Dipole induction by structural engineering of supports for Fe single-atom photocatalysts toward excellent photocatalytic ozonation

Abstract

Efforts in designing efficient polymer-based single atom photocatalysts (SAPs) have primarily focused on selecting specific metal atoms with tailored geometries and properties to control functionality. However, the impact of the light-harvesting units that bridge these single metal atoms, crucial for light absorption and energy transfer, has been largely overlooked. In this work, two carbon nitride (CN)-based iron SAPs with a similar FeN₄ coordination environment are synthesized: triazine-based CN (C₃N₄) and nitrogen-rich triazole-based CN (C₃N₅), differing in the unit cell structure. C₃N₅ exhibits better photocatalytic ozonation performance than C₃N₄ due to its unit cell asymmetry, which induces a dipole field that facilitates charge transfer. The addition of iron single atoms breaks the symmetry of C₃N₄ to enhance the dipole moment, while they weaken the separation and migration of bulk charge carriers in the Fe-C₃N₅ SAP. The iron atoms act as active sites in both Fe-C₃N₄ and Fe-C₃N₅ SAPs, accelerating interfacial reaction kinetics. These findings demonstrate the importance of the light-harvesting unit structures of CN-based SAPs in regulating photogenerated charge kinetics and offering valuable insights for the rational design of effective photocatalysts.

---

Introduction

Polymer-based semiconductors, with abundant anchoring sites and visible light responsiveness, are promising supports for single-atom photocatalysts (SAPs) in environmental pollution mitigation through reactive oxygen species (ROS) production.^1–3 Traditionally, enhancing SAP functionality and structural diversity has focused on manipulation of the metal atom and its coordination environment, that is, the type of metal and the atoms it coordinates with, to establish clear structure–activity relationships.^4–8 However, the influence of the unit cell structure of the support, where the metal anchoring sites reside, has received less attention despite its critical role in SAP design and performance. While single atoms (SAs) can induce charge transfer states between metal centers and ligands within the forbidden band of the photocatalyst material, thus broadening light absorption to longer wavelengths, the properties of the polymer-based support fundamentally dictate the light-harvesting capacity.^9–11 In addition, the SAs boost the separation and transfer of photogenerated charges by acting as charge pumps or introducing trap states, which facilitate the accumulation of surface charges and thus accelerate the target reactions.¹² Nevertheless, the specific role of SAs in the charge dynamics may vary depending on the structural unit of the support, leading to distinct photocatalytic behavior. Subtle variations in the structural unit of the support can also influence the metal–support interaction, altering surface reaction kinetics. Despite the profound impact of polymer support structures on photocatalytic mechanisms, a systematic study into how unit cell variations affect SAP performance remains underexplored.

Carbon nitrides (CNs) are promising polymer-based semiconductors with versatile structural tunability, rendering them ideal for a range of applications, such as photocatalytic hydrogen evolution and photocatalytic H₂O₂ generation.^13–15 Among CN materials, C₃N₄ with a tri-s-triazine symmetry unit structure has been widely used as a support for SAPs, as the nitrogen atoms within the framework provide abundant anchoring sites for single metal atoms.¹⁶ In contrast, nitrogen-rich triazole-based carbon nitride (C₃N₅) with a lower C/N ratio exhibits an asymmetrical structure resulting from the polymerization of triazole and triazine frameworks, which can lead to distinct carrier behaviors compared to C₃N₄.¹⁷ For example, recent studies have demonstrated that C₃N₅ with its asymmetrical structure overcomes the random distribution state of photogenerated carriers by inducing self-polarization to facilitate rapid, directional electron transfer, which is absent in C₃N₄.¹⁸ Moreover, both C₃N₄ and C₃N₅ possess similar rich N sites, allowing for comparable single-atom coordination environments.¹⁹ Therefore, comparing C₃N₄ and C₃N₅ as supports in SAPs provides a valuable platform to elucidate how the incorporation of symmetric triazine versus asymmetric triazole units in CN materials influences their charge transfer dynamics and photocatalytic activity, which is crucial for advancing CN-based photocatalysts in environmental remediation and other photocatalytic applications.

To promote the effectiveness of photocatalytic applications in environmental remediation, facilitating the formation of photo-induced radicals, particularly hydroxyl radicals (˙OH, E₀ = 2.80 V_NHE), is paramount.²⁰ Strategies for producing ˙OH via the reaction between electrons or holes and various acceptors in photocatalysis, such as O₂, O₃, H₂O, and H₂O₂ oxidants, have been developed.^21–23 Among these, photocatalytic ozonation stands out due to its high efficiency and selectivity by utilizing the oxidant (O₃) and photogenerated electrons to produce ˙OH.²⁴ Furthermore, known for its high oxidizing ability, O₃ (E₀ = 2.07 V_NHE) can pretreat easily degradable organic compounds, thereby reducing reliance on ˙OH and improving the overall efficiency of reaction systems in practical applications.²⁵ The reduction of O₃ follows a one-electron pathway (O₃ → ˙O₃⁻ → HO₃˙ → ˙OH), demonstrating a direct correlation between the yield of available electrons and ˙OH, making photocatalytic ozonation a powerful approach to elucidate structure–activity relationships and serving as an exemplary advanced oxidation process to alleviate environmental pollution.^26,27

We first used density functional theory (DFT) calculations to design a highly efficient CN-based SAP. To validate the design and investigate how the integration of asymmetric triazole versus symmetric triazine units in the carbon nitride supports influences optoelectronic properties and photocatalytic performance, two CNs with distinct structures were synthesized: symmetric triazine-based C₃N₄ and asymmetric triazole-based C₃N₅, along with their SAPs containing iron with FeN₄ sites. Experimental characterization and theoretical calculations revealed that the anchoring of iron SAs altered the charge distribution in the CN supports and modulated the dipole field, which in turn impacted charge transfer, charge separation and reactant adsorption. Additionally, the iron SAs introduced midgap states, enhanced light absorption and modified charge decay behaviors. This study shows that the role of SAs in photocatalytic reactions is intricately linked to the structure of the support and proposes fundamental guidelines for designing active photocatalysts.

Results and discussion

CN-based SAP design via DFT calculation

In general, efficient photocatalysts should exhibit the following characteristics: (1) broad light-harvesting corresponding to a narrow band gap, (2) effective charge carrier separation, which can be facilitated by an internal field, and (3) abundant active sites for efficient catalytic reactivity.^9,28,29 In SAPs, the supports serve as light-harvesting units, while the SAs act as active sites. However, metal SA–support interaction significantly affects the three key reaction steps by regulating the electronic properties of the SAPs.¹² To investigate how the structural differences between symmetric triazine- and asymmetric triazole-based supports affect photocatalytic performance, we calculated the band structures and dipole moments of the SAPs and their supports to assess their light-harvesting capabilities and charge separation efficiencies, respectively, and thereby predict their theoretical photocatalytic activity. For this purpose, two commonly used CN supports, C₃N₄ and C₃N₅, and Fe as the SA were selected, due to the high photocatalytic ozonation activity of C₃N₄-based SAPs for ˙OH formation reported in our previous work³⁰ (DFT calculation is detailed in the ESI†).

The structure of C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅ was theoretically modeled as shown in Fig. S1.† DFT calculations reveal that C₃N₄ and C₃N₅ exhibit band gaps of 2.6 and 2.4 eV at the Γ point, respectively (Fig. 1a and b). The conduction band (CB) of both C₃N₄ and C₃N₅ primarily consists of N 2p orbitals, while the hybridization of N 2p and C 2p orbitals contributes to the formation of the valence band (VB) (Fig. S2†). When Fe is loaded in C₃N₄ and C₃N₅, the hybridization of the N 2p and Fe 3d orbitals results in an increased density of states mainly near the conduction band minimum (CBM) and valence band maximum (VBM), respectively (Fig. 1c, e, d and f). Additionally, the addition of iron SAs also generates intermediate states in the band gap of both C₃N₄ and C₃N₅, which enhances light absorption, increasing the production of photo-generated charges.

Fig. 1 The calculated electron energy bands of (a) C₃N₄; (b) C₃N₅; (c) Fe-C₃N₄; and (d) Fe-C₃N₅ (elements are represented by different colors: carbon is blue, nitrogen is green, and iron is orange, and the size of each circle represents the contribution magnitude of this element to the band). Total density of states (TDOS) of (e) Fe-C₃N₄ and (f) Fe-C₃N₅. (g) Mechanism about the accelerated separation of photocatalytic carriers induced by the dipole field effect; dipole moments (per unit cell) of (h) Fe-C₃N₄ and (i) Fe-C₃N₅, calculated based on Berry phase expressions.

Additionally, the charge-transfer dynamics during photocatalytic reactions are also closely related to the interaction between iron SA and the light-harvesting support. The internal field is regarded as the common driving force for charge separation; therefore, DFT was used to estimate the dipole moment (Fig. 1g) of C₃N₄, Fe-C₃N₄, C₃N₅, and Fe-C₃N₅.³¹ The dipole moment of triazine-based C₃N₄ (with a crystallographic plane group of Pm) was 2.09 D (Fig. 1h). Upon introducing iron SAs to form Fe-C₃N₄, the symmetry of C₃N₄ was broken, resulting in a substantial increase in the dipole moment to 97.39 D according to Fig. 1h, indicating better charge separation efficiency. In contrast, C₃N₅, composed of both triazole and triazine units, inherently exhibited a much higher dipole moment of 42.57 D than that of C₃N₄, due to its local structural asymmetry (Fig. 1i). When iron SAs were anchored onto C₃N₅ to form Fe-C₃N₅, however, the crystallographic plane group remained unchanged, but the dipole moment sharply decreased to 5.46 D. This reduction was probably due to electron redistribution caused by the introduction of Fe atoms, which have a lower electronegativity than C, thereby partially offsetting the charge distribution differences between the triazole and triazine units.³² These results demonstrate that the effect of iron atoms on bulk charge carrier dynamics strongly depends on whether the CN support features symmetric triazine or asymmetric triazole units. It indicated that Fe-C₃N₄ would be the best photocatalyst; hence, C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅ were further synthesized to verify this hypothesis.

Synthesis and characterization of CN-based SAPs

To load iron SAs on different CN matrices, dicyandiamide for synthesizing C₃N₄ or 3-amino-1,2,4-triazole for C₃N₅ (Fig. S3†),^16,19 was mixed with FeCl₃·6H₂O. The atomically dispersed Fe was anchored on C₃N₄ (labeled Fe-C₃N₄) and C₃N₅ (Fe-C₃N₅) by a simple thermal polymerization strategy (see the Experimental section in the ESI†). Control samples of pristine C₃N₄ and C₃N₅ were also prepared for use as references. The Fe contents in Fe-C₃N₄ and Fe-C₃N₅ were 3.7 wt% and 3.8 wt%, respectively, determined by inductively coupled plasma triple quadrupole mass spectrometry (ICP-MS). Element analysis confirmed that the atomic ratios of C/N in C₃N₅ and Fe-C₃N₅ were around 0.60, whereas those in C₃N₄ and Fe-C₃N₄ were 0.67 (Table S1†), which are consistent with the literature.^17,19 Fig. 2a shows the X-ray diffraction (XRD) patterns of C₃N₅, which exhibit characteristic diffraction peaks at 13.5° and 26.9°, corresponding to the (100) and (002) planes of nitrogen-rich triazole-based CN. These peaks indicate ordered in-plane structural and interlayer stacking of aromatic systems in graphitic materials.^27,30 In comparison, C₃N₄ exhibits peaks at 13.1° and 27.4°, indicating a slightly smaller in-plane distance between each structural unit and a larger interlayer spacing in C₃N₅. The intensity of both peaks decreases greatly when iron is incorporated, presumably due to a decrease in structural orderliness of both the stacked layer and the in-planar structure of CN supports. Significantly, no iron species-related diffraction peaks are observed in both Fe-C₃N₅ and Fe-C₃N₄, indicating the homogeneous dispersion of iron species within the CN supports.

Fig. 2 Structural characterization of catalysts. (a) XRD patterns of C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅. (b) C K-edge NEXAFS of C₃N₅ and Fe-C₃N₅. (c) N K-edge X-ray NEXAFS of C₃N₅ and Fe-C₃N₅. (d) C 1s XPS spectra of C₃N₅ and Fe-C₃N₅. (e) N 1s XPS spectra of C₃N₅ and Fe-C₃N₅. (f) EPR spectra of C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅.

Fourier transform infrared spectroscopy (FT-IR) analysis reveals characteristic peaks at 740–775 cm⁻¹, 810–891 cm⁻¹, 1200–1700 cm⁻¹, 2180 cm⁻¹, and 3200–3400 cm⁻¹ (shown in Fig. S4†), which correspond to the heterocyclic N–N bond in the triazole group, the condensed C–N heterocycles in the triazine group, stretching modes of C–N heterocycles in triazine, cyano groups (–C≡N) formed from the terminal –C–NH₂ in the melon structural unit, and the terminal amino group of the CN framework, respectively.^17,18 These spectra confirm that both C₃N₅ and Fe-C₃N₅ comprised triazole units and triazine groups. Similarly, the FTIR analysis of C₃N₄ and Fe-C₃N₄ also corroborated the integral melon skeleton of the carbon nitride structure. Interestingly, a new peak at 2158 cm⁻¹ appeared in both Fe-C₃N₄ and Fe-C₃N₅, attributed to cyano groups (–C≡N) derived from the conversion of terminal –C–NH₂ in the melon unit.³³ This change is probably induced by the presence of iron species that affected the polymerization process.³⁴ Calculations from the nitrogen sorption measurements (Fig. S5†) show that C₃N₅ (0.9 m² g⁻¹) has a significantly lower specific surface area than C₃N₄ (7.1 m² g⁻¹), whereas Fe-C₃N₅ has a higher surface area (3.8 m² g⁻¹) than Fe-C₃N₄ (1.8 m² g⁻¹), implying that Fe-C₃N₅ could possess more active sites for surface reactions than Fe-C₃N₄.

Synchrotron-based near-edge X-ray absorption fine structure (NEXAFS) was employed to investigate the bonding structures of C₃N₅ and Fe-C₃N₅, compared with the structures of C₃N₄ and Fe-C₃N₄. The C K-edge NEXAFS of both C₃N₅ and Fe-C₃N₅ exhibits three typical peaks of triazole-based CN (Fig. 2b): C=C (C1) at 288.1 eV, C–N–C (C2) at 289.0 eV, and C–C (C3) at 295–300 eV.^17,35 These peaks slightly shift towards higher energies for C₃N₄ and Fe-C₃N₄ (Fig. S6a†), and the weaker corresponding bonds in C₃N₅ and Fe-C₃N₅ indicate that the binding affinity between electrons in the 1s orbital and C decreases due to the formation of a triazole moiety within the tri-s-triazine unit of C₃N₅. The N K-edge NEXAFS further reveals three main resonances in both CN supports and SAPs (Fig. 2c and S6b†): C–N–C (N1), graphitic three-fold nitrogen atom N-3C (N2), and sp³ N-3C (N3) bridging among the three structural units.^17,35 These results suggest that the structure of C₃N₅ closely resembles the heptazine-based structure of C₃N₄, but with subtle variations attributed to the integration of triazole units.

X-ray photoelectron spectroscopy (XPS) was utilized to further analyze the chemical states of the samples. The survey spectra confirm that C₃N₄ and C₃N₅ are primarily composed of C and N, with a minor peak of iron observed in Fe-C₃N₄ and Fe-C₃N₅ (Fig. S7†). In the C 1s XPS spectra of C₃N₅ and Fe-C₃N₅, peaks corresponding to N₂–C=N– (288 eV), adventitious C (285.7 and 284.8 eV) and the sp²-hybridized C in the aromatic ring attached to the NH₂ group (289.1 eV) are identified in Fig. 2d.^18,36 The peaks at 289.1 eV specifically suggest the incorporation of triazole moieties within the CN framework of C₃N₅ and Fe-C₃N₅.¹⁷ The peak at 288 eV in C₃N₄ (288.3 eV) slightly shifted to higher energy (Fig. S8a†), which is consistent with the NEXAFS results. In the N 1 s XPS spectra of C₃N₅ and Fe-C₃N₅ (Fig. 2e), the peak at 398.5 eV is assigned to the C–N=C bond, while the peak at 400.2 eV is attributed to a combination of signals from bridging N (e.g., tertiary N) and amino groups (–NH_x), indicating that C₃N₅ and Fe-C₃N₅ have a similar tri-s-triazine framework to C₃N₄ (Fig. S8b†).^17,37 Furthermore, signals at 403.8 eV in the N 1s XPS spectra indicate the delocalization of p electrons in CN heterocycles, likely due to the graphitic stacking of CN layers in both C₃N₅ and Fe-C₃N₅.¹⁷

The presence of unpaired electrons in the structural framework was investigated using electron paramagnetic resonance (EPR) spectroscopy.³⁶ As shown in Fig. 2f, C₃N₅ exhibited a more intense Lorentzian line compared to C₃N₄, indicating enhanced electron transport due to the presence of the triazole ring in the CN unit cell.¹⁷ Interestingly, the introduction of iron SAs into both C₃N₄ and C₃N₅ significantly reduces this EPR signal intensity, with Fe-C₃N₄ and Fe-C₃N₅ showing comparable diminished intensity. This reduction was attributed to the unpaired electrons likely transferring from the aromatic rings of carbon atoms to the iron atoms.

The TEM image of Fe-C₃N₅ (Fig. 3a) revealed a typical lamellar structure with minor occurrences of microcracks and pores, similar to the morphology observed on Fe-C₃N₄ (Fig. 3c). No metal nanoparticles were detected on Fe-C₃N₅ or Fe-C₃N₄ in the high-resolution TEM images (Fig. S10†). The atomic-level distribution of Fe species was analyzed by using aberration-corrected high-angle annular dark field scanning transmission electron microscopy (AC-HAADF-STEM). In Fig. 3b and d, bright spots (marked by red circles) corresponding to Fe single atoms are observed across the matrices of both C₃N₄ and C₃N₅. Moreover, the energy-dispersive spectroscopy (EDS) mapping images demonstrated that both C, N, and Fe species are uniformly distributed on Fe-C₃N₄ and Fe-C₃N₅, further verifying that Fe was predominantly present as SAs rather than forming nanoparticles (Fig. S11 and S12†). This corroborates the XRD findings, which show no detectable diffraction peaks corresponding to Fe oxide or metallic Fe.

Fig. 3 Characterization of catalysts. TEM images of (a) Fe-C₃N₅ and (c) Fe-C₃N₄. Magnified HAADF-STEM images of (b) Fe-C₃N₅ and (d) Fe-C₃N₄. (e) Fe K-edge XANES spectra of Fe-C₃N₅, Fe-C₃N₄, Fe foil, Fe₂O₃, and FeCl₃. (f) The XANES spectra at the Fe L-edge of Fe-C₃N₅ and Fe-C₃N₄. (g) The FT-EXAFS spectra of the EXAFS spectra. (h) The best-fit analysis results of EXAFS for Fe-C₃N₅ and Fe-C₃N₄. The wavelet transform EXAFS analysis of (i) Fe-C₃N₅ and (j) Fe-C₃N₄.

To further identify the iron coordination environment in Fe-C₃N₄ and Fe-C₃N₅, Fe K-edge X-ray absorption near edge structure (XANES) measurement was conducted. As shown in Fig. 3e, the absorption edges of Fe-C₃N₄ and Fe-C₃N₅ are positioned between those of metallic Fe foil and Fe₂O₃, suggesting that the average oxidation state of Fe lies between 0 and +3. Despite similar near-edge structures, the absorption edge of Fe-C₃N₅ is slightly shifted to higher energies compared to Fe-C₃N₄, indicating that the oxidation state of Fe in Fe-C₃N₅ is higher. Similarly, the Fe L-edge XANES spectra results (Fig. 3f) show a peak at higher energy for Fe-C₃N₅ compared to Fe-C₃N₄. Fourier transformed (FT) EXAFS spectra of Fe-C₃N₄ and Fe-C₃N₅ display a single prominent peak at 1.6 Å with no detectable Fe–Fe interactions at 2.2 Å (Fig. 3g), implying that the Fe in Fe-C₃N₄ and Fe-C₃N₅ is atomically dispersed. The atomic dispersion of Fe was also supported by the wavelet-transform (WT) contour plot (Fig. 3h and i), revealing a single intensity maximum around 4 Å⁻¹ for both Fe-C₃N₄ and Fe-C₃N₅, characteristic of Fe–N coordination. EXAFS best-fit analysis (Table S2†) indicates that a single Fe atom in Fe-C₃N₄ and Fe-C₃N₅ was coordinated with four N atoms, forming an Fe-N₄ coordination structure, with Fe-N bond lengths of 2.10 and 2.07 Å, respectively (Fig. 3j and S13†). The proposed Fe-N₄ coordination models for Fe-C₃N₄ and Fe-C₃N₅ are illustrated in Fig. S1.† The characterization conducted confirms that the DFT predicted structure aligned well with experimental results.

Photocatalytic activity and ROS identification over different CN-based SAPs

The photocatalytic ozonation activity of catalysts for ˙OH production was assessed using oxalic acid (OA) as the target pollutant, which exhibits resistance to direct O₃ oxidation (k < 0.04 M⁻¹ s⁻¹) but high reactivity with ˙OH, making it an ideal probe molecule for ˙OH.^30,38 As shown in Fig. S14,† no significant OA degradation occurs over all catalysts in either photocatalytic oxidation or catalytic ozonation processes. However, when visible light and ozone were applied simultaneously, obvious OA degradation was observed across all catalysts (Fig. 4a and S15†). The performance of the catalysts after 60 min reaction follows the sequence Fe-C₃N₄ > Fe-C₃N₅ > C₃N₅ > C₃N₄. In the case of pure supports, C₃N₅ demonstrated higher O₃ activation activity for OA degradation under visible light compared to C₃N₄. This suggests that, despite having similar active sites, the incorporation of symmetric triazine or asymmetric triazole units into CN-based SAPs can significantly affect their photocatalytic activity by altering charge carrier kinetics through differences in dipole moment. Notably, materials with Fe SAs outperformed their pristine supports, suggesting that incorporating iron atoms into C₃N₄ and C₃N₅ effectively promotes the activation of O₃ by photogenerated electrons to facilitate ˙OH generation. Additionally, Fe-C₃N₄ and Fe-C₃N₅ demonstrate superior photocatalytic ozonation activity in degrading other refractory pollutants, including phenolic compounds (bisphenol and diclofenac), antibiotics (sulfamethoxazole and cephalexin), and dyes (methylene blue) (Fig. 4b and c). Notably, Fe-C₃N₄ exhibits superior activity to Fe-C₃N₅ in total organic carbon (TOC) removal of all pollutants, consistent with OA degradation and ozone decomposition rates (Fig. 4d).

Fig. 4 (a) Degradation efficiency of OA in photocatalytic ozonation with different photocatalysts; degradation of various pollutants on (b) vis/O₃/Fe-C₃N₄ and (c) vis/O₃/Fe-C₃N₅ systems; (d) dissolved ozone decomposition rate in the photocatalytic ozonation processes. (e) Quenching using TBA in photocatalytic ozonation processes. (f) EPR spectra in the photocatalytic ozonation with DMPO as a probe. Stability of (g) Fe-C₃N₅ and (h) Fe-C₃N₄ in successive cycling (photocatalytic ozonation system). (i) Experimental Fe K-edge FT-EXAFS spectra for fresh and used Fe-C₃N₄. Conditions: [catalyst]: 0.1 g L⁻¹, [OA]: 2 mM, [O₃]: 30 mg L⁻¹, [TBA]: 100 mM, solution pH: 2.7, and solution volume: 0.3 L. Error bars represent the data from triplicate tests.

The stability of Fe-C₃N₄ and Fe-C₃N₅ in photocatalytic ozonation was assessed through recycling investigations (Fig. 4g and h), showing no noticeable decline in photocatalytic activity over five consecutive reaction cycles, indicating excellent stability. Consistently, inductively coupled plasma (ICP) analysis revealed that Fe leaching after each reaction cycle was as low as 0.03% for Fe-C₃N₄ and 0.08% for Fe-C₃N₅, further confirming the strong anchoring of Fe single atoms and the high structural stability of the catalysts. XANES spectra of the Fe K edge for used Fe-C₃N₄ after five repetitive reaction cycles exhibit a slight right-shift in the absorption edge in comparison with the fresh sample, indicating a small rise in the oxidation state of the iron species (Fig. S16†). Additionally, a negligible change in major peak positions is observed in the FT-EXAFS spectra when comparing used Fe-C₃N₄ with fresh Fe-C₃N₄ (Fig. 4i). Notably, no peaks related to an Fe–Fe bond appeared for used Fe-C₃N₄, further confirming its structural stability. Consistent with these findings, AC-HAADF-STEM images of the post-reaction Fe-C₃N₄ and Fe-C₃N₅ samples reveal that the Fe species remain atomically dispersed without any sign of aggregation or clustering (Fig. S17†). This atomic dispersion after catalysis is in good agreement with the EXAFS results and the excellent cycling stability observed in the catalytic performance tests. As summarized in Table S3,† Fe-C₃N₄ exhibits superior photocatalytic performance compared to previously reported materials for the degradation of various organic pollutants, demonstrating its clear advantage and potential in environmental applications.

Quenching experiments were performed to identify the dominant ROS in the photocatalytic ozonation system. Tert-butanol (TBA) was employed as a scavenger for ˙OH radicals.³⁹ As shown in Fig. 4e, the degradation of OA was completely inhibited when excess TBA was added, indicating the dominant role of ˙OH in OA removal over all catalysts in the photocatalytic ozonation system. EPR spectroscopy was used to characterize the formation of ˙OH radicals and 5,5-dimethyl-1-pyrroline N-oxide (DMPO) was used as a probe.^26,40 Fig. 4f shows the characteristic DMPO-˙OH adduct signal (a clear 1 : 2 : 2 : 1 quartet signal, g = 2.006, and a_N = a_Hβ = 14.9 G) observed in all photocatalytic ozonation systems, with signal intensities positively correlating with the catalytic activity of the samples. These results collectively confirm that ˙OH radicals are the primary reactive species responsible for OA degradation in the photocatalytic ozonation process.

Systematic photocatalytic mechanism in the photocatalytic ozonation process

A typical photocatalytic process involves three main steps: light harvesting, charge separation and transfer, and surface catalytic reaction. For photocatalysts, improving the light absorption capacity by narrowing the band gap, introducing a driving force for charge carrier separation and transfer, and designing active sites for the target reaction are key factors for the enhancement of photocatalytic activity. Thus, a systematic mechanistic investigation was conducted to determine how the incorporation of symmetric triazine versus asymmetric triazole units in CN materials influences their charge transfer dynamics and photocatalytic activity.

Light harvesting

As depicted in Fig. 5a, the UV-vis absorbance spectrum of C₃N₄ exhibits a characteristic absorption edge at 460 nm. In contrast, C₃N₅ presents an absorption edge around 540 nm, likely resulting from the p–p* electronic transitions associated with the sp² hybridization of C and N within the CN framework (Fig. S2†).⁴¹ Consequently, the band gap of C₃N₅ (2.30 eV) is notably lower than that of C₃N₄ (2.70 eV) (Fig. 5b), which is consistent with the DFT calculation (Fig. 1a and b). Upon anchoring iron atoms, Fe-C₃N₄ shows stronger absorption and a light absorption onset at 650 nm attributed to electronic transitions induced by iron, but retains a peak at 460 nm similar to pristine C₃N₄. Similarly, the incorporation of iron into C₃N₅ further enhances the visible light absorption compared to C₃N₅, with the optical absorption edge shifting to approximately 785 nm. The DFT results show that the loading of iron introduces an Fe 3d peak within the band gap, leading to the emergence of mid-gap states (Fig. 1e and f). It is noteworthy that Fe-C₃N₄ and Fe-C₃N₅ exhibit stronger background absorbance at longer wavelengths beyond 880 nm compared to C₃N₄ and C₃N₅. This indicates the formation of additional sub-bandgap states induced by Fe doping. Meanwhile, the diminished EPR signal suggests a transformation of paramagnetic defects (e.g., nitrogen vacancies) into EPR-silent or diamagnetic states. This apparent discrepancy implies a change in the nature or electronic configuration of the defects, rather than a straightforward increase in defect density.²⁶

Fig. 5 (a) The diffuse reflectance spectra, (b) band diagrams, (c) steady-state PL spectra, (d) transient-state PL spectra, and (e) photoelectrochemical tests of C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅. TA kinetic plots and representative fitting curves of (f) C₃N₄, (g) Fe-C₃N₄, (h) C₃N₅, and (i) Fe-C₃N₅.

The VB position of the catalysts was explored based on XPS analysis and then the CB position can be determined by combining the band gap value. As shown in Fig. 5b, C₃N₅ exhibits a lower CB position and higher VB position than C₃N₄. The addition of iron narrows the band gap by shifting the VB upwards and the CB downwards in both C₃N₄ and C₃N₅. For Fe-C₃N₅, the higher levels of CB and VB than those of Fe-C₃N₄ may be due to the distinct density of states near the CBM and VBM originating from the Fe 3d orbital. For all catalysts, the reductive ability of the photo-generated electrons enables the conversion from O₃ to ˙OH.

Charge transfer dynamics

The charge transfer dynamics between metal SAs and CN is of essential importance for the entire photocatalytic system. A systematic investigation was conducted to explore the role of iron SAs and the effect of triazine–triazole structural motifs in CN materials on charge separation and transfer.

Steady-state and transient-state photoluminescence (PL) spectroscopy were conducted to explore the carrier migration via radiative recombination in the photocatalysts.³² As shown in Fig. 5c, C₃N₅ has a lower PL intensity than C₃N₄, indicating that the radiative recombination of charges on C₃N₅ is inhibited by the dipole field.¹⁸ Upon loading iron SAs onto the CN, the PL intensity of both Fe-C₃N₄ and Fe-C₃N₅ is lower than that of the support, indicating that Fe SAs effectively suppress radiative recombination. Time-resolved PL (TRPL) in Fig. 5d illustrates the decay kinetics of emissive states. The two components (fast decay component τ₁ and slow decay component τ₂) reflect the intraplanar recombination and interplanar or intrachain migration, respectively (Table S4†).^42,43 C₃N₅ exhibits a significantly shorter lifetime for both τ₁ and τ₂ compared to C₃N₄, which is indicative of faster charge migration in C₃N₅. This can be attributed to its larger conjugated π network due to the presence of azo bonds extending π delocalization or stronger nonradiative transitions because higher carrier mobility accelerates charge migration into trap sites, leading to increased nonradiative recombination.⁴⁴ Intriguingly, the introduction of Fe SAs results in similar decay profiles for Fe-C₃N₄ and Fe-C₃N₅ (Fig. 5d), indicating that Fe atoms have a greater effect on the decay behavior than the structure of the CN supports. For Fe-C₃N₄, both τ₁ and τ₂ decrease compared to C₃N₄, while the τ₁ and τ₂ of Fe-C₃N₅ are longer than those in C₃N₅.

To further illustrate the effects of the CN support and Fe SAs on charge decay kinetics, femtosecond transient absorption spectroscopy (fs-TAS) was performed to monitor the electron trapping process (Fig. 5f–i). The probed wavelength was 4000 nm, allowing differentiation between the shallow trapping process, band gap recombination (visible light range), and deep trapping (near-infrared range).⁴⁵ For C₃N₄ and C₃N₅, the transient absorption (TA) time profiles were fitted using two and three exponential components, respectively (Fig. 5f and h). The fast decay component (τ₁) for C₃N₄ was attributed to intra-band electron transitions, the intermediate component (τ₂) to the trapping process,⁴⁶ and the slow decay component (τ₃) to recombination in trap states.^47,48 For C₃N₅, τ₁ corresponds to electron transition/trapping and τ₂ to trapped electron recombination. Notably, the percentage of the slow decay component in C₃N₅ was markedly reduced compared to C₃N₄, indicating that a greater proportion of electrons in C₃N₅ could participate in photocatalytic reactions.⁴⁹

Upon introducing Fe SAs, the TA fitting results for Fe-C₃N₄ and Fe-C₃N₅ are given in Fig. 5i and j. In Fe-C₃N₄, all three decay components show shorter lifetimes than those in C₃N₄, with increased contributions from the fast and intermediate decay components. Similarly, Fe-C₃N₅ exhibits shorter τ₁ and τ₂ compared to C₃N₅. These results suggest that electrons are captured by the Fe SAs.⁵⁰ This aligns with the TRPL results, confirming that Fe atoms significantly alter the charge decay process. Since the slow decay component was reduced in Fe-C₃N₄ and Fe-C₃N₅ compared to their pristine supports, both SAPs exhibit superior photocatalytic performance to their supports.⁹ In addition, the accelerated migration of electrons to Fe SAs suggests that Fe atoms may serve as the active sites in SAPs, facilitating interfacial electron transfer and enabling more electrons to access the surface to participate in O₃ reduction.

The distinct impact of symmetric triazine versus asymmetric triazole structures in CN-based SAPs on the separation of electrons and holes is further supported by the transient photocurrent responses of samples.⁵¹ As shown in Fig. 5e, C₃N₅ exhibits a higher photocurrent density than C₃N₄, most likely arising from the dipole field caused by the asymmetric structure. Fe-C₃N₄ produces significantly higher photocurrent density than C₃N₄. In contrast, the photocurrent density of Fe-C₃N₅ is lower than that of C₃N₅. Generally, an Fe–N bond forms between tri-s-triazine units, accelerating the transfer of electrons.³⁰ Considering that the symmetrical structure of C₃N₄ is broken when iron SAs are doped, the enhanced dipole field would facilitate charge separation. Nevertheless, the addition of iron SAs decreases the dipole field on C₃N_5, which has an asymmetrical structure, and thus an opposite trend appears. It is also noted that Fe-C₃N₅ exhibits better OA degradation efficiency than C₃N₅, most likely due to the role of the iron SAs in the surface reaction (discussed later). As shown in Fig. S18,† the Nyquist circle diameters of both C₃N₄ and C₃N₅ are smaller than those of Fe-C₃N₄ and Fe-C₃N₅, respectively. Hence, the anchoring of iron SAs between light-harvesting units facilitates the conductivity and mobility of electrons.

To gain deeper insight into the reaction mechanism, the structure and frontier orbital of the catalyst were studied (Fig. 6a–d). The highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) represent the electron distribution states before and after light excitation, respectively.⁵² For C₃N₄, a uniform spatial distribution of electrons on the LUMO and holes on the HOMO was observed (Fig. 6a), indicating a high tendency for electron–hole recombination. In contrast, C₃N₅ exhibited a more uneven spatial charge distribution due to the formation of a dipole field that facilitated charge separation (Fig. 6b). With the presence of iron SAs, Fe-C₃N₄ displays a stronger electron distribution asymmetry, confirming the role of iron atoms in electron accumulation and accelerated charge separation. However, Fe SAs in C₃N₅ did not lead to a more asymmetric charge distribution (Fig. 6c and d). In other words, Fe-C₃N₅ exhibits a more symmetric electron distribution than Fe-C₃N₄, indicating more severe carrier recombination and consequently lower photocatalytic ozonation activity. These findings reveal that the CN support structure influences the role of the iron SAs in charge carrier dynamics by regulating dipole moment. It is noted that charge density is concentrated around iron SAs in both Fe-C₃N₄ and Fe-C₃N₅, implying that effective interfacial charge transfer was primarily eased via the formation of strong Fe–N coordination bonds.

Fig. 6 Top and side perspectives of the calculated charge distribution for the HOMO and LUMO of (a) C₃N₄, (b) C₃N₅, (c) Fe-C₃N₄, and (d) Fe-C₃N₅. The yellow color highlights electron accumulation at an iso-surface value of 0.001 Å e⁻³. Each color is associated with a specific element: white represents carbon, blue stands for nitrogen, red for oxygen, and orange for iron.

Surface catalytic reaction

In the photocatalytic ozonation process, multiple reaction pathways occur because of the high reactivity of O₃ and the coexistence of O₃ and O₂, as shown in Fig. 7a.^27,53,54 Achieving high ˙OH production efficiency requires boosting the 1e⁻ ozone reduction reaction. The number of electrons transferred in the photocatalytic ozonation process was determined by rotating ring disc electrode analysis. The estimated electron transfer number (n) for all catalysts under light irradiation was close to 1 (Fig. 7b), indicating a dominant light-driven 1e⁻ reduction pathway.

Fig. 7 A schematic diagram illustrating the intermediates involved in the multielectron-transfer processes in photocatalytic ozonation. (a) The most efficient reaction for OH generation is the one-electron-reduction reaction; (b) the O₃ adsorption energy and electron transfer number of C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅; (c) the adsorption configuration of O₃ in C₃N₅ and Fe-C₃N₅ from top and side views; (d) the calculated Fermi levels (E_f) of C₃N₄, C₃N₅, Fe-C₃N₄, and Fe-C₃N₅.

To further understand this behavior, DFT calculation was conducted to gain insight into the adsorption of O₃ on the surface of the catalysts (Fig. 7c). The Fermi level (E_f) of C₃N₅ shifts positively to −4.22 eV compared to that of C₃N₄ (−5.05 eV), probably due to the presence of the electron-rich triazole groups. When introducing iron SAs, the E_f shows a large upward shift (Fig. 7d). Since a more positive Fermi level contributes to electron migration to the catalyst surface, C₃N₅ enables more efficient interfacial electron transfer than C₃N₄, as reflected by the higher energy of O₃ adsorption (−0.27 eV for C₃N₅ and −0.16 eV for C₃N₄) (Fig. 7b). Similarly, the incorporation of iron SAs further increases the Fermi level (Fig. 7d) and provides active sites, enhancing O₃ adsorption energy on both Fe-C₃N₄ and Fe-C₃N₅. Their less negative Fermi level promotes electron transfer towards O, facilitating the formation of an Fe–O bond (Fig. 7c). As a result, the photogenerated charges reach the surface more readily, making them more accessible to be captured by O₃ to generate ˙OH, and thus improving the photocatalytic ozonation performance of Fe-C₃N₄ and Fe-C₃N₅.

A systematic catalytic mechanism in the photocatalytic ozonation process was proposed based on the above characterization and analysis. For the CN supports, C₃N₅ exhibits a narrower band gap, a stronger dipole field that accelerates charge carrier separation and transfer, and stronger O₃ adsorption capability than C₃N₄. Consequently, better photocatalytic ozonation performance was observed for C₃N₅ compared with C₃N₄. Upon loading iron, the separation and migration of bulk charge carriers in C₃N₄ are enhanced, while those are weakened in C₃N₅ due to the distinct manipulation of the dipole moment. Additionally, photo-generated electrons are concentrated around iron SAs in both Fe-C₃N₄ and Fe-C₃N₅, and effective interfacial charge transfer occurs via the formation of strong Fe–N coordination bonds, leading to stronger interaction with O₃ and accelerated photocatalytic reactions. Therefore, despite more charge carrier recombination and a similar E_f in Fe-C₃N₅ compared to C₃N₅, the enhanced interfacial electron transfer contributed to superior O₃ adsorption and photocatalytic activity of Fe-C₃N₅. Meanwhile, Fe-C₃N₄, featuring both uneven charge distribution and efficient interfacial electron transfer, achieved the highest photocatalytic performance in ozonation.

Conclusions

The impact of the structural properties of a polymer-based support on the role of SAs was explored based on DFT predictions and experimental verification. Band structures and dipoles were calculated as key factors to identify high photoactivity, and the experiment verified excellent performance of the optimal sample determined by DFT calculation. Combining in-depth structural and optoelectronic characterization, the results revealed that the iron SAs play a distinct role in charge separation and transfer on different supports. The addition of iron SAs weakens the dipole field in C₃N₅ but strengthens it in C₃N₄. Consequently, the addition of iron SAs onto C₃N₄ leads to more even spatial distribution of electrons to accelerate the separation of photo-generated charge carriers, while on C₃N₅ it decreases the recombination of electrons and holes. This work underscores the impact of the structure of the light-harvesting unit on regulating the role of SAs in the photocatalytic mechanism and proposes foundational guidelines for designing active photocatalysts.

Data availability

The data supporting this article have been included as part of the ESI.†

Author contributions

H. C. and Y. X. supervised the research. J. W. Y. X. and H. C. conceived the ideas and designed the experiments. J. W., J. Q., Y. X., Z. W., S. L. and Y. W. performed the experiments, electrochemical measurements, materials characterization and data analysis. H. M. and R. C. performed the DFT calculation. J. W., H. M., R. C. and Y. X. wrote the manuscript. All authors discussed the experiments and commented on the manuscript.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work was financially supported by the National Natural Science Foundation of China (52470096 and 22478426) and the strategic priority research program of the Chinese Academy of Sciences (XDA0430105). It was carried out with the support of the 4B9A beamline at the Beijing Synchrotron Radiation Facility. This work was also supported by the Australian Research Council (ARC) Discovery Early Career Researcher Award (DE250100886) and computational resources provided by the Australian National Computational Infrastructure (NCI) under the National Computational Merit Allocation Scheme 2025 (NCMAS 2025).

References

1. D. Nguyen and T. Do, Adv. Mater. Interfaces, 2023, 10, 2201413.
2. P. Chen, M. Zhou, Y. Liu, B. Li, C. Chen, X. Duan and Y. Wang, Appl. Catal., B, 2024, 346, 123767.
3. H. Su, H. Yin, R. Wang, Y. Wang, W. Orbell, Y. Peng and J. Li, Appl. Catal., B, 2024, 348, 123683.
4. P. Zhou, M. Luo and S. Guo, Nat. Rev., 2022, 6, 823–838.
5. A. Wang, J. Li and T. Zhang, Nat. Rev., 2018, 2, 65–81.
6. S. Mitchell and J. Pérez-Ramírez, Nat. Commun., 2020, 11, 4302.
7. Y. Guo, M. Wang, Q. Zhu, D. Xiao and D. Ma, Nat. Catal., 2022, 5, 766–776.
8. M. Gawande, P. Fornasiero and R. R. Zbořil, ACS Catal., 2020, 10, 2231–2259.
9. C. Gao, J. Low, R. Long, T. Kong, J. Zhu and Y. Xiong, Chem. Rev., 2020, 120(21), 12175–12216.
10. A. Moutsiou, A. Olivati, L. Cipriano, A. Sivo, S. Collins, Q. Ramasse, I. Kwon, G. Liberto, M. Kanso, R. Wojcieszak, G. Pacchioni, A. Petrozza and G. Vilé, ACS Catal., 2025, 15, 5601–5613.
11. B. Singh, M. Gawande, A. Kute, R. Varma, P. Fornasiero, P. McNeice, R. Jagadeesh, M. Beller and R. Zbořil, Chem. Rev., 2021, 121, 13620–13697.
12. J. Ran, A. Talebian-Kiakalaieh and S. Qiao, Adv. Energy Mater., 2024, 2400650.
13. W. Ong, L. Tan, Y. Ng, S. Yong and S. Chai, Chem. Rev., 2016, 116, 7159–7329.
14. Y. Akinaga, T. Kawawaki, H. Kameko, Y. Yamazaki, K. Yamazaki, Y. Nakayasu, K. Kato, Y. Tanaka, A. Hanindriyo, M. Takagi, T. Shimazaki, M. Tachikawa, A. Yamakata and Y. Negishi, Adv. Funct. Mater., 2023, 33, 2303321.
15. Q. Luo, Y. Li, X. Huo, L. Li, Y. Song, S. Chen, H. Lin and N. Wang, Adv. Sci., 2022, 9, 2105346.
16. Z. Chen, S. Mitchell, E. Vorobyeva, R. Leary, R. Hauert, T. Furnival, Q. Ramass, J. Thomas, P. Midgley, D. Dontsova, M. Antonietti, S. Pogodian, N. Lopez and J. Perez-Ramirez, Adv. Funct. Mater., 2017, 27, 1605785.
17. G. Mane, S. Talapaneni, K. Lakhi, H. Ilbeygi, U. Ravon, K. Al-Bahily, T. Mori, D. Park and A. Vinu, Angew. Chem., Int. Ed., 2017, 56, 8481–8485.
18. Z. Li, Y. Zhou, Y. Zhou, K. Wang, Y. Yun, S. Chen, W. Jiao, L. Chen, B. Zou and M. Zhu, Nat. Commun., 2023, 14, 5742.
19. J. Zhang, B. Jing, Z. Tang, Z. Ao, D. Xia, M. Zhu and S. Wang, Appl. Catal., B, 2021, 289, 120023.
20. C. Martine-Huitle and L. Andrade, Quim. Nova, 2011, 34, 850–858.
21. Y. Tan, W. Chen, G. Liao, X. Li, J. Wang, Y. Tang and L. Li, Appl. Catal., B, 2022, 306, 121133.
22. M. Li, P. Wang, Z. Ji, Z. Zhou, Y. Xia, Y. Li and S. Zhan, Appl. Catal., B, 2021, 289, 120020.
23. J. Lin, W. Tian, Z. Guan, H. Zhang, X. Duan, H. Wang, H. Sun, Y. Fang, Y. Huang and S. Wang, Adv. Funct. Mater., 2022, 32, 2201743.
24. J. Xiao, Y. Xie, J. Rabeah, A. Bruckner and H. Cao, Acc. Chem. Res., 2020, 53, 1024–1033.
25. G. Yu, Y. Wang, H. Cao, H. Zhao and Y. Xie, Environ. Sci. Technol., 2020, 54, 5931–5946.
26. J. Xiao, Q. Han, H. Cao, J. Rabeah, J. Yang, Z. Guo, L. Zhou, Y. Xie and A. Bruckner, ACS Catal., 2019, 9, 8852–8861.
27. J. Wang, G. Yu, Y. Wang, S. Liu, J. Qiu, Y. Xu, Z. Wang, Y. Wang, Y. Xie and H. Cao, Adv. Funct. Mater., 2023, 33, 2215245.
28. X. Li, H. Mai, J. Lu, X. Wen, T. Le, S. Russo, D. Winkler, D. Chen and R. Caruso, Angew. Chem., Int. Ed., 2023, 62, e202315002.
29. H. Mai, D. Chen, Y. Tachibana, H. Suzuki, R. Abe and R. Caruso, Chem. Soc. Rev., 2021, 50, 13692–13729.
30. J. Wang, J. Qiu, Z. Wang, Z. Zhuang, S. Liu, Y. Wang, Y. Weng, D. Wang, Y. Xie and H. Cao, J. Catal., 2024, 439, 115772.
31. Z. Liu, C. Zhang, L. Liu, T. Zhang, J. Wang, R. Wang, T. Du, C. Yang, L. Zhang, L. Xie, W. Zhu, T. Yue and J. Wang, Adv. Mater., 2021, 33, 2104099.
32. T. Wang, C. Hsiao, S. Chen, Y. Cheng and L. Chen, Polyhedron, 2015, 102, 216–223.
33. Y. Zhao, P. Zhang, Z. Yang, L. Li, J. Gao, S. Chen, T. Xie, C. Diao, S. Xi, B. Xiao, C. Hu and C. Choi, Nat. Commun., 2021, 12, 3701.
34. Y. Li, Z. Wang, T. Xia, H. Ju, K. Zhang, R. Long, Q. Xu, C. Wang, L. Song, J. Zhu, J. Jiang and Y. Xiong, Adv. Mater., 2016, 28, 6959–6965.
35. I. Kim, S. Kim, X. Jin, S. Premkumar, G. Chandra, N. Lee, G. Mane, S. Hwang, S. Umapathy and A. Vinu, Angew. Chem., Int. Ed., 2018, 57, 17135–17140.
36. Q. Han, B. Wang, J. Gao and L. Qu, Angew. Chem., Int. Ed., 2016, 55, 10849–10853.
37. C. Wu, G. Yu, Y. Yin, Y. Wang, L. Chen, Q. Han, J. Tang and B. Wang, Small, 2020, 16, 2003162.
38. P. Faria, J. Orfao and M. Pereira, Appl. Catal., B, 2008, 79, 237–243.
39. Z. Zhang, P. Duan, J. Zheng, Y. Xie, C. Bai, Y. Sun, X. Chen, F. Chen and H. Yu, Nat. Commun., 2025, 16, 115.
40. X. Zhu, Y. Wang, X. Duan, P. Wang, S. Zhong, S. Ren, X. Xu, B. Gao, J. Vongsvivut and S. Wang, Adv. Mater., 2024, 36, 2401454.
41. C. Wu, G. Yu, Y. Yin, Y. Wang, L. Chen, Q. Han, J. Tang and B. Wang, Small, 2020, 16, 2003162.
42. B. Choudhury, K. Pual, D. Sanyal, A. Hazarika and P. Giri, J. Phys. Chem. C, 2018, 122, 9209–9219.
43. S. Cao, H. Li, T. Tong, H. Chen, A. Yu, J. Yu and H. Chen, Adv. Funct. Mater., 2018, 28, 1802169.
44. P. Kumar, E. Vahidzadeh, U. Thakur, P. Kar, K. Alam, A. Goswami, N. Mahdi, K. Cui, G. Bernard, V. Michaelis and K. Shankar, J. Am. Chem. Soc., 2019, 141, 5415–5436.
45. R. Kuriki, H. Matsunaga, T. Nakashima, K. Wada, A. Yamakata, O. Ishitani and K. Maeda, J. Am. Chem. Soc., 2016, 138, 5159–5170.
46. A. Yamakata, T. Ishibashi and H. Onishi, Phys. Lett., 2001, 333, 271–277.
47. A. EI-Zohry, B. Turedi, A. Alsalloum, P. Maity, O. Bakr, B. Ooi and O. Mohammed, Commun. Chem., 2022, 5, 67.
48. V. Nikolaou, C. Govind, E. Balanikas, J. Bharti, S. Diring, E. Vauthey, M. Robert and F. Odobel, Angew. Chem., Int. Ed., 2024, 63, e202318299.
49. R. Godin, Y. Wang, M. Zwijnenburg, J. Tang and J. Durrant, J. Am. Chem. Soc., 2017, 139, 5216–5224.
50. L. Lin, Y. Ma, J. Vequizo, M. Nakabayashi, C. Gu, X. Tao, H. Yoshida, Y. Pihosh, Y. Nishina, A. Yamakata, N. Shibata, T. Hisatomi, T. Takata and K. Domen, Nat. Commun., 2024, 15, 397.
51. Z. Zhang, L. Ren, H. Li, D. Jiang, Y. Fang, H. Du, G. Xu, C. Zhu, H. Li, Z. Lu and Y. Yuan, Small, 2023, 19, 2207173.
52. S. Gumber, S. Agrawal and O. Prezhdo, J. Phys. Chem. Lett., 2022, 13, 1033–1041.
53. Y. Tan, W. Chen, G. Liao, X. Li, J. Wang, J. Wang, Y. Tang and L. Li, Appl. Catal., B, 2024, 351, 124005.
54. H. Cao, J. Wang, J. Kim, Z. Guo, J. Xiao, J. Yang, J. Chang, Y. Shi and Y. Xie, Appl. Catal., B, 2021, 296, 120362.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5sc02256h

---