Discussion of experimental characterization. NDCN was successfully prepared by adding NaOH to the precursor of CN (see Methods for details).The chemical structure of CN and NDCN were first characterized by X-ray diffraction (XRD) patterns. The XRD pattern of CN (Fig. 1a, top line) showed two distinct diffraction peaks at 13.0° and 27.4°, which can be assigned to the in-plane packed heptazine units (100) and interfacial stacking (002) of graphitic CN, respectively27. It can be observed that the two characteristic peaks of NDCN were weakened and broadened, implying the loss of ordered structures within the framework to some extent. The Fourier transform infrared spectroscopy (FTIR) spectrum for CN (Fig. 1b) showed that the peak at 810 cm− 1 can be assigned to the out-of-plane bending mode of heptazine rings, and the peaks in the range of 900–1800 cm− 1 refers to N-C = N heterorings in CN framework. Multiple broad peaks located in the region of 3000–3500 cm− 1 can be attributed to N-H stretching vibrations. After adding NaOH, the peak at 2156 cm− 1 typical for the cyano groups (–C ≡ N) was emerged28. Meanwhile, the intensity of the N-H stretching peaks was significantly reduced. The results indicate that the addition of NaOH leads to the decrease of N-H groups and introduction of cyano groups.
Organic elemental analysis (OEA) was used to confirm the changes in bulk elemental composition of the samples. The N/C atomic ratio of NDCN was slightly reduced in compartion to that of CN (Supplementary Fig. 2). Next, the changes of surface element composition were further studied by XPS. Compared with that of bulk, the addition of NaOH made the surface N/C atomic ratio decreased more significantly (Supplementary Fig. 2). It is worth noting that the introduction of cyano groups and reduction of N-H groups does not cause the reduction of N elements. Therefore, this indicates that the treatment of NaOH resulted to the generation of surface N defects. Meanwhile, the O/C atomic ratio in NDCN increased (Supplementary Fig. 1a), which may be attributed to the increase of oxygen-containing species adsorbed on the surface29. On the other hand, valence band (VB) XPS spectra (Supplementary Fig. 1b) showed that the treatment of NaOH did not affect energy level position of VB. In addition, the morphology of NDCN had not changed significantly compared to that of CN (Supplementary Fig. 3).
Narrow scan C1s and N1s XPS spectra further proved the introductions of cyano group and N defects. For the C1s XPS spectra (Fig. 1c), it contained three peaks of 287.89, 286.23, and 284.5 eV, representing N-C = N, C-NHx (x = 1, 2) and adventitious hydrocarbons, respectively. The corresponding peak of C-NHx of NDCN was higher than that of CN, which proves the generation of C ≡ N, because the cyano group has similar binding energy of C1s30. For the N1s XPS spectra (Fig. 1d), the intensity of bicoordinated nitrogen (N2C) in NDCN was reduced in comparison to that of CN. Combined with the decrease of N/C in XPS (Supplementary Fig. 2), it can be inferred that the generation of nitrogen defect was caused by the absence of N atoms in N2C. The intensity of tricoordinated (N3C) increased and the peak shifted to lower binding energy. This can be attributed to the introduction of C ≡ N groups, because their binding energy lies between the binding energy of N3C and N2C30. Based on the above characterization, we can conclude that after NaOH treatment, N2C vacancies (NVs) were formed. Meanwhile, with the decrease of N-H groups, cyano groups were generated.
The optical properties of the samples were further characterized by DRS. Compared with CN, the absorption edge of NDCN has a significant red shift (Supplementary Fig. 4a) and the band gap is reduced from 2.64 to 2.04 eV (Supplementary Fig. 4b). Combining the above results of the VB XPS, the narrowing of the band gap is due to the decrease of conduction band (CB) position, while the VB position remains unchanged.
Evaluation of photocatalytic performance. The photocatalytic performances of CN and NDCN were evaluated. Under the irradiation of visible light, the 60 min production of H2O2 for NDCN was 476 µM (Supplementary Fig. 5a), which is 9.7 times higher than that of CN (47 µM) and better than many reported yields in recent years (Supplementary Table 1). In addition, after 5 cycle experiments, the photocatalytic performance did not decrease significantly (Supplementary Fig. 6). Comparing the XRD patterns of samples before and after 5 cycles, there was no obvious difference (Supplementary Fig. 7), indicating that NDCN has excellent stability in the photocatalytic H2O2 production. Therefore, it is meaningful to investigate the reaction mechanism of photocatalytic H2O2 production for NDCN.
The effects of defects on separation efficiency of photoexcited charge carriers were investigated. Photoluminescence (PL) spectroscopy measurements were performed to examine the charge separation ability31 of samples. As shown in Supplementary Fig. 5b, a intense PL signal was observed on CN, due to the serious recombinations of carriers. In contrast, the PL intensity of NDCN was weaker, indicating a signally suppressed electron-hole recombination. Supplementary Fig. 5c showed the transient photocurrent response curve of the samples, which is directly related to their charge separation ability. Compared with CN, NDCN exhibited a larger photocurrent density, indicating that nitrogen defects can greatly accelerate the separation of photoexcited carriers32. The electrical conductivity and interfacial charge transfer capability of the samples were evaluated by electrochemical impedance spectroscopy (EIS). As shown in Supplementary Fig. 5d, the EIS slop of NDCN decreased significantly in compartion to that of CN, implying that NDCN has better electrical conductivity, which is conducive to effective charge separation and transfer in the photocatalytic process.
Experimental research on reaction mechanism. To understand the reaction mechanism of photocatalytic reaction, a series of control experiments and measurements were conducted. Figure 2a showed the ESR spectra of DMPO-•O2− adduct for the CN and NDCN samples. It can be seen that the DMPO-•O2− adduct signal of NDCN was much stronger than that of CN, suggesting that NDCN can promote the generation of •O2− more effectively. The results of control experiment further illustrated the importance of •O2− in the photocatalytic reaction. After adding •O2− scavenger, p-benzoquinone (PBQ), the yield of H2O2 was 0 (orange curve in Fig. 2b). It is worth mentioning that the iodometry for determining H2O2 production do not be affected by PBQ33. Hence, it can be concluded that the consumption of •O2− can completely inhibit generation of H2O2, indicating that the generation of H2O2 is a two-step single electron transfer process (O2 → •O2− → H2O2)17,33. In the dark, the H2O2 production was also 0 (blue curve in Fig. 2b), implying that light irradiation is a necessary condition for the catalytic reaction. In the absence of isopropanol (IPA), the yield of H2O2 was significantly reduced (yellow curve in Fig. 2b), because IPA is the main provider of H source for H2O2 evolution. In the photocatalytic reaction, the holes oxidize IPA to generate two H+ (CH3CHOHCH3 + 2h+ → CH3COCH3 + 2H+), which can further combine with the reduced O2 to generate H2O2. Without IPA, holes can only react with water to produce H+, resulting in an excessively high energy barrier. Interestingly, when the reaction conditions changed from oxygen to air, the yield of H2O2 was only slightly reduced, implying that NDCN could still adsorb and reduce O2 well even in air, which will be further discussed below. Since the formation of H2O2 involves two-electron O2 reduction process, the average number of electrons (n) involved in the oxygen reduction is a vital indicator for the selectivity of reaction products. Figure 2c and Supplementary Fig. 8 showed the linear sweep voltammetry (LSV) curves of NDCN and CN on a rotating disk electrode at different rotate speeds, respectively. n is the slope of Koutecky-Levich plots based on the LSV curve9. As shown in Fig. 2d, the measured n value for CN was only 1.5, while that of NDCN reached 1.95, close to 2, indicating that NDCN is highly selective for the two-electron reduction of O2.
Through the above characterization and experiments, we have a basic understanding of the role of N defects in photocatalytic reaction and the path of O2 reduction. However, there are still some problems to be solved: (1) NDCN contains two MSs, N2C vacancy (NV) and cyano group. Which site is causing the enhanced photocatalytic performance? (2) Is there a synergy between the two MSs? (3) What is the essential reason for O2 activation? To solve the problems, DFT and TDDFT calculations were performed.
Confirmation of the optimal model. First, the most stable and reasonable geometry was constructed and verified. Eleven possible models (M1-M11 in Supplementary Fig. 9) were considered, and the most stable geometric structure of NDCN was selected by comparing their Gibbs free energy. As shown in Supplementary Fig. 10, model with the lowest Gibbs free energy is M3 whose cyano group and NV are located at the edge of the two melon frameworks, respectively. The calculated FTIR spectrum for M3 was similar to the experimentally measured spectrum (Fig. 3a), which again proved the rationality of M3. Meanwhile, the structure of M3 was also consistent with what Zhang et al. reported21. Thus, we will directly label the M3 as NDCN for discussion below. In order to clearly describe each site of NDCN, the NDCN model was divided into three regions: NV, cyano, and primitive melon. The key atoms were numbered to clearly identify each site (Fig. 3b).
Confirmation of •O 2 − and active site. In order to initially confirm the possible O2 activation sites, the calculated ultraviolet visible (UV-vis) spectrum and electron-hole distribution of each excited state were analyzed. The oscillator strength (f) in the UV-vis spectrum represents the transition probability of electrons from ground state to each excited state. The f of 0.01 is generally the critical point of transition24. Specifically, if the f is lower than 0.01, the corresponding transition is generally considered to be forbidden, otherwise the transition can occur with a probability determined by f. For NDCN, the f of D0→D1, D0→D2, and D0→D4 excitations were all greater than 0.01 (Fig. 4a), indicating that photo-generated e− can indeed be excited to the green regions of Fig. 4c1, 4c2 and 4c4, respectively, that is, these excitations are bright. The f of the D0→D3 excitations was almost zero (Fig. 4a), implying that the excitation is dark. On the basis, the electron-hole distribution of each excited state was visualized. As shown in Fig. 4c, for the D0→D1, electrons were mainly concentrated at 1,4 site near the NV (NV 1,4 site). In the D0→D2, electrons were clustered at 1,4 sites near the cyano group (cyano 1,4 site) and primitive melon (primitive 1,4 site). For the D0→D4, the electrons were distributed on the NV site. Since the D0→D3 excitation was dark, photo-generated electrons could not be directly excited to corresponding electron distribution area. Based on the above results, it is found that the activation sites of O2 may be NV 1,4 and NV sites, as well as cyano 1,4 and primitive 1,4 sites.
To further confirm the real active site, O2 were adsorbed on the 4 possible sites mentioned above, respectively. When O2 was adsorbed on the NV 1,4 site of NDCN [O2@NDCN (NV 1,4 site)], electrons might be excited to O2 in the excitations of D0→D1 and D0→D2, while holes might remain the NV region (Supplementary Fig. 11a). The IFCT method implemented in Multiwfn34 can further calculate the specific number of electrons transferred from NDCN to O2 in each excitation. As shown in Supplementary Table 2, for O2@NDCN (NV 1,4 site), there might be 0.96 and 0.95 electrons transferred from NDCN to O2 in the D0→D1 and D0→D2 excitations, respectively. Similarly, the probability of electron excitation needs to be evaluated by f. The f of D0→D1 and D0→D2 excitations for O2@NDCN (NV 1,4 site) were 0.0017 and 0.0136, respectively [O2@NDCN (NV 1,4 site) in Supplementary Table 2], suggesting that it is feasible to transfer 0.96 e− from the NV 1,4 site to O2 through the D0→D2 excitation, indicating that electrons can be directly excited from NDCN to O2 through the D0→D2 excitation while D0→D1 was forbidden. Therefore, combining the electron-hole distribution, IFCT and UV-vis spectrum results, it can be concluded that O2 adsorbed on NV 1,4 site could be directly excited by light to generate •O2−, which is consistent with the results of ESR, RDE and control experiments (Fig. 2). Using the same method, the adsorption of O2 on cyano 1,4 [O2@NDCN (cyano 1,4 site)] and primitive 1,4 sites [O2@NDCN (primitive 1,4 site)], as well as NV sites [O2@NDCN (NV site)] were also analyzed. Although the electron-hole distribution and IFCT results showed that in the D0→D1 and D0→D2 excitations, the electronic transitions of O2@NDCN (cyano 1,4 site) and O2@NDCN (primitive 1,4 site) were all from the NDCN to O2 (Supplementary Fig. 11b, 11c and Supplementary Table 2), but the corresponding oscillator strengths were too low (Supplementary Fig. 12b, 12c and Supplementary Table 2). Only 0.2 e− could be transferred from primitive 1,4 site to O2 through D0→D3 excitation, but this was not enough to activate O2. Photogenerated e− cannot be directly excited to dark excited state, but can only be promoted to bright excited state with higher energy, and then may relax to low energy dark state. However, in this indirect process, the possibility of carrier recombination or structural relaxation is greatly increased, which is not conducive to O2 reduction. Hence, the cyano 1,4 and primitive 1,4 sites had no major contribution to the generation of •O2−. When O2 was adsorbed on the NV site of NDCN [O2@NDCN (NV site)], it might get electrons through the D0→D1 and D0→D2 excitations (Fig. 4d). Although the f of D0→D1 excitation was negligible, through the D0→D2 excitation with f of 0.0428 (Fig. 4b), NDCN could transfer 0.85 e− to O2 (Supplementary Table 2). A higher f indicates that the NV site has a greater probability of generating •O2− compared to that of the NV 1,4 site. The results of vertical absorption of O2 on the surface of NDCN were summarized in Supplementary Table 2. The highlighted part indicates that the excitation is bright, which allows photogenerated electrons to transfer from NDCN to O2. Obviously, only NV and NV 1,4 site can effectively activate O2, promoting the generation of •O2−, while cyano 1,4 and primitive 1,4 site cannot.
Confirmation of endoperoxide. In electronic excitation, the transition of electrons occurs first, leading to a change in the electronic structure, which inevitably cause the evolution of geometry. In this section, the kinetic process of reaction system was investigated. Adding excess electrons to the photoreaction system for simulating light conditions is a common method35,36. First, two excess electrons were added to the O2@NDCN (NV site) and O2@NDCN (NV 1,4 site) models, respectively. After the structure was fully relaxed, O2 in the O2@NDCN (NV site) and O2@NDCN (NV 1,4 site) models finally both stabilized on the NV site (inset of Fig. 4e and Supplementary Fig. 13a), and the bond length of O-O increased from 1.199 and 1.194 Å to 1.383 and 1.382 Å, respectively. Moreover, the density of states (DOS) of two models were almost the same. Meanwhile, the two β-LUMO orbitals that were originally degenerate in the isolated O2 (Supplementary Fig. 13b) were split, one of which became β-HOMO-1 and the other β-LUMO + 1 (Fig. 4e and Supplementary Fig. 13a), suggesting that O2 got an electron37,38. This indicate that after the O2 adsorbed on the NV or NV 1,4 site becomes •O2− by vertical absorption, the further relaxation of structure lead to an increase of the O-O bond length, which generate more stable •O2−. The e− stabilized in the •O2− are effectively separated from the h+ left on the NV surface, effectively avoiding the recombination of photogenerated carriers, which explain the experimental result that NDCN exhibited a more effective charge separation ability (Supplementary Fig. 5b-d). When the excess electrons in the two models increased to 3, the O-O bond length in the O2@NDCN (NV site) and O2@NDCN (NV 1,4 site) models further increases to 1.600 and 1.591 Å, respectively. Meanwhile, the two β-LUMO orbitals that were originally degenerate in the isolated O2 all became occupied orbitals (Supplementary Fig. 14), indicating that O2 obtained two electrons, which became endoperoxide on the NDCN surface38,39.
For CN, the S0→S3 and S0→S4 excitations were confirmed to be bright by f (Supplementary Fig. 16a). The photogenerated electrons were mainly distributed in the 1,4 site of melon (Supplementary Fig. 15a3, a4), indicating that this position may be the O2 activation site in CN, which is consistent with the previous report4,19. However, compared to the 2.08 eV excitation energy of the first bright state of NDCN (596 nm for D0→D1 in Fig. 4a), that of CN was as high as 3.7 eV (330 nm for S0→S3 in Supplementary Fig. 16a), indicating that nitrogen defects can significantly expand the light response range of CN, in agreement with UV-vis DRS experimental results (Supplementary Fig. 4). When O2 was adsorbed on the 1,4 site of CN, the S0→S1 and S0→S2 excitations (Supplementary Fig. 15b) that might allow O2 to obtain photogenerated e− were all dark states (Supplementary Fig. 12d). The photogenerated electrons could only be promoted to the S0→S25 excitation with an energy as high as 3.8 eV (Supplementary Fig. 12d), and then relaxed inefficiently to the dark state that can activate O2 (Supplementary Fig. 15b and Supplementary Table 2). In contrast, on the NV or NV 1,4 site of NDCN, the excitation energy for converting O2 to •O2− was only 1.61 eV. It is worth mentioning that the calculated excitation energies were only used for qualitative comparison due to the influence of multiple factors, such as theoretical methods40, spin orbit coupling41 and overlap between absorption peaks of different transitions42, etc.
Investigation on the role of NV and cyano. The above discussion proved that after introducing cyano and NV, the NV and NV 1,4 site were both the real active sites for photocatalytic H2O2 production. Although the cyano cannot efficiently activate O2, whether the high catalytic activity of NDCN is due to the synergy between cyano group and NV need to be verified. For this purpose, we constructed nitrogen deficient CN containing only NV (NDCN-NV) and cyano group (NDCN-cy), respectively. The results indicated that the excited energy in the same excitation (Fig. 4f), UV-vis spectrum (Fig. 4a vs. Supplementary Fig. 16b) and electron-hole distribution (Fig. 4c vs. Supplementary Fig. 17a) of NDCN and NDCN-NV were almost identical, while those of CN and NDCN-cy were very similar (Fig. 4f, Supplementary Fig. 16a vs. Supplementary Fig. 16c, and Supplementary Fig. 15a vs. Supplementary Fig. 18a). Subsequently, based on the electron-hole distribution of bright excited state, O2 was adsorbed to the corresponding sites of NDCN-NV (Supplementary Fig. 17b-d) and NDCN-cy (Supplementary Fig. 18b, c), respectively. By observing the number of transferred electrons and f for each excitation (Supplementary Table 3), it can be confirmed that the NV and NV 1,4 sites of NDCN-NV can also promote the generation of •O2−, which are the same as the two sites of NDCN. However, NDCN-cy cannot effectively transfer photogenerated electrons to O2 (Supplementary Table 4). It can be concluded that the introduction of NV is the only reason to promote high-efficiency production of H2O2, while cyano group do not make any contribution to this.
The essential reason for O 2 activation. In order to reveal the essential reason for O2 activation, the electronic structures of each substance before and after adsorption were analyzed. To facilitate discussion of the role of each region and molecular orbitals (MO), NDCN was divided into three regions: cyano modified melon (melon-cy), NV modified melon (melon-NV) and primitive melon (melon-pri), as shown in Fig. 5. The key molecular orbitals (MO1-MO10 in Fig. 5) were marked in DOS (Fig. 5a-c). It can be seen from Supplementary Table 1 that no matter where O2 was adsorbed on NDCN, the transition orbital pairs corresponding to the excited state all belonged to β spin. Thus, the focus of analysis was on the β spin orbits. For NDCN, β-HOMO (MO5 in Fig. 5) was concentrated on melon-NV, while β-LUMO (MO4 in Fig. 5) was distributed on melon-cy and melon-pri, indicating that NV and cyano have impacts on the original optical response range of CN, which is consistent with previous report21. It is worth distinguishing that photogenerated e− could be excited to the cyano 1,4 site of NDCN (Fig. 4a, 4c2), but they could not further effectively activate O2 due to the negligible f mentioned above.
Before interacting with NDCN, the β-LUMO of O2 is degenerate (MO9 and MO10 in Fig. 5). The formations of •O2− and endoperoxide are the results of β-LUMO being occupied by one and two electrons, respectively. When O2 was adsorbed on the NV site of NDCN, MO9 and MO10 of O2 interacted with MO5 of NDCN to generate MO8 (Fig. 5d). Note that MO8 was an occupied orbital. The contribution of non-occupied MO9 and MO10 to MO8 indicated that the NV site was chemically adsorbed with O2 in the ground state, making O2 obtain a small amount of e−. Moreover, the adsorption energy of O2 at NV site was − 0.12023 eV, which is the lowest value among the four sites of NDCN (Supplementary Fig. 19a), indicating that the NV site could effectively adsorb O2, and thus is beneficial for subsequent O2 reduction process. The spin density map (Supplementary Fig. 19b) showed that there was unpaired β e- on melon-NV, so that the NV site could be acted as an electron donor to effectively adsorb O2 in the ground state. On the other hand, the key point of activating O2 was the electron excitation from MO8 to MO7 (see “Hb -> Lb 97.4%” in Supplementary Table 2). The formation of MO7 was also due to the interaction of MO5, MO9 and MO10. The lower energy MO5 contributed 6.24% to the MO7 (Fig. 5d), effectively reducing the excitation energy for activating O2. O2@NDCN (NV 1,4 site) was basically the same as O2@NDCN (NV site). Since oxygen was located on the NV 1,4 site instead of the NV site, the corresponding adsorption energy increased slightly, but the value was still negative (Supplementary Fig. 19a). Moreover, the contribution of MO5 to MO7 was also slightly reduced (Fig. 5e), leading to a slight increase in the excitation energy of O2 activation, but the NV 1,4 site could still effectively transfer photogenerated e− to O2. Combining the results of the control experiment in air (green curve in Fig. 2b), it can be concluded that when the saturated O2 environment was replaced with air, the reason why NDCN could still maintain high catalytic activity was the strong adsorption capacity of NV for O2. On the other hand, cyano 1,4 and primitive 1,4 sites showed complete inertness for O2 adsorption due to the positive adsorption energy (Supplementary Fig. 19a). It is worth mentioning that in the simulation of structural optimization, even if O2 was placed on top of the cyano group as the initial structure, O2 eventually stabilized at the cyano 1,4 site, indicating that cyano cannot adsorb O2. This was due to its own quite stable C ≡ N. Moreover, since O2 was not on melon-NV, MO9 and MO10 could not interact with MO5, but could only interact with higher energy orbitals corresponding to cyano 1,4 (Supplementary Fig. 20a) or primitive 1,4 site (Supplementary Fig. 20b), making the corresponding excitation energy higher.
For materials containing more modification sites, there are more possible geometries and O2 activation sites, resulting in an increase in the workload and complexity of simulation. But it should be emphasized that even so, the amount of work required is still far less than the currently widely used first-principles molecular dynamics43 and transition state calculations44. The more complex simulations resulting from more MSs can be completed accurately and quickly through high-throughput calculations45 and machine learning46. Therefore, the proposed method is universal and extensible.
We proposed a universal method combining experiments and DFT-TDDFT calculations to investigate the in-depth reaction mechanism of NDCN containing two MSs for photocatalytic H2O2 evolution. PL technology and electrochemical measurements proved that the introduction of defects could effectively enhance charge separation. The results of ESR, RDE and control experiments indicated that •O2− was the key active specie and photocatalytic reaction was a two-step single electron transfer process. After constructing accurate models based on characterization results, DFT-TDDFT calculations proved that NV could efficiently convert O2 to •O2−, while cyano group could not directly activate O2 and had no synergy with NV. Meanwhile, the introduction of NV promoted O2 adsorption and reduced the excitation energy for activating O2, significantly facilitating H2O2 evolution. This work provides a reference for the joint connection of experiments and DFT-TDDFT calculations and has general guiding significance for photocatalytic research involving O2 activation.