As a versatile reagent, hydrogen peroxide (H2O2) is widely used in the fields of chemical synthesis, energy storage and water treatment1,2. At present, oxidation of anthraquinone, electrochemical synthesis, and noble metal catalysis are the common methods used to produce H2O22,3. However, these methods require high energy input and cause environmental pollution by the release of toxic by-products4,5. Due to the use of naturally abundant water and oxygen in air as raw materials and solar light as energy input, H2O2 photosynthesis, especially without using sacrificial agents, is regarded as one of the green and sustainable methods, which yet is plagued by the severe side reactions such as the decomposition of produced H2O2 into oxygen and water due to the metastable feature of H2O2 during the synthesis process6,7,8,9,10,11,12. As one type of novel metal-free crystalline polymers that can form suitable intermediates and inhibit side reactions for H2O2 generation, covalent organic frameworks (COFs) recently show great promise in the field of H2O2 photosynthesis13,14. The insufficient formation or separation of excitons (bound state of electron−hole pairs) for the free charge generation in COFs yet seriously constrains the high-efficiency of H2O2 photocatalytic generation15,16. Some interfacial modulating strategies such as constructing heterojunction17 and incorporating single atoms18 are beneficial for enhancing the photoexcitation and charge separation/transfer of COFs photocatalysts, which yet require multistep, complex and time/energy-consuming synthesis process. Hence, developing a facile and cost-effective strategy to improve the generation and separation of photogenerated charges without introducing severe side reactions is in great demand for H2O2 photosynthesis, yet remains a great challenge.

Herein, we provide a facile and economical strategy for the design of metal-free donor-acceptor (D-A)-type COFs with optimal intramolecular polarity by introducing suitable amounts of phenyl groups as electron donors for excitonic regulation to boost the direct photocatalytic H2O2 production from water, air and sunlight without using sacrificial agent. By using triazine-cored triamine with different amounts of phenyl groups (n = 0, 1, 2) as the precursors, a class of D-A COFs with different intramolecular polarity were successfully synthesized. We find that weak intramolecular polarity in D-A COFs constrains excitons dissociation, yet too strong intramolecular polarity inhibits excitons formation via weakened π-conjugated effect as well as decreases the photo-stability of COFs. The well-designed COFs with the optimal intramolecular polarity (named as COF-N32) can facilitate excitons formation and dissociation, leading to the high and stable H2O2 yield (605 μmol g−1 h−1) with solar-to-chemical efficiency of 0.31% in water without additional sacrificial reagent. COF-N32 can also efficiently produce H2O2 in real water samples including tap water, river water and sea water even with natural solar light irradiation. Moreover, COF-N32 can be assembled into practical devices for facile consecutive uses with high photocatalytic stability under natural solar irradiation. In addition, we demonstrate that the produced H2O2 aqueous solution (without further separation) can be directly employed in water decontamination, indicating the potential application feasibility of COF-N32. Via in-situ Fourier transform infrared (FTIR) characterization and density functional theory (DFT) calculation, we reveal that the suitable N 2p states and C 2p states in COF-N32 with optimal intramolecular polarity effectively reduce the energy barrier for H2O activation and oxygen reduction, respectively, contributing to the high efficiency.

Results and discussion

Characterization of synthesized COFs

The COFs photocatalysts were synthesized by a solvothermal method. Figure 1a illustrates the theoretical chemical structures of three triazine-based COFs, which is also confirmed by the corresponding solid state 13C nuclear magnetic resonance (NMR) spectra (Fig. 1b). Clear peaks at ~180 ppm, ~170 ppm, ~145 ppm and ~115 ppm in the 13C NMR spectra of three COFs can be attributed to carbonyl carbon, triazine carbon, C-NH (amine linkage) carbon and olefin carbon, respectively19. The stretching vibration bands of C=C bond at ~1573 cm−1 and C-N bond at 1255 cm−1 in the FTIR spectra of three COFs (Fig. 1c) suggest the occurrence of Schiff base reaction and enol-to-keto tautomerism during the fabrication process of three COFs19,20. The formation of carbonyl group and amine linkage in three COFs is further confirmed by X- photoelectron spectroscopy (XPS) analysis (Fig. S1). Based on the simulation and Pawley refinement, the obvious peaks at 2θ = 9.6°, 5.5° and 3.9° corresponding to (100) planes and 2θ = ~ 27° corresponding to (002) planes in the X-ray diffraction (XRD) pattern (Figure S2) suggest that three COFs contain crystalline structure20,21,22. Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) images show that the three COFs are tiny granular particles with diameter of 2–3 μm, which are assembled by numerous nanorods (Figs. S3 and S4). The above results confirm the successful fabrication of three COFs.

Fig. 1: Chemical structure.
figure 1

a Schematic illustration of synthesis process of COF-N31, COF-N32 and COF-N33. b 13C nuclear magnetic resonance (NMR) spectra and c Fourier transform infrared spectroscopy (FTIR) of COF-N31, COF-N32 and COF-N33. d The schematic illustration of the octupolar structure in COF-N31, COF-N32 and COF-N33. e DFT calculation on charge distribution of COF-N32 structure.

To explore whether the incorporated phenyl group on triazine rings would affect the formation and separation of excitons in COFs, the intramolecular polarity of electronic structure in three COFs was determined. The theoretical calculation of charge distribution indicates that in the hexatomic rings (one type of octupolar subunit), carbonyl group with meta position acts as electron acceptor, while olefin group serves as donor (Figs. 1d and S5)23,24,25. For another type of octupolar subunit, due to the well-delocalizing π-electron over the three aromatic carbon atoms, 1,3,5-triazine ring with electron deficiency in all three COFs can also serve as electron acceptor center26,27. While enamine group in three COFs as well as phenyl group in COF-N32 and COF-N33 can act as electron-donating groups28,29. Clearly, all three COFs contain octupolar conjugated structure with two subunits (Fig. 1e), which is expected to facilitate the efficient charge separation especially in each subunit coupling with appropriate intramolecular polarity27. Accordingly, stable radicals with strong signal intensities are observed in the solid-state electron spin resonance (ESR) spectra of three COFs under dark condition at room temperature (Fig. S6). The paramagnetic absorption signal intensities of three COFs follow the order of COF-N31 > COF-N32 > COF-N33 (Fig. S6), indicating that different amounts of unpaired electrons exist in three D-A COFs under dark condition30. With increasing amount of phenyl groups as electron donors from 0 (COF-N31) to 1 (COF-N32) and further to 2 (COF-N33), the intramolecular polarization of composition fragments in three COFs decrease from 0.072 e Å−1 for COF-N31 to 0.032 e Å−1 for COF-N32 and further to 0.020 e Å−1 for COF-N33 (Fig. S7 and Table S1). The direction of intramolecular polarity in these COFs is from enamine and/or benzene groups to triazine groups (Fig. S7). Furthermore, the molecular polarity index (MPI) also follows the same order of COF-N31 (0.53 eV) > COF-N32 (0.50 eV) > COF-N33 (0.48 eV) (Table S2).

The solvatochromic behaviors of COFs were employed to further determine their polar properties. Obvious shift of emission peaks (~48 nm) is observed in fluorescence spectra of COF-N31 dispersed in water (as a polar solvent) relative to those in methanol, dichloromethane and ethyl acetate (with weaker polarity than water), indicating the strong local dipolar nature of COF-N31 with weak electron donor in π-conjugated system31. Similar observation has also been previously reported for COFs with high polarity27,32. In contrast, for both COF-N32 and COF-N33, relatively smaller positive shifts of emission peaks (<25 nm) are observed in water relative to other organic solvents (Fig. S8). The weak solvatochromic behaviors of both COF-N32 and COF-N33 can be attributed to the sufficient electron donor with phenyl or diphenyl group in the octupolar π-conjugated framework. The COFs with higher polarity usually exhibit higher affinity to polar H2O33. Note the intramolecular polarity follows the order of COF-N31 > COF-N32 > COF-N33. Accordingly, the contact angle of three COFs follows the order of COF-N31 (121°) < COF-N32 (139°) < COF-N33 (145°) (Fig. S9). Meanwhile, the unit water adsorption capacity is also consistent with the order of COF-N31 > COF-N32 > COF-N33 (Fig. S10).

In general, the excessively strong intramolecular polarity in COFs would restrain the π conjugated effect in sp2-hybridized orbit structure of COFs, leading to the inhibition of electron excitation34,35. The O atom with high electronegativity in COF-N31 with strong polarity can attract electrons surrounding amine N atoms, the delocalization on triazine rings in the lowest unoccupied molecular orbital (LUMO) of COF-N31 thus is limited across three atoms (N and C atoms) in each direction (Fig. S11a). In contrast, due to the less polarity relative to COF-N31, LUMO of COF-N32 and COF-N33 can be well delocalized across nine atoms surrounding the triazine rings (Figs. S11b and S11c), indicating the enhanced π conjugation in COF-N32 and COF-N33. This would facilitate the excitation of electrons in these two COFs under light irradiation. The less intramolecular polarity of COFs has shown to result in the overlap of the highest occupied molecular orbital (HOMO) and LUMO36,37. Similarly, for COF-N32 and COF-N33 with less polarity, the overlap of HOMO and LUMO especially on the C atoms in keto structure and diphenyl group is also observed (Fig. S11). Sm index (parameter denoting the hole-electron recombination degree38) of three COFs in the excited state based on DFT calculation is found to follow the order of COF-N31 (0.33) < COF-N32 (0.36) < COF-N33 (0.38) (Table S2). The observation indicates that hole-electron in COF-N32 and COF-N33 is relatively readily to be recombined than COF-N31. Overall, the strong intramolecular polarity in COF-N31 theoretically inhibits the π conjugation, the subsequent excitation of electrons, as well as the recombination of charges. In contrast, the weak polarity in COF-N33 yet facilitates the π-conjugated effect, the excitation of electrons, and charge recombination. COF-N32 with moderate polarity among three COFs is expected to display the best photocatalytic property.

The formation, transfer and separation of photo-induced excitons in three COFs were investigated. Tauc plot based on UV-vis diffused reflectance spectra (DRS) reveals that all three COFs exhibit n–π* transition in N atoms39, while COF-N33 with more benzene units also contains more obvious signals of π–π* transition40. The band gaps of COF-N31 (2.72 eV) > COF-N32 (2.42 eV) > COF-N33 (2.40 eV) (Fig. 2a) are also consistent with the results of calculated ones (Fig. S12). The observation indicates that COF-N32 and COF-N33 exhibit higher light absorption efficiency relative to COF-N31, which can be attributed to the efficient π-conjugated effect with relatively weaker intramolecular polarity of these two COFs than that of COF-N3127. The results of electrical impedance spectra (EIS) indicate that COF-N32 and COF-N33 have relatively smaller charge transfer resistance than COF-N31 (Fig. S13), contributing to the efficient interfacial electron transfer speed of these two COFs. The observation agrees with the efficient π-conjugation through electronic push-pull effect in COF-N32 and COF-N33 due to their relatively weaker intramolecular polarity discussed above. Both the exciton binding energy (Eb) determined by temperature-dependent photoluminescence (PL) spectra and the emission peaks intensities in steady-state PL spectra of three COFs follow the order of COF-N33 > COF-N32 > COF-N31 (Figs. 2b, S14 and S15), indicating that the dissociation of formed excitons in three COFs is in the order of COF-N31 > COF-N32 > COF-N33.

Fig. 2: Polarity property.
figure 2

a Tauc plot of three COFs; b Temperature-dependent PL spectra of three COFs excited at 365 nm for the determination of binding energy (Eb). c Photoinduced absorption decay dynamics of three COFs with the excitation of 360 nm pump pulse (P = 4.3 mJ cm-2 per pulse); d The signal intensity of 2,2,6,6-tetramethyl-1-piperidinyloxy (TEMPO) for charge detection in COF-N32 under light irradiation.

The transient absorption (TA) was performed to analyze charge carrier dynamics in three COFs. The broad and positive absorbance changes (ΔA) observed around 500–700 nm can be attributed to the photoinduced absorption of photoexcited electrons in the conduction band (CB) of three COFs (Fig. S16)41,42. Among three COFs, photoexcited electrons in COF-N32 exhibits smaller average lifetime (16 ps) relative to those in COF-N31 (46 ps) and COF-N33 (49 ps) (Fig. 2c), which is consistent with the average lifetimes obtained in PL decay curves in the nanosecond domain (0.31 ns for COF-N32 relative to 0.75 ns for COF-N31 and 1.34 ns for COF-N33) (Fig. S17). The observations suggest the more prominent non-radiative rate in COF-N32 than COF-N31 and COF-N3327,43. ESR analysis (Fig. 2d) shows that under visible light irradiation, COF-N32 can generate more photo-induced free charges (via the excitons formation and dissociation) than COF-N31 (Fig. S18a) and COF-N33 (Fig. S18b). The generation of photo-induced free charges in COF-N32 is confirmed by detecting the reduction product tetramethylpiperidine (TEMP, m/z 142.15826, ESI+) via reaction of electrons and 2,2,6,6-tetramethyl-1-piperidinyloxy (TEMPO) (Fig. S19). The above results show that among all three COFs, COF-N32 with moderate intramolecular polarity in the octupolar conjugated structure can generate the greatest amount of charges (i.e. electrons and holes) under visible light irradiation (Eq. S4). The energy band positions of three COFs derived from XPS valence band spectra combined with Tauc plot indicate the thermodynamic feasibility of oxygen reduction reaction (ORR) to photosynthesize H2O2 (−0.33 V vs. NHE) by three COFs (−1.09 V vs. NHE for COF-N31, −0.55 V vs. NHE for COF-N32 and −0.48 V vs. NHE for COF-N33) (Figs. S20 and S21)8, while the two-electron water oxidation reaction (WOR) directly to H2O2 (1.77 V vs. NHE) is thermodynamically feasible by COF-N32 (1.88 V vs. NHE) and COF-N33 (1.92 V vs. NHE) but not by COF-N31 (1.63 V vs. NHE) (further discussion is provided below).

Photocatalytic H2O2 production by COFs

The photosynthesis of H2O2 by COFs was evaluated in pure water without using any sacrificial agent under visible light irradiation (λ > 420 nm, 100 mW cm−2). During 12 h reaction duration, COF-N32 with moderate intramolecular polarity in the octupolar conjugated structure exhibits significantly improved performance for H2O2 photosynthesis relative to COF-N31 (with strong intramolecular polarity) and COF-N33 (with weak intramolecular polarity). Specifically, after 12 h of visible light irradiation, H2O2 yield by COF-N32 reaches 7092 μmol g−1 (605 μmol g−1 h−1), which is greatly higher than that by COF-N31 (4316 μmol g−1, 442 μmol g−1 h−1) and COF-N33 (1736 μmol g−1, 155 μmol g−1 h−1) (Figs. 3a and S22). Negligible amount of H2O2 (<5%) could be degraded by metal-free COF-N32 under visible light irradiation (Fig. S23), indicating the inhibited side reaction. This contributes to the stable and high yield of H2O2 in water by COF-N32. H2O2 yields by COF-N32 is also higher than those by conventional photocatalysts including TiO2, g-C3N4 and WO3 (Figs. 3a and S22). Moreover, COF-N32 can yield over 3.17 mmol g−1 h−1 with the addition of 1 mg COF-N32 in 50 mL ultrapure water after 3 h of visible light irradiation (Fig. S24), which is much higher than those of recently reported photocatalysts in pure water under similar measurement conditions (Fig. 3b and Table S3). In addition, COF-N32 exhibits high apparent quantum yield (AQY) of 6.2% at 459 nm (Fig. 3c). The solar-to-chemical efficiency of COF-N32 (0.31%) under visible light irradiation (details are provided in Supporting information) is greatly higher than solar-to-biomass efficiency by plants (~0.1%).

Fig. 3: Photocatalytic performance.
figure 3

a Photosynthesis of H2O2 by COF-N31, COF-N32 and COF-N33. Experimental conditions: λ > 420 nm (298 K; xenon lamp, light intensity:100 mW cm−2), ultrapure water (50 mL), photocatalyst (25 mg). b The comparison of H2O2 photosynthesis rate by COF-N32 with other reported photocatalysts without sacrificial agent under the similar measurement conditions (shape of symbol refers to the dosage of photocatalysts, sphere: 1 mg; circle: 10 mg; star: 20–30 mg; triangle: 50 mg; square: 100 mg; hexagon: 375 mg; color of symbol refers to the reaction duration, green: 1 h; blue: 1.2–2 h; pink: 3 h; purple: 6 h; orange: 10–24 h). c The apparent quantum yield (AQY) of COF-N32 as a function of wavelength (purple light at 400 nm, blue light at 459 nm, green light at 519 nm, yellow light at 580 nm, red light at 625 nm). d The reusability of COF-N32 for H2O2 photosynthesis. Experimental conditions: λ > 420 nm (298 K; xenon lamp, light intensity:100 mW cm−2), ultrapure water (50 mL), photocatalyst (25 mg). Error bars in a represent the average values (mean ± s.d., n = 3).

Due to the inhibited side reaction, COF-N32 also exhibits stable H2O2 yield and excellent reusability for 5 reused cycles (Fig. 3d). No obvious structural change is observed in COF-N32 after use (Fig. S25), indicating its excellent photo-stability under visible light irradiation. The stable H2O2 yield during the 5 reused cycles and no obvious change of crystalline structure after use are also achieved for COF-N33 (Figs. S26a and S26b). In contrast, the H2O2 yield by COF-N31 dramatically decreases with increasing reused cycles (Fig. S26c), indicating the relatively low photo-stability of COF-N31 during the reaction duration. The decreased crystallinity of COF-N31 after photocatalytic reaction (Fig. S26d) confirms its low photo-stability during the reaction process. The photo-stability observed for both COF-N32 and COF-N33 during the reaction duration indicates that h+ generated by COF-N32 and COF-N33 directly oxidizes water to H2O2 instead of attacking COFs. The WOR directly to H2O2 by COF-N32 and COF-N33 (thermodynamically feasible stated above, Fig. S21) is confirmed by isotopic experiments using H218O and 16O2 as precursors in a sealed reactor (Figs. S27b and S27c). Unlike that obtained for COF-N32 and COF-N33 (two COFs with relatively weak intramolecular polarity), WOR directly to H2O2 by COF-N31 with strong intramolecular polarity yet is not thermodynamically feasible (Fig. S21), which is also confirmed by isotopic experiments (Fig. S27a). Instead, h+ generated by COF-N31 can attack COF itself, leading to the low photo-stability of COF-N31 during the reaction duration. Note the quench of h+ by COF-N31 can promote the separation of e-h+ pairs, which is beneficial for the H2O2 production44,45. Therefore, even though the amount of free charges (i.e. e and h+) by COF-N31 is much lower than that by COF-N32 (Figs. S18a and 2d), H2O2 production rate by COF-N31 is similar to that by COF-N32 at the beginning of photocatalytic process (Fig. 3a). The decreased H2O2 production rate observed with the increasing reaction duration can be attributed to the self-decomposition of COF-N31 during the reaction duration, which is also confirmed by the decreased crystallinity of COF-N31 after photocatalytic reaction (Fig. S26d). Similar observation about self-decomposition of COFs during H2O2 photosynthesis in water has also been reported previously46. The results clearly show that in pure water, COF-N31 with strong intramolecular polarity has low photo-stability of COF-N31 during the reaction duration, whereas COF-N32 and COF-N33 with relatively weak intramolecular polarity especially COF-N32 owns excellent photo-stability under visible light irradiation and can be consecutively reused for the photo-generation of H2O2.

The H2O2 yield (~3.5 mM) by COF-N32 after 12 h with visible light irradiation (Fig. 3a) can meet the H2O2 concentration required for water purification47. The filtrate of H2O2 generated by COF-N32 can be directly used to efficiently inactivate antibiotic resistant bacteria under dark condition, indicating that the produced H2O2 can be employed for water disinfection (Fig. S28). In addition to ex-situ disinfection, COF-N32 can also in-situ disinfect antibiotic resistant bacteria and degrade organic pollutant with emerging concerns (diclofenac) under visible light irradiation (Fig. S29). More importantly, COF-N32 is able to produce H2O2 at wide ranges of initial solution pH (3–11) (Fig. S30). Due to the consumption of radicals by dissolving ions and natural organic matter (NOM), the H2O2 production in real waters has been previously found to be inhibited48,49. However, we experimentally demonstrate that the efficient H2O2 photosynthesis by COF-N32 can also be achieved in more available real water samples with complex water matrix conditions. The H2O2 photosynthesis rate by COF-N32 reaches 667, 648 and 554 μmol g−1 h−1 in tap water, river water and sea water within 3 h of visible light irradiation, respectively (Fig. S31). The H2O2 photosynthesis in a cheap commercial membrane filter reactor (facile for separation of COF-N32 after use) was also investigated under simulated visible light irradiation. Regardless without or with using rubber plug to prevent the penetration of water through membrane in reactor under gravity, COF-N32 can efficiently produce ~30 μmol H2O2 in 2 h for the successive four cycles (Fig. S32).

The H2O2 photosynthesis by COF-N32 was further examined under natural sunlight irradiation (Figs. 4a–d). In two different reactor systems (double-walled beaker and membrane filter reactor), COF-N32 can efficiently produce H2O2 in different types of water samples under natural sunlight both in cloudy and sunny days (Figs. 4a, c, d, and S3335). Although the used COF-N32 can be easily recovered by membrane filtration in double-walled beaker or can be separated in membrane filter reactor with the removal of H2O2 solution under gravity, COF-N32 powders were immobilized onto indium tin oxide (ITO) glass slide to further ease its recovery and reuse for practical application (Fig. 4b). Pre-experiments show that ~3.5 μmol of H2O2 can be produced in every 2 h for the consecutive five cycles by COF-N32-immobilized glass slide (2 cm × 2 cm, containing 5 mg COF-N32) under visible light irradiation (Fig. S36). In a scaled-up reactor with working volume of 200 mL, ~85 μmol of H2O2 can be generated in 3 h on five sunny days by a larger COF-N32-loaded glass side (10 cm × 10 cm glass slide with the loading of 60 mg COF-N32) under natural sunlight irradiation (Fig. 4b). The results suggest the feasibility of panel-level H2O2 photosynthesis based on COFs photocatalysts.

Fig. 4: Feasibility investigation for practical applications.
figure 4

a Photocatalytic H2O2 production by COF-N32 in 50 mL of ultrapure water, tap water, river water and sea water under natural sunlight irradiation on a sunny day. b Immobilization of COF-N32 powders (60 mg) onto ITO glass slide (10 cm × 10 cm) for H2O2 production in 200 mL of pure water under natural sunlight irradiation from 10:30 to 13:30 on sunny days. c The schematic illustration of H2O2 photosynthesis process in a membrane reactor. d The photocatalytic production of H2O2 by COF-N32 in a membrane reactor under natural sunlight irradiation. Experimental conditions: ultrapure water (50 mL), photocatalyst (25 mg).

Photocatalytic mechanisms of COFs

The generation process of H2O2 by COF-N32 under visible light irradiation was investigated by quenching experiments and in-situ ESR analysis. Negligible H2O2 is generated within 12 h when dissolved O2 in water is eliminated by N2 purging (Figs. 5a and S37), indicating O2 is essential for H2O2 production. Rotating ring-disk electrode (RRDE) analysis shows that the number of electrons transferred from COF-N32 to O2 is estimated to be 2.17 (Fig. S38), indicating that O2 is generally reduced to generate H2O2 via the apparent 2-electron reaction. The intermediates during oxygen reduction were further investigated by the trapping experiments. The addition of p-benzoquinone (p-BQ, the scavenger of ·O2) into reaction system significantly inhibits the production of H2O2 with negligible yield, suggesting ·O2 is an intermediate product crucial for the H2O2 photosynthesis. Note that ·O2 is the reduction product of O2 by electrons (e) photo-generated by COFs (Eq. S5). As stated above, among three COFs, the amount of e generated by COF-N32 is the largest (Fig. 2d). The DMPO-·O2 intensity in the in-situ ESR spectra for COF-N32 thus is higher than the other two COFs (Fig. S39). This indicates that regardless the amount of O2 adsorbed by COF-N32 was not the highest among three COFs (Figs. S40 and S41), the electron transfer efficiency from COF-N32 to O2 for the production of ·O2 species was higher than the other two COFs. Accordingly, the H2O2 yield by COF-N32 is the highest among three COFs. As mentioned above, e generated from the dissociation of formed excitons by COFs with light irradiation can react with O2 to produce ·O2 and subsequently to H2O2 (Eqs. S5 and S6). h+ generated from the separation of e-h+ pairs (excitons) yet can oxidize H2O to form adsorbed *OH (Eq. S7) and then to H2O2 (Eq. S8). Consistently, diffusing ·OH is not detected by ESR in COF-N32 reaction system (Fig. S42), suggests the fast oxidation process of *OH to H2O2. Meanwhile, the introduction of tertiary butanol (TBA) has negligible effect on H2O2 yield by COF-N32 (p > 0.1), indicating that diffusing ·OH does not have contribution to the photocatalytic process of H2O2 production.

Fig. 5: Photocatalytic mechanisms for H2O2 production process.
figure 5

a Quenching experiments for H2O2 photosynthesis. Experimental conditions: λ > 420 nm (298 K; xenon lamp, light intensity:100 mW cm−2), volume (50 mL), photocatalyst (25 mg), [p-BQ]0 = 3 mM, [TBA]0 = 3 mM. b Time-course in-situ FTIR spectra of O2 on COF-N32 under visible light irradiation with O2. c PDOS of COF-N31, COF-N32 and COF-N33. The dashed lines stand for the Fermi level. Calculated energy profile for (d) oxidation of water into H2O2 and (e) reduction of oxygen into H2O2 on COF-N31, COF-N32 and COF-N33 at U = 0 V vs. SHE at pH = 7. Error bars in a represent the average values (mean ± s.d., n = 3).

In-situ FTIR spectra in sealed chamber was further employed to reveal the reactive sites on COF-N32 during the H2O2 generation process. The results show that H2O adsorbed onto surfaces of COFs without light irradiation can be dissociated into C-OH (1094 cm−1) and TzH+ (1508 cm−1) in triazine rings on COF-N32 (Fig. S43)50. To further elucidate the water oxidation process by COF-N32 with light irradiation, the in-situ FTIR spectra with the presence of O2 and H2O was also achieved. With the increase of reaction duration, both peaks at 1182 cm−1 corresponding to N-O and 1379 cm−1 corresponding to O-H bonds51 in COF-N32 are found to be increased. Moreover, TzH+ (at 1522–1557 cm−1) is also generated (Figs. 5b and S44). The observation indicates that the water oxidation surrounding the N atoms in triazine rings leads to the formation of adsorbed *OH and H+. This allows the occurrence of water oxidation with relatively low overpotential of VB compared with E(H2O2/H2O) (1.77 V vs. NHE)52. The production of H2O2 in half-reaction (with the removal of e by NaBrO3) under the N2 atmosphere (Fig. S45) confirms the presence of two-electron water oxidation process in COF-N32 reaction system, while the negligible O2 generation in oxygen evolution experiment excluded four-electron water oxidation (Fig. S46). The isotopic experiment (using H218O and 16O2 as precursors in a sealed reactor) shows that the amount of 16O2 and 18O2 generated from the decomposition of H2O2 is generally equivalent (Fig. S27b), which further confirms the presence of two-electron water oxidation as well as the charge conservation with oxygen reduction during the H2O2 production process. It is worth pointing out that replacing triazine rings in COF-N32 by benzene rings (COF-C32) can decrease the yield of H2O2 under visible light irradiation (Fig. S47). The observation indicates that introduction of triazine structure into COFs is favorable for H2O2 production via facilitating water oxidation. The vibration peak at 990 cm−1 is corresponded to adsorbed ·O2 and that at 861 cm−1  is corresponded to peroxy species (-O-O-) in the in-situ FTIR spectra of COF-N32 (Figs. 5b and S48)53. The formation of C-O (1419 cm−1) and C = C-O (1264 cm−1) in the in-situ FTIR spectra of COF-N32 further suggest the reduction of O2 occurs at C atoms of COF-N32.

The H2O2 photosynthesis process by COFs was further revealed by theoretical calculation. The calculated PDOS plots show that the valence band top and conduction band bottom in all three COFs mainly consist of N 2p and C 2p states, respectively (Fig. 5c). This implies that during the photocatalytic reaction in all three COFs, the N sites act as the oxidizing centers, while the C sites serve as reducing centers. Comparing with COF-N31 with weakened π conjugation, the N 2p states in COF-N33 move toward a more negative region. N 2p states of COF-N32 locate between those of COF-N31 and COF-N33. Note that more negative N 2p states can easily lead to the formation of unstable chemical bond between catalyst and reactant molecule. In contrast, more positive N 2p states readily causes the formation of inert chemical bond, which is not beneficial for the catalytic reaction either54. Instead, a moderate location of N 2p can effectively improve the photocatalytic activity. Likewise, the location of C 2p states in COF-N32 is also between those in COF-N31 and COF-N33. Hence, the H2O2 yield by COF-N32 is greatest among three COFs. The energy profiles of H2O2 production by three COFs were further determined. As shown in Fig. 5d, e, the water oxidation is the rate-determining reaction in the photocatalytic water and oxygen reforming. Moreover, the conversion of *OH into *+H2O2 is the rate-determining step in water oxidation55. Specifically, COF-N32 owns a lowest energy barrier (1.76 eV) among three COFs. Meanwhile, the energy barrier of oxygen reduction into H2O2 by COF-N32 is only 0.24 eV, which is also the lowest among COFs. Hence, the highest H2O2 yield is obtained by COF-N32 with light irradiation.

In summary, in order to enhance the yield of H2O2 by COFs under light irradiation, we propose a strategy to facilitate the formation and dissociation of excitons in COFs through optimizing the intramolecular polarity of COFs by introducing suitable amount of phenyl group as electron donors. We fabricate a class of COFs with different intramolecular polarity by using triazine-cored triamine with different amount of phenyl group (n = 0, 1, 2) as the precursors. We find that among all three COFs, COF-N32 with moderate intramolecular polarity in the octupolar conjugated structure can generate greatest amount of charges (i.e. electrons) for H2O2 photosynthesis under light irradiation. Without the requirement of sacrificial agent, the H2O2 yield by COF-N32 reaches 7092 μmol g−1 (605 μmol g−1 h−1) after 12 h of visible light irradiation with solar-to-chemical efficiency of 0.31% and high AQY of 6.2% at 459 nm. Moreover, we find that COF-N32 can also efficiently produce H2O2 in more available real water samples including tap water, river water and sea water. COF-N32 can be assembled into practical devices for consecutive uses with high photocatalytic stability. COF-N32 either dispersed in membrane filter reactors or immobilized onto glass slides can efficiently produce H2O2 under natural sunlight irradiation. During the H2O2 photosynthesis process, suitable N 2p states and C 2p states in COF-N32 with optimal intramolecular polarity reduce the energy barrier for H2O activation and oxygen reduction, respectively, contributing to the high efficiency. This study not only provides deep insight into the design of COFs via regulating its intramolecular polarity to boost the H2O2 photosynthesis without using sacrificial agent, but also paves the way for the practical application of COFs-based photosynthesis of H2O2.


COFs synthesis

Unlike COF-N32 and COF-N33 that can be fabricated by using one-pot solvothermal method based on the Schiff-base reaction between two types of precursors in mesitylene/dioxane/acetic acid solvent, COF-N31 yet can not be successfully synthesized by the mixture of 1,3,5-triformylphloroglucinol (Tp) and melamine in mesitylene/dioxane/acetic acid solvent due to its instability in this solvent. Instead, COF-N31 were fabricated in dimethyl sulfoxide/N,N-dimethylacetamide (DMAc)/acetic acid following the method reported in previous literature22. Specifically, 0.3 mmol of Tp (63 mg) and 0.3 mmol of melamine (38 mg) were added into a reactor, followed by the addition of 2 mL of dimethyl sulfoxide, 1 mL of DMAc and 0.3 mL of 6 M acetic acid. After ultrasonication and degassed by three consecutive freeze-pump-thaw cycles, the reactor was sealed under vacuum condition, which was then heated at 120 °C for 72 h. The collected product was firstly rinsed by DMAc, which was then solvent exchanged with DMAc, pure water and washed with acetone for three times. The final product was dried at 120 °C under vacuum.

COF-N32 and COF-N33 were synthesized by mixing Tp with the precursors with triazine and triamine in mesitylene/dioxane/acetic acid solvent, respectively. Briefly, 0.9 mmol (190 mg) of Tp and 0.9 mmol of 4,4',4''-(1,3,5-triazine-2,4,6-triyl)-trianiline (318.6 mg, for COF-N32) or 4,4',4''-(1,3,5-triazine-2,4,6-triyl)tris(([1,1'-biphenyl]−4-amine)) (524 mg, for COF-N33) were added into a Teflon lining with the volume of 20 mL. Then, dioxane (4.5 mL), mesitylene (4.5 mL) and 3 M acetic acid (1.5 mL) were added. The mixture was ultrasonicated and bubbled with N2 for 20 min. After that, the Teflon lining was sealed in an autoclave and heated at 120 °C in the oven for 72 h. The mixture was separated through filtration and the solid was rinsed by acetone for five times, which was finally dried at 60 °C. The precursors-to-COFs efficiency is 76% for COF-N31, 78% for COF-N32 and 75% for COF-N33, respectively.

Photocatalytic experiments

The photosynthesis of H2O2 was first performed in a double wall quartz reactor with 50 mL of ultrapure water and 25 mg of photocatalysts (1, 5 and 10 mg of COF-N32 were also considered in particular experiments). The initial solution pH of the mixture was adjusted by NaOH or HClO4. The reaction suspension was irradiated by using a 300 W Xenon lamp (λ > 420 nm, 100 ± 1 mW cm−2). The temperature was fixed at 25.0 ± 0.2 °C by circulating water system during the photocatalytic experiments. The reaction suspension was extracted and filtered for H2O2 measurement at specific time intervals. The reusability of three COFs was investigated. The H2O2 production by COF-N32 in various real waters (e.g. tap water, river water and sea water) was evaluated under both simulated visible light and natural solar light. COF-N32 were employed in commercial membrane reactors and further immobilized onto ITO glass with the size of 10 cm × 10 cm for the continuous H2O2 photosynthesis.

Catalyst characterization

XRD (DMAX-2400, Rigaku, Japan), XPS (Axis Ultra, Kratos, UK), SEM (JSM-F100, JEOL, Japan), TEM (Tecnai F30, USA), 13C NMR (Bruker-400 AVANCE III, Bruker, Switzerland), FTIR (Nicolet is50, Thermo Fisher, USA) were employed to reveal the chemical and structural information of COFs. In-situ FTIR spectra measurement (Bruker Tensor, Bruker, Switzerland), UV-vis DRS (UV-2400, Shimadzu, Japan), steady state PL spectra, time-resolved PL decay curve, temperature-dependent PL spectra (FLS980, Edinburgh, UK), TA spectrometer (Helios, Ultrafast System, USA), ESR analysis (Bruker EMX, Bruker, Switzerland), water adsorption analysis (3Flex, Micromeritics, USA), RRDE (PINE E6, USA), 18O isotopic experiment and EIS (CHI760E, Chenhua, China) were performed to investigate the mechanisms of H2O2 photosynthesis by prepared COFs.