCB. Source data are provided as a Source Data file.

The Fourier transformed-extended X-ray absorption fine structure (FT-EXAFS) spectra of Sn₂/NCB and Sn₁/NCB exhibited major peaks at 1.57 Å and 1.58 Å, corresponding to the Sn-N coordination similar to the SnPc standard (Fig. 3d). The Sn₂/NCB did not exhibit the peak of Sn-Sn coordination at 2.79 Å of Sn foil, indicating the absence of direct Sn-Sn interactions. The coordination number of Sn pair-atom in Sn₂/NCB was estimated to be 4.1 ± 0.2 based on EXAFS fitting (Fig. 3e and Table S1), which suggested the formation of Sn-N₄ structure for both pair-atoms. The k-space and R-space of Sn₂/NCB, Sn₁/NCB, SnPc, and Sn foil were additionally analyzed via wavelet transform (WT) to obtain 2D and 3D contour maps (Fig. 3f–i). The maximal peak near 4.9 Å⁻¹ of Sn₂/NCB was attributed to the Sn-N bond, which shifted slightly from the 5.2 Å⁻¹ of Sn₁/NCB. These indicated the differences in the Sn–N coordination in Sn₂/NCB due to the coupling compared to fully isolated single-atom sites in Sn₁/NCB.

Performance of Sn₂/NCB in NO₃RR and mechanistic insights

We first employed a batch cell to evaluate the NO₃RR performance of Sn₂/NCB. Sn₂/NCB showed a higher current density compared to Sn₁/NCB and NCB during linear sweep voltammetry (LSV) measurements, indicating its superior NO₃RR activity than Sn₁/NCB and NCB (Fig. 4a). The Sn₂/NCB showed a comparable NO₃⁻ conversion rate (401.6 mg-N h⁻¹ m⁻²) with Sn₁/NCB (Fig. 4b). However, N₂ selectivity was significantly higher with Sn₂/NCB (86%) than Sn₁/NCB (50%), consistent with our hypothesis that the Sn pair-atom architecture would be beneficial for enhancing the N₂ generation over NH₃. Sn₂/NCB also exhibited higher NO₃⁻ conversion rate and N₂ selectivity compared to other benchmark Sn materials such as nanocluster (1 wt% and 5 wt%), nanoparticle (10 wt% and 20 wt%), and bulk powder (Fig. S6). These results suggested that the pair-atom structure rather than the Sn loading played a critical role in enhancing the NO₃RR performance. When applying Sn₂/NCB for NO₃RR under varied potentials, the NO₃⁻ conversion rate gradually increased with higher potentials, while the N₂ selectivity started to decline above −1.2 V vs. RHE (Fig. 4c).

Fig. 4: Performance of Sn₂/NCB in NO₃RR and mechanisms.

a LSV curves of Sn₂/NCB with and without NO₃⁻, Sn₁/NCB with NO₃⁻, and NCB with NO₃⁻. The voltages were not iR corrected. b NO₃⁻ conversion rates and N₂ selectivity of Sn₂/NCB, Sn₁/NCB, and NCB. c NO₃⁻ conversion rates and N₂ selectivity of Sn₂/NCB under varied potentials. The voltages were not iR corrected (electrode surface area = 1 cm², resistance value = 3 Ω). The data in (b) and (c) are presented as mean values ± s.d. (n = 3). In situ ATR-SEIRAS and the corresponding contour maps of (d) Sn₁/NCB and (e) Sn₂/NCB during NO₃RR. f The charge density difference plots of NO₃⁻ on Sn₂/NCB and Sn₁/NCB. g The adsorption energies of NO₃⁻ on CB, NCB, Sn₁/NCB, and Sn₂/NCB. h Gibbs free energy curves of NO₃⁻ reduction on Sn₂/NCB. The inset molecular structures represent the most thermodynamically favorable N₂ generation path on Sn₂/NCB. Transition state energy diagrams of (i) *H and (j) *N spillover on Sn₂/NCB and Sn₁/NCB. Red, blue, white, and grey atoms represent O, N, H, and C elements respectively. Source data are provided as a Source Data file.

In situ attenuated total reflectance surface-enhanced infrared absorption spectroscopy (ATR-SEIRAS) was applied to identify the key N intermediates during NO₃RR for Sn₂/NCB and Sn₁/NCB (Fig. 4d, e). Both groups displayed a negative peak at 1224 cm⁻¹ corresponding to the Si–O stretching vibration⁵¹ and a bending vibration mode of H₂O at 1640 cm^−1 52. The peaks at 1330 cm⁻¹, 1190 cm⁻¹, and 1560 cm⁻¹ were assigned to *NO₃^‒, *NO₂^‒, and *NO, respectively⁵³. For Sn₁/NCB, a clear peak at 1460 cm⁻¹ from the stretching vibration of the N−H bond in *NH₄⁺ was observed, along with the intermediates including *NH₂ at 1140 cm⁻¹ and *NH₂OH at 1110 cm^−1 52. A peak at 1030 cm⁻¹ corresponded to the generation of *N₂^54,55,56. In contrast, for Sn₂/NCB, the peak intensity of *NH₄⁺ at 1460 cm⁻¹ along with other intermediates for NH₃ production decreased significantly compared to Sn₁/NCB, which was also observed from the corresponding contour maps. Further, the peak intensity of *N₂ at 1030 cm⁻¹ was more prominent, indicating that Sn₂/NCB promoted the conversion of NO₃⁻ to N₂ over NH₃ compared to Sn₁/NCB.

We further performed density functional theory (DFT) calculations to scrutinize the mechanisms of the enhanced N₂ selectivity on Sn pair-atoms over single-atoms. We first evaluated the adsorption strength and charge transfer of NO₃^‒ on different substrates. The adsorption configurations of NO₃^‒ on Sn₁/NCB and Sn₂/NCB were optimized and shown in Fig. S7. Instead of the conventional vertical adsorption configuration on Sn single-atom, on Sn₂/NCB, NO₃^‒ interacted with two adjacent Sn pair-atoms, with an inclined omega angle (N-Sn-O₂-O₁) of 87.1°. The coordinates of all atoms in the NO₃^‒ adsorbed Sn₁/NCB and Sn₂/NCB models were presented in Supplementary Data 1. The charge density difference plots of NO₃^‒ on NCB, Sn₁/NCB, and Sn₂/NCB (Figs. 4f, S8) indicated that the charge transfer from Sn pair-atoms to NO₃^‒ (1.02 e^‒) was more significant than Sn single-atoms (0.93 e^‒). The adsorption of NO₃^‒ on Sn₂/NCB and Sn₁/NCB was predominantly chemical adsorption, with adsorption energies of ‒1.54 eV and ‒1.89 eV, respectively (Fig. 4g), confirming the adsorption of NO₃^‒ on Sn₂/NCB was stronger than that on Sn₁/NCB. The projected density of states (PDOS) diagram showed that after Sn₁ introduction, the band gap decreased from 0.68 eV of NCB to 0.23 eV of Sn₁/NCB (Fig. S9). After further adding Sn atom to form Sn₂/NCB, the contribution of the electronic peak was more obvious, and the band gap was further reduced to 0.14 eV, which would enhance the charge transfer across the catalyst. The Sn₂/NCB possessed a slightly more positive d-band center and consequently a stronger adsorption strength than that of Sn₁/NCB.

We further compared the lowest energy NO₃RR pathways for Sn₂/NCB versus Sn₁/NCB (Fig. 4h and Fig. S10) and other control materials (Fig. S11). The reduction of NO₃^‒ to *NO followed the NO₃ → *NO₃ → *NO₃H → *NO₂ → *NO₂H → *NO hydrogenation route. For NCB, the weak adsorption of NO₃^‒ was inconducive to the occurrence of the reaction. Further hydrogenation of *NO followed three pathways, generating *NOH, *ONH, and *N₂O₂H, respectively. On Sn₁/NCB, the production of *NOH was inhibited by a positive ΔG_max of 1.0 eV. The N₂ generation occurred through energetically favorable pathway of *NO → *N₂O₂H → *N₂O → *N₂ + OH → *N₂. In contrast, the path toward NH₃ (*NO → *ONH → *ONH₂ → *OHNH₂ → *NH₂ → *NH₃) only had a small energy barrier of ΔG_max of 0.1 eV from *NO to *ONH. Therefore, both two paths could happen, while the N₂ generation was slightly more favorable. This was consistent with the selectivity of 50% for N₂ and the selectivity of 38% for NH₃ in batch NO₃RR tests.

In contrast, the adsorbed *NO₃ and *NO₃H interacted with two Sn pair-atoms on Sn₂/NCB. Consequently, their free energies were more negative than those in the case of Sn₁/NCB. This indicates that the pair-atom interaction is more conducive to the initiation of the NO₃RR. For the *NOH, *ONH, and *N₂O₂H pathways, the most thermodynamically favorable route was *NO → *NOH → *N → *N₂. The Sn pair-atom design altered the dominant intermediate of N₂ generation to *NOH, which contrasts from *N₂O₂H on Sn single-atoms. In comparison, the *ONH path had a ΔG_max of 0.9 eV of *ONH₂ → *OHNH₂, while the *N₂O₂H path had a ΔG_max of 0.7 eV of *N₂O₂H → *N₂O, both inhibiting these pathways. The above calculations collectively suggest that Sn pair-atoms would preferentially lead to the production of N₂ compared with NH₃, consistent with the experimental results.

Finally, we examined the energetics of transition state involving *H and *N spillover, preceding N₂ formation. The energy barriers for water dissociation on both Sn₂/NCB and Sn₁/NCB were similar (Fig. S12). For Sn₂/NCB, *H formed on a Sn site could easily migrate and attack the *NO on the adjacent Sn site (Fig. 4i). In contrast, for Sn₁/NCB, *H on a Sn atom needed to spillover onto and migrate across NCB support with an energy barrier of >0.8 eV to react with *NO on another distant Sn atom site. This difference accounts for more effective hydrogenation of *NO with Sn₂/NCB than Sn₁/NCB. The simulation also suggested that the energy barrier of coupling two *N atoms on Sn pair-atoms to form N₂ was 0.8 eV (Fig. 4j), which was lower than the diffusion (spillover) barrier (0.9 eV) of *N from a Sn single-atoms to NCB support. These results confirm that efficient N₂ formation through *N coupling is due to close proximity of *N atoms placed on neighboring Sn pair-atoms.

Design principle, fabrication, and characterization of bilayer electrified membrane

Based on N₂-selective NO₃RR demonstrated in the batch cell, we incorporated the Sn₂/NCB onto a bilayer electrified membrane (bilayer-EM) to further improve the NO₃RR kinetics and selectivity. As shown in Fig. 5a, the cathodic layer was placed on the top of the flow cell (i.e., toward the influent side) to selectively reduce NO₃⁻ to N₂. The mass transfer of NO₃⁻ to catalytic sites would be enhanced due to convection across the porous electrode. Through the anodic bottom layer (i.e., placed below the cathode), free chlorines generated in situ would oxidize the small amount of NH₃ generated from the cathode to N₂^57,58. This bilayer-EM was designed to achieve sequential cathode-to-anode electrofiltration and to realize near-complete removal of low-concentration NO₃⁻ with ~100% N₂ selectivity.

Fig. 5: Operating principle and characterization of the bilayer electrified membrane (bilayer-EM).

a Schematic showing the concept of applying the bilayer-EM with sequential cathode-to-anode electrofiltration for realizing near-complete NO₃⁻-to-N₂ reduction with ~100% selectivity. b Photograph and SEM images of (c) the top and (d) the cross-sectional views of the top-layer bilayer-EM (i.e., Sn₂/NCB-EM). e Photograph, (f) top view SEM images, and (g) XPS of the bottom-layer bilayer-EM (i.e., TiO_2-x-EM). h Pore size distribution and the average value of the Sn₂/NCB-EM and the CNT-EM. i Water flux of the Sn₂/NCB-EM, the CNT-EM, and the TiO_2-x-EM. Water permeabilities of the membranes are indicated by the fitted slopes. j EIS spectra of the Sn₂/NCB-EM, the CNT-EM, and the TiO_2-x-EM over a frequency range of 1–10⁶Hz in 10 mM Na₂SO₄ solution. Source data are provided as a Source Data file.

We prepared a free-standing electrified membrane (Sn₂/NCB-EM) that serves as the top layer of the bilayer-EM by coating the Sn₂/NCB on the CNT interwoven framework. The membrane was morphologically uniform across the effective membrane area of 12.6 cm² and mechanically robust as a flow-through electrode (Fig. 5b). The comparison of the top view SEM images between the catalyst-loaded membrane (Fig. 5c) and a control CNT electrified membrane (CNT-EM, Fig. S13) indicated that the CNT interwoven framework was well-coated with the Sn₂/NCB. The cross-section SEM images showed a lamellar structure of the membrane with a thickness of 0.15 mm (Fig. 5d). We also prepared a TiO_2-x ceramic electrified membrane (TiO_2-x-EM) that serves as the bottom anodic layer of the bilayer-EM by reducing a TiO₂ ceramic membrane in pure H₂ atmosphere at 1100 °C (see Fig. S14a for photos of the ceramic membrane before and after reduction). The TiO_2-x-EM had the same effective membrane area as the Sn₂/NCB-EM (Fig. 5e) but exhibited a much rougher surface (Fig. 5f). XPS spectra demonstrated the slight reduction of the TiO₂ ceramic membrane to TiO_2-x (x = 0.11) with the appearance of Ti³⁺ (Figs. 5g, S14b). We finally placed a thin macroporous polyester fabric as a spacer (photo see Fig. S15) between the top and bottom layers to construct the bilayer-EM.

The bilayer-EM was overall hydrophilic (Fig. S16) with the water contact angle of the bottom TiO_2-x-EM layer of only 9.0°. The Sn₂/NCB coat on the CNT framework improved the hydrophilicity of the top Sn₂/NCB-EM layer to a low water contact angle of 30.3°. The coating structure reduced the average pore size and the water permeability of the membrane from 31.3 nm and 148 L m⁻² h⁻¹ bar⁻¹ (CNT-EM) to 24.9 nm and 77 L m⁻² h⁻¹ bar⁻¹ (Sn₂/NCB-EM, Fig. 5h, i), respectively. The smaller pore size and lower water permeability demonstrated a dense interwoven nature of the membrane, which was essential to intensify the utilization of the catalysts and the mass transfer of low concentration NO₃⁻ during electrofiltration. We noted that the water flux of the bilayer-EM was mainly determined by the top Sn₂/NCB-EM layer due to its much smaller water permeability compared with the bottom TiO_2-x-EM layer (Fig. 5i). Additionally, the electrochemical impedance spectroscopy (EIS) amplitudes of the semicircles for both the Sn₂/NCB-EM and the TiO_2-x-EM were only 23 Ω in 10 mM Na₂SO₄ solution (Fig. 5j). These results demonstrated the high conductivity of the bilayer-EM for sufficient electron transfer under a relatively low electrolyte concentration.

Electrochemical NO₃ ⁻-to-N₂ reduction using bilayer electrified membrane

We evaluated the performance of the bilayer-EM for removing low-concentration NO₃⁻ (10 mg-N L⁻¹) by electrofiltration of neutral feed solutions containing either 10 mM Na₂SO₄ or 10 mM NaCl + 5 mM Na₂SO₄ (Fig. S17). The loading amount of Sn₂/NCB per membrane was first optimized based on NO₃⁻ removal efficiency in 10 mM Na₂SO₄ (Fig. S18). Under a current density of 2.5 mA cm⁻² and a permeate flow rate of 1.2 mL min⁻¹, NO₃⁻ removal reached 95.3% at 50 mg of Sn₂/NCB. Further increase of loading to 100 mg only slightly increased the removal to 96.3%. Therefore, we used 50 mg catalyst loading for further experiments. We compared the NO₃⁻ reduction performance of a single-layer Sn₂/NCB-EM in flow-through mode (Fig. 6a, middle panel) with that of a Sn₂/NCB-loaded carbon paper electrode in flow-by mode (Fig. 6a, left panel). Under the same current density of 2.5 mA cm⁻², the N₂ selectivity was similar between flow-through (90.1%) and flow-by (89.3%). However, the flow-through mode showed a 3-fold enhancement of NO₃⁻ conversion rate (546.1 mg-N h⁻¹ m⁻² and 95.3% removal efficiency) compared to that of the flow-by mode (186.3 mg-N  h⁻¹ m⁻² and 31.1% removal efficiency), suggesting the significant role of the flow-through system in stimulating mass transfer and reaction kinetics.

Fig. 6: NO₃RR performance and theoretical analysis of the bilayer-EM.

a Effect of electrified operation modes, including flow-by (left, grey shading), single-layer flow-through (middle, blue shading), and bilayer flow-through (right, pink shading) modes, on the distribution of nitrogen species in the permeate (left axis) and N₂ selectivity (right axis) using the membrane. Black bars indicate the residual nitrogen concentration (i.e., the sum of NO₃⁻, NO₂⁻, and NH₃) in the permeate. The voltages were not iR corrected (membrane electrode surface area = 12.6 cm², resistance value = 23 Ω). CFD simulating the NO₃⁻ concentration distributions under (b₁) flow-through and (b₂) flow-by modes and (b₃) the concentration changes with flow distance. c Generation of free chlorine by the TiO_2-x-EM as a function of current density using a feed solution with 5 mM Na₂SO₄ and 10 mM NaCl under a permeate flow rate of 1.2 mL min⁻¹. d NO₃⁻ removal performance of the bilayer-EM when treating different feed solutions containing 10 mg-N L⁻¹ NaNO₃ in 10 mM Na₂SO₄, 1 mM Na₂SO₄, or simulated surface water (containing 2 mM Cl⁻, other constituents listed in Table S2). The data in (a) and (b) are presented as mean values ± s.d. (n = 3). e NO₃⁻ removal efficiency and loss of Sn of the bilayer-EM as a function of filtration cycles. After operating for 1.5 h, the membrane was rinsed and dried and then used for the next cycle. f Sn FT-EXAFS spectra of the Sn₂/NCB-EM before and after long-term operation. Source data are provided as a Source Data file.

We performed computational fluid dynamics (CFD) simulations to examine how flow-through operation enabled more efficient mass transfer and NO₃⁻ conversion than flow-by mode (Fig. 6b). Flow-through and flow-by modes were simulated considering water flow over a nanofiber interwoven framework and a flat plate, respectively (Fig. S19). The simulated velocity fields showed that the flow of water was significantly intensified inside the nanofiber framework (Fig. S20). Catalysts coated on the framework could increase the thickness of the nanofibers to further enhance the local velocity, resulting in a 12-fold difference in the average velocity fluctuation compared to the plate. This is crucial for improving the utilization efficiency of the highly exposed catalysts coated on the framework. When considering a constant NO₃⁻ reduction rate at the exposed surface, the NO₃⁻ concentration decreased rapidly under flow-through mode (Fig. 6b₁), whereas the NO₃⁻ concentration decrease under flow-by mode was moderate (Fig. 6b₂). This resulted in a three-time difference in the amount of NO₃⁻ removal between the flow-through and flow-by modes (Fig. 6b₃).

We then applied sequential cathode-to-anode electrofiltration using the bilayer-EM to determine whether the existence of Cl⁻ could further enhance the selectivity of N₂ generation. In the absence of Cl⁻, similar NO₃⁻ removal efficiency and N₂ selectivity were observed under single-layer (Fig. 6a, middle panel) and bilayer modes (Fig. 6a, right panel). However, when 10 mM Cl⁻ was added to the feed, the N₂ selectivity increased significantly to 97.7% under the bilayer mode (Fig. 6a, right panel). The free chlorine was efficiently generated from electrofiltration of Cl⁻ through the TiO_2-x-EM (e.g., 6.7 mg-Cl₂ L⁻¹ at 2.5 mA cm⁻², Fig. 6c). The free chlorine or hypochlorous acid (HClO) from chlorine dissolution would oxidize the small amount of NH₃ released from the top Sn₂/NCB-EM layer into N₂^57,58, which resulted in a low residual total nitrogen concentration (the sum of NO₃⁻, NO₂⁻, and NH₃) of only 0.5 mg-N L⁻¹ in the permeate even under the high water treatment capacity of 72 mL h⁻¹. We note that our proposed method is suitable only for removing NH₃ with low concentrations (<1.5 mg-N L⁻¹), considering that free chlorine (dosage of 6–15 mg-Cl₂ L⁻¹) is generally added in the treated municipal water for disinfection^59,60. Applying chlorination to oxidize high-concentration NH₃ can significantly increase the potential for harmful chlorinated by-product formation.

The applicability of the bilayer-EM for removing low concentration NO₃⁻ under different electrolyte conditions was further evaluated. Under the single-layer mode at neutral pH, when altering the electrolyte from 10 mM Na₂SO₄ to lower ionic strength (1 mM Na₂SO₄) and simulated surface water (see Table S2 for the composition), similar NO₃⁻ removal performances were achieved (Fig. 6d). Under the bilayer mode when treating the simulated surface water, the N₂ selectivity further increased to 98.1% as the surface water contained 2 mM Cl⁻. The simultaneous achievement of high NO₃⁻ removal efficiency and N₂ selectivity of Sn₂/NCB bilayer-EM with only ~8 s resident time was superior when compared to the previous NO₃RR system for NO₃⁻ decontamination to N₂ (Table S3). When treating water with NO₃⁻ concentration at municipal wastewater level of 50 mg-N L⁻¹ (Fig. S21), the bilayer-EM also achieved high removal efficiency (86%), conversion rate (2457.6 mg-N h⁻¹ m⁻², a 6-fold enhancement compared to the batch system), and Faradaic efficiency (91%). Notably, the bilayer-EM maintained a stable NO₃⁻ removal efficiency of ~96% with negligible dissolution of the Sn (<0.05%) throughout a long-term testing (6 repeated cycles which corresponds to 9 h of operation, Fig. 6e). The Sn FT-EXAFS spectra for the Sn₂/NCB-EM before and after the long-term operation showed that the Sn maintained the Sn-N coordination structure without aggregation of Sn (Fig. 6f). The Sn XANES spectra for the Sn₂/NCB-EM before and after reaction indicated that the oxidation state of Sn was also maintained (Fig. S22). These results collectively demonstrate the high stability of Sn pair-atoms in the flow-through operation, which was ascribed to the strong binding through the nitrogen coordination^61,62. The robustness of the Sn pair-atom loaded membrane for removing low concentration NO₃⁻ at realistic water conditions further confirms its superiority over previous nanoparticle-based NO₃RR systems.

Discussion

In this work, we developed a bilayer electrified membrane incorporating Sn pair-atoms as the NO₃RR catalysts. We successfully demonstrated that the membrane could realize superior removal efficiency and N₂ selectivity for converting NO₃⁻ with low concentrations in single-pass electrofiltration. Our design principles on the catalysts and the electrofiltration system present three major advancements. First, without using noble metals, the Sn pair-atoms enhanced NO₃⁻ adsorption and electron transfer and promoted *N-*N coupling to result in a high N₂ selectivity via facilitating *N spillover, which would be difficult to achieve via fully-isolated single-atom sites. Second, coating the pair-atom catalysts with high N₂ selectivity on the interwoven framework of the electrified membrane enhanced the exposure and utilization of the catalysts during flow-through operation to facilitate the NO₃RR compared to directly functionalizing the catalysts on a plate electrode for conventional flow-by operation. Third, the sequential cathode-to-anode NO₃⁻ reduction and residual NH₃ oxidation during the single-pass electrofiltration achieved a near 100% N₂ selectivity with near zero total dissolved N in the effluent when treating water with an low NO₃⁻ concentration (10 mg-N L⁻¹). Further efforts should focus on the mitigation of side reactions when applying electrofiltration to remove contaminants at low concentrations from complex water matrices^63,64. The findings of this study not only advance the NO₃RR catalyst design but also pave the way for scalable treatment of water containing NO₃⁻ with low concentrations using electrified membranes.

Methods

Chemicals and materials

Tin chloride (SnCl₄), tin powder, sodium nitrate (KNO₃), 1,10-phenanthroline (1,10-phen), carbon nanotubes (CNT, multi-walled), Nafion solution, isopropanol, N,N-dimethylformamide (DMF), polyacrylonitrile (PAN), and sodium sulfate (Na₂SO₄) were purchased from Sigma-Aldrich. Carbon black was obtained from Cabot Corporation (CB, EMPEROR 2000). All chemicals used in the experiments were reagent grade or higher and used as received without further purification. Experimental solutions were prepared using deionized water (DI, >18.2 MΩ·cm) from the Milli-Q system.

Synthesis of Sn₂/NCB and Sn₁/NCB

The synthesis step first involved mixing SnCl₄ and 1,10-phen at a molar ratio of 1:3 in ethanol. The CB was further added into the mixed solution for impregnation. The nominal loading of Sn was 5% and 1% for Sn₂/NCB and Sn₁/NCB respectively. The suspensions were dried at 80 °C for 24 h. The catalysts were then calcined at 600 °C for 3 h in a tube furnace (STF1200, Across International) under N₂ at a flow rate of 250 sccm.

Preparation of Sn loaded electrodes

5 mg of Sn₂/NCB (or NCB, Sn₁/NCB, Sn NC-1%, Sn NC-5%, Sn NP-10%, Sn NP-20%, and Sn metal powder) and 20 µL of 5 wt% Nafion solution were dispersed 1 mL isopropanol, followed by sonication for 1 h to prepare a catalyst ink. 200 µL of the ink was drop-casted on carbon fiber paper with an effective working area of 1 × 1 cm², giving a mass loading of 1 mg cm⁻².

Fabrication of Sn₂/NCB-EM

The as-received CNTs were pre-treated in a concentrated HCl solution (36.5–38.0%) at 90 °C under refluxing for 12 h, followed by washing with DI water until neutral and drying at 60 °C for 12 h for preparing the pristine CNTs. A certain amount of Sn₂/NCB (10 to 100 mg) and 50 mg of pristine CNTs were dispersed in 40 mL DMF containing 0.1 wt% PAN, followed by sonication for 20 min using an ultrasonic probe. We applied layer-by-layer assembly by vacuum filtering 2 mL of the as-prepared suspension onto a ceramic membrane (0.45 μm, Sterlitech). The prepared membrane with 20 layers in total was then rinsed with DI water followed by drying at 90 °C for 12 h. The Sn₂/NCB-FEM was obtained by directly peeling the whole carbonaceous layer off from the ceramic membrane after drying.

Fabrication of TiO_2-x-EM

The TiO_2-x-EM was obtained by calcining a pristine TiO₂ membrane (Sterlitech) under pure H₂ atmosphere at 1100 °C for 12 h using the tube furnace at a heating rate of 5 °C min⁻¹.

Preparation of bilayer-EM

The bilayer-EM was obtained by stacking the free-standing Sn₂/NCB-EM on the TiO_2-x-EM with a thin macroporous polyester fabric as the spacer in between to separate the membranes.

Characterizations of catalysts

The Sn loaded onto the catalyst was measured using inductively-coupled plasma mass spectrometry (ICP-MS). The morphology of synthesized Sn₂/NCB catalysts was characterized using transmission electron microscopy (TEM, Tecnai Osiris 200 kV, FEI) and aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC-HAADF-STEM) at 300 kV. X-ray Diffraction (XRD, Rigaku SmartLab) using Cu Kα radiation (λ = 1.542 Å) was carried out to identify the crystallinity of carbon and the existence of metal phases. Raman spectroscopy (LabRAM HR Evolution, Horiba) was carried out to observe the characteristic D and G bands of the catalysts. The Sn 3d and N 1 s were measured by X-ray photoelectron spectroscopy (XPS) with a Versa Probe II scanning XPS microprobe (PhysicalElectronics) using monochromatic Al Kαradiation (1486.6 eV).

XAFS measurement of Sn₂/NCB

To characterize the coordination environment of Sn, X-ray absorption spectra (XAS) at Sn K-edge were recorded at the ISS beamline of the National Synchrotron Light Source II at Brookhaven National Laboratory⁶⁵, using a Si(111) double-crystal monochromator and a passivated implanted planar silicon (PIPS) fluorescence detector. Near-edge X-ray absorption spectroscopy (XANES) and extended X-ray absorption fine structure (EXAFS) data were collected at room temperature, with energy calibrated using a Sn foil. EXAFS data was fitted by using Demeter 0.9.25 software package. Athena software was used to calibrate energy and normalize the data. The edge energy was determined using the maximum of the first peak in the first derivative of the XANES spectra⁶⁶. EXAFS spectra were fit using a least-squares fit in R-space of k² weighted Fourier transform (FT) to determine the coordination number (CN) and bond distances (R) between Pd and Cu using the Artemis software⁶⁶. The first shell was used to fit the EXAFS spectra. Least-squares fit in R-space of the k²-weighted Fourier transform data from 3.0 to 12.0 Å⁻¹ was used to obtain the EXAFS coordination parameters. The amplitude reduction factor (S0²) was determined as 0.78 by fitting a reference spectrum of the Pd foil, and then it was used for fitting all the other EXAFS spectra.

Characterization of membranes

SEM (SU8230, Hitachi) was employed to investigate the surface and cross-section morphologies of the membrane. Membrane pore size distribution was estimated by analyzing the surface SEM images using Nano Measurer software. Water contact angles were measured by the sessile drop method using a contact angle goniometer (OneAttension, Biolin Scientific). Water flux was calculated by dividing the permeate flow rate by the effective membrane area (i.e., 12.6 cm²). Membrane pore volume was determined by the weight difference between a wet and dry membrane. Water residence time was calculated by dividing the pore volume by the permeate flow rate.

Electrochemical measurements

Electrochemical measurements were performed by an electrochemical workstation (CHI 660E, CH Instruments) in a typical three-electrode batch cell containing the working electrode, a mixed metal oxide (MMO) electrode as the counter electrode, and an Ag/AgCl electrode as the reference electrode. All potentials were expressed relative to the RHE scale and were calculated using the following equation: E (vs RHE) = E (vs Ag/AgCl) + 0.197 + 0.0591 × pH. LSV curves were collected at a scan rate of 20 mV s⁻¹ in 50 mM Na₂SO₄ solution with or without 50 mg-N L⁻¹ NaNO₃ addition. EIS was conducted by applying frequencies ranging from 1 to 10⁶Hz in a 10 mM Na₂SO₄ solution at open circuit voltage. The electrolyte was prepared before each test.

Batch NO₃RR experiments

The NO₃⁻ reduction performance of the catalysts was investigated through batch experiments. The batch system consisted of a reactor with the catalyst coated electrode and an MMO mesh electrode with a spacing of 1 cm. Batch experiments were performed for treating 20 mL feed solution with 50 mg-N L⁻¹ NaNO₃ and 50 mM Na₂SO₄ for 2 h at a string rate of 500 rpm. The reason for choosing 50 mg-N L⁻¹ instead of 10 mg-N L⁻¹ was that the low mass transfer limitations in the batch system would result in no visible effect among different catalysts.

Electrofiltration NO₃RR experiments

Electrofiltration experiments were performed using a cross-flow membrane filtration system. The membrane and a counter MMO mesh electrode were placed in an electrofiltration cell with 1 cm spacing. Electrolytes containing 10 mg-N L⁻¹ NaNO₃ in 10 mM Na₂SO₄ or in 10 mM NaCl + 5 mM Na₂SO₄ at neutral pH were used as the feed solution. The feed was circulated at a flow rate of 200 mL min⁻¹ by a peristaltic pump and the permeate flow rate was controlled to 1.2 mL min⁻¹.

Nitrogen speciation and free chlorine measurement

NO₃⁻, NO₂⁻, NH₄⁺, and free chlorine concentrations in both the permeate and the feed were quantified after 1 h of operation. The concentrations of NO₃⁻, NO₂⁻, NH₄⁺, and free chlorine were determined according to the cadmium reduction method, the diazotization method, the Nessler method, and the N,N-diethyl-1,4-phenylenediamine sulfate method respectively, using assay kits (HI93728 for NO₃⁻, HI93707 for NO₂⁻, HI93715 for NH₄⁺, and HI97710 for free chlorine, Hanna Instruments). The standard curves used for the colorimetric quantification of NO₃⁻, NO₂⁻ and NH₄⁺ are shown in Fig. S5. Ion chromatography (IC, 930 Compact, Metrohm) was also employed to double-check the concentrations of the nitrogen ions. The N₂ selectivity was calculated by the following equation, where C represents the concentration of a certain ion:

$${N}_{2}\%=\frac{[{C}_{{\left[{NO}3-N\right]}_{0}}{-C}_{{\left[{NO}3-N\right]}_{t}}-{C}_{{\left[{NO}2-N\right]}_{t}}-{C}_{{[{NH}4-N]}_{t}}]}{{C}_{{[{NO}3-N]}_{0}}{-C}_{{[{NO}3-N]}_{t}}}$$

(1)

The Faradaic efficiency (FE) calculation

The FE for the conversion of NO₃⁻ to NO₂⁻, NH₄⁺/NH₃, and N₂ were calculated according to the following equation:

$$F{E}_{{NO}3{RR}}=\frac{{n\times F\times C}_{{product}}\times V}{M{W}_{{product}}\times i\times t}$$

(2)

where n is the number of charge equivalents required to produce 1 mole of product (n = 2 for NO₂⁻, n = 8 for NH₄⁺/NH₃, n = 5 for N₂). F is the Faraday’s constant (96485 C/mol), ${C}_{{product}}$ is the concentration of product (g/L), V is the volume of the electrolyte (0.04 L), MW_product is the molecular weight of products (MW_NH3 = 17 g mol⁻¹, MW_NO2 = 46 g mol⁻¹, MW_N2 = 28 g mol⁻¹), i is the operating current density, and t is the electrolysis time.

In situ attenuated total reflectance surface-enhanced infrared absorption spectroscopy

ATR-SEIRAS was performed using a Nicolet iS50 FTIR spectrometer equipped with an MCT detector and a reflectance unit for the electrochemical cell at an incidence angle of 60°. The in situ electrochemical cell consisted of two chambers with three electrodes and the chambers were filled with 10 mL of 50 mg-N L⁻¹ NaNO₃ and 50 mM Na₂SO₄ electrolyte. Platinum sheet (1 cm²) and Ag/AgCl were used as the counter electrode and reference electrode, respectively. For the working electrode, 30 μL of catalyst ink (5 mg mL⁻¹) was dropped onto a silicon crystal and dried in air before testing. The detector was cooled with liquid nitrogen for at least 30 min before testing to maintain a stable signal. NO₃RR testing was performed using the chronoamperometry method, and the spectral signal was collected after 1 min of reaction. Scans were conducted from the open circuit potential to −1.0 V (vs. RHE). All spectra were given by absorbance with a spectral resolution of 4 cm⁻¹ for each curve.

Density functional theory calculations

Spin-polarized first-principles calculations were performed by using the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional as implemented in the Vienna ab initio simulation package (VASP). The projector augmented wave (PAW) method and the corresponding pseudopotentials were employed. A cutoff energy of 500 eV was set for the plane wave basis set. The adsorption configurations of NO₃^‒ on Sn₁/NCB and Sn₂/NCB and the structures of reaction intermediates were optimized by controlling the convergence thresholds for the energy and force to be 10⁻⁵ eV and 0.02 eV/Å, respectively. During the relaxation, all atomic positions were allowed to relax. The Grimme’s DFT-D3 correction method was included to describe the weak dispersion interactions during surface adsorption. A 2 × 2 × 1 Monkhorst-Pack k-point sampling was set for all models. The corresponding adsorption Gibbs free energies were then calculated as

$$\Delta G=\Delta E+\Delta {{{\rm{ZPE}}}}-T\Delta S$$

(3)

where ΔZPE and ΔS are the changes in zero point energy and in entropy, respectively, between the adsorbed state and free state, which can be obtained from the vibrational frequencies calculations (with adsorbates relaxed and substrates fixed) and standard thermodynamic data. T is the temperature. The adsorption energy of species on the surface was calculated as:

$${E}_{{{{\rm{ads}}}}}={E}_{{{{\rm{x}}}}/{{{\rm{surface}}}}}-{E}_{{{{\rm{x}}}}}-{E}_{{{{\rm{surface}}}}}$$

(4)

where E_x/surface, E_x, and E_surface are the energies of the slab with adsorbed species, the adsorbed species calculated in a cubic cell, and the energies of the slab, respectively.

Computational fluid dynamics simulation

CFD was performed for flow field simulation inside the membrane using COMSOL Multiphysics software. The geometry was constructed with a cube fluid domain with a side length of 400 nm. The cube was swept with random rotation and movement multiple times to obtain multiple gaps and the geometric structures outside the fluid domain were removed. Three models were built inside the fluid domain, including an interwoven framework with thick fibers, an interwoven framework with thin fibers, and a flat plate control. The simulations were solved based on the Navier-Stokes equations using laminar flow physics. On the basis of the convection velocity calculation, the conservation calculation of convective mass transfer was performed. The calculations meet the dilute species transfer interface under different pore conditions and the mass transfer was simulated through the diffusion equation:

$$\nabla \cdot \left(-D\nabla c\right)+{{{\bf{u}}}}\cdot \nabla {{{\rm{c}}}}={{{\rm{R}}}}$$

(5)

where D is diffusivity (m²/s), c is concentration (mol/m³), u is fluid velocity vector (m/s), and R is the rate expression of matter (mol/m³·s). Considering that the reaction happens mainly on the surface rather than in the body of the model, the R term here is zero and the overall reaction satisfies Fick’s diffusion law. The inlet flow rate at the entrance of the mass transfer region was set to 2.4 μm/s and the surface reaction rate was set to 2 × 10⁻⁴ m/s as the field boundaries.