Spatially separating redox centers on 2D carbon nitride with cobalt single atom for photocatalytic H₂O₂ production

Significance

Photocatalysts frequently require simultaneous loading of oxidative and reductive cocatalysts to achieve both efficient half-reactions within a single material. Nevertheless, unregulated loading and distribution of two cocatalysts will result in direct contact between oxidation and reduction centers, leading to detrimental charge recombination. This research presents a center/edge approach to load two redox cocatalysts with controlled physical separation in atomistic scale using single-atom architecture. This spatial separation is critical for enhancing surface charge separation and achieving efficient H₂O₂ production. We report that redox cocatalysts are spatially separated on a two-dimensional (2D) photocatalyst, which opens an approach for achieving both efficient oxidation and reduction reactions on 2D photocatalysts.

Abstract

Redox cocatalysts play crucial roles in photosynthetic reactions, yet simultaneous loading of oxidative and reductive cocatalysts often leads to enhanced charge recombination that is detrimental to photosynthesis. This study introduces an approach to simultaneously load two redox cocatalysts, atomically dispersed cobalt for improving oxidation activity and anthraquinone for improving reduction selectivity, onto graphitic carbon nitride (C₃N₄) nanosheets for photocatalytic H₂O₂ production. Spatial separation of oxidative and reductive cocatalysts was achieved on a two-dimensional (2D) photocatalyst, by coordinating cobalt single atom above the void center of C₃N₄ and anchoring anthraquinone at the edges of C₃N₄ nanosheets. Such spatial separation, experimentally confirmed and computationally simulated, was found to be critical for enhancing surface charge separation and achieving efficient H₂O₂ production. This center/edge strategy for spatial separation of cocatalysts may be applied on other 2D photocatalysts that are increasingly studied in photosynthetic reactions.

Harvesting solar photon energy to drive redox reactions involving water and oxygen is the most espoused strategy for the green synthesis of alternative fuels such as H₂ and H₂O₂ (1–4). Yet, solar-to-energy conversion efficiencies achieved using current semiconductor photocatalysts remain relatively low (5, 6), due to inherent limitations in material properties such as prevalent charge recombination in low-bandgap materials and the insufficient selectivity toward the fuel synthesis reaction (7). One promising material engineering strategy is to decorate the semiconductor surface with cocatalysts (1, 8), ideally both reductive and oxidative cocatalysts within a single photocatalytic material. Nevertheless, randomly loading two cocatalysts often results in direct contact between oxidation and reduction centers, worsening the charge recombination that is detrimental to photosynthetic reactions (Fig. 1A) (9).

Fig. 1.

(A) Randomly loading two cocatalysts leads to detrimental sequences of reactions involving oxidant (Ox) and reductant (Red). (B) Core/shell structured photocatalysts and (C) photocatalysts with different exposed crystalline facets to achieve controlled spatial separation of oxidative and reductive cocatalysts. (D) Spatial separation of Co single atom (as oxidation center) and AQ (as reduction center) cocatalysts by anchoring them in the center (i.e., pyridinic N) and on the edge (i.e., primary/secondary amine N) of 2D ultrathin C₃N₄, respectively.

Placing two cocatalysts without direct contact requires sophisticated material architecture and synthesis strategy. One cocatalyst, typically oxidative, can be loaded on a substrate in trace amounts to minimize such contact but only at the expense of the available catalytic sites and thus the overall efficiency (10, 11). A more promising strategy is to design the substrate photocatalysts to provide physically separated sites for cocatalyst hosting. For instance, Wang et al. (9) recently fabricated a core/shell photocatalyst that can host reductive and oxidative cocatalysts separated inside and outside of the shell surfaces (Fig. 1B). Following this seminal work, various core/shell structures have been prepared, which typically require complicated synthesis procedures involving the use of sacrificial templates (e.g., SiO₂) (12–15). Alternatively, different facets of photocatalytic materials were found to selectively load different cocatalysts, allowing spatial segregation (Fig. 1C) (16–18).

The existing strategies to prepare spatially separated cocatalysts, however, exclusively rely on the three-dimensional nature of the substrate structure and cannot be readily extended to 2D materials such as graphitic carbon nitride (C₃N₄). C₃N₄ has often been used as the semiconductor material of choice for the photocatalytic synthesis of H₂O₂ (3, 19, 20), an emerging substitute for compressed H₂ due to recent advances in H₂O₂ fuel-cell technology (21). C₃N₄ exhibits valence-band (VB) and conduction-band (CB) potentials that span those of H₂O/O₂ and H₂O₂/O₂ redox pairs and is capable of harnessing broad spectrum of sunlight due to its low-bandgap energy. However, solar-to-fuel conversion efficiencies remain, in general, relatively low due to limitations that are commonly found in other materials: 1) ineffective hole scavenging via water oxidation and the resulting charge recombination (3, 19), which often necessitates the addition of organic electron donors (22–25), and 2) low selectivity toward H₂O₂ synthesis via two-electron reduction of O₂ (O₂ + 2H⁺ + 2e⁻ → H₂O₂) as compared to four-electron reduction of O₂ (O₂ + 4H⁺ + 4e⁻ → 2H₂O) or two-electron H₂ evolution (2H⁺ + 2e⁻ → H₂) (3, 19).

Here we introduce an innovative strategy to load two cocatalysts onto 2D C₃N₄, with controlled physical separation in atomistic scale (Fig. 1D). We use cobalt and anthraquinone (AQ) as cocatalysts that are crucial for efficient photocatalytic synthesis of H₂O₂. Co is anchored to void center of the C₃N₄ as a single atom (Co₁) and serves to facilitate the water oxidation (26–30). At the same time, AQ is attached to amine anchors that are present only on the edge of C₃N₄, ensuring that it is not in direct contact with the Co centers. The AQ enhances the selectivity of O₂ reduction to H₂O₂, following the mechanism widely exploited in current industrial H₂O₂ production process (31). The composite catalyst, Co₁/AQ/C₃N₄, photocatalytically produces H₂O₂ at high efficiency under simulated solar irradiation without the supply of a sacrificial agent.

We first prepared ultrathin C₃N₄ nanosheets by exfoliating bulk C₃N₄ under probe sonication (32). The C₃N₄ nanosheets appeared to be only a few layers thick according to high-resolution transmission electron microscopy (HRTEM) images (Fig. 2A). We then loaded Co onto the ultrathin C₃N₄ using a two-step synthesis: attachment of Co precursors to anchor sites followed by pyrolysis (33). Co ions are attached to the void center of C₃N₄ nanosheets through forming stable coordination with pyridinic N atoms in surrounding heptazine units of C₃N₄ (Fig. 1D), as suggested by this site having the lowest relative energy (34, 35). After pyrolysis under N₂ atmosphere, Co ions were further phosphodized under PH₃ atmosphere to enhance their activity for water oxidation (26, 33, 36).

Fig. 2.

(A and B) HRTEM and EDS images of Co₁/AQ/C₃N₄. (C) Photooxidative deposition of Mn on Co₁/C₃N₄. (D and E) HAADF-STEM image of C₃N₄ and Co₁/AQ/C₃N₄. (F) FT-IR spectra of C₃N₄ and Co₁/AQ/C₃N₄. (G) Photoreductive deposition of Au on AQ/C₃N₄.

Energy-dispersive X-ray spectroscopy (EDS) elemental mapping suggests that Co is uniformly distributed across the C₃N₄ surface (Fig. 2B). To further provide a visual confirmation, we photooxidatively deposited Mn²⁺ to grow MnO_x nanoparticles (Mn²⁺ + xH₂O + [2x-2]h⁺ → MnO_x + 2xH⁺) on Co as seed sites (16, 37, 38). The formation of MnO_x across the C₃N₄ surface readily observed by low-resolution TEM (Fig. 2C) suggests that Co atoms are also distributed across the C₃N₄ surface and serve as oxidation centers. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images indicates that the Co is likely atomically dispersed [Fig. 2E as compared to C₃N₄ before Co loading (Fig. 2D)], since the radii of Co were estimated to be ∼0.5 Å. Notably, all Co single atoms identified were located at least ∼1 nm away from the edge of C₃N₄ (Fig. 2E and SI Appendix, Fig. S2), indicating the selective loading of Co on the surface, not the edge, of C₃N₄ nanosheet.

Consistent with the absence of Co metallic clusters in HAADF-STEM images, another strong piece of evidence for the atomic dispersion of Co is provided by the absence of Co-Co coordination in K-edge spectrum from Fourier-transformed extended X-ray absorption fine-structure spectroscopy (FT-EXAFS, Fig. 3A). The spectrum also indicates that Co atoms are primarily coordinated by P (i.e., peak at 1.8 Å in FT-EXAFS) (33), which confirms the complete phosphidation. The coordination with P is further supported by the occurrence of a prominent Co-P peak at 129.6 eV in the X-ray photoelectron spectroscopy (XPS) spectrum (Fig. 3D). A P-N peak at 133.6 eV also suggests that P atoms coordinate with N atoms in heptazine rings of C₃N₄. Best-fit parameters extracted from the FT-EXAFS spectra (Fig. 3B) suggest an average Co-P coordination distance at 2.29 Å and coordination number of 4.1 (SI Appendix, Table S1), consistent with previous observations (33). Co atoms are found to be positively charged with partially unoccupied 3d orbitals. Comparison of the Co K-edge normalized near-edge X-ray absorption spectroscopy (XANES) of Co₁/AQ/C₃N₄ with those of reference compounds shows that the spectral line shape and the absorption edge position closely resemble those of CoO (Fig. 3C), indicating that the oxidation state of the Co single atoms is close to +2. Density-functional theory calculations (DFT; see SI Appendix, section S6 for details) confirm that Co atoms are positively charged. The above results collectively suggest that Co cocatalysts are uniformly loaded in the center of C₃N₄ nanosheet as positively charged single atoms.

Fig. 3.

(A) FT EXAFS spectra of Co₁/AQ/C₃N₄ at the Co K edge. The intensity of Co₃O₄ and Co₁/AQ/C₃N₄ was normalized to the same maximum of Co foil to facilitate the comparison of radical distances by multiplying with a factor of 2.1 and 3.6, respectively. (B) Fit of Co₁/AQ/C₃N₄ EXAFS spectra using Co foil and Co₃O₄. (Inset) Corresponding K-space curves. (C) Normalized XANES of Co₁/AQ/C₃N₄ at the Co K edge. (D) Binding energy of N 1s, C 1s, P 2p, and O 1s for ultrathin C₃N₄ and Co₁/AQ/C₃N₄ by high-resolution XPS.

XAFS measurements at the P K-edge further provide a clue on the structure of Co single atom and its surrounding. EXAFS spectrum at the P K-edge indicates that P atoms are primarily coordinated by Co atoms (i.e., peak at 1.87 Å in FT-EXAFS; SI Appendix, Fig. S3A), which is consistent with the corresponding FT-EXAFS data recorded at Co K-edge (Fig. 3B). These P atoms are further coordinated with N atoms in heptazine rings of C₃N₄ as well as O atoms, as evidenced by P-N peak at 133.5 eV and P-O peak at 134.5 eV in XPS spectrum (Fig. 3D). XANES spectrum at the P K-edge shows that the preedge region is dominated by the strong feature at 2,143.6 eV, which is assigned to low-valence phosphidic species (SI Appendix, Fig. S3B) (39). The higher-energy, broader peak centered around 2,152.0 eV is consistent with high-valence P (SI Appendix, Fig. S3B) (40). Comparison of intensities of the two maxima suggests that 20–30% of the phosphorus atoms exist in high-valence state. Alternatively speaking, Co atoms are coordinated with three low-valence P atoms (i.e., coordinated with N atoms in heptazine rings of C₃N₄) and one high-valence P atom (i.e., coordinated with O atoms). Geometry optimization conducted under these constraints using DFT confirms that the proposed Co center structure is stable (Fig. 1D, top and side view of Co center), in which Co is placed out of C₃N₄ plane. Other configurations, particularly in-plane P substitutions and Co insertions, resulted in sheet disintegration or massive structural rearrangement.

Loading Co single atoms significantly enhanced C₃N₄ for water oxidation (2H₂O → O₂ + 4H⁺ + 4e⁻), as indicated by an 8.4-fold enhancement on 4-h O₂ production (Fig. 4A and see SI Appendix, section S5 for details). According to our DFT calculations (see SI Appendix, section S6 for details), the enhanced water oxidation is attributed to strong adsorption of water molecule on Co single atom (exothermic by 2.7 eV when replacing the phosphite moiety), while no adsorption is observed on the plane C₃N₄ sheet. In addition, atomically dispersed Co produces two distinct, occupied, midgap states ∼0.5 and 0.9 eV above the VB maximum when the H₂O is absorbed, promoting the localization of photoexcited holes and subsequent charge separation (Fig. 5 A and B), whereas Co nanoparticles completely fill the bandgap and thus act as charge recombination centers (Fig. 5C). All of these results indicate that Co loading enhances the hole quenching by water and therefore the overall charge-separation, i.e., more electrons are available for the reductive H₂O₂ synthesis.

Fig. 4.

(A) Time course of O₂ evolution measured under 0.6-kPa Ar pressure and 300-W xenon lamp irradiation with 0.5 g/L of catalyst, 1 g/L La₂O₃, and 20 mM AgNO₃ in 100 mL water. (B) Selectivity of H₂O₂ production. (C and D) Time course of H₂O₂ production measured under simulated sunlight irradiation (xenon lamp solar simulator, 100 mW/cm², AM 1.5G) with 0.5 g/L of catalyst under O₂-saturated condition. Solid lines are the fitting result of the kinetic model. Dotted lines are H₂O₂ productions estimated assuming additive enhancement of each cocatalyst. (E and F) H₂O₂ formation and decomposition rate constants. Error bars represent the SDs of triplicates.

Fig. 5.

Density of states computed with DFT for (A) C₃N₄, (B) C₃N₄ loaded with Co single atom cocatalyst (SAC), (C) C₃N₄ loaded with Co nanoparticle (showing Co₄ as an example), and (D) Co loaded with AQ.

Secondly, we loaded AQ cocatalyst onto Co₁/C₃N₄ by forming amide bonds between carboxylic groups in anthraquinone-2-carboxylic acid and primary/secondary amine groups on the edge of C₃N₄ (Fig. 1D) (8, 35). Successful loading of AQ was confirmed by XPS in which Co₁/AQ/C₃N₄ exhibits strong peak corresponding to C-C fragments (284.7 eV) that mostly originate from AQ molecules (Fig. 3D). The AQ molecules remained bound to C₃N₄ after intensive solvent washing, suggesting that they are chemically attached rather than physically adsorbed (23). The successful loading of AQ was also confirmed by Fourier-transform infrared spectroscopy (FT-IR) spectroscopy. As shown in Fig. 2F, the intensities of the FT-IR peaks corresponding to the amide functionalities, including the C=O stretching vibration peak at 1627 cm⁻¹ and the N-H stretching vibration peak at 3,076 cm⁻¹, increased dramatically with AQ loading. The quantitative analysis of XPS spectra indicates that AQ was loaded at 16% (wt/wt).

To provide a visual confirmation of the site-selective loading of AQ, we photoreductively deposited noble metals by reducing metal precursors (i.e., H₂AuCl₄ or H₂PtCl₆) on AQ as seed sites (Mⁿ⁺ + ne⁻ → M⁰) (16). TEM images clearly showed that the Au and Pt nanoparticles were selectively deposited on the edge of C₃N₄ nanosheets (Fig. 2G and SI Appendix, Figs. S4 and S5), which were in stark contrast to random deposition of Au nanoparticles on pristine C₃N₄ surface without AQ functionality (SI Appendix, Fig. S6) (41). These results confirm that AQ cocatalysts were selectively loaded on the edge of C₃N₄ nanosheets and serve as reduction center. DFT calculations confirmed the electron withdrawal by the AQ cocatalyst, where AQ molecule generates an empty state that is only 0.3 eV below the CB of C₃N₄; while filled AQ states, where a hole would occupy, sits more than 0.8 eV below the VB of C₃N₄ (Fig. 5D). Therefore, transfer of a photoexcited electron to AQ is allowed but transfer of a photoexcited hole is prohibited, leading to enhanced charge separation.

Loading AQ cocatalyst onto C₃N₄ had a significant impact on enhancing the selectivity of H₂O₂ synthesis from ∼30% by pristine C₃N₄ to over 60% (Fig. 4B); H₂O₂ production selectivity is defined as the ratio of electrons utilized for H₂O₂ synthesis to the total number of electrons consumed (SI Appendix, section S3 and ref. 22). In contrast, C₃N₄ exfoliation or Co loading had limited impact on H₂O₂ production selectivity (Fig. 4B). The enhanced H₂O₂ production selectivity is attributed to the two-step reaction catalyzed by AQ: 1) reductive hydrogenation of AQ to hydroxyanthraquinone (AQH₂) utilizing 2 e⁻ from photoexcited C₃N₄ followed by 2) H₂O₂ formation from concurrent oxygen reduction and dehydrogenation of AQH₂ back to AQ (SI Appendix, Fig. S7).

The photocatalytic H₂O₂-production performance of the as-prepared catalysts was evaluated under simulated sunlight irradiation in the absence of organic electron donor. Exfoliation of bulk C₃N₄ to nanosheets enhanced the photocatalytic H₂O₂ production performance of C₃N₄ (Fig. 4C) due to a larger number of exposed reaction sites and improved light-harvesting capability (8, 32). The light-harvesting capability of C₃N₄ was further improved with Co loading, as indicated by the lowered bandgaps [refer to the band-structure diagram (SI Appendix, Fig. S8)] constructed from XPS valence spectra (SI Appendix, Fig. S9) and diffuse reflectance spectra (SI Appendix, Fig. S10). Co cocatalyst loading promoted the water oxidation reaction and consequentially reduced detrimental exciton recombination (26, 33, 34, 42), leading to enhanced H₂O₂ production (Fig. 4C). For instance, when Co was loaded as nanoparticles (see SI Appendix, section S2 for synthesis details), i.e., not as single atoms, H₂O₂ production was enhanced, albeit slightly; when Co was loaded as single atoms, H₂O₂ production was enhanced by 4.0-fold. On the other hand, loading AQ cocatalyst onto ultrathin C₃N₄ improved H₂O₂ production selectivity, resulting in a 1.9-fold enhancement in H₂O₂ production (Fig. 4C).

Simultaneous loading of Co single atoms and AQ cocatalyst significantly enhanced H₂O₂ production by a factor of 7.3 (Fig. 4D). For the solar photocatalytic H₂O₂ production performed in the absence of electron donor, the initial production rate of 62 µM/h (apparent quantum efficiency = 0.054% over the full spectrum of sunlight; see SI Appendix, sections S8 and S9 for calculation details) and the cumulative production of 230 µM over 8-h period (SI Appendix, Fig. S11) achieved by Co₁/AQ/C₃N₄ in this study are among the highest reported (SI Appendix, Table S2) (43, 44). When the suspension was N₂-purged, the H₂O₂ production was mostly inhibited, confirming that O₂ reduction was the major pathway for H₂O₂ production (SI Appendix, Fig. S12). The stability of Co₁/AQ/C₃N₄ was demonstrated by its stable catalytic performance through repetitive use up to five cycles (SI Appendix, Fig. S13). XPS (SI Appendix, Fig. S14) and TEM (SI Appendix, Fig. S15) analyses show no significant change in chemical composition or ultrathin layered structure of Co₁/AQ/C₃N₄ after 8-h irradiation. We note that a better intersheet packing control may be achieved by immobilizing Co₁/AQ/C₃N₄ to facilitate its application in large-scale photolysis device setup (43, 44).

The enhancement on H₂O₂ production by coloading of Co single atoms and AQ is close to the multiplication of individual enhancements (Fig. 4D); i.e., the 4.0-fold enhancement by Co single atom cocatalyst times the 1.9-fold enhancement by AQ cocatalyst is close to the observed 7.3-fold enhancement on 60-min H₂O₂ production. This collaborative effect suggests that two cocatalysts contribute to H₂O₂ production enhancement independently without any negating effect. In contrast, when Co was loaded as nanoparticles onto AQ-C₃N₄, H₂O₂ production was much lower than that expected by considering the individual effects of Co nanoparticles and AQ (i.e., the observed 1.2-fold enhancement is much lower than the 1.5-fold enhancement by Co nanoparticle times the 1.9-fold enhancement by AQ on 60-min H₂O₂ production; Fig. 4D). This is due to random distribution of large Co particles across C₃N₄ and likely contact between Co cocatalyst and AQ cocatalyst. Such a contact would facilitate direct electron transfer without H₂O₂ evolution or water splitting (Fig. 1A). Loading Co as single atom, in contrast, ensures that Co is spatially separated from AQ. Since Co atom occupies the void center of C₃N₄ and AQ is located only at the edge of C₃N₄, molecular structure of C₃N₄ inherently maintains their separations at the minimum of 0.8 nm (SI Appendix, Fig. S16). Consistently, the steady-state photoluminescence from Co₁/AQ/C₃N₄ was markedly reduced compared not only to C₃N₄ but also to Co_nano/AQ/C₃N₄, indicating that the radiative recombination is more effectively retarded with Co single atom than Co nanoparticle (Fig. 6A). Time-resolved photoluminescence (TRPL, Fig. 6B) analyses show that the lifetime of excited-state Co₁/AQ/C₃N₄ (1.73 ns) is shorter than that of C₃N₄ (3.23 ns) and Co-_Nano/AQ/C₃N₄ (2.23 ns), highlighting the improved photoexciton dissociation in Co₁/AQ/C₃N₄ that was realized by spatial separation of AQ and Co single atoms.

Fig. 6.

(A) Steady-state PL emission spectra (excitation at 375 nm) of C₃N₄, Co_nano/AQ/C₃N₄, and Co₁/AQ/C₃N₄. (B) TRPL spectra monitored for entire emission between 400 and 800 nm. The curves were fitted to the equation y = y₀ + A₁ exp(−t/τ₁) + A₂ exp(−t/τ₂) + A₃ exp(−t/τ₃).

We further analyze the H₂O₂ production by evaluating the rate of H₂O₂ formation (k_f) separately from the rate of H₂O₂ decomposition (k_d) (see SI Appendix, section S7 for the kinetic analysis). The results show that H₂O₂ formation rate constant increased upon individual loading of Co single atom, Co nanoparticle, or AQ. While simultaneous loading of Co single atom and AQ lead to additive enhancement on k_f, simultaneous loading of Co nanoparticle and AQ had an antagonistic effect on k_f (Fig. 4E), once again highlighting the importance of controlled physical separation between Co and AQ. It is also noteworthy that Co may negatively impact H₂O₂ synthesis performance by enhancing the oxidative decomposition of H₂O₂ (Fig. 4F). This catalyzed H₂O₂ decomposition was minimized by separating H₂O₂ production centers (i.e., AQ) from Co decomposition sites, as indicated by much lower k_d in Co₁/AQ/C₃N₄ system as compared to Co₁/C₃N₄ system (Fig. 4F).

Results of our study suggest a facile strategy to anchor two spatially separated cocatalysts on a 2D photocatalyst. Such spatial separation ensures that the functions of both cocatalysts (i.e., Co₁ for enhanced water oxidation activity and AQ for improved H₂O₂ production selectivity) are fully utilized, resulting in additive enhancement in H₂O₂ photosynthesis. Here, atomic dispersion of metal cocatalyst presents advantage over conventional nanoparticles because the small size and strong ligand–metal coordination of single atom allow for facile manipulation of loading sites. The stark contrast on the performance of Co single atoms versus nanoparticles emphasizes the exclusive benefits of single atom catalysts in this material design. This center/edge strategy for loading two spatially separated cocatalysts may be also applicable on other 2D photocatalysts for achieving efficient charge separation while maintaining the effectiveness of both cocatalysts.

Methods

Preparation of Photocatalysts.

Bulk C₃N₄ was prepared following a thermal polymerization procedure by heating melamine powder in a ceramic crucible at a heating rate of 1 °C/min to 550 °C and annealing for 5 h in a muffle furnace. As-prepared bulk C₃N₄ was grounded, exfoliated under probe sonication for 8 h, separated by centrifugation, washed with deionized water, and dried at 80 °C overnight. As-prepared ultrathin C₃N₄ (160 mg) was dispersed in 50 mL water under ultrasonication for 30 min, followed by addition of 1.5 mL Co(NO₃)₂ solution (2 g/L). The mixture was stirred and heated at 70 °C for 18 h, separated by centrifugation, dried at 80 °C overnight, and annealed at 400 °C for 2 h in a tube furnace under N₂ gas. The obtained powder was grounded, mixed with NaPO₂•H₂O (twice the weight of obtained powder), and heated at 300 °C for 2 h in a tube furnace under N₂ gas. As-prepared Co₁/C₃N₄ was washed with water and ethanol, and dried at 80 °C overnight. The Co loading amount was determined to be 0.13% (wt/wt; 0.34% of voids occupied) by inductively coupled plasma mass spectrometry (ICP-MS, PerkinElmer SCIEX Elan DRC-e) analysis after acid digestion. As-prepared Co₁/C₃N₄ (100 mg) was mixed with 10 mg anthraquinone-2-carboxylic acid, 7.7 mg diisopropylethylamine, 8.1 mg 1-hydroxybenzotriazole hydrate, and 11.5 mg N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride. The mixture was dispersed in 50 mL dichloromethane under ultrasonication for 5 min and stirred for 48 h. The Co₁/AQ/C₃N₄ product was separated by centrifugation, washed with dichloromethane and water, and dried at 80 °C overnight. Co and AQ were loaded at a molar ratio of 1:37.

Catalyst Characterizations.

HAADF-STEM images were taken using a Titan Themis Z STEM (ThermoFisher Scientific) operated at 200 kV, coupled with a probe aberration corrector to improve imaging spatial resolution to less than 1 Å. XPS measurements were performed with a Versa Probe II Scanning XPS Microprobe [Physical Electronics (PHI)]. We performed a survey analysis across the entire energy range and higher-resolution analyses in the N 1s, C 1s, P 2p, and O 1s regions. The XAS spectra at Co K edge were measured at Beamline 8-ID (ISS) of the National Synchrotron Light Source II (NSLS-II) at Brookhaven National Laboratory, using a Si (111) double-crystal monochromator and a passivated implanted planar silicon fluorescence detector. XANES data were collected at room temperature, with energy calibrated using Co foil. The P K-edge XAS data were collected at 8-BM of NSLS-II under fluorescence mode, employing a Si (111) crystals monochromator (45). Each sample was scanned for seven times to get better signal to noise ratio. Photoluminescence (PL) emission spectra were obtained using a fluorescence spectrophotometer (Shimadzu RF-5301). PL lifetime decays were measured by an inverted-type scanning confocal microscope (MicroTime-200, Picoquant) with a 20× objective. A single-mode pulsed diode laser (excitation wavelength at 375 nm with an instrumental response function of ∼240 ps in pulse width) was used to as an excitation source to excite the samples coated on glass substrate. A dichroic mirror (490 DCXR, AHF), a long-pass filter (HQ500lp, AHF), and an avalanche photodiode detector (PDM series, MPD) were used to collect the entire emission (400–800 nm) from the samples. Photon counting and exponential fitting of the obtained PL decays were performed using the SymPhoTime software (version 5.3).

Photocatalytic Activity Tests.

Photocatalytic production of H₂O₂ was assessed by irradiation of photocatalyst suspension (12 mL, 0.5 g/L) using a xenon lamp solar simulator (model 10500; Abet Technologies, Inc.). The light intensity was adjusted to 100 mW/cm² (AM 1.5G; irradiation area = 1.77 cm²). The suspension was purged with O₂ before (for 5 min) and during irradiation. At designated time points, small aliquots from suspensions were taken for analysis of H₂O₂ productions.

Data Availability.

All data of this study are included in the text and SI Appendix.