Sustained hydrogen peroxide production via MXene-functionalized supramolecular docking

Abstract

Direct photosynthesis of hydrogen peroxide (H₂O₂) from air and water using metal/covalent-organic frameworks offers a sustainable alternative to the conventional energy-intensive anthraquinone process. However, current systems suffer from short operational lifetimes typically 10–100 h due to mismatches between O₂ capture efficiency and multielectron redox kinetics. Here, we report a supramolecular platform that integrates O₂ capture, H₂O₂ synthesis, and in situ utilization, enabling continuous H₂O₂ production for over 1000 h without sacrificial agents. Mesoporous bromine-substituted COFs are hydrogen-bonded to photothermal MXene, providing organized O₂ docking sites and columnar charge transport via σ-σ interactions and π-π stacking. Through a dual-pathway mechanism, the architecture achieves a competitive H₂O₂ production rate of 2878 μmol g⁻¹ h⁻¹, alongside complete pollutant removal and durable operation across diverse water sources and outdoor conditions. This work demonstrates a supramolecular design featuring programmable O₂ docking, directional charge transport, and localized H₂O₂ utilization toward decentralized chemical manufacturing.

Programmable supramolecular COF/MXene assemblies integrate convenient H₂O₂ photoproduction with photothermal units, realizing long-term O₂ capture, in situ H₂O₂ synthesis and utilization toward decentralized green chemical manufacturing.

Introduction

Hydrogen peroxide (H₂O₂) has been extensively used as a clean oxidant in bleaching, disinfection, environmental remediation, and organic synthesis¹. More recently, it has emerged as a promising zero-emission liquid fuel, capable of converting chemical energy into electricity with minimal pollution and high cost-efficiency². Today’s industrial production of H₂O₂ is dominated by the conventional anthraquinone oxidation (AO) process, which requires multistep hydrogenation–oxidation cycles at elevated temperatures to activate reagents³ (Supplementary Fig. 1a). This not only incurs substantial energy penalties but also generates large volumes of wastewater, exhaust gases, and solid waste, undermining its environmental credentials and raising downstream treatment costs. Therefore, photocatalytic H₂O₂ synthesis with green and mild reaction conditions has been considered as an alternative approach. Although inorganic catalysts (TiO₂, CdS, and g-C₃N₄) have been widely explored for electro/photo synthesis^4,5, their periodic lattice lacks robust docking sites and efficient transport channels, hindering the material’s ability to capture O₂ from open air and rapidly transfer H₂O molecules, thereby limiting its conversion efficiency (Supplementary Fig. 1b). Covalent organic frameworks (COFs) with donor-acceptor structures offer high porosity, crystallinity, and tunable band structures, and have shown promise for photocatalytic H₂O₂ production. Despite conventional strategies including band structure tuning and suppression of photogenerated carrier recombination, their photocatalytic activities remain hampered by an inherently weak photoresponse. In particular, harnessing near-infrared light (NIR, >760 nm), which accounts for ~53 % of solar irradiance, could drastically enhance catalytic performance⁶, yet few COFs successfully combine NIR photo-absorbers with efficient, well-defined charge transfer pathways.

To mimic the multi-unit architecture of respiratory enzymes in living organisms optimized for O₂ and H₂O molecules utilization and maintain the same dynamic response as the conventional metal-organic frameworks (MOFs), we proposed a versatile supramolecular system that can be universally employed to integrate various frameworks and functional building units. In specific, we incorporate metallic amine-functionalized MXene (Ti₃C₂T_x-NH₂) photothermal units into rigid COF backbones to construct an integrated supramolecular (COF)_x-NH₂-M system (where x refers to the mass ratio between COF and MXene, supported by the elemental analysis in Supplementary Tables 1 and 2) for the capture–synthesis–utilization of H₂O₂ (Fig. 1a), exhibiting a MOFs-like light harvesting characteristics enhanced by the presence of metallic units. Hydrogen-bonded interactions on highly electronegative N atoms links conjugated layers via π-π stacking interactions, while σ-σ stacking interactions⁷ connect multilayered nanosheets, creating densely organized O₂-docking sites and rapid, directional charge pathways (Fig. 1b). To guide supramolecular assembly for controlled crystallinity, surface area and carrier mobility⁸, we surmise that TAPB (1,3,5-tris(4-aminophenyl)benzene) monomer, among the common ligands⁹, maximizes hydrogen-bonded interactions by providing the highest density of N sites, thereby promoting intermolecular charge transfer (Fig. 1c). Compared to triazine, the decentralized spatial distribution of N atoms in TAPB further promotes charge separation throughout catalytic reactions.

Fig. 1. Molecular design of metallic unit-functionalized hydrogen-bonded supramolecular assemblies with docking sites.

a Schematic of integrated supramolecular design featuring enzyme-like docking sites for various frameworks and functional units. b σ-π assembly contributed by the connection of σ-σ and π-π stacking interactions (dash line on the chain represents hydrogen bonds). c Molecular design of N-rich ligands with steric effects (blue, metallic MXene units; red, COF ligands). d Electrostatic potential distribution maps of TAPB-Ta COF in AA stacking, TaPB-TaBr₂ COF in AA stacking, and TAPB-TaBr₂ COF in AB stacking, the unit of scale bar is Hartree/e.

The highly tunable pore structures in COFs confine guest molecules within their channels through multiple van der Waals interactions, theoretically supporting efficient molecular recognition for selective gas capture. However, O₂ and N₂ molecules in air have remarkably similar kinetic diameters (3.46 and 3.64 Å, respectively)¹⁰, limiting the specificity of O₂ capture. To elaborate, O₂ is a paramagnetic molecule possessing two unpaired electrons in its π* orbital, which allows it to engage in coordination or charge-transfer complexes with hosts featuring vacant orbitals or donor sites. In contrast, the chemically inert, diamagnetic N₂ molecule with its stable triple bond exhibits minimal propensity for similar electron interactions and docking affinity toward supramolecular framework. In addition to the intrinsically higher electron affinity of O₂, the presence of electron-deficient centers (such as aromatic nitrogen heterocycles¹¹ and conjugated quinones¹²) in aromatic molecules have been shown to preferentially adsorb O₂. Besides, water molecules serves as proton donors for H₂O₂ production renders their interaction with the COF structure essential for tailoring the overall reactivity and durability¹³. To address these challenges, we herein introduce a docking strategy based on the π-conjugated structure of our supramolecular materials and the electron configuration of O₂. Although recent advancements have underscored the potential of molecular docking in drug design for predicting ligand-receptor interactions^14–16, a comprehensive understanding of docking mechanisms in catalytic reactions remains to be fully established to enable their efficient application.

In this work, we incorporate bromine, a more electronegative substitute for carbon to induce the delocalization of π-electrons, thereby lowering the electron density of benzene rings. The AA-stacked TAPB-TaBr₂ COF (where TaBr₂ is 2,5-dibromoterephthalaldehyde) exhibits the largest electrostatic potential variation, optimizing distinct molecular docking sites for O₂ (Fig. 1d and Supplementary Fig. 2). Our theoretical calculations confirm that O₂ adsorption at theses aromatic docking sites is energetically favored within the supramolecular framework. Notably, variations in binding affinity across different sites of supramolecular assemblies are closely related to the adsorption-desorption dynamics underpinning target-molecule capture and release. The intrinsic mesoporous architecture of COF frameworks exposes abundant, spatially defined sites for O₂ and H₂O adsorption, maximizing opportunities for molecular reactant docking. Thereby, we establish a well-defined supramolecular catalyst, enabling two key advances (i) optimized charge transfer through σ-π hydrogen-bonded connection and (ii) efficient catalytic turnover through strategic reactant-catalyst docking. Simultaneously, (COF)_x-NH₂-M exhibits a broadened light absorption into the NIR region and harnesses photothermal heating of hot electrons to drive the photosynthesis of H₂O₂ at elevated local temperatures.

Results

Structure of COF and supramolecular photocatalysts

In the (COF)_x-NH₂-M supramolecular system, TAPB-TaBr₂ COF is introduced with Br substituents derived from TAPB-Ta COF (where Ta is terephthalaldehyde), a selected prototype for H₂O₂ photocatalysis¹⁷. Solvothermal condensation of TAPB with TaBr₂ yields an imine-linked COF designed to feature (i) mesopores for efficient mass transfer without requiring additional activation, (ii) electron-donating TAPB “knots” to drive four-electron water oxidation reaction (WOR), and (iii) electron-accepting TaBr₂ linkers for two-electron oxygen reduction (ORR)¹³. The rigid Br substituents in TAPB-TaBr₂ COF induce a distortion on Ta backbones (Supplementary Fig. 3), resulting in stronger interplanar interactions, enhancing thermal stability and crystallinity¹⁸. To stabilize the MXene and improve its supramolecular interactions, Ti₃C₂T_x nanosheets are functionalized with 3-aminopropyl-triethoxysilane (APTES), yielding NH₂-terminated surfaces that circumvent phase transformation¹⁹ under photocatalytic process and boost hydrogen-bonded bridging with TAPB-TaBr₂ COF (Supplementary Fig. 4a). To elucidate the local intermolecular interaction on imine linkages, we employed the Independent Gradient Model based on Hirshfeld partitioning (IGMH) to analyze weak interaction (Fig. 2a). The density functional theory (DFT) optimized cluster model clearly reveals the formation of N–H···N hydrogen bonds (2.20 Å). The green IGMH isosurface with faint blue center localized at the hydrogen bonding sites indicates the coexistence of van der Waals interaction and weak hydrogen bonding interaction²⁰. The distribution of most data points between −0.02 and 0.01, together with the relatively larger δg values near 0.02 in the scatter plot, further supports our proposal. N 1 s X-ray photoelectron spectroscopy (XPS) results verify the bridging on N atoms, with N–H peak (401.6 eV) significantly enhanced by amino hydrogen bonding, while the N = C peak (399.5 eV)²¹ remains unchanged (Fig. 2b). Moreover, the Fourier transform infrared (FTIR) spectra of Ti₃C₂T_x-NH₂ composite under TaBr₂ monomers presence at varied duration exhibit no detectable enhancement on C = N absorption, confirming the absence of covalent bond formation between Ti₃C₂T_x-NH₂ and TAPB-TaBr₂ COF (Supplementary Fig. 5).

Fig. 2. Structural characteristics of TAPB-TaBr₂ COF and supramolecular (COF)_x-NH₂-M.

a IGMH analysis on molecular cluster models of TAPB-TaBr₂ COF and Ti₃C₂T_x-NH₂ (gray for C; blue for N; red for Br; light orange for Si; blue dash box for zoom-in details of hydrogen bonds). b N 1 s XPS spectra of Ti₃C₂T_x-NH₂, TAPB-TaBr₂ COF, and (COF)_2.5-NH₂-M. c Experimental PXRD pattern of TAPB-TaBr₂ COF with Pawley refinement fit (Bragg positions are indicated with blue ticks). d, f Solid-state NMR spectra of Ti₃C₂T_x-NH₂ (dark blue), TAPB-TaBr₂ COF (red), and (COF)_2.5-NH₂-M (blue) (d) ¹H and (f) ¹³C. e Variable-temperature ¹H solid-state NMR spectra of (COF)_2.5-NH₂-M (from −40 °C to 20 °C with an increment of 20 °C). g, h HR-TEM image of (COF)_2.5-NH₂-M in different magnification. i Fourier transform of the COF domains, showing lattice fringes of the COF networks and MXene multilayers. j EDS line scan (along the dash line) profiles of selected elements across the (COF)_2.5-NH₂-M.

The highly ordered π-π stacked COF backbone is evident from its theoretically consistent crystal configuration and mesoporous architecture. The crystal structure of TAPB-TaBr₂ COF in AA stacking (Supplementary Fig. 4b) was optimized by DFT (Supplementary Tables 3–5) and reconstructed by Pawley refinement with lattice parameters of a = b = 3.62 nm, c = 0.419 nm, α = β = 90°, and γ = 120°. Powder X-ray diffraction pattern (PXRD) patterns (Fig. 2c) confirm the AA-stacked hexagonal framework, while other symmetries and stacking modes fail to reproduce the observed intralayer atom distributions and peak positions (Supplementary Figs. 3 and 6). Solid-state ¹³C cross-polarization magic angle spinning nuclear magnetic resonance (CP/MAS NMR) spectroscopy further verifies the distinct carbon environments along the imine linkages (Supplementary Fig. 7). To assess COF quality, Raman spectroscopy shows the absence of TaBr₂ aldehyde C=O stretching vibration around 1690 cm⁻¹, alongside an intense phenyl C=C band at ~1570 cm⁻¹, and a C=N band at 1616 cm⁻¹ (Supplementary Fig. 8), indicating monomer depletion and imine bond formation. Similarly, characteristic peaks for imine COF with highly conjugated aromatic rings are observed at 1615 cm⁻¹ and 1594 cm⁻¹ in FTIR spectrum of TAPB-TaBr₂ COF (Supplementary Fig. 9). Nitrogen sorption measurements at 77 K determine the high Brunauer-Emmett-Teller (BET) surface areas of 1289.56 m² g⁻¹ for TAPB-TaBr₂ COF and 1125.70 m² g⁻¹ for (COF)_2.5-NH₂-M. Both reveal type VI isotherms accompanied by H2-type hysteresis, confirming the extensive presence of columnar pores and interparticle voids (Supplementary Fig. 10). Barrett-Joyner-Halenda (BJH) analysis shows a dominant micropore width of 3.79 nm (Supplementary Fig. 11), which is consistent with the pore size predicted from the crystal structure (37.274 Å from Supplementary Fig. 3a). The resulting structural stability and mesoporosity expose abundant binding sites, enabling efficient docking of O₂ molecules.

In addition, detailed analysis of the chemical states associated with σ-σ units was performed for both the MXene and the supramolecular assembly. APTES grafting on single-layered Ti₃C₂T_x MXene was verified by solution-state ¹H NMR (Supplementary Fig. 12) and Raman spectroscopy (Supplementary Fig. 8b), which show signals attributed to silane hydrolysis and amino (–NH₂) groups. XPS survey scans of Ti₃C₂T_x-NH₂, TAPB-TaBr₂ COF, and the assembled (COF)_x-NH₂-M supramolecular catalysts reveal additional N 1 s and Si 2p peaks (Supplementary Figs 13–17). The N–H peak in the N 1 s XPS spectrum (Supplementary Fig. 15f) and the Si–O/Si–C peak (101.9 eV) in the Si 2p XPS spectrum (Supplementary Fig. 15g) suggest the coexistence of silane chains and amino groups. In addition, the increased intensity of the C–O peak around 286.2 eV in the C 1 s XPS spectrum (Supplementary Fig. 15b) and the Ti(VI)–O doublet in the Ti 2p XPS spectrum (Supplementary Fig. 15c) confirms the formation of an oxygen bridge between MXene surface and the silane chain via hydrolysis reaction. The C 1 s XPS spectrum of (COF)_2.5-NH₂-M exhibits characteristic peaks composed of C–Ti (281.8 eV), C=C/C–C (284.6 eV) and C=N (285.4 eV), which were respectively assigned to metallic MXene units, aromatic rings, and imine linkages (Supplementary Figs. 16 and 17). Detection of the Ti 2p XPS spectrum in (COF)_2.5-NH₂-M sample (Supplementary Fig. 17e), along with the increased intensity of the Ti₃C₂T_x-NH₂ (002) peak at 2θ = 4.3° in the PXRD patterns of (COF)_x-NH₂-M further confirms the retention of both MXene and COF phases (Supplementary Fig. 18).

Specifically, the reliability of the supramolecular assembly was confirmed through the validation of hydrogen-bonded bridges and morphological characterization. FTIR spectroscopy and NMR analysis distinguish hydrogen bond formation within the supramolecular assembly from covalent linking. As shown in Supplementary Fig. 9, the splitting and shifting²² of the C=N/C–N stretching peaks around 1693/1393 cm⁻¹ in the FTIR spectra reveal the redistribution of electron density around the N center induced by N–H···N hydrogen bonding at the imine linkage in (COF)_2.5-NH₂-M. Similarly, the signal around 0.6 ppm (–NH₂) in ¹H NMR of (COF)_2.5-NH₂-M becomes broader in TAPB-TaBr₂ COF and sharper in Ti₃C₂T_x-NH₂, indicating the extensive formation of N–H···N hydrogen bonding at the interface to offset the electron shielding effect by conductive matrix of MXene (Fig. 2d). This assertion is further evidenced by the variable-temperature NMR results under ultrafast MAS conditions, presenting an increasing intensity of –NH₂ peaks due to the weakening hydrogen bonds with temperature rising²³ (Fig. 2e and Supplementary Fig. 19). Furthermore, the distinct signal around 40 ppm in ¹³C NMR spectrum of Ti₃C₂T_x-NH₂ is slightly shifted in (COF)_2.5-NH₂-M (Fig. 2f), reflecting the effect of hydrogen bonding on the carbon atoms adjacent to the amino terminals. Scanning electron microscopy (SEM) images show that the TAPB-TaBr₂ COF forms granules ~2 μm in diameter (Supplementary Fig. 20a, b). With the incorporation of Ti₃C₂T_x-NH₂, COF particles uniformly assembled and stacked onto the metallic MXene nanosheets via supramolecular interactions (Supplementary Figs. 20c–f). In Fig. 2g, the polycrystalline domains of TAPB-TaBr₂ COF appear widely distributed and interspersed with darker Ti₃C₂T_x-NH₂ fringes by high-resolution transmission electron microscopy (HR-TEM). Each COF domain displays a honeycomb-like hexagonal framework with a pore size of 3.52 nm, while the MXene lattice fringes exhibit an interplanar spacing of 3.61 nm (Fig. 2h, i). At higher resolution (Supplementary Fig. 21), well-defined boundaries indicate effective supramolecular bridging. Correspondingly, energy-dispersive spectroscopy (EDS) mappings (Supplementary Figs. 22–25) and line scans (Fig. 2j) on different sites corroborate the existence of uniform supramolecular structure arising from hydrogen-bonded connections.

Energy levels and photogenerated carrier dynamics

The electronic structure governs charge carrier dynamics of COFs, which is crucial for evaluating the performance of supramolecular catalysts in practical photosynthesis applications. Theoretical calculations show the highest occupied molecular orbital (HOMO, Fig. 3a) is localized on the donor knots (TAPB), while the lowest unoccupied molecular orbital (LUMO, Fig. 3b) resides on the acceptor linkers (TaBr₂), establishing segregated oxidation and reduction sites. To analyze its band structure, the semiconductor energy levels of TAPB-TaBr₂ COF and the work function of Ti₃C₂T_x-NH₂ were separately measured (Fig. 3c). The TAPB-TaBr₂ COF yields an optical bandgap of 2.50 eV, as determined from the Tauc plot (Supplementary Fig. 26a). The conduction band minimum (CBM) of TAPB-TaBr₂ COF was determined to be −1.58 V (versus normal hydrogen electrode (NHE), pH 6.8) from the Mott-Schottky plot (Supplementary Fig. 26b), which indicates sufficient driving force for ORR into H₂O₂ via the superoxide anion intermediate (O₂/^•O₂⁻, − 0.35 V). Meanwhile, its valence band maximum (VBM) lies at 0.92 V (NHE, pH 6.8), suggesting the involvement of 4e⁻ WOR (0.83 V). Ultraviolet photoemission spectroscopy (UPS) (Supplementary Fig. 26c, d) reveals a work function of 3.97 eV for metallic Ti₃C₂T_x-NH₂, promoting the formation of Schottky junction with accelerated charge transfer²⁴. Based on the band structures predicted by DFT (Fig. 3d, Supplementary Fig. 27, and Supplementary Data 1), the AA-stacked TAPB-TaBr₂ COF exhibits the largest bandgap and highest reduction potential to trigger dual H₂O₂ production pathways. According to the projected density of states (PDOS) for TAPB-TaBr₂ COF (Fig. 3d), the energy states near the band edge are primarily contributed by p-orbital electrons. Specifically, the Br, N, C atoms on the acceptor linkers exhibit greater electron deficiency, as evidenced by their significantly higher p-orbital electron density at the conduction band edge (Supplementary Figs. 28 and 29). Moreover, the electrostatic potential calculations reveal that these acceptor linkers carry the most negative average charges which induce photo-generated carrier localization and O₂ adsorption (Fig. 1d).

Fig. 3. Band structure and photogenerated carrier dynamics of supramolecular (COF)_x-NH₂-M.

a HOMO distribution of TAPB-TaBr₂ COF. b LUMO distribution of TAPB-TaBr₂ COF. c Experimentally derived energy band alignments of TAPB-TaBr₂ COF and Ti₃C₂T_x-NH₂. d Simulated band structure and PDOS of TAPB-TaBr₂ COF. e Electrochemical impedance spectra of (COF)_x-NH₂-M (solid circle for dark condition; hollow circle for 100 mW cm⁻² light irradiation; orange for x = 50; green for x = 5; blue for x = 2.5). f Steady-state photoluminescence of (COF)_x-NH₂-M. g Transient fluorescence decay lifetime of the TAPB-TaBr₂ COF and (COF)_2.5-NH₂-M (open circle for data point; solid line for fitting results).

Next, we investigated the dynamics of photogenerated carriers in (COF)_x-NH₂-M. Nyquist plots from electrochemical impedance spectroscopy (EIS) analysis indicate an obvious light response for (COF)_x-NH₂-M (Fig. 3e). By further fitting with an equivalent circuit, the charge transfer resistance is determined to reduce due to enhanced charge migration with increasing MXene content, under both dark conditions and 1 sun (standard solar illumination of 100 mW cm⁻²) irradiation (Supplementary Fig. 30 and Supplementary Table 6). Steady-state photoluminescence (PL) spectra in Fig. 3f show a progressive reduction in emission intensity of (COF)_x-NH₂-M’s PL around 600 nm to 640 nm, indicative of more efficient exciton dissociation and suppressed electron-hole recombination. Correspondingly, time-resolved fluorescence measurements further corroborate these findings with the average carrier lifetime extending from 0.69 ns to 2.54 ns as MXene loading increases, reflecting enhanced charge extraction by the conductive MXene domains (Fig. 3g, Supplementary Fig. 31, and Supplementary Table 7). Conclusively, integrating metal-like MXene with the COF creates a supramolecular heterostructure with tailored band structures and optimized carrier dynamics, driving dual-route H₂O₂ production.

Photo-to-thermal conversion performance

Among the (COF)_x-NH₂-M series samples, obvious color changes of the yellow COF are observed (Supplementary Fig. 32a). Ultraviolet-visible-near-infrared diffuse reflectance spectroscopy (UV-Vis-NIR DRS) was employed to investigate the full-spectrum optical properties of (COF)_x-NH₂-M. As shown in Fig. 4a, metallic Ti₃C₂T_x-NH₂ nanosheets exhibit a strong absorption bands around 670 nm and 1100 nm due to the LSPR effect²⁵. Compared to the TAPB-TaBr₂ COF, (COF)_x-NH₂-M shows a substantial broadening of absorption across the entire NIR region, which indicates an improved sunlight utilization beyond 250–500 nm. In addition, NIR absorption increases with MXene content in the supramolecular system (Supplementary Figs. 32b–d). Therefore, the MXene units functionalized supramolecular material that we developed provides opportunities for tailoring light response in the NIR region.

Fig. 4. Light response and photocatalytic performance of supramolecular (COF)_x-NH₂-M.

a Light absorption spectra of Ti₃C₂T_x-NH₂ (dark blue),TAPB-TaBr₂ COF (red), and (COF)_x-NH₂-M (orange for x = 50; green for x = 5; blue for x = 2.5) produced by Kubelka-Munk function. b Photothermal temperature profiles of thin film samples under 100 mW cm⁻² light irradiation. c Localized temperature variation induced by light-to-heat conversion at a selected spot of (COF)_2.5-NH₂-M. d Topography and surface potential maps of a selected area of (COF)_2.5-NH₂-M at dark condition and UV-NIR irradiation (scale bar: 200 nm). e H₂O₂ production of TAPB-TaBr₂ COF and (COF)_x-NH₂-M over 90 min. Reaction condition: deionized water (10 mL), catalyst (1 mg), 300 W Xe lamp (full spectrum), peak light intensity: 200 mW cm⁻², without aeration or sacrificial agent. f H₂O₂ production of TAPB-TaBr₂ COF and (COF)_2.5-NH₂-M over 60 min under different gas conditions (seal: reaction solution is purged with relevant gas for 30 min before reaction and then sealed during the reaction; flow: reaction solution is purged with relevant gas for 30 min before reaction and continuously purged throughout the reaction process).

A thermal imaging camera and scanning photothermal microscopy (SPThM) were used to evaluate the enhanced photothermal heating capability of (COF)_x-NH₂-M at both macroscopic and localized scales (Supplementary Fig. 33a). As illustrated in Fig. 4b and Supplementary Figs. 33b–f, the temperature of a self-supported Ti₃C₂T_x-NH₂ film rapidly reached around 72 °C after exposure to simulated 1 sun irradiation. Under identical conditions, the surface temperature of (COF)_x-NH₂-M rose from ~38 °C (pure COF) to 57 °C (x = 2.5), attributed to the photothermal effect of the Ti₃C₂T_x-NH₂ nanosheets. Under such intentionally controlled MXene loading, the elevated temperature remains within a favorable range for H₂O₂ synthesis kinetics³. This optimized temperature range, together with the enhanced stability of Ti₃C₂T_x-NH₂, effectively suppresses the secondary decomposition (Supplementary Figs. 34 and 35). For localized photo-to-thermal measurements, SPThM equipped with a UV-NIR light source probed illuminated areas on TAPB-TaBr₂ COF and (COF)_2.5-NH₂-M samples. The MXene photothermal units induce a more pronounced temperature rise, which in turn produced proportional changes in the tip voltage due to heat flow from the heated samples to the SPThM tip²⁶. By precisely positioning the SPThM tip on the catalyst surface, a higher light-induced temperature increase of 1.8 °C for (COF)_2.5-NH₂-M nanoparticles were observed versus 1.0 °C for TAPB-TaBr₂ COF under the same irradiation conditions, indicating a superior photothermal response (Fig. 4c).

To visualize localized heating effects throughout the supramolecular catalysts, Kelvin probe force microscopy (KPFM) with an integrated light source was employed to map light-induced variations on surface potential²⁷. LSPR-induced hot-electron generation on the MXene surface, together with interfacial charge transfer within (COF)_x-NH₂-M increases free-electron density and the measured surface potential. As shown in the topography and surface potential maps (Fig. 4d), UV-NIR light illumination on (COF)_2.5-NH₂-M sample produced a significant increase in surface potential by 51 mV relative to dark conditions, while only marginal change of 12 mV was observed on TAPB-TaBr₂ COF sample (Supplementary Fig. 36). Together, the KPFM and SPThM results reveal enhanced localized photo-to-thermal conversion in (COF)_x-NH₂-M, which indicates both elevated temperature for accelerated O₂ activation and enhanced charge transfer for driving photocatalytic reaction.

Photocatalytic H₂O₂ production

The photocatalytic activities of (COF)_x-NH₂-M were assessed under simulated sunlight in an open batch reactor by monitoring H₂O₂ formation directly from water and air via colorimetry (Supplementary Fig. 37). H₂O₂ production commenced spontaneously upon the addition of (COF)_x-NH₂-M dispersion and light irradiation, with no induction period, gas bubbling, pH adjustment, or sacrificial reagents. Under optimized catalyst dosage (Supplementary Fig. 38a) and reaction temperature, (COF)_2.5-NH₂-M achieved an enhanced H₂O₂ production of 1.77 mmol g⁻¹ within 30 min, around 1.81 times that of TAPB-TaBr₂ COF (Fig. 4e). Notably, H₂O₂ production rates scaled with MXene content, whereas a pure Ti₃C₂T_x-NH₂ suspension yielded negligible 15 μM over 3 h (Supplementary Fig. 39), highlighting the superior performance contributed by our supramolecular design on catalysts. Photocatalytic H₂O₂ yield experiments conducted under irradiation with different wavelength regions revealed a distinct decrease in activity when the NIR and UV components were excluded (Supplementary Fig. 40). This behavior confirms the existence of a synergetic photo- and photothermal-catalytic effect that accelerates the reaction and enhances the overall catalytic efficiency.

Considering the energy band alignment in Fig. 3c, we proceeded to investigate the dual-route pathways with 2e⁻ ORR and 4e⁻ WOR throughout the catalytic process. To probe the role of O₂ in the photocatalytic H₂O₂ production, reactions were performed under different gas atmospheres (Fig. 4f). Continuous Ar/N₂ flow to deoxygenate the dispersion suppressed H₂O₂ formation for both TAPB-TaBr₂ COFs and (COF)_2.5-NH₂-M dispersions, while O₂ flow significantly increased the H₂O₂ yield. Together with the likewise quenched H₂O₂ production with electron scavenger (AgNO₃) addition (Supplementary Fig. 38b), a more specific rotating ring-disk electrode (RRDE) measurements in an O₂-saturated phosphate buffer solution (PBS, pH = 7) confirmed the highly selective 2e⁻ ORR pathway for catalytic process (Supplementary Fig. 41a, b). Noticing the persisted H₂O₂ generation when the reactor was purged with inert gas only before reaction, ORR is expected to sustain owing to the in situ O₂ produced via the 4e⁻ WOR (oxygen evolution reaction, OER, 0.83 V) which is thermodynamically favored over the 2e⁻ WOR (1.36 V). Beyond this, the distinct current response for RRDE curves in Ar-saturated PBS (Supplementary Fig. 41c) coincides with the increasing ¹⁸O₂/¹⁶O₂ ratios with irradiation time in isotope labeling experiment for sealed photocatalytic system starting with ¹⁶O₂ saturated H₂¹⁸O solution (Fig. 5a), confirming the existence of 4e⁻ WOR. Additionally, the ¹⁸O₂ signal remains detectable when the extracted aqueous phase after irradiation was then purged with Ar and injected into a vial containing MnO₂, depicting that the in situ generated O₂ from WOR could subsequently participate in ORR toward H₂O₂.

Fig. 5. In situ analysis and reaction pathways.

a ¹⁸O₂ isotope labeling experiment for (COF)_2.5-NH₂-M system dispersed in ¹⁶O₂ saturated H₂¹⁸O. b In situ EPR spectra of TAPB-TaBr₂ COF and (COF)_2.5-NH₂-M reaction solution under dark and light irradiation for different time to capture ^•O₂⁻ intermediate with DMPO. c In situ DRIFTS spectra of (COF)_2.5-NH₂-M collected at different photoirradiation time under O₂ and H₂O environment (blue arrows: characteristic peaks of oxygen species). d O₂ adsorption energies on aromatic carbon (linker), imine linkage, and oxygen vacancies of MXene (red column, TAPB-TaBr₂ COF; blue column, (COF)_x-NH₂-M). e Gibbs free energy diagram for all intermediates in the oxygen reduction reaction over aromatic carbon (linker) site (red line, TAPB-TaBr₂ COF; blue line, (COF)_x-NH₂-M). f Gibbs free energy diagram for all intermediates in the oxygen evolution reaction over aromatic carbon (knot) site (red line, TAPB-TaBr₂ COF; blue line, (COF)_x-NH₂-M).

To verify the possible intermediates involved in photocatalytic production, in situ electron paramagnetic resonance (EPR) spectroscopy was conducted in methanol/H₂O using 5,5-dimethyl-1-pyrroline N-oxide (DMPO) as spin-trapping reagent. As depicted in Fig. 5b, the EPR spectra of both TAPB-TaBr₂ COF and (COF)_2.5-NH₂-M dispersion exhibited intensified characteristic peaks of O₂⁻ with increasing irradiation time, while no signals were detected under dark condition, indicating light-induced generation of O₂⁻ intermediate. Meanwhile, the more distinct signals in (COF)_2.5-NH₂-M system suggest accelerated O₂ activation upon light irradiation. In situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) in a sealed chamber with the presence of O₂ and H₂O was employed to further elucidate H₂O₂-formation pathways and identify reactive sites on (COF)_x-NH₂-M (Fig. 5c). A broad peak at 886 cm⁻¹ corresponding to O–O bonding^13,28,29 gradually increases with irradiation time, suggesting the accumulation of H₂O₂. Moreover, the concurrent emergence and intensification of the OOH bending mode of surface-adsorbed H₂O₂ (*H₂O₂) at 1396 cm⁻¹ further reveal in situ H₂O₂ generation^30,31. Peaks at 1160 cm⁻¹ and 1260 cm⁻¹ assigned to ^•O₂^{− 28,32} and *OOH^30,31,33,34, respectively, further support the formation of key intermediates in a two-step ORR pathway (Supplementary Fig. 27e). Compared to DRIFTS without O₂ flow (Supplementary Fig. 42a), enhanced aromatic signals between 1400–1600 cm⁻¹ indicate that the aromatic rings serve as O₂ docking site. Notably, in situ Raman spectra corroborate the occurrence of ^•O₂⁻, evidenced by the signal magnification at 1091 cm⁻¹ over time³⁵ (Supplementary Fig. 42b).

To elucidate the docking mechanism underlying the enhanced H₂O₂ photosynthesis, we combined DFT analysis with spectroscopic observations and temperature-programmed measurements on TAPB-TaBr₂ COF and (COF)_x-NH₂-M assemblies. The DRIFTS results show time-dependent variations in the peak intensities of aromatic carbon moieties under photoirradiation, suggesting their likely involvement as active sites (Fig. 5c). Based on this observation and the LUMO distribution of TAPB-TaBr₂ COF (Fig. 3b), the imine linkages and adjacent aromatic carbons are selected as candidate O₂ adsorption sites for comparative analysis (Supplementary Figs. 43–45). O₂ adsorption energy calculations on TAPB-TaBr₂ COF (Fig. 5d, red columns) indicate that O₂ docking at aromatic carbons (0.36 eV) is energetically more favorable than at imine linkages (0.86 eV). The incorporation of Ti₃C₂T_x-NH₂ units in (COF)_x-NH₂-M contributes to the electron abundance on COF framework with lower O₂ adsorption energy at aromatic carbons (0.06 eV), suggesting the facilitated O₂ docking at COF framework (Fig. 5d, blue columns). In comparison, oxygen vacancies of Ti₃C₂T_x-NH₂ exhibit a high adsorption energy (0.81 eV), implying limited likelihood of effective O₂ activation at these sites. To experimentally evaluate the O₂ docking capability of materials, oxygen temperature-programmed desorption (O₂-TPD) was performed to probe the adsorption-desorption behavior of surface oxygen species. Prior to the O₂-TPD measurements, thermogravimetric analysis verified the thermal stability of the TAPB-TaBr₂ COF over the relevant temperature range (Supplementary Fig. 46). The TPD profiles exhibit a pronounced low-temperature desorption feature (~150 °C), which can be attributed to reversible oxygen species that are weakly bound at π-conjugated aromatic carbon sites³⁶ (Supplementary Fig. 47). Notably, the increasing MXene content leads to a pronounced enhancement of the low-temperature desorption, indicating the improved O₂ docking capability in (COF)_x-NH₂-M. Furthermore, Gibbs free energy calculations disclose a lower energy barrier toward forming the *O₂ intermediate at the aromatic carbon site of (COF)_x-NH₂-M (Fig. 5e), which is the rate-determining step of ORRs. Similarly, OER is expected to proceed via a single-site (*O/*OOH) pathway at the aromatic carbons located at the TAPB edge, where (COF)_x-NH₂-M exhibits a lower energy barrier (1.993 eV) than TAPB-TaBr₂ COF (2.492 eV) (Fig. 5f). It should be noted that the calculated free-energy profiles serve as thermodynamic descriptors of elementary steps and do not directly represent absolute photocatalytic reaction kinetics. Nevertheless, the consistency between the computed energetic trends and the experimentally established reaction pathways, together with the enhanced catalytic performance of the supramolecular systems, provides qualitative support to the proposed ORR/WOR mechanism and more favorable reaction energetics upon MXene incorporation.

Practical potential assessment

To investigate the stability of (COF)_2.5-NH₂-M, a recycling test was performed, presenting a negligible H₂O₂ yield decay and minor structural changes within 5 cycles (Supplementary Fig. 48). To evaluate the scalability and real-world applicability of our photocatalytic system, we designed an integrated flow reactor incorporating (COF)_x-NH₂-M dispersions. Coupled with a peristaltic pump and a collection reservoir, the aqueous (COF)_2.5-NH₂-M dispersion was continuously circulated through modular tandem reactors. This setup enabled uninterrupted H₂O₂ generation over 24 h under laboratory conditions without noticeable activity loss (Supplementary Fig. 49), which can be attributed to the robust π-conjugated framework of (COF)_x-NH₂-M supramolecular assembly. To further validate practical viability, we deployed a self-powered outdoor system driven by a photovoltaic cell (Fig. 6a), which sustained a steady production of H₂O₂ at a peak rate of 658 μmol g⁻¹ h⁻¹ under ambient conditions (light intensity <1 sun) as shown in Supplementary Fig. 50.

Fig. 6. Practical application and performance of (COF)_x-NH₂-M.

a Digital photograph of scale-up reactor under ambient environment. b A schematic demonstration of an integrated industrial system for capture-synthesis-utilization. c Catalytic stability of flow-system with operation over 1000 h. Test condition: deionized water (0.5 mL min⁻¹), catalyst (15 mg), 300 W Xe lamp (full spectrum), peak light intensity: 100 mW cm⁻², without sacrificial agent. d H₂O₂ production of TAPB-TaBr₂ COF (red) and (COF)_2.5-NH₂-M (blue) over 60 min using different water source. e Removal efficiencies of MO and TFP pollutants by (COF)_2.5-NH₂-M over 60 min (w/o: without catalyst; w/: with catalyst).

Importantly, in many decentralized applications, such as environmental disinfection, food preservation, and green chemistry, the required H₂O₂ concentration at the point of use is typically below 0.4 wt%³⁷. This stands in stark contrast to conventional AO processes, which generate highly concentrated H₂O₂ solution that requires subsequent distillation and poses substantial safety risks during transport and storage. To address this, we proposed a modular collection platform (Fig. 6b) to directly produce controllable concentrations of H₂O₂ on-site, thus eliminating the demands for energy-intensive post-processing and mitigating explosion hazards. In addition, immobilizing (COF)_2.5-NH₂-M supramolecular catalyst on a solid substrate facilitated continuous replenishment of the reaction medium and prolonged the operational lifespan while maintaining consistent productivity. As a result, the total amount of H₂O₂ accumulated linearly over time (Supplementary Figs. 51 and 52). Among state-of-the-art photocatalytic H₂O₂ production system^{11,13,17,29,34,38–46}, our system delivers a competitive production rate of 2,878 μmol g⁻¹ h⁻¹ without requiring O₂ bubbling, pH adjustment, or sacrificial agent. Notably, this is the first demonstration of continuous operation for 1,008 h in a fully integrated setup (Fig. 6c and Supplementary Table 8). This combination of operational simplicity, long-term durability, and point-of-use compatibility offers a promising pathway for advancing photocatalytic H₂O₂ production toward real-world applications.

We also evaluated the photocatalytic performance of (COF)_2.5-NH₂-M in deionized water, simulated seawater, and real seawater. Only a slight decrease in H₂O₂ production was observed (Fig. 6d), suggesting that the material retains high activity despite potential interference from competing ions. To further demonstrate practical applicability, we conducted in situ pollutant removal tests using methyl orange (MO) and 2,4,6-triformylphloroglucinol (TFP) as representative dye and phenolic pollutants (Supplementary Fig. 53). Under 1 sun irradiation, rapid degradation of MO (95%) and TFP (77%) was achieved within 60 min (Fig. 6e), confirming not only the reactivity of the produced H₂O₂ but also the system’s broad potential for decentralized water treatment scenarios. The on-demand generation of dilute H₂O₂ solutions minimizes the need for transporting, storing, and handling concentrated oxidants, thereby offering an inherently safer and more sustainable disinfection route. As evidenced by the global solar irradiance map (Supplementary Fig. 54), the widespread distribution of solar resources supports the feasibility of deploying (COF)_x-NH₂-M-integrated systems across diverse regions. Operating under ambient conditions without external energy input or hazardous reagents, this platform offers an environmentally benign, decentralized, and scalable alternative to the conventional AO H₂O₂ industry.

Discussion

Our work proposes a metallic unit-functionalized supramolecular docking strategy for targeted reactant adsorption to overcome longstanding challenges in the intrinsic activity and catalytic selectivity of COFs. By engineering precise σ–π stacking interactions in (COF)_x-NH₂-M, we not only synergistically enhance of charge carrier dynamics and light absorption but also facilitates site-specific capture of O₂ and H₂O, enabling H₂O₂ production under ambient conditions without external gas feeds, pH-regulation, or sacrificial agents. The demonstrated activity of 2878 μmol g⁻¹ h⁻¹ in batch reactors and over 1008 h of stable production in continuous flow highlights robustness and scalability of the system. The generated H₂O₂ can be utilized in situ for efficient pollutant removal, without limitations from the water source. Beyond these benchmarks, our supramolecular docking strategy provides a modular and generalizable platform for designing next-generation COF-based catalysts with programmable structure–function relationships. By integrating reactant capture, chemical conversion, and product utilization “capture-synthesis-utilization” into a single system, this approach unlocks precision-driven and distributed chemical manufacturing.

Methods

Materials

TAPB, TaBr₂, mesitylene, 1,4-dioxane, acetic acid, ethanol, acetone, lithium fluoride, hydrochloric acid (HCl, 37%), APTES, potassium hydrogen phthalate (KHP), potassium iodide (KI), and Nafion (5 wt% in the mixture of lower aliphatic alcohols and water) were purchased from Adamas. D₂O (≥99.9 at% D), Water-¹⁸O (≥97 at% ¹⁸O), and DMPO (≥97%) were purchased from Aladdin. Titanium aluminum carbide (Ti₃AlC₂, MAX) was purchased from Laizhou Kai Kai Ceramic Materials Co. Ltd. All chemicals were used as received without further purification.

Synthesis of TAPB-TaBr₂ COF

To synthesize TAPB-TaBr₂ COFs, 0.4 mmol TAPB (140.58 mg) and 0.6 mmol TaBr₂ (175.15 mg) were added into a Teflon reactor and ultrasonically dispersed for 20 min in 10 mL mixed solvent of mesitylene/1,4-dioxane (v: v = 1:4). Subsequently, 1 mL acetic acid solution (6 M) was added and followed by another 5 min ultrasonication. The reactor was then sealed and transferred to an oven at 120 °C. After 3 days of solvothermal reaction, the precipitate was washed three times with acetone and ethanol respectively and collected by suck filtration through hydrophobic membranes. Finally, the product was ground with ethanol and dried in vacuum oven at 60 °C for 12 h.

Synthesis of Ti₃C₂T_x-NH₂

The Ti₃C₂T_x suspension (30 mg in 3 mL ethanol and 12 mL H₂O), obtained by wet etching of HCl/LiF and ethanol addition, was stirred with 1 mL APTES for 12 h at ambient temperature. After centrifuging sequentially in ethanol and H₂O, the precipitate was freeze-dried for 2 days to obtain Ti₃C₂T_x-NH₂.

Synthesis of (COF)_x-NH₂-M

To synthesize (COF)_x-NH₂-M, Ti₃C₂T_x-NH₂ powder was finely ground and then sonicated to disperse in the mixture of 0.4 mmol TAPB (140.58 mg), 0.6 mmol TaBr₂ (175.15 mg), and 10 mL mixed solvent of mesitylene/1,4-dioxane (v: v = 1:4), where the Ti₃C₂T_x-NH₂ content was determined by the mass ratio to the feeding amount of COF monomers (x). After the addition of 1 mL acetic acid solution (6 M), the following sonication time was prolonged to 20 min to further ensure thorough mixing of the COF monomers and Ti₃C₂T_x-NH₂ nanosheets. Finally, the product after subsequent solvothermal reaction and collection were conducted inductively coupled plasma optical emission spectroscopy (ICP-OES) and CHNS measurements for elemental analysis to verify the repeatability.

Photocatalytic H₂O₂ synthesis and quantification

For batch reactor, 1 mg of photocatalyst was dispersed in 10 mL deionized water. A 300 W Xe lamp (CEL-PE300-3A, China Education Au-light) was calibrated with a light intensity meter and used to drive the photocatalytic reaction with magnetic stirring at 150 rpm. The reaction solution was sampled at certain intervals and passed through hydrophilic syringe filters. The yield of H₂O₂ was determined by iodometry method (0.5 mL reaction solution + 0.5 mL deionized water + 1 mL 0.1 M KHP + 1 mL 0.4 M KI) and followed by UV-Vis spectroscopy to determine the concentration of I₃⁻ oxidized by H₂O₂. For flow reactor, 15 mg photocatalyst was dispersed in the mixture of 4.5 mL ethanol and 0.5 mL Nafion (binder), subsequently deposited on the substrate. Deionized water was continuously pumped into the reactor and the diluted H₂O₂ solution was collected in the source container. A controlled amount of catalyst was periodically replenished to offset stripping.

Materials characterization

The valence state and work function of as-prepared samples were measured by XPS (Al Kα source) and UPS (ESCALAB 250Xi, Thermo Fisher Scientific). The PXRD patterns were recorded on a diffractometer (D8 Advance, Bruker, Cu Kα radiation), together with BET test (3Flex, Micromeritics) to further analyze the crystal structure and mesoporosity. The chemical bonding was determined by Raman spectroscopy (WITec alpha300, Oxford Instruments) and FTIR (IRPrestige-21, Shimadzu). The morphology was revealed by SEM (JSM-7001F, JEOL) and HRTEM (JEM-F200, JEOL). Steady-state PL spectra and time-resolved fluorescence lifetime were recorded by spectrofluorometer (FLS1000, Edinburgh) under 364 nm excitation light. The reflectance and absorbance spectra were obtained using an UV-Vis-NIR spectrophotometer (UV-3600, Shimadzu). The chemical adsorption characteristics were evaluated by thermogravimetric analysis (STA200, HITACHI) and O₂-TPD (TP-5080B, Xianquan).

Independent gradient model based on Hirshfeld partition (IGMH)

The DFT calculations on molecular cluster model was performed using the Gaussian 16 package⁴⁷. Geometry optimizations and vibrational frequency calculations were carried out with the BL3YP-D3^48–50, employing the lanl2dz basis set for Br atom and the 6-31 G (d)⁵¹ basis set for other elements. Using optimized structure, the IGMH⁵² was employed to analyze with Multiwfn⁵³ and visualize with Visual Molecular Dynamics⁵⁴ software.

Nuclear magnetic resonance (NMR) measurements

The solution-state ¹H NMR (Varian mercury 400, Agilent, D₂O as solvent) was applied to analyze the successful grafting of –NH₂ groups on Ti₃C₂T_x. The solid-state ¹H and ¹³C CP/MAS NMR (Avance Neo 400 M, Bruker) were performed to validate the existence of N–H···N hydrogen bonding interactions, which were further verified by the variable-temperature ¹H solid-state NMR (Avance Neo 400WB, Bruker) measurements at MAS = 15 kHz.

Electrochemical measurements

The electrochemical impedance spectroscopy (EIS) and Mott-Schottky measurements were conducted using a CHI660E electrochemical workstation configured in a standard three-electrode system. The working electrode was prepared by coating the catalyst on fluorine-doped tin oxide substrates, while a platinum plate and an Ag/AgCl electrode served as the counter and reference electrodes, respectively. All measurements were carried out in the 0.2 M NaSO₄ solution¹³ (pH = 6.8). EIS was analyzed with a frequency ranging from 100 kHz to 0.1 Hz. The position of conduction band minimum (CBM) can be converted into normal hydrogen electrode (NHE) using the following equation:

$$E_{NHE\left(pH=6.8\right)}=E_{AgCl/Ag}+0.197V$$ 1

Scanning photothermal microscopy (SPThM) and Kelvin probe force microscopy (KPFM)

SPThM and KPFM were carried out on a commercial scanning probe microscopy system (MFP-3D, Asylum Research, Oxford Instruments). During the measurements, the samples were irradiated with a metal halide fiber optic illuminator (Model MH-100), covering a full UV-NIR spectrum.

¹⁸O₂ isotope labeling experiments

The ¹⁸O₂ isotope was measured on a gas chromatograph mass spectrometer (GC-MS, GCMS-QP2010Ultra, Shimadzu). (COF)_2.5-NH₂-M photocatalyst (2 mg) and H₂¹⁸O ( ≥ 97 atom% ¹⁸O, 1 mL) were added into a sealed quartz tube and pumped with ¹⁶O₂ in the dark for 30 min. After 2 h and 12 h irradiation, the 60 μL of gas products in the headspace were extracted and analyzed by GC-MS. The remaining liquid phase of mixture was pumped with Ar for 30 min to remove the remaining O₂ and subsequently added with MnO₂ to generate O₂ by the decomposition of photogenerated H₂O₂.

Rotating ring-disk electrode (RRDE) measurements

The ORR and WOR were studied by using a RRDE with (COF)_2.5-NH₂-M, which was coated using a uniformly dispersed mixture (5 mg photocatalyst in 20 μL 5% Nafion and 180 μL isopropanol). The RRDE assembly included a glassy carbon disk (r = 4 mm) and a platinum ring (r = 4.3 mm/6.3 mm). For ORR, 0.1 M PBS (pH = 7) saturated with O₂ was used as electrolyte, with platinum plate as counter electrode and Ag/AgCl electrode as reference electrode. The ring/disk currents were measured by an SP-150e potentiostat (Biologic). The H₂O₂ product was detected at the ring potential of 0.6 V (vs. Ag/AgCl). The electron transfer number (n) is calculated as follows:

$$n=\frac{4I_d}{I_d+\frac{I_r}{N}}$$ 2

where I_d is the disk electrode current, I_r is the ring electrode current, and N is the collection efficiency.

Similarly, the WOR was analyzed in Ar saturated solution and detected at the ring potential of −0.23 V (vs. Ag/AgCl).

Electron paramagnetic resonance (EPR) spectroscopy

The catalysts (2 mg) were separately dispersed into a methanol/H₂O mixture (9/1 v/v, 500 μL) containing 0.1 mmol DMPO as a spin-trapping reagent to detect ^•O₂⁻ intermediate. Before reaction, the dispersion was purged with O₂ gas for 5 min. The resulting mixture was extracted and sealed in a capillary tube, followed by Xe lamp irradiation for subsequent testing.

In situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS)

In situ DRIFTS measurements were conducted in an in situ reaction cell (Hefei in situ technology) and detected by a modified FTIR spectrometer (Nicolet iS50, Thermo Fisher Scientific) equipped with an MCT detector cooled by liquid N₂. Before the measurement, the photocatalyst with a small amount of water was filled into the holder and the chamber was purged with He/O₂ gas mixture at a flow rate of 10 mL min⁻¹ for 30 min to record the baseline. Subsequently, the Xe lamp was used to irradiate the reaction cell and the IR spectra were collected with time.

Computational details

The simulation of periodic COF structure was conducted on the BIOVIA Materials Studio suite of programs to perform molecular dynamics and DFT calculations within the generalized gradient approximation (GGA)⁵⁵ using the Perdew-Burke-Ernzerhof (PBE) functional. The geometry structure was initially established in the Forcite module and followed by optimization in the DMoL³ module. During the structural optimization, we employed an energy convergency smaller than 2 × 10⁻⁵ Ha. Pawley refinement was then predicted using Reflex module with convergent R_wp and R_p. For energy calculation, we used the k-point set of 1 × 1 × 4 to sample the electronic structure.

To simulate the H₂O₂ generation, we employed the Vienna Ab initio Simulation Package (VASP)^56,57 at GGA-PBE level with k-points of 1 × 1 × 1. The interaction between valence electrons and ionic cores was treated with the projected augmented wave method⁵⁸ with an energy cutoff of 450 eV. To account for long-range dispersion interactions, the DFT-D3 empirical correction was incorporated⁵⁹.

More discussions of the DFT calculation are listed in the supplementary information.

Author contributions

G.W.H. and Y.M.Z. guided the project. J.X.S. and Y.M.Z. conceived the research and designed the experiments. J.X.S. prepared the materials and carried out the experiments. W.H.L. performed the AFM and KPFM experiments. J.X.S., Y.M.Z., and W.L.O. contributed to the batch reactor and flow reactor design. X.L.P. contributed to the FTIR and Raman measurements. J.X.S., Y.M.Z., and X.L.P. performed the materials studio calculations. W.T.S. conducted the photoluminescence tests. Y.M.Z., Z.H.L., and Z.S.L. performed the in situ DRIFTS test. J.X.S., Y.M.Z., and G.W.H. wrote the paper. All authors read and commented on the paper.

Peer review

Peer review information

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

Data availability

All data that support the findings of this study are available within the article and Supplementary Information. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70693-9.