Introduction

Hydrogen peroxide (H2O2), an eco-friendly oxidant and an emerging energy carrier, has been widely implemented in a variety of fields, such as disinfection, sewage treatment and chemical synthesis1,2. Given that the industrial anthraquinone process is capital and energy intensive (~US$ 1200 tonne−1 and ~67 GJ tonne−1)3, the development of a facile and eco-friendly avenue for H2O2 production is urgent. Artificial photosynthesis of H2O2 is a potentially more sustainable alternative to the traditional anthraquinone method since it uses H2O and O2 as reactants and sunlight as the only energy supply. However, this technology has not been used for scalable H2O2 production. The primary challenge lies in the fact that the efficiency and stability of current state-of-the-art photocatalysts have not yet fulfilled industrial requirements. Consequently, more efforts have been devoted to developing low-cost, durable and highly efficient catalysts for photocatalysis4. Graphitic carbon nitride (g-C3N4) has drawn extensive attention in H2O2 photosynthesis because of its negative conduction band, which drives two-electron oxygen reduction reactions (ORRs) (O2 + 2H+ + 2e−→H2O2, 0.695 V vs. NHE)5. However, a two-electron water oxidation reaction (WOR) (2H2O + 2 h+→H2O2 + 2H+, 1.76 V vs. NHE) is not easy to achieve because of the rather positive potential6. Another possible route is a four-electron WOR with O2 as the product, possibly followed by the subsequent consumption of O2 in the ORR. In fact, the photogenerated holes of mostly reported g-C3N4 photocatalysts are not thermodynamically favourable for the evolution of O2, which severely restricts overall H2O2 production7,8. Notably, Shen et al. innovatively introduced boron dopants into g-C3N4 (BCN) to remarkably modulate the valence band structure and hence alter the water oxidation capacity9. Thus, the resulting BCN with improved O2 evolution renders it a promising candidate for H2O2 photosynthesis from O2 and H2O. Unfortunately, the current photocatalytic performance of H2O2 synthesis for BCN is still hindered by sluggish ORR kinetics and the short lifetime of electrons.

Generally, manipulating metallic sites could modulate both the activity and selectivity of the ORR. As reported earlier, the Pauling-type (end-on) O2 adsorption configuration on a metal surface exhibited a highly selective two-electron ORR due to minimized O-O bond breaking10,11,12. More importantly, modulating the direct one-step two-electron ORR route is crucial for achieving H2O2 production with high atomic utilization but is challenging. Currently, Fe-O and Fe-N sites, etc., have been demonstrated as promising ORR cocatalysts in photocatalysis and electrocatalysis13,14. However, how to achieve a direct two-electron pathway for H2O2 production at such Fe sites remains obscure.

Stabilization of endoperoxide species is deemed to be the key for direct one-step two-electron ORR15. The Gibbs free energies of endoperoxide species can be regulated at O2 adsorption centres16, which is closely related to the configuration and coordination of the introduced metallic sites. Hence, it is reasonable to deduce that the inherent structure of Fe species can greatly influence the ORR pathway. Compared with single-atom Fe or Fe clusters, iron oxyhydroxide clusters with subnanoscales have more misaligned atoms, potentially resulting in preferable O2 activation properties. For example, the FeO6 moiety of FeOOH was revealed as an effective active centre in the electrocatalytic ORR, and a theoretical study demonstrated that FeOOH with surface vacancies has faster dynamics for the ORR than perfect FeOOH17. Inspired by these findings, modulating the defective structure of FeOOH is expected to expose more active unsaturated metallic sites to alter O2 adsorption and stabilize the endoperoxide species18. This makes it an ideal candidate for H2O2 production via the selective direct two-electron pathway. In general, topochemical transformation with pyrolysis has been regarded as a classic method to support single atoms or clusters on g-C3N4-related substrates19,20. However, it is ill-suited to retain the structure of atomically dispersed FeOOH since the oxyhydrogen groups are deprived during thermal treatment. In recent years, a freezing-assisted approach has been proposed as an ingenious route to anchor single atoms or ultrasmall clusters since uniformly dispersed metal precursors at low temperatures can be confined to a limited space, and subsequent drying treatment avoids surface tension at the gas‒liquid interface, preventing the aggregation of metal atoms synergistically21,22. Hence, it is reasonable to speculate that highly atomically dispersed FeOOH with unsaturated coordination could be successfully supported on BCN via a freeze‒drying strategy.

In parallel, arranging the WOR in H2O2 photosynthesis is equally crucial since the hole-involved half-reaction is quite sluggish and strongly influences charge separation23,24. Nevertheless, current concerns of hole quenching by water to liberate long-lived electrons to evoke the ORR are often overlooked. A representative Co oxide species remains one of the most effective cocatalysts capable of trapping photogenerated holes and lowering the energy barrier of water oxidation. CoOx, a classic category of Co oxide species with tuneable Co3+/Co2+ ratios, has long been reported as an effective modifier for capturing holes and accelerating water oxidation kinetics25. Hence, engineering highly active and selective sites of FeOOH with proper defects for two-electron ORR and CoOx centres for water oxidation simultaneously offers a promising strategy to achieve direct one-step two-electron H2O2 production on BCN; however, this strategy has seldom been reported to date, and the precise manipulation of dual sites remains a great challenge in spatial separation.

In addition, identifying the individual roles of modified FeOOH and CoOx and understanding the charge carrier dynamics in the ORR for H2O2 production are important. Transient absorption spectroscopy (TAS), an advanced time-resolved spectroscopic technique, has now been widely applied to elucidate the charge carrier dynamics of semiconductors in photocatalysis, such as water splitting and CO2 reduction26. Clearly, it is highly desirable to clarify the behaviours of charge carrier dynamics in the ORR during H2O2 formation by using in situ TAS, especially under identical reaction conditions. Unfortunately, there have been no reports on the spectroscopic investigation of such an ORR procedure to date.

Here, we report a class of catalysts (CoOx-BCN-FeOOH) for H2O2 photosynthesis in a water and oxygen mixture without any sacrificial agents under visible light. The optimal one delivers a 30-fold activity enhancement compared with g-C3N4, and achieves an SCC efficiency up to 0.75%. CoOx clusters were found to facilitate hole‒water oxidation and prolong the electron lifetime of BCN. While FeOOH was proven to accept electrons and promote oxygen activation. Moreover, the one-step two-electron ORR pathway regulated by coordination-unsaturated Fe sites was elucidated via experimental and theoretical methods. This work provides a design guide to develop efficient photocatalysts with dual-active-site engineering for H2O2 production and an approach to investigate electron kinetics during the ORR.

Results

Synthesis protocol and structural characterization of CoOx-BCN-FeOOH

A schematic of the synthetic protocol for the well-designed photocatalyst is depicted in Supplementary Fig. 1. Starting from g-C3N4 nanosheets (CN) synthesized by the thermal polymerization of urea, boron dopants were incorporated into the CN through thermal treatment with NaBH4 to prepare boron-doped CN (BCN) with sufficient water oxidation thermodynamics9,27. Subsequently, the Fe species were loaded onto BCN by facile impregnation and freeze-drying in a two-step procedure, followed by the in situ photoassisted oxidation of Co2(CO)8 in an inert atmosphere to fabricate Co and Fe species co-modified BCN. The B dopant and N defects in BCN are revealed by the solid-state 11B MAS-NMR spectrum and X-ray diffraction (XRD) patterns (Supplementary Fig. 2), which are consistent with the previous report9. Compared with CN, BCN has a sheet-like morphology with thinner sheet structural features (Supplementary Figs. 3–4). This might be related to the broader interlayer spacing caused by the second calcination process28. Compared with that of CN, the absorption threshold wavelength of BCN has a redshift, with a narrowed bandgap (Supplementary Fig. 5). The conduction bands (CB) of CN and BCN are determined to be approximately −1.23 and −0.59 V vs. NHE, respectively (Supplementary Fig. 6). Therefore, the valence bands (VB) of CN and BCN are deduced to be 1.47 and 1.81 V, respectively. Undoubtedly, the deeper VB position of BCN caused by boron doping makes it more favourable for water oxidation, an equally crucial step in H2O2 production.

The introduction of Fe and Co species barely changed the structural characteristics, bandgap and skeleton features of BCN (Supplementary Figs. 7–9, Supplementary Tables 1–2). The overall morphology and the atomic structure of the Fe and Co species over CoOx-BCN-FeOOH as well as the individual Fe species over BCN-1FeOOH were distinguished via transmission electron microscopy (TEM) and aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (AC HAADF-STEM) images (Fig. 1a, Fig. 1b and Supplementary Fig. 10). Highly dispersed atomic Fe and Co species on BCN are resolved, predominantly very small subnanoscale (marked with red circles) and partial of isolated single atoms (marked with yellow circles). Some species that a little > 1 nm have also been observed, perhaps owing to the intense electron beam during TEM or uncontrollable agglomeration in the preparation process. In addition, Fe, Co, C, N, O and B are clearly observed in the EDX mapping images (Fig. 1c), indicating the successful loading of Fe and Co species. This is further verified by the slightly decreased specific surface area of CoOx-BCN-FeOOH compared with that of pristine BCN (Supplementary Table 3). Based on the calculation from the inductively coupled plasma-atomic emission spectrometry (ICP‒AES) measurements, the loading amounts of Fe and Co on the BCN are 0.47 and 0.37 wt.%, respectively (Supplementary Table 4).

Fig. 1: Morphology and structural characterization of CoOx-BCN-FeOOH.
Fig. 1: Morphology and structural characterization of CoOx-BCN-FeOOH.
Full size image

a Aberration-corrected high-angle annular dark-field scanning TEM (AC-HAADF-STEM) image, b TEM image, c HAADF-STEM image of CoOx-BCN-FeOOH and the corresponding EDX mapping images of C, N, O, B, Fe and Co (scale bar: 500 nm). d XPS analyses of Fe 2p for BCN-FeOOH and CoOx-BCN-FeOOH. e XPS analyses of Co 2p for CoOx-BCN and CoOx-BCN-FeOOH. f Fe K-edge XANES profiles of Fe foil, FeO, Fe2O3, FeOOH and CoOx-BCN-FeOOH. g Fitting curves for the Fe K-edge EXAFS spectra of CoOx-BCN-FeOOH. h Co K-edge XANES profiles of Co foil, CoO, Co3O4 and CoOx-BCN-FeOOH. i Fitting curves for Co K-edge EXAFS spectra of CoOx-BCN-FeOOH. Source data are provided as a Source Data file.

X-ray photoelectron spectroscopy (XPS) analysis was conducted to determine the chemical states of the modified Fe and Co species as well as the potential interactions between BCN and the cocatalysts. The B 1 s XPS signal with a binding energy (BE) of 191.1 eV is recorded for BCN, suggesting the successful introduction of boron dopants. Notably, the BE of B 1 s and N 1 s for BCN is identical to that of the Fe and Co comodified BCN (Supplementary Fig. 11 and Fig. 12a), suggesting that the introduced Fe and Co species are anchored neither on the B nor N sites. The C 1 s peak at 288.5 eV assigned to N-C=N shifts slightly positive after loading with Fe species5, and the characteristic peak further shifts in the direction of high BE after modification with Co species (Supplementary Fig. 12b). Moreover, the O 1 s peak ascribed to C-OH of BCN shifts negatively after FeOOH modification29, and then shifts backwards after further modification with CoOx (Supplementary Fig. 13). In combination with the weakened vibration of surface hydroxyl groups (3500–3000 cm−1) on BCN detected via FT-IR spectra (Supplementary Fig. 9), it is reasonable to deduce that the Fe and Co species could interact with the hydroxyl groups grafted on the C atoms of BCN and then be immobilized. The Fe 2p peaks could be further deconvoluted into six subpeaks, in which the BEs at 710.7 and 724.0 eV are associated with an Fe(III) oxidation state along with two satellite peaks as the fingerprint of Fe(III). The other two peaks at 714.0 and 725.9 eV are attributed to hydroxide groups30 (Fig. 1d). Note that the Fe 2p XPS profiles of the Fe and Co species comodified with BCN are nearly the same as those of the individual Fe-modified BCN, implying that the chemical state of Fe has not been influenced after further decoration with Co. On the other hand, the Co 2p peaks could be further fitted to several subpeaks assigned to the Co(II) (781.9 eV) and Co(III) (779.5 eV) oxidation states, as well as the satellite peaks of the Co(II) oxidation state (786.3 and 802.9 eV)31 (Fig. 1e). The fine structure and atomic coordination of the Fe and Co species introduced in BCN were further ascertained via X-ray absorption spectroscopy (XAS). X-ray absorption near-edge structure (XANES) spectra were collected at the Fe K-edge of CoOx-BCN-FeOOH and several references (Fe-foil, FeO, Fe2O3 and FeOOH) (Fig. 1f). Notably, the absorption edge of the investigated sample is close to that of FeOOH and Fe2O3, suggesting that the valence state of Fe in the catalyst is close to +3. This is consistent with the XPS analysis. Figure 1g presents the Fourier transform (FT) of the Fe K-edge extended X-ray absorption fine structure (EXAFS) of Co and Fe species comodified with BCN. The peak at ~ 1.5 Å (phase shift not corrected) is ascribed to the first oxygen neighbours around the Fe3+ ions32, and no scattering path signals assigned to the Fe-Fe bond (2.47 Å) from the Fe foil can be observed. The peak positions of the Co and Fe species comodified with BCN are closer to those of the FeOOH reference, suggesting that the introduced Fe species on the surface of BCN are more similar to those of FeOOH with respect to FeO and Fe2O3. In particular, based on the analysis of the EXAFS data (Supplementary Tables 5 and 6), the coordination number of Fe-O is less than six, implying that FeOOH is potentially endowed with an unsaturated coordination environment. On the other hand, the absorption edge of CoOx-BCN-FeOOH falls between those of CoO and Co3O4 (Fig. 1h), indicating that the oxidation state of Co is between +2 and +2.67. The local coordination environment of the Co atom in the catalyst was further revealed by Fourier-transform cobalt K-edge extended EXAFS spectra (Fig. 1i). The fitting results reveal that the peak located at ~ 1.44 Å (phase shift not corrected) is assigned to the Co-O scattering path of ultrasmall CoOx clusters and isolated Co atoms33, however, no predominant peaks ascribed to Co–Co bonds from the Co foil are observed. This implies that the modified Co species are not in the form of metallic Co nanoparticles. The lack of Co-Co scattering at ~2.5 Å is probably due to the modified Co species being quite small34,35.

Photocatalytic activities of H2O2 production and charge separation

The photocatalytic H2O2 generation performance of the catalysts was evaluated under visible-light irradiation (λ ≥ 420 nm, 100 mW cm−2) in pure water (18.4 MΩ·cm, without any sacrificial agent, pH = 7.0). The H2O2 yields of all the investigated samples increased with increasing irradiation time (Fig. 2a and Supplementary Figs. 14-15). The average H2O2 generation rate of BCN is higher than that of pristine CN, and it ascends steadily as the loading amount of FeOOH gradually increases from 0.5 wt.% (85 μmol g−1 h−1) to 1.0 wt.% (111 μmol g−1 h−1). However, further loading of FeOOH deteriorates the photocatalytic H2O2 production performance, which is associated with the aggregation of modified Fe species and larger sizes caused by their overloading (Supplementary Fig. 16). Surprisingly, the photocatalytic activity further improved after CoOx was introduced into the prepared BCN-1FeOOH. The optimal CoOx-BCN-FeOOH delivers the highest H2O2 generation rate of up to 0.34 mmol g−1 h−1 when the CoOx feed loading is controlled to 0.75%, which is ~ 30 times greater than that of CN (11.59 μmol g−1 h−1). Notably, no H2O2 is detected without light irradiation or without catalysts, demonstrating that it is a photo-driven reaction. Moreover, no competitive H2 is recorded on CoOx-BCN-FeOOH. Meanwhile, the SCC efficiency of CoOx-BCN-FeOOH reached 0.95% in the first hour and finally stabilized at 0.75% in 3 h (Supplementary Fig. 17 and Supplementary Table 7). The AQY of CoOx-BCN-FeOOH at 420 nm also reached 8.36% (Supplementary Table 8). Impressively, the performance of CoOx-BCN-FeOOH is ranking among the forefront of g-C3N4-based catalysts12,36, even surpasses most of the relevant materials reported recently37,38 under comparable experimental conditions, with SCC efficiency and AQY as the evaluation indicators of photocatalytic activity (Fig. 2b and Supplementary Table 9).

Fig. 2: Performance of H2O2 photosynthesis and the roles of CoOx and FeOOH in charge modulation.
Fig. 2: Performance of H2O2 photosynthesis and the roles of CoOx and FeOOH in charge modulation.
Full size image

a Photocatalytic performance of CN, BCN, BCN-FeOOH, CoOx-BCN and CoOx-BCN-FeOOH towards H2O2 production with pure water under visible-light irradiation (pH= 7.0). b Comparison of the SCC efficiencies of CoOx-BCN-FeOOH and recently reported photocatalysts. The relevant catalysts and the specific reaction conditions are listed in Supplementary Table 9. c Five consecutive runs of H2O2 production by CoOx-BCN-FeOOH with pure water under visible-light irradiation. d The formation rate constant (Kf) and decomposition rate constant (Kd) of H2O2 for CN, BCN, BCN-FeOOH, CoOx-BCN and CoOx-BCN-FeOOH. e Time-resolved PL spectra and f TPV responses of BCN, BCN-FeOOH, CoOx-BCN and CoOx-BCN-FeOOH in an O2 atmosphere. g-h In situ-irradiated XPS analyses for Fe 2p and Co 2p of CoOx-BCN-FeOOH. The error bars (mean ± standard deviation) were obtained based on three independent photocatalytic experiments. Source data are provided as a Source Data file.

Furthermore, the H2O2 generation rate of CoOx-BCN-FeOOH is disclosed to maintain > 95% of the initial activity after five consecutive runs (Fig. 2c), and the structural features of FeOOH and CoOx are well retained (Supplementary Figs. 18–20), confirming that the catalyst is quite stable during the photoreaction. Given that photocatalytic H2O2 production is a dynamic process, the ultimate generation rate depends on the formation and decomposition of H2O2 over the catalysts. Consequently, the H2O2 degradation behaviours of the investigated samples were also explored. The rate constants of H2O2 formation (Kf, μM min−1) and decomposition (Kd, 10−2 min−1) for the abovementioned catalysts were calculated (Fig. 2d and Supplementary Fig. 21). CoOx-BCN-FeOOH presents the highest Kf value while the lowest Kd. This result suggests that FeOOH and CoOx modification promotes H2O2 formation and simultaneously suppress the backwards decomposition of H2O2. Photoluminescence (PL) spectra were adopted to explore the influence of FeOOH and subsequent CoOx modification on the charge transfer behaviours of BCN. Evidently, the PL intensities of BCN-xFeOOH are weaker than that of pristine BCN, suggesting that FeOOH modification is favourable for the charge separation of BCN. Notably, the PL signals are further quenched on yCoOx-BCN-FeOOH ones (Supplementary Figs. 22–23), implying that the charge recombination of BCN is well restricted by the collaboration of CoOx and FeOOH. This result is further supported by time-resolved PL spectra. The average PL lifetimes of BCN, CoOx-BCN, BCN-FeOOH and CoOx-BCN-FeOOH are calculated to be 9.12, 8.16, 7.83 and 7.21 ns, respectively (Fig. 2e and Supplementary Table 10). The shortened PL lifetime of BCN after FeOOH and CoOx modification, especially for the comodified one unambiguously demonstrates that an efficient charge transfer channel is established between BCN and FeOOH and that the subsequent introduction of CoOx further promotes charge transfer from BCN to FeOOH.

The charge transfer dynamic behaviour of the typical catalysts was further revealed via transient-state surface photovoltage spectroscopy (TPV) in an O2 atmosphere (Fig. 2f). Given that the detected TPV signal of BCN is positive, it is reasonable to deduce that the adsorbed O2 would capture photogenerated electrons so that the photogenerated holes could preferentially diffuse to the electrode surfaces. In this case, the recorded TPV response is predominantly ascribed to the photogenerated holes. The charge carrier lifetime of BCN is shortened from 1300 – 1200 μs after CoOx decoration, which means the photogenerated holes are captured by the CoOx and hence enhance the charge separation of BCN. However, an inverse case is observed for the BCN-FeOOH catalyst. FeOOH modification prolongs the charge carrier lifetime of BCN, indicating that the electrons of BCN are captured by the modified FeOOH to facilitate the diffusion of photogenerated holes to the surface39,40. Notably, the charge carrier lifetime of BCN-FeOOH is shortened from 1470 – 1340 μs after the introduction of CoOx, manifesting that the photogenerated holes survived in BCN-FeOOH are further captured by CoOx clusters. Accordingly, it can be concluded that the charge separation of BCN has evidently promoted with the simultaneous modulation of electrons and holes by the dual-site engineering. In situ-irradiated XPS analyses were conducted to determine the charge transfer direction on the CoOx-BCN-FeOOH catalyst (Fig. 2g, h). The BEs of C 1 s and N 1 s barely changed before and after light irradiation (Supplementary Fig. 24). A negative shift of ~ 0.25 eV for Fe 2p is clearly recorded after light irradiation, indicating an increase in its electron density. This further suggests that the modified Fe species could accept electrons from BCN during the photoreaction. In contrast, a positive shift of ca. 0.24 eV is identified for Co(II), while the BE of Co(III) is essentially retained. Moreover, the peak area of Co3+/Co2+ increases from 0.76 – 1.30 upon exposure to light irradiation, implying that the photogenerated holes of BCN are captured by the CoOx clusters.

Electron decay dynamics and dual cocatalysis

To shed light on the specific roles of the introduced FeOOH and CoOx, TAS at the microsecond timescale, which matches well with the real reaction time, was employed to investigate the charge carrier dynamics of the as-fabricated photocatalysts. To distinguish the charge carrier nature of the excited states of BCN, via TAS, the individual BCN was initially probed at 900 nm with and without triethanolamine (TEOA), a proverbial hole-trapping agent for long-lived electrons. As shown in Supplementary Fig. 25, the half-lifetime (t50%) detected on the BCN in the presence of TEOA is much longer than that in the case without TEOA. Consequently, the collected signal could be assigned to photogenerated electrons in BCN, which is also supported by previous studies on CN41. The electron t50% of BCN (19.5 μs) is shortened after FeOOH modification (17.4 μs) under a N2 atmosphere (Fig. 3a), suggesting that the electron is captured by FeOOH. In contrast, CoOx-BCN has a longer electron lifetime (t50% = 25.9 μs) than BCN does, suggesting that long-lived electrons are preserved because CoOx traps photogenerated holes. Predictably, a moderate t50% of the electron signal is recorded for CoOx-BCN-FeOOH (22.7 μs).

Fig. 3: Investigation of the electron kinetics of the ORR via in situ TAS.
Fig. 3: Investigation of the electron kinetics of the ORR via in situ TAS.
Full size image

a μs-TAS decay kinetics of BCN, BCN-FeOOH, CoOx-BCN and CoOx-BCN-FeOOH in N2. b μs-TAS decay kinetics of BCN and BCN-FeOOH under O2(H2O) conditions. c μs-TAS decay kinetics of BCN and CoOx-BCN under N2(H2O) conditions. d μs-TAS decay kinetics of CoOx-BCN-FeOOH in different atmospheres. e Plots of the relationships among the half-life time (t50%), electron transfer efficiency (ETE) and H2O2 yield of the investigated catalysts. f Electrochemical reduction curves of CN, BCN and BCN-FeOOH in the O2-saturated electrolyte (the oxidation curves of CN, BCN and CoOx-BCN are presented in the inset). Source data are provided as a Source Data file.

To elucidate the crucial roles of the modified FeOOH and CoOx, in situ exploration of electron kinetics was conducted by mimicking real reaction conditions, where O2 (g) containing H2O (O2 (H2O) for short) was introduced into the test system. Moreover, measurements involving the substitution of O2 with N2 were also carried out as a reference. Figure 3b depicts the μs-TAS decay kinetics of BCN- and FeOOH-modified samples under O2 (H2O) conditions. Compared with that of pristine BCN, the electron t50% in BCN-FeOOH is shortened by 6.1 μs. The electron t50% variation between these two samples in N2 (H2O) is only 1.8 μs (Supplementary Fig. 26). This finding infers that the electrons can be efficiently captured by the anchored FeOOH in an oxygen-saturated environment and, in turn, accelerate the oxygen reduction reaction. From another point of view, the electron t50% of BCN is significantly prolonged after CoOx decoration (from 25.1 μs – 49.2 μs) under N2 (H2O) conditions. Notably, the electron t50% variation (24.1 μs) is in stark contrast to the case of N2-filled one (6.4 μs), suggesting that CoOx introduction could effectively extract the photogenerated holes of BCN and hence facilitate water oxidation (Fig. 3c).

In addition, the electron t50% of CoOx-BCN under O2(H2O) conditions is similar to that under N2 (H2O), implying that CoOx contributes little to oxygen activation (Supplementary Fig. 27). As expected, the electron t50% of CoOx-BCN is shortened from 49.2 – 41.4 μs after further modification with FeOOH under conditions of N2 (H2O) (Supplementary Fig. 28), and a much shorter lifetime is observed in the case of O2 (H2O) (Fig. 3d). Following the above indications, one can readily see that the electron decay kinetics of BCN with FeOOH modification can be further regulated by CoOx introduction under simulated real reaction conditions. To make it clear, the electron transfer efficiency (ETE) for O2 reduction is presented by the following equation42:

$${{\boldsymbol{\eta }}}=1-\frac{{{{\bf{t}}}}_{50\%}({{{\rm{O}}}}_{2})}{{{{\bf{t}}}}_{50\%}({{{\rm{N}}}}_{2})}$$
(1)

The ETE for O2 reduction of the investigated samples is calculated on the basis of the acquired electron t50% (Supplementary Table 11), and the plots of the relationships of t50%, ETE and H2O2 evolution are depicted in Fig. 3e. Note that the calculated ETE of BCN is only 8.7%, whereas the ETE for BCN-FeOOH reaches 27.5%, suggesting that electron extraction by O2 is accelerated by the modified FeOOH. The ETE also indicates that the anchored FeOOH can function as a good cocatalyst for O2 reduction. Interestingly, CoOx-BCN delivers an ETE of 13.3%, which is far less than that of the FeOOH-modified sample (27.5%), but the t50% of CoOx-BCN detected under N2(H2O) conditions is quite long (49.2 μs). By comparison, the H2O2 yields for these two catalysts are comparable (101 μmol g−1 h−1 for CoOx-BCN and 110 μmol g−1 h−1 for BCN-FeOOH). This observation illustrates that the ETE for O2 reduction reflects catalysis based on a certain electron lifetime, whereas the photocatalytic activity is closely related to the ETE value and the substantially transferred electron quantity, which is determined by the electron lifetime. Hence, the maximum ETE of 34.1% and the highest H2O2 yields are recorded for CoOx-BCN-FeOOH, which is due to the long-lived electron lifetime of BCN and the preferable catalysis for O2 activation brought by the modified CoOx and FeOOH, respectively. The above TAS results are further supported by the electrochemical measurements (Fig. 3f). The onset potential of CoOx-BCN for electrochemical oxidation is obviously lower than that of pristine BCN in N2-saturated electrolyte, verifying that CoOx introduction is beneficial for reducing the energy barrier of water oxidation. On the other hand, the onset potential of BCN for electrochemical reduction greatly decreases after FeOOH modification under O2-saturated conditions, revealing the crucial role of Fe species in O2 activation.

Direct one-step two-electron reaction pathway

Rotating ring-disk electrode (RRDE) tests were performed to determine the selectivity of the ORR and the number of transferred electrons (n) in the H2O2 production reaction. The selectivity of H2O2 production on the investigated catalysts was monitored in an O2-saturated electrolyte (0.1 M KOH). The disk current that comes from the reduction reaction of O2 for BCN and CoOx-BCN-FeOOH progressively increases as the applied potential decreases from 1.0 V vs. RHE (Fig. 4a, bottom). Concurrently, the H2O2 produced on the disk electrode can quickly diffuse to the ring electrode and be further oxidized to form a positive ring current. The ring current of CoOx-BCN-FeOOH is much greater than that of BCN, BCN-FeOOH and CoOx-BCN, demonstrating that more H2O2 is produced on CoOx-BCN-FeOOH (Fig. 4a, top and Supplementary Figs. 29a and 30a). Additionally, the n calculated for CoOx-BCN-FeOOH is closer to 2, and the H2O2 selectivity (nearly 85%) is greater than that of the other investigated catalysts (Fig. 4b, Supplementary Fig. 29b and Supplementary Fig. 30b). The results reveal that FeOOH modification greatly improves the 2e− selectivity of O2 reduction and hence effectively promotes the surface reaction of the ORR to H2O2.

Fig. 4: One-step two-electron ORR pathway for H2O2 photosynthesis.
Fig. 4: One-step two-electron ORR pathway for H2O2 photosynthesis.
Full size image

a RRDE polarization curves over BCN and CoOx-BCN-FeOOH at 1600 rpm in an O2-saturated 0.1 M KOH solution with a ring current (upper part) and disk current (bottom part). b H2O2 selectivity as a function of the applied potential (the inset shows the calculated average number of transferred electrons). c H2O2 generation rates of a series of samples under different reaction gases or different sacrificial agents. d DMPO spin-trapping EPR spectra recorded for •O2− under light irradiation for BCN, BCN-FeOOH, CoOx-BCN and CoOx-BCN-FeOOH. e H2O2 generation rates of BCN, BCN-FeOOH, CoOx-BCN and CoOx-BCN-FeOOH with different superoxide dismutase (SOD) concentrations. f Mass spectra of luminol after oxidation with hydrogen peroxide generated by photocatalysis with H218O over CoOx-BCN-FeOOH. The error bars (mean ± standard deviation) were obtained based on three independent photocatalytic experiments. Source data are provided as a Source Data file.

Following the indication of a two-electron reduction pathway for H2O2 production, the half reactions on the investigated catalysts and the possible reactive species in the ORR were investigated via variable comparison and trapping experiments (Fig. 4c). The H2O2 yields for all samples apparently decrease when N2 is used as an alternative to O2, implying that the ORR route is the dominant pathway to produce H2O2. However, no H2O2 could be detected in a NaIO3 aqueous solution (NaIO3 as an electron capturer) saturated with N2, excluding the two-electron WOR pathway for H2O2 production. It can be well understood that the detected H2O2 in N2 is produced by the ORR, with the O2 originating from the photogenerated holes initiating the WOR, which is further confirmed by the photocatalytic O2 evolution test (Supplementary Fig. 31). This is in line with previous studies9. While the H2O2 yields for the investigated samples essentially remain unchanged when adding the tert-butanol (TBA), suggesting that •OH is not involved in H2O2 production. Interestingly, based on a set of trapping experiments with benzoquinone (BQ) as the •O2− sacrificial agent43,44 (Supplementary Fig. 32), it turns out that the H2O2 yields for BCN and CoOx-BCN decrease precipitously, which indicates •O2− is an essential intermediate for the formation of H2O2. By comparison, the activities of BCN-FeOOH and CoOx-BCN-FeOOH are substantially maintained with the addition of BQ, even a little bit decrease is observed. However, the degree of decrease in the photocatalytic activity of BCN-FeOOH and CoOx-BCN-FeOOH is far less than that of BCN and CoOx-BCN, implying that FeOOH modification potentially changes the •O2−-associated reaction pathway for H2O2 production. The above indication is further elucidated by the weakened •O2− radicals after FeOOH modification, as revealed by electron paramagnetic resonance (EPR) spectra (Fig. 4d). In contrast, the DMPO-•O2− signal detected on CoOx-BCN is more pronounced than that on BCN, demonstrating that CoOx modification facilitates the generation of •O2− radicals. Given that •O2− is an important intermediate in the stepwise 1e− ORR for 1–4 endoperoxide formation, the decreased DMPO-•O2− signals in BCN-FeOOH and CoOx-BCN-FeOOH suggest rapid O2 reduction to produce H2O2 via a direct 2e− ORR pathway.

As reported earlier, superoxide dismutase (SOD) tends to accelerate the •O2− dismutation reaction to increase •HO2 formation and thus improve H2O2 generation45. Accordingly, SOD was introduced into the photocatalytic system to reveal the contribution of the •O2− intermediates to the reaction pathway. The H2O2 yields of pristine BCN and CoOx-BCN increase with increasing SOD concentration, whereas the H2O2 yields for BCN-FeOOH and CoOx-BCN-FeOOH are basically maintained with increasing SOD concentration (Fig. 4e), which is in line with the results of the EPR measurements. Taken together, these results imply that the introduction of FeOOH on BCN is likely to undergo a one-step 2e− ORR pathway and that the •O2−-associated transfer pathway contributes little to H2O2 production. This finding is further supported by the results of the NBT experiments (Supplementary Figs. 33–34). Moreover, H218O isotope experiments were also conducted under identical conditions to those of H216O (Fig. 4f); however, luminol was added to the system after the photoreaction and acted as a detector to trace the source of the produced H2O2 on the optimized CoOx-BCN-FeOOH catalyst. An isotope luminol oxide−16O (~180 m/z) peak is detected when luminol solution and horseradish peroxidase (HRP) are added to the CoOx-BCN-FeOOH suspension (after 1 h of irradiation), along with a luminol oxide−18O (~182 m/z) peak that shows relatively low intensity. The above results reveal that the synthesized H2O2 mainly originates from O2 reduction at the FeOOH sites.

In situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) measurements were employed to monitor O2 adsorption behaviour and record the H2O2 production process under visible-light irradiation. The intensity of the peaks assigned to the breathing mode of the triazine structure (~820.3 cm−1)46 gradually increase with increasing adsorption time in the dark, indicating that O2 is adsorbed mainly on the triazine moiety of BCN (Fig. 5a). With respect to the CoOx-BCN-FeOOH catalyst, the infrared absorption peak at 891.4 cm−1, which is attributed to the bending vibration of Fe-OH47, gradually strengthens during the dark adsorption procedure, whereas the variation in the triazine structure is quite small (Fig. 5b, c), suggesting that the dominant O2 adsorption sites changed to those of the modified FeOOH on BCN in this case. In parallel, the O2 adsorption feature detected on BCN-FeOOH is consistent with that of CoOx-BCN-FeOOH (Supplementary Fig. 35). To further identify the O2 adsorption configurations, quasi-operando Raman spectroscopy was performed (Supplementary Fig. 36). The results indicate that a new absorption band at ~855 cm−1 increases with increasing irradiation interval. This band can be assigned to the O–O stretching mode of an Fe-OOH species with a Pauling-type adsorption configuration48.

Fig. 5: Mechanism of H2O2 photosynthesis on CoOx-BCN-FeOOH.
Fig. 5: Mechanism of H2O2 photosynthesis on CoOx-BCN-FeOOH.
Full size image

a–c In situ DRIFTS of O2 adsorption over BCN and CoOx-BCN-FeOOH. d-e Insitu DRIFTS of BCN and CoOx-BCN-FeOOH under light irradiation. f Relative peak intensity of H2O2 production in different samples. g–i Top and side views of the structure with different charge densities for O2 adsorption on BCN, BCN-FeOOH and CoOx-BCN-FeOOH. Eads and Δq are the total adsorption energy of O2 and the charge transfer on O2, respectively. Yellow indicates electron accumulation, and light blue indicates depletion. j Free energy diagrams for 2e− ORR processes on BCN, BCN-FeOOH and CoOx-BCN-FeOOH with optimized configurations for each step, where the asterisk (*) denotes the active site of the catalyst. k Schematic of the charge transfer and redox reactions involved in CoOx-BCN-FeOOH. Source data are provided as a Source Data file.

On the other hand, the absorption bands (ranging from 3200 to 3600 cm−1) assigned to the joint stretching vibration coupling the H-bound hydroxyl groups with the adsorbed water49 detected on CoOx-BCN-FeOOH are significantly stronger than those of BCN and BCN-FeOOH. In contrast, the vibrations of the hydroxyl groups on BCN and BCN-FeOOH are roughly the same (Supplementary Fig. 37). Accordingly, it can be deduced that the decorated CoOx serves as a water adsorption site in the CoOx-BCN-FeOOH catalyst. One can see that the intensity of the signals at 1159 cm−1 attributed to •O2−50 gradually increased with increasing illumination time for pristine BCN (Fig. 5d), suggesting a two-step single-electron reaction route. This feature has not been observed for CoOx-BCN-FeOOH (Supplementary Fig. 38), unravelling the ORR procedure, which is different from the BCN procedure. Interestingly, the IR peaks for typical Fe‒OH vibrations on CoOx-BCN-FeOOH slightly shifted under light illumination, along with an obvious consumption in contrast to the case in the dark (Fig. 5e). This implies that O2 participates in the photocatalytic ORR. By comparison, the variation in the IR peaks assigned to the triazine structure of CoOx-BCN-FeOOH can be neglected (Supplementary Fig. 39). Moreover, H2O2 formation on the investigated samples was also monitored via DRIFTS. As depicted in Supplementary Fig. 40, the peaks at 2848 cm−1 are assigned to the typical (ν2 + ν6)/2ν6 OH bending characteristic of H2O251. The peak intensities for all the catalysts gradually increased over the course of 20 min of light irradiation. On the one hand, this observation provides direct evidence of the formation of H2O2 on these catalysts with photoirradiation; on the other hand, CoOx-BCN-FeOOH results in the fastest increase in H2O2 production, as revealed by the normalized relative peak intensities at 2848 cm−1 for all the investigated samples (Fig. 5f). Taken together, the above results unambiguously verify the synergistic effects of delicate CoOx and FeOOH as redox centres for high-efficiency photocatalytic H2O2 production.

Density functional theory (DFT) calculations were adopted to obtain an in-depth understanding of O2 adsorption characteristics and the pivotal steps involved in H2O2 formation on BCN, BCN-FeOOH and CoOx-BCN-FeOOH. The corresponding optimized geometries are shown in Supplementary Fig. 41. Considering that the adsorption of O2 at the active site is a precondition for H2O2 synthesis, the adsorption energy of O2 at different sites in these three models was calculated. The adsorption energy of O2 in the in-plane triazine ring of melon units of BCN was calculated to be −0.13 eV (Fig. 5g). In comparison, it becomes much stronger in the unsaturated coordinated Fe site of BCN-FeOOH (Eads = −0.40 eV) with the end-on configuration (Fig. 5h), indicating that FeOOH modification is more beneficial for O2 adsorption. This is also supported by O2 temperature programmed desorption (O2-TPD) (Supplementary Fig. 42). Interestingly, the adsorption energy of O2 on CoOx-BCN-FeOOH is essentially equal to that on BCN-FeOOH (Fig. 5i). However, the adsorption energy of H2O at the CoOx site is greater than that at the unsaturated coordinated Fe sites (Supplementary Fig. 43), indicating the positive contribution of CoOx to H2O adsorption. Moreover, Bader charge analysis suggests that more electrons can be transferred to O = O in the case of CoOx-BCN-FeOOH (Δq = 0.32 e), in sharp contrast to that of pristine BCN (Δq = 0.08 e), implying that FeOOH introduction plays a significant role in the subsequent proton-coupled electron transfer process for the 2e− ORR.

The Gibbs free energy profiles of the preferred 2e− ORR pathways on BCN, BCN-FeOOH and CoOx-BCN-FeOOH were further calculated. (Fig. 5j and Supplementary Fig. 44). The trend of the Gibbs free energy for all the photocatalysts decreases in the two-electron pathway, suggesting that the generation of H2O2 on the BCN-based catalysts is thermodynamically feasible. For the initial step of O2 adsorption, BCN-FeOOH (ΔG = −0.40 eV) and CoOx-BCN-FeOOH (ΔG = −0.41 eV) display much lower energy barriers than pristine BCN (ΔG = −0.12 eV), which can be attributed to the superior O2 adsorption ability of the coordination-unsaturated FeOOH site. The formation and hydrogenation of the OOH* intermediate are generally considered to be the rate-limiting steps to the 2e− ORR52. The ΔG values of OOH* formation on BCN-FeOOH and CoOx-BCN-FeOOH are remarkably reduced to −1.57 eV and −1.91 eV, whereas the value of OOH* formation on BCN is only −0.77 eV, indicating that OOH* formation is more favourable on FeOOH-modified BCN. The lowest ΔG value of the protonation steps for CoOx-BCN-FeOOH confirms that the anchored unsaturated coordinated FeOOH and CoOx clusters accelerate the reaction dynamics of the 2e− ORR.

Discussion

On the basis of the above results and analysis, a plausible mechanism involving charge transfer and redox reactions for H2O2 production over the CoOx-BCN-FeOOH catalyst is proposed (Fig. 5k). Upon light irradiation, the photogenerated holes of BCN are extracted by the anchored CoOx clusters and then accelerate water oxidation to produce O2. In parallel, O2 adapts to coordination-unsaturated FeOOH with a Pauling-type adsorption configuration, and energetic electrons with greatly prolonged lifetimes initiate the one-step 2e− ORR to facilitate H2O2 photosynthesis. Apparently, the synergy of the long lifetime of photogenerated electrons and preferable oxygen activation contributes to the commendable H2O2 production on the well-designed CoOx-BCN-FeOOH.

In summary, we have presented a strategy involving the direct one-step 2e− ORR for efficient H2O2 production without any sacrificial agents. Coordination-unsaturated FeOOH and CoOx clusters have been anchored precisely on BCN, and the optimized CoOx-BCN-FeOOH photocatalyst delivers an impressive H2O2 production rate of 0.34 mmol g−1 h−1 under visible light irradiation, with an SCC efficiency of 0.75%. An electron transfer efficiency of 34.1% for the ORR is achieved on CoOx-BCN-FeOOH, as revealed by the in situ TAS method. The experimental and theoretical results confirmed that the superior activity of the catalyst towards H2O2 production is due to the synergy of the introduced FeOOH for oxygen activation and the modified CoOx clusters for long electron lifetime. Importantly, the coordination-unsaturated structure of FeOOH allows O2 to adapt to the Pauling-type adsorption configuration, further stabilizing the peroxide species and limiting the formation of superoxide radicals, which contributes to the high selectivity of the direct one-step 2e− ORR pathway. This study provides a dual-active-site engineering strategy to arrange electron lifetime and oxygen activation of a single-component catalyst for H2O2 production with high efficiency.

Methods

Synthesis of boron doped g-C3N4 (BCN)

BCN was synthesized via a modified method based on a previous report9, which simply calcined the mixture of NaBH4 and g-C3N4 (CN) in an N2 atmosphere. Fabrication of ultrathin CN is the first step. Briefly, 35 g of urea was placed into an alumina crucible with a lid, and followed by calcining in air at 550 °C for 3 h with a rate of 0.5 °C min −1. Then, 0.4 g of the as-prepared CN and 0.24 g of NaBH4 were ground finely and then calcined at 500 °C in a N2 atmosphere for 2 h with a ramping rate of 10 °C min−1. The obtained powder was denoted as BCN.

Synthesis of FeOOH modified BCN (BCN-FeOOH)

In a typical synthesis, the as-prepared BCN (0.2 g) was dispersed in 50 mL of deionized water, then a certain amount of Fe(NO3)3 solution was added dropwise with vigorous stirring in ice bath for 4 h. Thereafter, the samples were collected and washed with water, and followed by a freeze-drying treatment. The obtained powder was denoted as BCN-xFeOOH, where x% represented the feeding mass percent of Fe dosage on the BCN. BCN-1FeOOH was denoted as BCN-FeOOH for short unless specified otherwise.

Synthesis of CoOx and FeOOH co-modified BCN (yCoOx-BCN-FeOOH)

A two-step synthesis was employed to fabricate CoOx-BCN-FeOOH. Typically, BCN-FeOOH was synthesized firstly as the method described above. Then, CoOx was modified on BCN-FeOOH by an in-situ photo-assisted oxidation strategy26. In detail, a certain dosage of Co2(CO)8 was dissolved in a mixture of 10 mL ethanol and 2 mL hexane to achieve the stablization. 0.2 g of BCN-FeOOH was dispersed in 20 mL of aqueous ethanol with continuous stirring under Ar protection at 40 °C for 30 min. Slowly Co2(CO)8 solution was introduced into the above suspension with consecutive stirring at 40 °C for 2 h to allow the Co source fully adsorbed on the catalyst surface. Thereafter, the Co4(CO)12 dimers would be obtained once the Co2(CO)8 is sublimate. Subsequently, the reaction system was irradiated with a Xenon lamp equipped with a filter (λ ≥ 420 nm) for 2 h in air, during which the active Co4(CO)12 could be decomposed to Co and CO, and the Co atom further oxidized into CoOx by oxygen with the assistance of photogenerated holes. Finally, the obtained samples were washed with ethanol to remove the dissociative Co source or weakly attached on BCN-FeOOH surface ones and then dried at 60 °C in an oven. The obtained sample was named as yCoOx-BCN-FeOOH, where y% represented the feed mass percent of Co dosage on the BCN. 0.75CoOx-BCN-FeOOH was denoted as CoOx-BCN-FeOOH for short unless specified otherwise.

Characterizations

Powder XRD patterns were collected with Bruker D8 Advance diffractometer, with CuKa radiation. UV-Vis DRS were taken on Model Shimadzu UV-2700 spectrophotometer with BaSO4 as the reference. AFM images were taken on a multimode nanoscope VIII instrument (Bruker) using mica as the base. TEM images were taken on an FEI Tecnai G2 S-Twin instrument operated at 200 kV. The HAADF-STEM images were obtained on a FEI Titan 60–300 electron microscope equipped with a spherical aberration corrector. XPS analyses of the catalysts were performed on a Kratos-AXIS ULTRA DLD instrument (Al (Mono) as the X-ray source). In-situ irradiated XPS analyses are identical the conventional one but equipped with a 405 nm monochromatic light as the illumination source. In-situ irradiated XPS analyses were conducted under a 405 nm monochromatic light illumination. The binding energies for the investigated sample were recorded in dark and under light irradiation, respectively. An inductively coupled plasma atomic emission spectroscopy (ICP-AES, AVIO 200, PerkinElmer) was used to determine the contents of iron and cobalt. Nitrogen adsorption-desorption isotherms were taken on a Quanta chrome-automated gas sorption analyzer (Autosorb-iQ-C-AG-TCD). The Fourier-transform infrared (FT-IR) spectra were collected with a Thermo Scientific Nicolet iS50 FT-IR spectrometer, KBr as the diluents. 11B solid-state MAS NMR was recorded on a Bruker Avance III HD 600 MHz WB. EPR measurements were taken on a Bruker EMX plus model spectrometer. PL spectra were carried out on a spectrofluorophotometer, LS55 Perkin-Elmer, with an excitation wavelength of 332 nm. Time-resolved PL spectra were recorded with a single photon counting spectrometer from (Edinburgh Instrument, FLS 920) with 1μs pulse lamp as the excitation.

Photocatalytic activities for H2O2 production

20 mg of photocatalyst was dispersed in 20 mL of pure water under ultrasonic treatment. Then, O2 was continuously introduced into the solution to reach the saturation. Thirty minutes later, the reactor was exposured to a 300 W Xe lamp equipped with a 420 nm cut-off filter. The light intensity was settled as 100 mW cm−2, and the reaction temperature was controlled to be 25 °C by a stirrer and circulating water. The H2O2 yields was quantified during cetain intervals. The reaction mixture (1.5 mL) was extract from the reactor every 20 min, followed by filtering the photocatalyst. The amounts of produced H2O2 was determined by the iodometry, the details could be found in supplementary Information.