Charge transfer in triphenylamine–tetrazine covalent organic frameworks for solar-driven hydrogen peroxide production

Abstract

Hydrogen peroxide is a clean and valuable oxidant that can be produced by photocatalysis under visible light. Developing efficient metal-free photocatalysts for this reaction is a key challenge. To address this challenge, we make two crystalline vinyl-linked covalent organic frameworks, ATP-COF-1 and ATP-COF-2, which contain triphenylamine and tetrazine units. ATP-COF-1 has a hydrogen peroxide production rate of 14,000 μmol g⁻¹h⁻¹ with an apparent quantum yield of 23.05%, while ATP-COF-2 reaches 12,700 μmol g⁻¹h⁻¹ and 20.38%. The higher activity originates from efficient photoinduced charge separation, driven by electron transfer between the electron-donating triphenylamine and electron-accepting tetrazine components. Ultrafast spectroscopy confirms prolonged excited-state lifetimes of 351.9 and 277.3 picoseconds, consistent with enhanced charge mobility. In this work, we show that integrating donor-acceptor building blocks within covalent organic frameworks enables efficient and stable visible-light-driven hydrogen peroxide production.

Metal-free ATP-COFs efficiently produce hydrogen peroxide under visible light, achieving high rates and selectivity via enhanced charge separation from donor–acceptor building blocks in stable covalent organic frameworks.

Introduction

Hydrogen peroxide (H₂O₂) is a valuable chemical oxidant widely used in industrial applications¹. Traditional methods of H₂O₂ production, such as the anthraquinone process, are energy-intensive and environmentally problematic². In the photo-driven two-electron oxygen reduction reaction (2e⁻ ORR) photocatalysts absorb light energy, initiating charge separation and facilitating the reduction of oxygen to H₂O₂. This approach aligns with the principles of green chemistry, offering several advantages over conventional methods, including milder reaction conditions, reduced carbon emissions, and higher selectivity for H₂O₂. However, achieving an efficient 2e⁻ ORR is hindered by slow reaction kinetics^3–5.

The key challenge in photocatalytic H₂O₂ production is to develop catalysts that selectively promote the two-electron oxygen reduction reaction (2e^– ORR) pathway while suppressing competing reactions, such as the four-electron oxygen reduction reaction (4e^– ORR), which results in the formation of water instead of H₂O₂⁶. A critical focus of research is the design of photocatalysts that increase the separation and transfer of photogenerated charge carriers. Covalent organic frameworks (COFs) have emerged as promising candidates to provide stable active sites for catalysis and provide the opportunity to intentionally incorporate specific functional groups that can modulate catalytic activity^7–9.

COFs with vinylene linkages have gained attention for their chemical stability^10,11. In comparison to traditional linkages like boroxine, boronate ester, or imine, which are prone to hydrolysis, vinylene-linked COFs are more resistant to degradation, making them suitable for use in photocatalytic applications where long-term stability is critical^12,13. This increased stability not only preserves the structural integrity of the COF but also supports its sustained photocatalytic performance.

Among the various building blocks available for COF synthesis, triphenylamine (TPA) and tetrazine have gained attention due to their electrochemical properties^14,15. TPA is a well-known electron-donator that can be easily oxidized to form radical cations (TPA^•+)¹⁶. While tetrazine serves as a strong electron-accepting unit that can be reduced to form radical anions¹⁷. The integration of the electron donor, TPA, and the electron acceptor, tetrazine, within a COF structure is expected to enhance electron transfer and facilitate more efficient charge carrier separation.

In this work, we report the synthesis of two covalent organic frameworks (ATP-COF-1 and ATP-COF-2) that integrate TPA as an electron-donating unit and tetrazine as a strong electron-accepting component. This donor-acceptor design aims to enhance charge separation and directional charge transfer during photocatalysis. The study focuses on understanding how these electronic interactions influence photocatalytic hydrogen peroxide production through the 2e⁻ ORR pathway. Comprehensive experimental and theoretical analyses were performed to correlate the structural and electronic features of ATP-COFs with their photocatalytic efficiency and selectivity toward H₂O₂ generation.

Results

A key aspect of this work is the strategic selection of linkers that influence both the electronic and structural properties of the resulting COFs. The aldehyde precursors, 4,4’,4”-nitrilotribenzaldehyde (NTBA) and 4’,4”’,4””’-nitrilotris(([1,1’-biphenyl]-4-carbaldehyde (NTBCA), introduce extended conjugation and enhanced charge delocalization within a COF framework due to their triphenylamine cores. These linkers provide a stable π-electron network while facilitating charge transport, which is critical for photocatalysis. Combined with the integration of 3,6-dimethyl-1,2,4,5-tetrazine (DMTAZ) which imparts a strong electron-accepting character, the electron transfer pathways are enhanced and reinforced. This combination within a vinylene-linked COF ensures both structural robustness and optimized electronic interactions, enabling effective H₂O₂ production via the 2e⁻ ORR pathway (Fig. 1a).

Fig. 1. Structural design and crystallinity of ATP-COFs.

a Representation of the synthetic pathway, reaction conditions, and structural framework of ATP-COFs. ACN Acetonitrile, TFA Trifluoroacetic acid. PXRD patterns of ATP-COF-1 (b) and ATP-COF-2 (c): experimental data (red), fitted patterns from Pawley refinement (black), simulated profiles for AA-eclipsed stacking (blue) and AB stacking (green), Bragg positions (purple bars), and residuals between observed and refined data (dark blue). Insets display top and side views of the ATP-COFs in the eclipsed (AA) stacking configuration (cyan: carbon; gray: hydrogen; blue: nitrogen).

The crystal structures of the synthesized ATP-COFs were analyzed both theoretically, using Materials Studio simulations (Supplementary Figs. 1 and 2), and experimentally through powder X-ray diffraction (PXRD). The experimental data aligned well with the simulated crystal structures, supporting the structural stability and ordered framework of ATP-COFs (Fig. 1a, b). The Pawley refinement of the PXRD patterns confirmed that both ATP-COF-1 and ATP-COF-2 adopt an AA-eclipsed stacking arrangement (Fig. 1b, c, black curves), supporting the peak assignments and yielding precise unit cell parameters. Additional details on the optimized structures and Pawley-refined results are provided in the Supplementary Information. Crystallographic data, including simulated structures (.cif format), Pawley refinement results, and experimental crystal data, are also provided as Supplementary Data 1.

High porosity and large surface area are critical design elements for COFs intended for catalytic applications, as they directly impact reactant diffusion, accessibility to active sites, and overall efficiency. In our study, ATP-COFs exhibited a microporous nature with exceptionally high surface areas (Fig. 2), making them well-suited for catalysis. The porosity of ATP-COFs was assessed through N₂ adsorption-desorption measurements (Fig. 2). For the multipoint Brunauer-Emmett-Teller (BET) analysis of ATP-COFs, the selection of the P/P₀ range (0.01–0.15) ensured a linear relationship that reflects accurate adsorption behavior (Supplementary Fig. 13). In this range, the BET theory assumes that the adsorption is a multilayer process, where the coverage of the surface by adsorbed nitrogen molecules increases with pressure until it reaches saturation¹⁸. The BET specific surface areas were found to be 1517 m² g⁻¹ for ATP-COF-1 and 1275 m² g⁻¹ for ATP-COF-2, corresponding to approximately 57.9% and 49.8% of their theoretical Connolly surface areas (2619 m² g⁻¹ for ATP-COF-1 and 2559 m² g⁻¹ for ATP-COF-2). The total pore volumes for ATP-COF-1 and ATP-COF-2 were measured at 0.570 cm³g⁻¹ and 0.466 cm³g⁻¹, respectively, representing about 36.2% and 20.5% of their theoretical total pore volumes (1.574 cm³g⁻¹ for ATP-COF-1 and 2.271 cm³g⁻¹ for ATP-COF-2). The observed type-IV isotherm behavior signifies the microporous nature of the structures, which is further supported by the pore size distribution data (Fig. 2). These structural attributes enhance mass transport and catalytic efficiency, making ATP-COFs promising materials for photocatalysis applications.

Fig. 2. Porosity and surface area analysis of ATP-COFs.

N₂ adsorption-desorption isotherm curves for ATP-COF-1 (a) and ATP-COF-2 (c). The isotherms exhibit type-IV behavior, indicating mesoporous structures. Insets in (a, c) present the Brunauer-Emmett-Teller (BET) specific surface area, derived from the adsorption data, and the Connolly surface area, calculated from the framework structure. Pore size distribution for ATP-COF-1 (b) and ATP-COF-2 (d).

Fourier transform infrared (FT-IR) spectroscopy, solid-state ¹³C cross-polarization magic-angle spinning (CP/MAS) NMR, and X-ray photoelectron spectroscopy (XPS) were used to confirm the structural integrity of ATP-COFs. Thermal, solvent, and chemical stability were evaluated. The microstructure of ATP-COFs was examined using field-emission scanning electron microscopy (FE-SEM) and high-resolution transmission electron microscopy (HRTEM) imaging. The results align with expected parameters for these materials. Additional details are available in Supplementary Figs. 3–16 and Supplementary Discussion.

The ultraviolet-visible diffuse reflectance spectroscopy (DRUV) analysis of the ATP-COFs revealed their ability to absorb a broad range of visible light, with absorption edges extending beyond 650 nm (Supplementary Fig. 17a). This extended absorption range indicates that these COFs are capable of harnessing photons across a significant portion of the visible spectrum, which is advantageous for photocatalytic applications as it maximizes the wide range of useful wavelengths afforded by sunlight. Such broad absorption suggests strong electronic interactions within the COF frameworks, due to extended π-conjugation or the innate electronic properties of the linker groups, which facilitate efficient light absorption and photon trapping.

The optical band gaps of 2.08 and 2.15 eV, derived from Kubelka-Munk plots, indicated that both ATP-COFs exhibit semiconducting properties (Supplementary Fig. 17b)¹⁹. In addition, the calculated HOMO-LUMO gap was 1.89 eV for ATP-COF-1 and 1.96 eV ATP-COF-2 which aligned closely with the experimental findings. TPA units, known for their strong electron-donating characteristics, host a higher electron concentration in HOMO, facilitating charge donation (Supplementary Fig. 18). Meanwhile, tetrazine’s strong electron-accepting behavior leads to greater electron density in LUMO. This spatial separation highlights an effective electron-donor (TPA) to electron-acceptor (tetrazine) interaction, promoting charge transfer efficiency and supporting the COFs’ semiconducting properties.

ATP-COFs fluoresced at maximum emissions at approximately 584 and 571 nm when excited at 395 nm, indicating efficient energy transfer processes (Supplementary Fig. 17c). The fluorescence lifetimes were recorded at 3.02 and 2.76 ns, respectively, indicating that both ATP-COFs exhibit similar excited-state decay dynamics, with neither dominated by rapid quenching nor long-lived trapping, thus reflecting balanced photophysical behavior (Supplementary Table 1 and Supplementary Fig. 17d). The higher emission peak and longer fluorescence lifetime in ATP-COF-1 are due to restricted rotational freedom and increased rigidity of the vinyl-linked donor-acceptor framework, which stabilize the electrons in the excited state. In contrast, the additional benzene rings in ATP-COF-2 increase rotational freedom, reducing π-delocalization, as a result the electrons are less stable in the excited state and the fluorescence lifetime is shorter. These structural features, combined with the extended π-conjugation from TPA and tetrazine linkers, further extend electron delocalization, lowering the band gap and facilitating visible-light absorption²⁰.

We used a casting method to deposit ATP-COFs onto ITO substrates, ensuring consistent film quality essential for photocatalytic applications. Cross-sectional and top-view FE-SEM analyses (Fig. 3a and Supplementary Fig. 19a) reveal that ATP-COF thin films exhibit uniform thickness, measuring around 5.3 µm for ATP-COF-1 and 6.8 µm for ATP-COF-2. This uniform layering is advantageous for device applications, as consistent thickness improves photon absorption and promotes charge separation efficiency²¹.

Fig. 3. Photophysical and electrochemical analysis.

a SEM image of drop-cast ATP-COF-1 film on ITO glass with cross-sectional and top-view insets. b Cyclic voltammetry (CV) curves of ATP-COF-1 against RHE in 0.2 M Na₂SO₄ aqueous solution (pH 6.8 ± 0.1). c Mott-Schottky plots of ATP-COF-1 at 1.0, 1.5, and 2.0 kHz, characterizing the ATP-COF-1 film in 0.2 M Na₂SO₄ solution (pH 6.8 ± 0.1), with RHE as the reference electrode in the dark. d UPS spectrum of ATP-COF-1. e Energy band diagram of ATP-COF-1, comparing HOMO-LUMO levels to the O₂/O₂^•− reduction potential (E_red). f Transient photocurrent responses of ATP-COFs measured at −0.8 V bias (vs. RHE), highlighting dynamic performance. g 2D and (h) 3D AFM images of ATP-COF-1 film, showing thickness and surface features. i Grazing Incidence Diffuse X-ray Scattering (GIDXS) pattern of the ATP-COF-1 film, showing distinct scattering arcs corresponding to crystallographic planes, confirming the film’s preserved crystallinity and preferred orientation. The diffuse arc at Q_z = 1.53 Å⁻¹ highlights the out-of-plane π–π stacking with an interlayer distance of ~ 4.10 Å, indicating perpendicular alignment of COF layers relative to the substrate. All voltages are uncorrected for ohmic drop (iR).

The three-dimensional atomic force microscopy (AFM) images show the film’s surface with an average roughness of 5–6 μm, indicative of a textured, continuous layer (Fig. 3g, h and Supplementary Fig. 19b, c). Such uniform surface morphology and thickness are critical in optimizing optical and electronic properties, directly influencing the films’ stability and catalytic performance under visible light²².

Grazing incidence diffuse X-ray scattering (GIDXS) confirmed the crystallinity of the ATP-COF films and the uniform molecular orientation of the COF layers. (Fig. 3i and Supplementary Fig. 19d). The GIDXS pattern displays well-defined scattering peaks at 0.19 Å⁻¹, 0.33 Å⁻¹, 0.47 Å⁻¹, 0.86 Å⁻¹, and 1.53 Å⁻¹, corresponding to the (100), (110), (200), (210), and (001) planes, respectively, in the out-of-plane direction²³. These peaks align well with those observed in the PXRD profile of ATP-COF-1 powder, indicating that the crystalline structure is preserved in the film. The scattering intensity concentrated near Q_z = 0 Å⁻¹ suggests that the c-axis of the COF framework is primarily oriented perpendicular to the substrate. Additionally, the (001) reflection presents as a diffuse arc at Q_z = 1.53 Å⁻¹, pointing to π–π stacking interactions with an interlayer distance of about 4.10 Ų⁴. This observation supports the conclusion of a vertical c-axis alignment relative to the substrate surface. Notably, the diffraction features do not show a strong preference for “edge-on” or “face-on” alignment, suggesting an isotropic distribution of the ATP-COF-1 sheets²⁵. This uniform arrangement likely facilitates consistent charge transport pathways, contributing to enhanced photocatalytic efficiency in H₂O₂ production.

The band structure of the ATP-COFs was confirmed using cyclic voltammetry (CV) which identified the oxidation potential (E_ox,onset) and estimate the highest occupied molecular orbital (HOMO), as well as the lowest unoccupied molecular orbital (LUMO) levels. This approach was based on the method proposed by Brédas et al.²⁶, which relates ionization potential and oxidation potential, along with electron affinity and reduction potentials (E_red) in organic materials. To obtain reliable measurements, we used slow scan rates, ensuring full oxidation or reduction of the ATP-COF films. This technique allowed us to pinpoint the initial oxidation potentials versus Ag/AgCl of +1.3 V (1.90 V vs. reversible hydrogen electrode (RHE)) for ATP-COF-1 and +0.89 V (1.49 V vs. RHE) for ATP-COF-2 (Fig. 3b and Supplementary Fig. 20).

Mott-Schottky analysis confirmed that ATP-COFs possess n-type semiconducting behavior with conduction band minimum (CBM) potentials significantly more negative than the reduction potential of O₂ to superoxide radicals (O₂/O₂^•−). This electronic structure provides a strong thermodynamic driving force for oxygen reduction, making ATP-COFs effective for photocatalytic applications²⁷. For n-type semiconductors, such as certain COFs with π-conjugated linkers like vinylene and tetrazine, the Mott-Schottky plot reveals a positive slope, indicating that electrons serve as the primary charge carriers²⁸. The E_fb were determined to be −0.34 V for ATP-COF-1 and −0.76 V for ATP-COF-2 vs. the RHE, measured in an aqueous 0.2 M Na₂SO₄ solution at pH 6.8 ± 0.1. These E_fb values, determined from the linear region of the Mott-Schottky plot, inform the relative positions of conduction and valence bands, aligning well with the COFs’ suitability for redox reactions under visible light (Fig. 3c and Supplementary Fig. 21). The alignment of their CBM with the oxygen reduction potential (E_{red, O2/O2•−}= 0.07 V vs. RHE) further supports their role as efficient photocatalysts for oxygen reduction reactions under visible-light activation²⁹.

The valence band maximum (VBM) measured at −6.39 eV for ATP-COF-1 and −5.98 eV for ATP-COF-2 by ultraviolet photoelectron spectroscopy (UPS), relative to the vacuum level further confirmed n-type semiconductor behavior (Fig. 3d and Supplementary Fig. 22). These VBM values were consistent with the Mott-Schottky analysis. Using the optical band gaps, the CBM values were calculated as −4.31 eV (−0.13 V vs. RHE at pH 6.8 ± 0.1) for ATP-COF-1 and −3.83 eV (−0.60 V vs. RHE at pH 6.8 ± 0.1) for ATP-COF-2. These values were in good agreement with the E_fb obtained from the Mott-Schottky analysis, reinforcing the strong alignment of the COFs’ band positions for efficient photocatalytic applications.

The energy level diagrams for ATP-COF-1 and ATP-COF-2 (Fig. 3e and Supplementary Fig. 23) show that the CBM positions are significantly more negative than the oxygen reduction potential required to reduce O₂ to superoxide radicals. These band alignments indicate that ATP-COFs are well designed to catalyze oxygen reduction reactions effectively by promoting electron transfer to oxygen molecules.

Upon exposure to visible light (λ > 400 nm) from a xenon lamp, both ATP-COFs generated transient photocurrents, showing clear responses during light on/off cycles (Fig. 3f). The chopped current-potential plots revealed a strong photo-response, where ATP-COF-1, under a bias potential of −0.8 V vs. RHE, produced a cathodic photocurrent with a maximum current density of ~ 32 μA cm⁻², surpassing that of ATP-COF-2, which achieved ~21 μA cm⁻². This higher photocurrent of ATP-COF-1 suggests a more efficient internal charge transfer mechanism³⁰. Compared to other COF-based photocathodes and recently developed carbon nitrides, ATP-COFs demonstrated much better stability and efficiency during repeated light cycling, with photocurrent densities remaining constant over a 400-second test period (Supplementary Table 2). This durability highlights the strong photochemical resilience of ATP-COFs, particularly in sustaining reliable charge separation and transfer under continuous light exposure. These properties suggest ATP-COFs are well-suited for applications in oxygen reduction reactions to generate hydrogen peroxide, where efficient and stable electron flow is essential for sustained catalytic performance.

In O₂-saturated water under visible light (λ > 420 nm), ATP-COF-1 photocatalytically produced 140.93 μmol of H₂O₂ within one hour while ATP-COF-2 produced 127.06 μmol of H₂O₂ within one hour (Fig. 4a). This means that the H₂O₂ production rates were 14,000 µmol g⁻¹ h⁻¹ for ATP-COF-1 and 12,706 µmol g⁻¹ h⁻¹ for ATP-COF-2 (Supplementary Table 3). These values are competitive with those reported for other COFs designed for photocatalytic H₂O₂ generation.

Fig. 4. Photocatalytic H₂O₂ production performance and stability of ATP-COFs.

a Comparison of photocatalytic H₂O₂ generation rates for ATP-COFs. b Stability assessments of ATP-COFs during photocatalytic H₂O₂ production in O₂-saturated pure water under visible light (λ > 420 nm). c Rate constants for H₂O₂ formation (k_f) and decomposition (k_d). d Apparent quantum yield (AQY) for H₂O₂ production across various wavelengths, overlaid with the diffuse reflectance UV-Vis (DRUV) spectra of ATP-COFs. Experimental setup: 2 mg photocatalyst in 5 mL PBS (pH 7.4), O₂-saturated, illuminated by monochromatic LED light. e Wavelength-dependent AQY for ATP-COFs in H₂O₂ production. f Influence of different O₂ concentrations on the photocatalytic H₂O₂ generation by ATP-COFs. g Comparison of H₂O₂ production by ATP-COFs under different conditions: water, water/IPA, and water/IPA with the addition of benzyl alcohol (BA), benzoquinone (BQ), and AgNO₃ (5 mL, 10 mM aqueous solution, 10 mg ATP-COF), all irradiated for 1 h (300 W Xe lamp, λ > 420 nm). h Koutecky-Levich plots from rotating disk electrode (RDE) analysis of ATP-COFs in 0.1 M Na₂HPO₄/NaH₂PO₄ solution, with O₂ at −0.80 V (vs. RHE). i EPR spectroscopy of DMPO-O₂^•− adducts in aqueous dispersions of ATP-COF-1.

Both COFs retained consistent H₂O₂ production rates over multiple recycling cycles (Fig. 4b). The PXRD analysis after 72 h of continuous photocatalysis (Supplementary Fig. 25) confirmed that the ATP-COFs maintained their structural integrity, and the FT-IR spectra of ATP-COFs showed no significant changes in the functional groups (Supplementary Fig. 26). The FE-SEM images showed that the ATP-COFs maintained their crystalline structure and morphology even after extended photocatalytic use (Supplementary Fig. 27). Structural stability is critical to ensure consistent catalytic performance through the photocatalysis. Previous studies have emphasized that preserving the catalyst frameworks is essential for their effectiveness and ATP-COFs durability under harsh reaction conditions makes it suitable chemical processes where stability plays a critical role³¹.

Extended photocatalytic H₂O₂ production experiments under continuous visible-light irradiation for 96 hours were done to evaluate the long-term operational stability of the ATP-COFs. ATP-COFs maintained a consistently high H₂O₂ generation rate throughout the duration, with only a slight decrease observed after 72 hours (Supplementary Fig. 28a). This minor decline is attributed to extrinsic factors such as reactant depletion, accumulation of intermediate products, or light-induced surface changes, rather than catalyst degradation. PXRD patterns recorded after the long-term test (Supplementary Fig. 29) showed no significant change in peak positions or intensities, confirming that the crystalline structure of ATP-COFs remained unchanged.

The pH tolerance of ATP-COFs was assessed by performing H₂O₂ production experiments in acidic (pH ~ 3.0, adjusted with 0.05 M HCl) and alkaline (pH ~ 12.0, adjusted with 0.05 M NaOH) media. While both COFs retained over 85% of their original activity in basic conditions after repeated cycles, ATP-COF-1 retained ~75% of its initial performance under strongly acidic conditions (Supplementary Fig. 28b). This moderate activity loss in low pH is likely due to partial protonation of functional groups or changes in surface wettability.

Strong interactions with reactants is important for photocatalytic processes where surface adsorption is important, ATP-COF-1 and ATP-COF-2, with formation constants (k_f) values of 18.6 and 16.0, respectively indicating the equilibrium is favorable toward adsorption (Fig. 4c).

A lower decomposition constant (k_d) indicates how readily the adsorbed species dissociates and implies a more stable complex, meaning that once the reactants are bound, they are less likely to decompose or desorb quickly. For ATP-COF-1 and ATP-COF-2, the k_d values were 0.39 and 0.34, respectively and suggests that the adsorbed species on the surfaces of these materials are relatively stable and will likely remain on the surface long enough to undergo photocatalytic reactions, contributing to a more effective catalytic cycle.

High apparent quantum yields (AQYs) indicate efficient charge separation and transfer within the COF which is necessary as a driving force for water oxidation and oxygen reduction to produce H₂O₂. The ATP-COFs exhibited strong photocatalytic performance for H₂O₂ production, achieving high AQYs of 23.05% and 20.38% for ATP-COF-1 and ATP-COF-2, respectively, under 420 nm illumination. These values correlate well with the visible-light absorption profiles of the COFs, suggesting that light harvesting plays a critical role in their activity (Fig. 4d)³². While the AQY trends generally mirror the absorption spectra, slightly elevated AQYs were observed at longer wavelengths (~700 nm), where absorbance is minimal. This anomaly likely arises from instrumental factors common in monochromator-based systems, such as second-order diffraction or light leakage at long-wavelength cutoffs. These effects can introduce higher-energy photons into the beam, artificially inflating AQY values³³. Similar trends have been observed in previous studies, where apparent quantum yields remained high at longer wavelengths¹⁰. Therefore, while AQYs at shorter wavelengths reliably reflect true photocatalytic activity, values at 700 nm should be interpreted with caution. In comparison to other top-performing COFs for H₂O₂ generation, ATP-COF-1 and ATP-COF-2 showed higher activity and efficiency (Supplementary Table 3).

Light-to-chemical energy conversion efficiency of the materials, was evaluated by calculating the solar-to-chemical conversion (SCC) efficiencies under visible-light irradiation (λ = 420 nm). ATP-COF-1 and ATP-COF-2 achieved SCC efficiencies of 4.70% and 4.23%, respectively, reflecting their high solar use and catalytic performance for sustainable H₂O₂ production (Supplementary Table 3).

A series of control and scavenger experiments were done to determine the influence of oxygen availability and the roles of reactive species. In O₂-saturated conditions, H₂O₂ production reached peak values (140.93 µmol h⁻¹ for ATP-COF-1 and 127.06 µmol h⁻¹ for ATP-COF-2), but under air or reduced O₂, H₂O₂ yield dropped (Fig. 4f). Control experiments confirmed that H₂O₂ generation was negligible in both N₂-saturated environments and in the dark, confirming both the necessity of visible light and that O₂ was a reactant.

The reactive oxygen species involved in the photocatalytic formation of H₂O₂ were identified by both scavenger experiments and electron paramagnetic resonance (EPR) spectroscopy. The addition of electron and O₂^•− scavengers markedly decreased H₂O₂ production, while hydroxyl radical scavengers had minimal impact. This indicated that both electrons and O₂^•− are essential to H₂O₂ formation, whereas ^•OH radicals contribute only negligibly (Fig. 4g). Additionally, the EPR spectra recorded using 5,5-dimethyl-1-pyrroline N-oxide (DMPO) as a spin trap showed no detectable DMPO–^•OH adducts under either dark or light conditions (Fig. 4i). While this does not categorically rule out the presence of trace ^•OH species due to the limitations of spin trapping under high or low pH conditions, it does strongly suggest that ^•OH does not play a dominant role in H₂O₂ production. By contrast, eight well-defined peaks corresponding to DMPO–O₂^•− adducts emerged under visible-light illumination, confirming the generation of superoxide radicals as the dominant reactive species (Fig. 4i and Supplementary Fig. 30). Thus, ATP-COFs promote the selective 2e⁻ ORR, wherein O₂ is reduced to H₂O₂ via the O₂^•− intermediate.

To further investigate the potential involvement of singlet oxygen (¹O₂), we performed EPR measurements using 2,2,6,6-tetramethylpiperidine (TEMPO) as the spin probe. No substantive variation was observed in the TEMPO–1O₂ signal under light exposure (Supplementary Fig. 31), indicating that ¹O₂ is not a contributing intermediate. This comprehensive EPR analysis supports a mechanistic pathway dominated by photogenerated electrons driving the 2e⁻ ORR through O₂^•− intermediates, without the involvement of ¹O₂ or ^•OH radicals.

To exclude contributions from catalyst leaching, reaction supernatants were analyzed by GC-MS. No evidence of ATP-COF fragments, linker species, or leached organic species was detected, confirming the structural stability and heterogeneity of ATP-COFs during photocatalysis (Supplementary Fig. 32).

The detection of O₂^•− as an intermediate confirms the oxygen reduction reaction (ORR) pathway leading to H₂O₂ production per Eq. 10 through 21 in the Supporting Information. The role of O₂^•− was further supported by control experiments using p-benzoquinone, a known scavenger of O₂^•−, which notably reduced H₂O₂ yield when added, confirming the importance of superoxide radicals in the reaction. Additionally, EPR spectroscopy showed that ATP-COF-1 generated a slightly stronger O₂^•− signal compared to ATP-COF-2 under similar light conditions. This difference suggests that more photo-generated electrons in ATP-COF-1 are involved in O₂ reduction, increasing the overall H₂O₂ yield. This suggests a higher efficiency in O₂^•− formation, likely due to a greater number of active sites or better electron transfer dynamics in ATP-COF-1’s structure.

To verify and measure O₂^•− generation in ATP-COFs suspensions, Nitroblue tetrazolium (NBT) was used to quantify the O₂^•− generation; it forms a purple insoluble formazan product with superoxide radicals. This reaction enables the detection and quantification of O₂^•− by tracking the decrease reduction in the NBT absorbance at 259 nm³⁴.

In experiments using ATP-COF suspensions containing NBT under simulated sunlight, UV-Vis spectra were recorded at different time intervals to monitor the decrease in NBT adsorption, and therefore, indirectly O₂^•− generation (Supplementary Fig. 33). ATP-COF-1 produced more O₂^•− than ATP-COF-2 under identical conditions, with absorbance changes in NBT allowing concentration tracking over time via a standard curve at 259 nm. Calculations based on the reaction stoichiometry revealed nearly linear O₂^•− generation rates, reaching 17.36 × 10⁻⁵ mol L⁻¹ for ATP-COF-1 and 14.98 × 10⁻⁵ mol L⁻¹ for ATP-COF-2 after 60 min of irradiation (Supplementary Fig. 34). These results highlight ATP-COF-1’s higher capability for superoxide production compared to ATP-COF-2.

The ORR was analyzed using rotating disk electrode (RDE) and rotating ring-disk electrode (RRDE) systems in an O₂-saturated phosphate buffer solution (PBS) at pH 7.4. Linear sweep voltammetry (LSV) measurements were taken at various rotation speeds (Supplementary Fig. 35), showing that higher rotation speeds led to increased current density highlighting the role of mass transport in influencing the ORR kinetics. The Koutecky-Levich plots at −0.80 V versus RHE (Fig. 4h) showed that the electron transfer kinetics and the influence of O₂ diffusion rates on the reaction mechanism resulted in better catalytic performance of the ATP-COFs for H₂O₂ production.

ATP-COF-1 showed higher selectivity for H₂O₂ production than ATP-COF-2, prompting us to investigate alternative reduction pathways. This led to an additional experiment to assess the impact of excess donor-acceptor interactions on H₂O₂ formation (Supplementary Fig. 36). The kinetic current densities (j_k) were determined as 16.07 and 12.18 mA cm⁻², respectively, with ATP-COF-1 exhibiting a higher j_k, indicative of more effective electron transfer for oxygen reduction³⁵.

In addition to the electronic structure and charge transport features, the porosity of COFs can influence photocatalytic performance by affecting mass transport and substrate accessibility. ATP-COF-2 possesses a slightly larger average pore size (4.70 nm) and higher pore volume (0.570 cm³g⁻¹) than ATP-COF-1 (3.18 nm and 0.466 cm³g⁻¹, respectively). Importantly, ATP-COF-1 exhibits a higher BET surface area (1517 m² g⁻¹) than ATP-COF-2 (1275 m² g⁻¹). The larger surface area of ATP-COF-1 facilitates greater substrate accessibility and more exposed active sites for the oxygen reduction reaction, which is reflected in its higher H₂O₂ production rate. Despite the role that physical attributes may play in the photocatalytic performance, we emphasize that factors such as donor-acceptor interactions, electronic delocalization, and ultrafast charge separation are more influential.

DFT calculations, together with in-situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS), help to distinguish between the 2-electron oxygen reduction reaction (2e⁻ ORR) and the 4-electron water oxidation reaction (4e⁻ WOR), providing a theoretical perspective that complements experimental observations. The limitations of DFT under realistic catalytic conditions should be considered when interpreting these results. During the photocatalytic process, the reduction of O₂ via the 2e⁻ ORR is driven by photogenerated electrons, while water undergoes the 4e⁻ WOR, facilitated by photogenerated holes. The calculated structures for critical intermediates involved in both the 2e⁻ ORR and 4e⁻ WOR on ATP-COF-1 (Supplementary Fig. 37) highlight the importance of specific adsorption sites. The vinylene linkage (site 1) emerges as the primary O₂ adsorption site, where O₂ interacts with nitrogen atoms on the tetrazine ring to form endo-peroxide species (OO), a key step in the selective 2e⁻ ORR pathway.

DFT analysis suggests that additional reactive sites may exist at nitrogen-rich areas within the tetrazine rings (site 2), potentially facilitating electron transfer for the ORR pathway. Carbon atoms adjacent to the tetrazine ring appear to carry partial positive charge, creating localized charge separation with the negatively charged tetrazine rings. This environment could enhance O₂ binding and activation, in line with prior studies on electron density distribution in catalytic sites of covalent organic frameworks^36,37. Additional DFT calculations, including localized orbital locator (LOL) analysis and the behavior of frontier molecular orbitals upon oxygen adsorption, are provided in the Supporting Discussion (Supplementary Figs. 38–40). We note that these DFT results are derived from idealized models and may not fully capture the dynamic catalytic conditions in the experimental system.

The in-situ diffuse reflectance infrared Fourier transform (DRIFT) spectra of ATP-COFs, collected in an oxygen atmosphere and in the dark had weak signals corresponding to O₂ adsorption (Fig. 5a, b). Peaks linked to endo-peroxide (1214 cm⁻¹), O₂^•− (1093 cm⁻¹), and O – O bonds (962 cm⁻¹) increased (Fig. 5a), suggesting O₂ uptake and a two-stage reduction mechanism that produces O₂^•− 38. Moreover, with increased photoirradiation, an increase in peak intensity was observed at 2817 cm⁻¹, corresponding to the O – H bending vibration of H₂O₂ (Fig. 5b). These experimental observations, paired with theoretical models, suggest that the ORR primarily occurs at the vinylene bond and adjacent tetrazine sites in ATP-COFs. This is likely influenced by the *OOH species adsorbing at the carbon-vinylene bond, leading to symmetry disruption of the C = C linkers and heightened polarization, which amplifies absorbance³⁹. The catalytic importance of the *OOH intermediate at ATP-COF’s vinylene-linked sites was confirmed by the vibrational modes indicating a strong interaction between H₂O₂ and the COF structure (Fig. 5c).

Fig. 5. Spectroscopic and mechanistic insights into charge transfer and active sites.

a In situ DRIFT spectra of ATP-COF-1 measured in the dark and during different periods of photoirradiation in an O₂ atmosphere, covering the (a) 2000–800 cm⁻¹ and (b) 2900–2500 cm⁻¹ ranges. c Catalytic site after the adsorption of the *OOH intermediate, along with the vibrational modes of H₂O₂. d 3D contour plot and (e) spectra of femtosecond transient absorption (fs-TA) for ATP-COF-1, recorded at specified delay times using a 350 nm laser pulse. f Structure showing organized pathways for charge and mass flow within a hexagon comprising ten layers: the tetrazine linkers (blue) facilitate photogenerated electron movement (red arrows), while the TPA units (blue) serve as sites for hole collection (green arrows). Micropores channel water molecules to the TPA units and supply oxygen to the tetrazine linkers. The overall process involves multilayered hexagonal ATP-COF enabling electron transfer (red dashed arrows) from TPA units to tetrazine linkers, water oxidation using four holes at the TPA sites, and oxygen reduction at tetrazine linkers with two electrons to produce H₂O₂ (blue arrow). Each TPA unit at the six vertices provides three pathways for donor knots, facilitating light absorption or electron transfer to neighboring vinyl-tetrazine linkers, completing charge separation in less than 6.0 ps.

In ATP-COFs, the vinylene bond and tetrazine linkages facilitate the adsorption of the *OOH intermediate, a key precursor in the reduction of oxygen to H₂O₂. Upon adsorption, this intermediate changes the vibrational modes of the O – H and O – O stretches of H₂O₂. The O – O stretch (typically around 960 cm⁻¹ in the DRIFTS spectrum) and the H − O – O − H bending vibrations (around 1384 cm⁻¹) are often sensitive to the local electronic environment, including bond polarization in the vicinity of the vinylene bond. That means that this bond in ATP-COFs, in particular, disrupts the symmetry of the carbon-carbon bond linkers, which increases the bond polarization and, in turn, strengthens the adsorption of O₂ and *OOH intermediates.

The tetrazine further contributes to the stabilization of *OOH intermediates. The nitrogen-rich tetrazine ring offers strong electron-withdrawing properties, which polarizes the adsorbed *OOH species and shifts the electron density within the H₂O₂ molecule. This change in electron distribution influences the vibrational modes of H₂O₂, making them more detectable and assists the selective formation of H₂O₂ rather than reduction to water.

The TPA group, with its electron-donating properties, works synergistically by donating electrons to the catalytic site, stabilizing reactive intermediates, and improving the efficiency of the oxygen reduction process. TPA also affects the overall electronic structure of the COF, influencing the vibrational modes of adsorbed H₂O₂ by modifying the local charge density around the vinylene and tetrazine sites

In summary, the interplay between the vinylene bond, tetrazine, and TPA groups alters the vibrational modes of H₂O₂. This alteration facilitates H₂O₂ formation by improving the interaction between the catalyst and the intermediate species, thereby promoting the 2e⁻ ORR pathway.

Femtosecond transient absorption (fs-TA) spectroscopy confirmed that ATP-COFs efficiently generate and localize photogenerated electrons, a key step in driving the 2e⁻ ORR and 4e⁻ WOR for H₂O₂ production. The strong and persistent ground-state bleach (GSB) signal observed at 440–640 nm (Fig. 5d, e, and Supplementary Fig. 40) indicates efficient light absorption and charge separation, while the broad excited-state absorption (ESA) at long

wavelengths reveals that photogenerated electrons are effectively trapped within tetrazine units. ATP-COF-1 exhibited stronger and more prolonged signals than ATP-COF-2, which we attributed to the shorter distance between tetrazine and TPA junctions, enhancing electron accumulation and promoting greater H₂O₂ yield (Fig. 5d, e and Supplementary Fig. 41). A prominent negative peak was observed in the range of 440–640 nm, indicative of ground-state bleach (GSB), a characteristic signature of electronic transitions from the ground state to excited states which confirms efficient light absorption and photoinduced charge generation in ATP-COFs. In addition, a broad positive signal appeared rapidly, within ~1 ps, in the 640–780 nm region and persisted throughout the delay window, corresponding to excited-state absorption (ESA). This feature is attributed to photogenerated electrons stabilized within the electron-accepting tetrazine units, reflecting effective charge separation and electron localization at these sites⁴⁰.

To validate this assignment, fs-TA experiments on a structurally similar COF without donor-acceptor units were done. The control COF was prepared from 1,3,5-triformylbenzene and p-phenylenediamine. The resulting spectra showed only a weak ESA signal in the 650–700 nm range, with minimal growth over time delays up to 1000 ps (Supplementary Fig. 42). This confirms that the pronounced and early-onset ESA signal observed in ATP-COFs arises from their donor-acceptor architecture and not from neutral excitons, hot carriers, or instrumental noise. Such ultrafast electron dynamics are consistent with behavior reported in other donor-acceptor COF systems^41,42.

ATP-COF-1 had stronger and more persistent signals for both GSB and ESA than ATP-COF-2, which we attribute to a shorter distance between the tetrazine and TPA junctions, which not increase the accumulation of photogenerated electrons in the system⁴¹.

To understand how charge trapping influences the photocatalytic efficiency of ATP-COFs for H₂O₂ production, we analyzed electron trapping dynamics through decay kinetics at 650 nm within the first 1000 ps. Detailed discussion, including tri-exponential fitting and time constants, is provided in the Supplementary Discussion (Supplementary Fig. 43).

Our ATP-COFs use water and oxygen from the atmosphere to drive a photosynthetic reaction and produce H₂O₂ by combining rapid charge separation with efficient charge transport (Fig. 5f). We have found that the photogenerated electrons are localized within the tetrazine acceptor linkers (Fig. 5f, red arrows), where they drive the reaction O₂ + 2H⁺ + 2e⁻ → H₂O₂ for hydrogen peroxide production. Meanwhile, the holes accumulate in the TPA units, promoting the oxidation of water through the reaction 2H₂O → O₂ + 4H⁺ + 4e⁻ (Fig. 5f, green arrows). This synergy completes the photosynthetic process, 2H₂O + O₂ → 2H₂O₂.

To verify the origin of the oxygen atoms in the produced H₂O₂, an isotopic labeling study was done using H₂¹⁸O. If water were the oxygen source, we would expect to detect ¹⁸O-labeled H₂O₂. A peak at 35.9917 m/z in the mass spectrum corresponding to ¹⁸O₂ rather than H₂¹⁸O₂ indicates that the oxygen in H₂O₂ originates from molecular O₂ rather than water (Supplementary Fig. 44). This was convincing evidence that the reaction follows an oxygen reduction pathway rather than direct water oxidation.

Further, calculations were used to model key intermediate states involved in both 2e⁻ ORR and 4e⁻ WOR pathways (Fig. 6). In the initial step of the 2e⁻ pathway model, a hydrogen atom adsorbs at a nitrogen site on the tetrazine ring of ATP-COFs (highlighted by a yellow dashed circle in Fig. 6a). This interaction causes radical anion formation to occur on the tetrazine unit and the nitrogens of the tetrazine site are protonated. Following this, molecular oxygen attaches to a carbon within the vinylene linkage, where it reacts with *H to produce the *OOH intermediate. This intermediate is further stabilized through intramolecular hydrogen bonding between the tetrazine nitrogen and the *OOH hydrogen atom (Supplementary Fig. 45, top). Supporting evidence from DRIFTS shows vibrational peaks at 994 cm⁻¹ (C–O–O bonds on vinylene carbons) and 1214 cm⁻¹ (associated with *OOH), with an additional peak at 1093 cm⁻¹ indicating O₂^•⁻ formation (Fig. 5a).

Fig. 6. DFT-derived reaction pathways for ORR and WOR on ATP-COFs.

a, b Structural representations illustrating key reaction intermediates on ATP-COFs during the oxygen reduction reaction (ORR) and water oxidation reaction (WOR) pathways, with active sites highlighted by yellow dashed circles. Carbon, nitrogen, and hydrogen atoms in ATP-COFs are shown as green, blue, and gray spheres, respectively, while the reacting species are represented by light cyan (H) and red (O) spheres. c Free Gibbs energy changes profiles for the two-electron ORR pathway at different active sites on ATP-COFs. d Free energy profiles for the four-electron WOR at various active sites on ATP-COFs.

Subsequent hydrogen adsorption at the tetrazine nitrogen site results in the *HOOH intermediate, which eventually converts to H₂O₂. This proposed 2e⁻ ORR pathway identifies the initial H formation as the rate-determining step, due to its high free energy demand, as shown by the free energy profiles (Fig. 6c). The energy barriers for *H adsorption, 0.69 eV for ATP-COF-1 and 0.84 eV for ATP-COF-2, align with observed variations in their photoreduction activities. These models and the experimental results define the role of tetrazine nitrogens and vinylene carbons in modulating electron transfer and hydrogen bonding within ATP-COFs.

With respect to the 4e⁻ WOR, an exploration of the role of vinylene groups and tetrazine units as active sites that facilitate proton interactions allows for a direct comparison of reaction efficiencies (Fig. 6b and Supplementary Fig. 45, bottom). The initial step of the 4e⁻ WOR after water adsorption involves water dissociation at the vinylene linkages to form *OH groups. The proximity of the vinylene linkages to the tetrazine nitrogen increases proton transfer and stabilizes intermediates, contributing to the adsorption energy landscape. This protonation influences the ΔG values, promoting the initial bond dissociation with ΔG values of 0.21 and 0.26 eV for ATP-COF-1 and ATP-COF-2 respectively, both of which are low enough to drive *OH formation readily.

As the reaction model progresses, the formed *OH undergoes deprotonation, leading to *O formation. The TPA units stabilize the epoxy-like structure through conjugative interactions, indicated in the ΔG profile for this step (−0.17 eV (ATP-COF-1) and −0.20 eV (ATP-COF-2)). The stabilization of the radical is likely due to the conjugation between the vinylene and tetrazine units.

In the final stage of the 4e⁻ WOR, which produces O₂, a slight increase in ΔG to −0.10 eV suggests that the reaction remains energetically favorable but suggests a stable, controlled release of O₂ from the catalyst. This modest ΔG shift at the O₂ formation stage aligns with the role of ATP-COFs’ vinylene and tetrazine active sites, which stabilize intermediates and facilitate electron transfer without substantial energetic hindrance. The balanced ΔG profile helps ensure effective electron transfer throughout the reaction, favoring the complete oxidation pathway. The comprehensive analysis of electron transport pathways illustrates how ATP-COFs facilitate the efficient separation and transfer of photogenerated charges (Fig. 7). The interaction between the electron-donating TPA and the electron-accepting tetrazine units in ATP-COFs is consistent with enhanced charge transfer and improved reaction efficiency, which correlates with the observed H₂O₂ production. This insight offers a clearer understanding of the mechanism driving H₂O₂ production and the role of ATP-COFs in this process.

Fig. 7. Proposed mechanism of selective photocatalytic H₂O₂ formation on ATP-COF-1.

Schematic of ATP-COF-1’s reaction steps for H₂O₂ production. This configuration highlights the role of ATP-COFs in optimizing steps for selective H₂O₂ production by promoting efficient charge transfer between TPA and tetrazine units. This interaction supports the selective two-electron reduction of O₂ to H₂O₂, advancing our understanding of the mechanisms driving ATP-COF-1’s high photocatalytic efficiency. The chemical structure of ATP-COF-1 was generated using Materials Studio software, with circular highlights added in Microsoft PowerPoint.

Discussion

In summary, ATP-COF-1 and ATP-COF-2, featuring vinylene linkages and electroactive TPA and tetrazine units, were found to be highly effective for photocatalytic hydrogen peroxide production. These COFs exhibited high light absorption properties, low HOMO-LUMO gaps, and high stability under challenging conditions. The electron-donating properties of TPA and the electron-accepting capabilities of tetrazine, which readily oxidize to TPA^•+ and reduce to tetrazine radical anions, promote rapid charge separation and efficient electron transfer during photocatalytic reactions. Their photocatalytic performance is competitive with previously reported photocathodes (Supplementary Table 3). Under visible-light irradiation, ATP-COF-1 and ATP-COF-2 show AQYs of 23.05% and 20.38%, respectively, with high selectivity for hydrogen peroxide production (~96% for ATP-COF-1 and 91% for ATP-COF-2).

Our investigation into the catalytic mechanisms, supported by DFT calculations and in-situ DRIFTS analysis, provides insights into the potential impact of electronic structure on photocatalytic performance, particularly through the 2e⁻ ORR and 4e⁻ WOR pathways. We note that DFT models represent idealized conditions and may not fully capture the complexity of the catalytic environment. This work highlights the promise of ATP-COFs as effective photocathodes for diverse applications, including sustainable H₂O₂ production and other photoelectrochemical processes.

Methods

Materials

All chemicals used in this study were used as received. Palladium on carbon (10 wt.% loading, matrix activated carbon), sodium hydroxide (≥99.0%), potassium hydroxide (≥85.0%, pellets), sodium chloride (99.0%), magnesium sulfate (≥99.5%), manganese (IV) oxide (MnO₂, ≥99.9%), acetaldehyde (>98.0%), benzaldehyde (>99.0%), hydrazine monohydrate (98.0%), sodium nitrite (≥99.0%), mesitylene (98.0%), 1,4-dioxane (99.8%), trifluoroacetic acid (TFA) (HPLC grade, ≥99.0%), potassium bicarbonate (KHCO₃, ≥99.5%), sulfuric acid (H₂SO₄, 98.0%, density of 1.85 g mL⁻¹), hydrochloric acid (HCl, 37.2%, density of 1.19 g mL⁻¹), methanol (99.9%, HPLC grade), acetic acid (glacial, ≥99.8%), trifluoroacetic acid (TFA, ≥99.0%, HPLC grade), tetrahydrofuran (THF, ≥99.9%), N,N′-dimethylformamide (DMF, anhydrous, 99.0%), N,N′-dimethylsulfoxide (DMSO, anhydrous, 99.0%), water-¹⁸O (H₂¹⁸O, 99 atom % ¹⁸O), 2-propanol (99.0%), nitroblue tetrazolium chloride (NBT, 98.0%), 5,5-dimethyl-1-pyrroline N-oxide (DMPO, ≥99.0%), N-bromosuccinimide (NBS, ≥99.0%), 2,2,6,6-tetramethylpiperidine (TEMPO, ≥99.0%), potassium hydrogen phthalate (KHP, 99.8%), sodium iodate (NaIO₃, ≥99.5%), potassium iodide (KI, 99.0%), ceric sulfate (Ce(SO₄)₂, 99.0%), benzyl alcohol (BA, 99%), p-benzoquinone (BQ, 99.0%) and silver nitrate (AgNO₃, ≥99.0%) were procured from Sigma-Aldrich, MO, USA. Diethyl ether (≥99.0%, Fisher Chemical, Nazareth, PA, USA), acetone (99.5%, Fisher Chemical, Nazareth, PA, USA), ethanol (anhydrous, 96.0%, Fisher Chemical, Nazareth, PA, USA), acetonitrile (ACN, HPLC grade, ≥99.5%, VWR Chemicals, Bridgeport, NJ, USA), 4-(diphenylamino)benzaldehyde (DPAB, 98.0%, Ambeed, Arlington Heights, IL, USA), 4,4’,4”-nitrilotribenzaldehyde (NTBA, 95.0%, Ambeed, Arlington Heights, IL, USA), and 4’,4”’,4””’-nitrilotris(([1,1’-biphenyl]-4-carbaldehyde)) (NTBCA, 98.0%, Ambeed, Arlington Heights, IL, USA) were used for the synthesis of starting materials, model compound, and ATP-COFs.

Physical methods

Powder X-ray diffraction (PXRD)

PXRD measurements were performed on a Bruker D8 Powder Diffractometer equipped with CuKα radiation (λ = 0.154056 Å) and a nickel filter. The LynxEye position-sensitive detector was used to record the diffraction patterns, scanning over a 2θ range of 1⎯40°.

Grazing incidence Diffuse X-ray Scattering (GIDXS)

GIDXS measurements were conducted using a Bruker D8 Discover diffractometer equipped with a CuKα radiation source (λ = 0.154056 Å) and a VÅNTEC-500 area detector. The incident angle was set to 0.2° to ensure surface sensitivity and minimize substrate interference. Scattering patterns were collected over a range of in-plane (Q_xy) and out-of-plane (Q_z) momentum transfer values to analyze the crystallinity and preferred orientation of the ATP-COF-1 thin film. Data processing and reciprocal space mapping were performed using Bruker EVA software to identify diffraction features corresponding to specific lattice planes.

Fourier-transform infrared (FT-IR) spectroscopy

FT-IR analysis was conducted with a Jasco-680 spectrometer from Japan. The instrument operated at a resolution of 4 cm⁻¹, scanning across a wavenumber range between 400 and 4000 cm⁻¹. The vibrational frequencies were measured in wavenumbers (cm⁻¹), and the intensity of the absorption bands was classified as weak (w), medium (m), or strong (s).

In situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS)

DRIFTS was performed using a Nicolet iS50 FT-IR spectrometer. The photocatalyst was loaded into a specialized in-situ IR holder inside the chamber. Before data collection, the chamber was flushed with argon at a flow rate of 20 mL min⁻¹ for one hour. Once a stable baseline was established at ambient temperature, both oxygen and water vapor were introduced into the chamber for another hour without light exposure. Following this, visible light (420–600 nm) was applied through the chamber window, and FT-IR spectra were recorded at specific time intervals.

Solid-state ¹³C cross-polarization magic-angle-spinning (¹³C CP/MAS) NMR

Solid-state carbon structures were analyzed using a Varian Unity-Inova 500 spectrometer equipped with a 3.2 mm Phoenix MAS NMR probe and rotors, operating at 125 MHz. Chemical shifts were referenced to tetramethylsilane (TMS, 0 ppm). For magic-angle spinning cross-polarization (CP/MAS) experiments, the samples were spun at a rate of 12 kHz at the magic angle, with carbon and proton RF fields optimized to satisfy the Hartmann-Hahn condition. The ¹³C pulse duration was set to 4 μs, and the decoupling frequency was maintained at 72 kHz. Additionally, magic-angle direct polarization experiments (DP/MAS) were performed to facilitate carbon quantification and enable comparison with CP/MAS spectra, using the same spinning rate.

Solution NMR

For solution-phase NMR studies, both ¹H and ¹³C spectra were recorded using a Bruker Avance 500 spectrometer (Billerica, MA, USA). The experiments were done at room temperature (~25 °C), with dimethyl sulfoxide-d₆ (DMSO-d₆) as the solvent. Tetramethylsilane (TMS) served as the reference standard for chemical shifts, which were reported in parts per million (ppm) with the reference point (δ) set at 0.00 ppm. Coupling constants (J) were provided in hertz (Hz), and the signal multiplicity was denoted by the following: s (singlet), d (doublet), t (triplet), and m (multiplet).

Diffuse reflectance solid state ultraviolet-visible (DRUV-Vis) absorption

DRUV-Vis absorption measurements were carried out at room temperature using a Shimadzu UV-Vis NIR diffuse reflectance spectrophotometer. Barium sulfate was used as the reference standard for optical measurements. The absorption spectra were recorded over the range of 400 to 800 nm. The Kubelka-Munk function⁴³ was applied to the reflectance data to estimate the band gap. For solution samples, UV-Vis spectra were obtained using a Shimadzu UV-2600 spectrometer (Japan), covering a wavelength range of 250 to 650 nm.

Photoluminescence (PL) spectroscopy

PL emission data were obtained at room temperature using a Shimadzu RF-6000 Spectro Fluorometer. A 450 W Xenon lamp within the instrument provided the excitation light, which was passed through a Czerny⎯Turner monochromator equipped with a single grating (1800 l/mm, 250 nm blaze) and a slit to control the bandwidth. The light emitted by the sample was then guided through another Czerny⎯Turner monochromator featuring a double grating (1200 l/mm, 500 nm blaze) and appropriate bandwidth adjustments, with detection performed by a cooled microchannel plate photomultiplier tube (MCP-PMT).

The fluorescence decay curves were fitted using the Eq. 1:

$$y=y_0+A_1{e}^{\frac{-t}{{\tau }_1}}+A_2{e}^{\frac{-t}{{\tau }_2}}$$ 1

To determine the average lifetime (τ_t), Eq. 2 was applied:

$${\tau }_t=\frac{A_1{\tau }_1^2+A_2{\tau }_2^2}{A_1{\tau }_1+A_2{\tau }_2}$$ 2

In these equations, y represents the fluorescence intensity at a given time t, while y₀ indicates the baseline offset. The parameters A₁ and A₂ are the respective amplitudes, and τ₁ and τ₂ are the time constants for the exponential decay terms. The calculated average lifetime (τ_t) provides an overall measure of the fluorescence duration (as listed in Supplementary Table 1).

Field-emission scanning electron microscopy (FE-SEM)

Secondary electron (SE) images were captured using either a Zeiss Merlin or a VEGA TS 5130MM (TESCAN), both of which were equipped with SEM-EDX capabilities, utilizing a Si/Li detector. The imaging was operated at an acceleration voltage of 10 kV (Oxford). For sample preparation, the COF material was dispersed onto a mica surface. The sample was then coated with a carbon layer using a Leica EM SCD 500 Sputter, under an argon atmosphere at a pressure of 10⁻⁴ mbar for 45 seconds, with a sputtering current set at 15 mA. SEM imaging was conducted at accelerating voltages between 1 and 15 kV, and all images were taken from the same sample preparation.

High-resolution transmission electron microscopy (HRTEM)

For HRTEM characterization, a Philips CM30 ST microscope was utilized, featuring a 300 kV LaB₆ cathode and operating at an acceleration voltage of 200 kV. Prior to imaging, COF suspensions in ethanol were sonicated for 5 min to ensure homogeneity. The prepared samples were then drop-cast onto TEM grids made of carbon-coated copper (TED PELLA, INC. 200 line/inch Hexagonal mesh).

Atomic Force Microscopy (AFM)

The topographical characteristics and surface roughness of the ATP-COF films were analyzed using atomic force microscopy (AFM) in a non-contact oscillating mode. Measurements were performed with an MFP-3D system from Asylum Research, a part of Oxford Instruments NanoAnalysis, based in High Wycombe, UK. This equipment features a 3D closed-loop scanner with a scanning range of 90 µm and an X and Y resolution of 5 Å, alongside a Z-axis range of 15 µm and a resolution of 0.6 Å. The experiments were conducted at room temperature in open air, utilizing a scanning frequency of 0.30 Hz, a travel speed of 7.51 μm s⁻¹, and a drive amplitude of 28.79 mV at a frequency of 169.63 kHz. A pixel resolution of 512 × 512 was used. A conical silicon nitride tip with a spring constant of 40 N·m⁻¹, sourced from Mikromasch, Ltd. in Tallinn, Estonia, was utilized for the measurements. Surface roughness was quantified by calculating the root mean square (RMS) values.

Gas Chromatography-Mass Spectrometry (GC-MS)

The GC-MS analysis was carried out using a Thermo Scientific Exactive GC Orbitrap system equipped with a TG-5SILMS column (30 m × 0.25 mm i.d., 0.25 µm film thickness). The injection was performed manually in splitless mode with an injection volume of 10 µL. The injection temperature was set at 250 °C, and helium was used as the carrier gas at a constant flow rate of 1.0 mL min⁻¹. The oven was maintained at a constant temperature of 40 °C during the analysis.

The mass spectrometer operated in electron impact (EI) ionization mode at 70 eV, with a scan range of m/z 35–50. The ion source temperature was set at 280 °C, and the instrument operated at a resolution of 60,000 in positive polarity mode. Data acquisition and analysis were conducted using Thermo Scientific software.

Nitrogen adsorption-desorption measurements

Nitrogen adsorption-desorption experiments were performed with a Quadrasorb SI MP instrument. Before initiating gas adsorption analysis, 50 mg of the synthesized samples were dried under a dynamic vacuum (pressure <10^–3 Torr) at room temperature. The samples were then heated to 120 °C for 24 h to ensure thorough preparation. Following this step, gas adsorption measurements were carried out at 77 K, covering a pressure range from 0 to 1 atm. To calculate the Brunauer-Emmett-Teller (BET) surface areas, we concentrated on the linear portion of the N₂ isotherm at 77 K, specifically in the pressure range of P/P₀ from 0.05 to 0.20. Pore size distributions were analyzed using non-local density functional theory (NLDFT) with the MicroActive software, using a model that describes the adsorption and desorption of N₂ at 77 K in cylindrical pores of a porous oxide surface.

Electron paramagnetic resonance (EPR)

EPR spectra were obtained using a Bruker ELEXSYS X-band spectrometer at ambient temperature. The spectrometer settings included a power of 2.0 mW, modulation amplitude of 2.0 G, a sweep width of 120 G, a sweep time of 43.14 seconds, and a time constant of 40.96 milliseconds. The spin-trapping agent used for capturing O₂^•− was 5,5-dimethyl-1-pyrroline N-oxide (DMPO).

X-ray photoelectron spectroscopy (XPS)

XPS measurements were performed using a Scienta Omicron ESCA-2SR instrument maintained at an approximate vacuum of 1 × 10^–9 Torr. Monochromatic AlKα X-rays (1486.6 eV) were generated at a power of 250 W (15 kV; 20 mA), targeting a 2 mm diameter analysis area. The collection of photoelectrons occurred at a 0° emission angle, with the angle between the source and analyzer set at 54.7°. A hemispherical analyzer determined the electron kinetic energy, using a pass energy of 200 eV for wide/survey scans and 50 eV for high-resolution scans. To prevent charge accumulation on non-conductive samples, a flood gun was utilized.

Ultraviolet photoelectron spectroscopy (UPS)

UPS measurements were performed using an ESCALAB 250Xi photoelectron spectrometer from Thermo Fisher Scientific, USA. The analysis used a gas discharge lamp that emitted unfiltered He (I) light, providing an excitation energy of 21.22 eV. To ascertain the conduction band minimum (CBM) of the covalent organic frameworks (COFs), the UPS data were analyzed by subtracting the spectral width of the He (I) UPS from the excitation energy⁴⁴. For instance, in the case of ATP-COF-1, the vacuum level of the valence band maximum (VBM) was computed using the equation: VBM = 21.22 eV − ($E_{cutoff}$ − $E_{onset}$), resulting in a value of 6.39 eV [21.22 − (18.78 − 3.95)]. The band gaps (E_g) were evaluated via the Kubelka-Munk plot, leading to the calculation of the CBM as CBM = VBM − E_g = 6.39 eV–2.08 eV = 4.31 eV. This processing approach was consistently applied to ATP-COF-2 as well.

Thermogravimetric analysis (TGA)

The TGA of the covalent organic frameworks (COFs) was performed using a TA Instruments 5500 Thermogravimetric Analyzer (New Castle, DE 19720, USA). The samples were contained in platinum pans and analyzed under a nitrogen atmosphere. The temperature ramp rate during the analysis was set to 10 °C min⁻¹.

Contact angle measurements

Contact angle measurements were performed utilizing the sessile drop method with a goniometer Model 500-U1 from Ramé-hart Instrument Co. in Succasunna, NJ, USA. The droplet shape was recorded using a CCD camera, and the integrated DROP Image Advanced software was used for image analysis using the tangent method.

Electrochemical measurements

Electrochemical measurements were done using a CHI-610E electrochemical workstation from CH Instruments (USA) for cyclic voltammetry experiments. A three-electrode configuration was utilized, comprising an indium tin oxide (ITO) glass plate coated with catalyst slurry (3 mm diameter) as the working electrode, a platinum wire (CHI 115) as the counter electrode, and an Ag/AgCl reference electrode (CHI 111, 1 M KCl). All voltages are reported as measured without iR correction. The experiments were carried out under conditions where the solution resistance is expected to have a negligible influence on the observed electrochemical behavior.

To prepare the working electrode, ITO glasses were first cleaned by sonication in acetone for 30 min, followed by drying under a nitrogen stream. A slurry was prepared by mixing 5 mg of COF powder with 1 mL of ethanol and 5 μL of 5 wt.% Nafion perfluorinated resin. This slurry was ultrasonicated for 30 min. Subsequently, 200 μL of the slurry was applied to the ITO glass (1 × 2 cm²), with the uncoated areas of the electrode isolated using epoxy resin. The resulting film was allowed to dry at ambient temperature and pressure for 12 h.

A 0.2 M aqueous solution of Na₂SO₄ (pH 6.8 ± 0.1) served as the electrolyte, which was purged with argon gas for 1 hour prior to measurements. The working electrode was immersed in the electrolyte for 60 seconds before conducting any measurements. The reduction onset potentials were determined from the x-intercepts of the linear fits in the voltammograms.

Photocurrent measurements were conducted using a BAS Epsilon workstation, with the working electrodes illuminated from the backside to minimize the impact of the semiconductor layer thickness. Visible light irradiation was supplied by a Xe arc lamp (PerfectLight, PLS-SXE300/300UV) rated at 300 W, coupled with a UV cutoff filter (λ > 420 nm).

Mott-Schottky analysis was performed to determine the flat band potential using a conventional three-electrode system in a 0.2 M Na₂SO₄ aqueous solution, using an IM 6 electrochemical system. Measurements were conducted with a CHI 760E electrochemical workstation from Chenhua Instrument, Shanghai, China. To prepare the catalyst suspension, 5 mg of the sample was mixed with 5 μL of Nafion (5 wt.%) and 200 μL of ethanol to create a uniform slurry. The catalyst suspension (5 μL × 3) was deposited onto a glassy carbon (GC) electrode, while a platinum plate served as the counter electrode, and an Ag/AgCl electrode acted as the reference electrode.

To achieve steady-state conditions, the potential was maintained prior to each spectral registration. Alternating current (AC) frequencies of 1000, 1500, and 2000 Hz were applied during the measurements. The applied potentials were subsequently converted to normal hydrogen electrode (NHE) and reversible hydrogen electrode (RHE) potentials using the equations outlined in Eqs. 3 and 4.

$$E_{NHE}=E_{Ag/AgCl}+E_{Ag/AgCl}^0\left(E_{Ag/AgCl}^0=+0.199V\right)$$ 3

$$E_{RHE}=E_{Ag/AgCl}+0.0591pH+E_{Ag/AgCl}^0\left(E_{Ag/AgCl}^0=+0.199V\right)$$ 4

Rotating disk electrode (RDE) measurements

For the RDE measurements, a rotating disk electrode was utilized as the working electrode substrate. A catalyst mixture was prepared by dispersing 10 mg of COF in 0.5 mL of ethanol and adding 5 µL of a 5 wt.% Nafion solution. This slurry was carefully applied to the disk electrode and allowed to dry at room temperature. A platinum foil was used as the counter electrode, and Ag/AgCl electrodes served as reference electrodes. Linear sweep voltammograms (LSV) were recorded in an O₂-saturated 0.1 M KOH solution (pH 13) at a scan rate of 10 mV s⁻¹, with varying rotation speeds. The average number of electrons (n) transferred was determined using the Koutecky-Levich equation (Eq. 5):

$$\frac{1}{J}=\frac{1}{J_k}+\frac{1}{J_l}=\frac{1}{J_k}+\frac{1}{B{\omega }^{1/2}},withB=0.62nF{C}_0{D}_0^{2/3}{v}^{-1/6}$$ 5

Here, J is the measured current density, J_k and J_l are the kinetic and diffusion-limiting current densities, respectively, ω is the rotation speed or angular velocity (rpm), n is the number of electrons transferred, F is the Faraday constant (96485 C mol⁻¹), C₀ is the bulk concentration of O₂ (1.26 × 10⁻¹ mol cm⁻³), D₀ is the diffusion coefficient of O₂ in 0.1 M phosphate buffer solution (2.7 × 10⁻⁵ cm² s⁻¹), and ν is the kinetic viscosity of the electrolyte (0.01 cm²s⁻¹).

Rotating ring-disk electrode (RRDE) measurement

Electrochemical oxygen reduction reaction (ORR) tests were conducted using a rotating ring-disk electrode assembly (AFE7R9GCPT, Pine Instrument Company, USA), which has a theoretical collection efficiency of 37%. The glassy carbon (GC) electrode (CHI 104; GC area: 0.0707 cm²; total area: 0.196 cm², mass loading ≈ 0.02 mg / 0.0707 cm² ≈ 0.283 mg cm⁻²) or the RRDE (GC disk area: 0.2475 cm²) was polished to a mirror finish using gamma alumina powder (0.05 μm, CH Instruments). After polishing, the electrode was rinsed and sonicated with double-distilled water to remove any alumina residues, followed by drying under nitrogen.

The experiments were carried out on an MSR electrode rotator at rotation speeds of 100, 400, 900, and 1600 rpm. The working electrode consisted of a glassy carbon electrode loaded with the catalyst. The catalyst ink was prepared by dissolving 1 mg of ATP-COF or a Pt/C sample in a solvent mixture of 992 μL of ethanol and 8 μL of 5 w/w% Nafion. The mixture was sonicated for at least 30 min, resulting in a concentration of 1 mg mL⁻¹ and 0.04 w/w% Nafion. Subsequently, 20 μL of the catalyst ink was drop-cast onto the GC electrode or the GC part of the RRDE and allowed to air dry.

The ORR activity and selectivity were evaluated using polarization curves and RRDE measurements performed in an oxygen-saturated electrolyte.

The selectivity for hydrogen peroxide (H₂O₂) and the number of electrons transferred were calculated from the disk current (I_disk) and ring current (I_ring) using the following equations (Eqs. 6 and 7)⁴⁵:

$$Selectivity\left(H_2{O}_2\right)=\frac{{200I}_r}{N\times I_d+I_r}$$ 6

$$n=\frac{4I_d}{I_d+\frac{I_r}{N}}$$ 7

where I_r is the ring current, I_d is the disc current, and N is the RRDE collection efficiency, determined to be 0.37 in this study.

Structure optimization and refinement

Structure optimization and refinement were conducted as follows: Initial guess structures were generated using the Accelrys Materials Studio software package 5⁴⁶. These structures were designed with the highest possible symmetry and optimized using force-field methods, specifically using the Dreiding force field.

For structure refinement, Pawley refinement was performed within the Materials Studio software. Initially, the unit cell dimensions were set to theoretical parameters. Iterative Pawley refinement was then conducted to optimize the lattice parameters until the R_wp value converged. The Pseudo-Voigt profile function was applied for whole profile fitting, utilizing the Finger-Cox-Jephcoat asymmetry correction during the refinement process. After refinement, the final R_wp and R_p values were achieved: 4.34% and 7.74% for ATP-COF-1, and 3.78% and 5.78% for ATP-COF-2.

The obtained crystallographic values closely align with the dimensions of the eclipsed model.

Computational methodology

All structural optimizations, frequency analyses, and energy computations were performed using the Gaussian 09 software package⁴⁷ via density functional theory (DFT). Excited-state properties and spin-orbit coupling (SOC) were investigated using time-dependent density functional theory (TD-DFT). Data post-processing was conducted with the Multiwfn 3.8 program⁴⁸.

The geometries of the COF structures were thoroughly optimized using the B3LYP hybrid functional along with a newly defined split valence polarization basis set (def2-SVP). Grimme’s D3 dispersion with Becke-Johnson damping (D3BJ) was incorporated for dispersion corrections. These optimized structures served as the basis for studying reaction mechanisms. The same computational level (B3LYP-D3BJ/def2-SVP) was used for analyzing free energies.

For TD-DFT and SOC evaluations, the ground state geometries (S₀) in water were optimized at the B97-3c level, followed by state-specific TD-DFT calculations at the ωB97X-D3/def2-SVP level, which is suitable for large systems (>300 atoms).

The adsorption energies for O₂ were computed using the following equation (Eq. 8):

$${\Delta E}_{O_2^{*}}=E_{conf}-\left(E_{COF}+E_{O_2}\right)$$ 8

where $E_{conf}$, $E_{COF}$ and $E_{O_2}$ represent the total energies of the catalyst with and without O₂, and the energy of the free O₂ molecule in its triplet state, respectively.

The Gibbs free energy changes for each step in the O₂ reduction process were calculated using Eq. 9:

$$\Delta G={\Delta E}_{DFT}+{\Delta G}_{cor}$$ 9

Here, ΔG_cor = ΔE_ZPE − TΔS, represents the correction term for the Gibbs free energy, accounting for zero-point energy (ZPE), thermal corrections, and entropy changes. The term ΔE_ZPE quantifies the inherent quantum energy at absolute zero.

Synthesis of 3,6-dimethyl-1,2,4,5-tetrazine (DMTAZ)

DMTAZ was synthesized following the reported procedure¹⁵.

Synthesis of model compound, 4,4’-((1E,1’E)-(1,2,4,5-tetrazine-3,6-diyl)bis(ethene-2,1-diyl))bis(N,N-diphenylaniline) (TDPA)

The model compound was synthesized through a coupling reaction between 4-(diphenylamino)benzaldehyde (DPAB) and 3,6-dimethyl-1,2,4,5-tetrazine (DMTAZ) in a 2:1 molar ratio (Supplementary Fig. 46). DMTAZ (50.0 mg, 0.454 mmol), DPAB (248.19 mg, 0.908 mmol), and KOH (50.9 mg, 0.908 mmol) were dissolved in 10 mL of methanol, then heated to 80 °C for 24 h under an argon atmosphere. Upon cooling, the crude product precipitated, and the solvent was removed by vacuum filtration. The resulting solid was washed with cold methanol and dried in a vacuum oven overnight at 60 °C. The purified product, obtained as yellow crystals, was air-stable. Yield: 82% (462.2 mg, 0.74 mmol), M.p. 255⎯256 °C. FT-IR (cm⁻¹): 3015 (w, =C–H), 1620 (s, C = C), 1450 (s, C = N tetrazine ring), 1418 (s), 1278 (s), 1182 (m, C–N = N), 969 (s, disubstituted trans C = C), 751 (s). ¹H NMR (500 MHz, DMSO-d₆) δ (ppm): 8.25–8.23 (d, J = 8.95 Hz, 4H), 7.73 (m, 8H), 7.63–7.61 (d, J = 9.80 Hz, 8H), 7.38–7.36 (d, J = 9.30 Hz, 4H), 7.26 (m, 4H), 7.04–7.02 (d, J = 6.10 Hz, 2H), 6.99–6.97 (d, J = 5.90 Hz, 2H); ¹³C NMR (125 MHz, DMSO-d₆) δ (ppm): 167.5, 147.7, 146.9, 140.6, 133.3, 130.6, 130.3, 129.9, 125.8, 124.1, 123.3. [M]⁺: Calcd. for C₄₂H₃₂N₆, 620.27 g mol⁻¹; Found, 620.25 m/z.

Synthesis of ATP-COFs

The ATP-COFs were synthesized through a Knoevenagel condensation reaction using the following generalized procedure:

For ATP-COF-1, 9.92 mg of 3,6-dimethyl tetrazine (DMTAZ) (0.072 mmol) and 15.8 mg of 4,4’,4”-nitrilotribenzaldehyde (NTBA) (0.048 mmol) were placed in a 10 mL Pyrex tube (dimensions: 270 mm length, 11.5 mm outer diameter, 9 mm inner diameter). A solvent mixture of mesitylene (700 μL), 1,4-dioxane (700 μL), acetonitrile (37 μL), and trifluoroacetic acid (292 μL) was added to the tube. The reaction mixture was degassed using three cycles of freeze-pump-thaw. Afterward, the tube was sealed and heated at 120 °C for seven days. Upon cooling to ambient temperature, the solid product was filtered and rinsed three times with 10 mL of methanol and three times with 10 mL of THF. To enhance crystallinity, the obtained product was treated with supercritical CO₂ to improve crystallinity⁴⁹. In this process, the sample was treated with supercritical CO₂ at 53 °C and 30 MPa for six hours. Following the CO₂ treatment, the system was carefully depressurized, and the sample was removed from the chamber. The final product was vacuum-dried overnight at 40 °C. The pure ATP-COF-1 was obtained as a yellow powder (yield: 38.0 mg, 88%). FT-IR (cm⁻¹): 3,003 (w, =C⎯H), 1,618 (s, C = C aromatic), 1,478 (s, C = N tetrazine ring), 1,139 (s, C⎯N = N), 945 (s, disubstituted trans C = C), 858 (s), 717 (s). Solid-state ¹³C CP/MAS NMR (125 MHz, 25 °C) δ (ppm): 167.2, 146.1, 136.9, 135.3, 132.3, 129.7, 123.8, 115.0.

For ATP-COF-2, the same procedure was followed, substituting NTBA with 4’,4”‘,4”“‘-nitrilotris(([1,1’-biphenyl]-4-carbaldehyde, NTBCA)) (26.8 mg, 0.15 mmol). The resulting product was a pale-yellow powder (yield: 41.7 mg, 83%). FT-IR (cm⁻¹): 3,007 (w, =C⎯H), 1,620 (s, C = C aromatic), 1,485 (s, C = N tetrazine ring), 1,122 (s, C⎯N = N), 942 (s, disubstituted trans C = C), 856 (s), 744 (s). Solid-state ¹³C CP/MAS NMR (125 MHz, 25 °C) δ (ppm): 167.2, 146.1, 136.9, 135.3, 132.3, 129.7, 123.8, 115.0.

Photocatalytic H₂O₂ production

In a 100 mL Schlenk vessel, 50 mL of deionized water was combined with 10 mg of the photocatalyst. To ensure even dispersion, the mixture sonicated. For oxygen saturation, the suspension was purged with oxygen gas at a flow rate of 100 mL min⁻¹ while stirring in darkness for 15 minutes. The dispersion was then illuminated with a 300 W Xe lamp (XBO 300 W/CR OFR, OSRAM) with light filtered to wavelengths above 420 nm. The reactor temperature was maintained using a continuous cooling fan, which ensured stable thermal conditions and prevented localized heating during extended illumination (Supplementary Fig. 47). The H₂O₂ concentrations were determined using the ceric sulfate method with Ce(SO₄)₂. The reaction proceeds in two steps under acidic conditions (pH 0–1.4) as described by Eqs. 10 and 11⁵⁰:

$${\mathrm{H}}_2{\mathrm{O}}_2+{\mathrm{Ce}}^{4+}\to {\mathrm{Ce}}^{3+}+{\mathrm{HO}}_2+{\mathrm{H}}^{+}$$ 10

$${\mathrm{HO}}_2+{\mathrm{Ce}}^{4+}\to {\mathrm{O}}_2+{\mathrm{H}}^{+}+{\mathrm{Ce}}^{3+}$$ 11

These reactions suggest that the quantity of Ce³⁺ generated correlates directly with the amount of

H₂O₂ present. During the reaction, H₂O₂ interacts with the yellow Ce⁴⁺, reducing them to the colorless Ce³⁺ form. The complete reaction is represented in Eq. 12⁵¹:

$${2Ce}^{4+}+H_2{O}_2+{2H}_{2}O\to {2Ce}^{3+}+O_2+{2H}_3{O}^{+}$$ 12

$$C_{H_2{O}_2}=\frac{1}{2}{\Delta C}_{{Ce}^{4+}}$$ 13

The concentration of H₂O₂ was evaluated based on the reduction in Ce⁴⁺ concentration. To perform this, a 0.1 mL sample of the reaction mixture was withdrawn with a syringe and passed through a 0.22 μm filter to remove any suspended solids. The filtered sample was then combined with 0.9 mL of a 0.5 M Ce(SO₄)₂ solution prepared in 0.25 mM H₂SO₄. For long-term experiments, a more concentrated 1.5 M Ce(SO₄)₂ solution was used to ensure sufficient oxidant remained throughout the extended reaction time. The decrease in absorbance at 316 nm was recorded using a UV-Vis spectrophotometer and compared against a calibration curve established from standard Ce⁴⁺ solutions. Prior to measurement, the reaction mixture was diluted to a final volume of 500 mL to bring the absorbance values within the instrument’s linear range (Supplementary Fig. 48).

In the scavenger experiments involving the addition of p-benzoquinone (BQ), which disrupts the Ce(SO₄)₂ method, concentration was instead determined via iodometry. In this method, H₂O₂ reacts with iodide ions (I^–) to produce triiodide ions (I₃^–), which exhibit a strong absorbance around 350 nm, as shown in the reaction below (Eq. 14):

$${\mathrm{H}}_2{\mathrm{O}}_2+{3\mathrm{I}}^{-}+{{2\mathrm{H}}_3\mathrm{O}}^{+}\to {\mathrm{I}}_3^{-}+{4\mathrm{H}}_2\mathrm{O}$$ 14

$$C_{H_2{O}_2}=C_{I_3^{-}}$$ 15

The H₂O₂ concentration was determined by measuring the amount of I₃^–. To prepare the reaction mixture, 0.5 mL of 2.0 M potassium iodide (KI) and 0.5 mL of 0.1 M potassium hydrogen phthalate (KHP) were combined, followed by the addition of 1 mL of the sample solution. This mixture was allowed to react for 30 minutes. The resulting I₃⁻ shows two main absorbance peaks at 288 nm and 350 nm, consistent with previous findings⁵². As H₂O₂ concentration increases, absorbance at 350 nm also increases, while the spectral peak position and curve shape remain unchanged. This strong linear relationship between I₃^– absorbance at 350 nm and H₂O₂ concentration supports its use in a spectrophotometric method for determining H₂O₂ levels (Supplementary Fig. 49).

Photocatalytic generation of O₂^•−

To investigate the photocatalytic generation of superoxide anion (O₂^•−), 20 mg of the catalyst was suspended in 10 mL of a 5.0 × 10⁻⁵ M nitro blue tetrazolium (NBT) solution. Before starting the reactions, the mixtures were sonicated for 10 min in the dark to ensure proper dispersion. A 300 W Xenon lamp (XBO 300 W/CR OFR, OSRAM) simulated solar light during the reaction, while a magnetic stirrer was employed to maintain a consistent mixing of the suspensions. The reactions were quenched by centrifuging, then filtering the mixtures through a Millipore membrane to eliminate the ATP-COF particles. The production of O₂^•− was evaluated by measuring the decrease in NBT concentration in the supernatant using a UV-Vis spectrophotometer. By using calibration curves of absorbance intensity, the concentrations of NBT and their correlation with O₂^•− were established, which allowed for the quantification of the superoxide generated. The reaction demonstrated that the mole ratio of produced O₂^•− to consumed NBT was 4:1 (Supplementary Fig. 50).

Recycling procedure for ATP-COFs after photocatalytic reaction

50 mg of ATP-COFs were dispersed in 50 mL of O₂-saturated water, and this mixture was exposed to visible light (λ > 420 nm) for 2 h to activate the photocatalytic reaction. Once the photocatalysis was complete, the ATP-COFs were separated from the solution through filtration, leaving a filtrate containing the H₂O₂ generated. The collected ATP-COFs were then rinsed thoroughly with 100 mL of water to remove any residual substances, followed by drying under vacuum at 80 °C for 6 h. After drying, the ATP-COFs were weighed to assess any mass change and prepared for use in additional cycles of photocatalytic testing.

Isotope labeling study

A glass vial (1.5 mL) containing 2 mg of ATP-COF-1 was prepared, and 1 mL of H₂¹⁸O was added. To ensure proper dispersion, the suspension was sonicated for 5 minutes and then exposed to air bubbling for 15 min in the absence of light. The vial was sealed using a rubber septum, and the mixture was irradiated under a 300 W Xe lamp (XBO 300 W/CR OFR, OSRAM) with a 420 nm to 600 nm wavelength range, achieved by using a cutoff filter. The system’s temperature was maintained using a continuous cooling mechanism via a coil fan. Light exposure was carried out for 12 h.

Following irradiation, the gas phase in the vial was analyzed using GC-MS (H768). The mixture was then purged with argon, and the aqueous phase was extracted. The photocatalyst was separated, and the solution was treated with MnO₂ to oxidize the H₂O₂ into O₂. A second gas-phase analysis was conducted via GC-MS, which confirmed the generation of ¹⁸O₂, as depicted in Supplementary Fig. 44.

Chemical stability evaluation of ATP-COFs

Chemical stability of ATP-COFs was assessed using a 10 mg sample that was submerged in 5 mL of different solvents and solutions—MeOH, TFA, 9 M NaOH, 9 M HCl, and water—at 25 °C for a period of 72 h. Afterward, each sample was thoroughly rinsed with THF, vacuum dried at 60 °C overnight then PXRD was used to measure crystallinity.

Calculation details of k_f and k_d

To calculate the rate constants for the formation (k_f) and decomposition (k_d) of H₂O₂, specific equations are used based on the kinetics of the processes. The formation of H₂O₂ is modeled as a zero-order reaction, while its decomposition is modeled as a first-order reaction. In these equations, t represents the reaction time, and $n_{H_2{O}_2}$ denotes the number of moles of hydrogen peroxide at a given time. For instance, considering ATP-COF-1, the mole numbers of H₂O₂ can be determined at two time points, t_a = 2 h and t_b = 4 h can be determined, denoted as $n_{H_2{O}_2,a}$ and $n_{H_2{O}_2,b}$. The values of k_f (18.6 μmol h⁻¹) and k_d (0.39 h⁻¹) are subsequently derived from the following equations:

$$n_{H_2{O}_2,a}=\left(\frac{k_f}{k_d}\right)\left[1-e^{-k_d{t}_a}\right]$$ 16

$$n_{H_2{O}_2,b}=\left(\frac{k_f}{k_d}\right)\left[1-e^{-k_d{t}_b}\right]$$ 17

Femtosecond transient absorption (fs-TA) spectroscopy

Femtosecond transient absorption (fs-TA) spectroscopy measurements were done using a specialized pump-probe system with a time resolution extending to 10 ns. The timing between pump and probe pulses was managed by an optical delay stage controlled via computer. To produce an 800 nm laser pulse of around 30 femtoseconds, a titanium-sapphire laser system was used, stabilized by an external oscillator. This laser beam was directed to an optical parametric amplifier, converting it to a 350 nm pump pulse. Various optical elements, such as beam splitters and lenses, were used to direct and focus the pulse accurately. Neutral density filters adjusted the pump pulse intensity to approximately 100 μW. The original 800 nm beam was also used to generate a white-light continuum ranging from 400 nm to 750 nm, achieved by passing it through a quartz element to broaden the spectrum. Transient absorption spectra were obtained by varying the delay between probe and pump pulses, capturing the excited state dynamics across multiple time delays. The data were collected with precision and high sensitivity to identify transient absorption features accurately.

Author contributions

A.Z.: Data Curation, Formal Analysis, Investigation, Methodology, Quantum calculations, Validation, Writing—Original Draft, Review and Editing. F.A.: Formal Analysis, Investigation, Review and Editing. D.K.: Investigation, Review and Editing, A.A.: Project Administration, Resources, Supervision, Writing—Review and Editing.

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

The source data generated in this study are included in the accompanying Source Data file. All source data supporting the findings of this study are provided with this paper. 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/s41467-025-66679-8.