In recent years, numerous research groups have worked on the photocatalysis for the environmentally friendly conversion of pollutants, including dyes, heavy metals, antibiotics, carbon dioxide, and nitric oxide, among others [1-5]. The photocatalysts in these reactions can be excited by incident light to produce charge carriers, which may transport to the surface of photocatalyst. Then, the generated charge carriers can react with external pollutants either directly or indirectly [6-8]. Among various photocatalysts, g-C₃N₄ attracted much attention from all over the world because it has many advantages, such as its low cost, stability, nontoxic properties, and easy preparation [9-11]. However, the photocatalytic activity of g-C₃N₄ is not high due to the fast recombination of photogenerated electron and hole pairs [12]. Various methods have been developed to improve the photocatalytic activity of g-C₃N₄, including element doping, self-doping, formation of a heterojunction, and sensitization [13-18]. Out of all of these modification methods, element doping is one that is simple and effective. In the operation process, researchers only need to add small doses of dopants in the raw materials of g-C₃N₄ without doing any other treatment [19]. As for improving the effectiveness of the materials, element doping can improve visible light absorption and the separation of photogenerated charge carriers by introducing an impurity level within the band gap of g-C₃N₄ [20].

In the past few years, scientists have doped various metallic and nonmetallic elements into the structure of g-C₃N₄. The doping methods of nonmetallic elements, such as B, P, and S, have been studied extensively. Previous studies have shown that nonmetallic elements generally substitute for the lattice atoms of g-C₃N₄ [15, 21]. However, the doping model of metal ions, such as Na⁺, Ni³⁺, Co³⁺, and Cu²⁺, still has not been studied clearly [13, 22-24]. Some previous works reported that metal ions are shown to embed in the in-planar cavities of g-C₃N₄, while others have suggested that metal ions insert into the interlayer of g-C₃N₄. As is known, different doping methods correspond to different effect mechanisms of metal ions for the photocatalytic performance of g-C₃N₄. Therefore, the unclear doping model of metal ions determines that the effect mechanisms of metal ions for the photocatalytic performance of g-C₃N₄ are also unclear [25]. Therefore, fully characterizing and understanding the doping model and effect mechanisms of metal ions on the photocatalytic performance of g-C₃N₄ is very important, despite being challenging.

As is known, the surface and edge of g-C₃N₄ contain a number of amino groups. Previous literatures have reported that metal ions can form coordination complexes with these amino groups [26-29]. Therefore, it is possible that the metal ions may bond with amino groups and form the coordination doping on the surface and edge of g-C₃N₄ structure. According to previous reports, amino groups can help with the transfer of photogenerated electrons. Meanwhile, metal ions can extract photogenerated electrons from semiconductors because of their lower work function [30, 31]. In this case, coordination doping may produce a synergistic effect existing between the amino groups and metal ions. This synergistic effect may further improve the transfer of photogenerated electrons and boost the photocatalytic activity of g-C₃N₄.

To verify the above theoretical analysis, ferric ion (Fe(III)) was chosen as a dopant because of the following reasons. First, iron is important to the human body, and is widely used in daily life and industrial processes. Second, iron is a variable valence metal and stably exists in the form of Fe(III), which is a common oxidant. Third, Fe(III) could capture the photogenerated electrons to be reduced to Fe(II), which triggers the Fenton reaction to produce more ^•OH [32]. After the Fenton reaction, Fe(II) can be recovered in the form of Fe(III). Therefore, the fourth reason is that Fe(III) can be used repeatedly without inactivation.

Recently, we prepared Fe(III)-doped g-C₃N₄ (CNFe_x) through the coordination between amidogen and Fe(III) ions. After activity tests, it was found that the coordination doping of Fe(III) could enhance the Rhodamine B (RhB) and Cr(VI) removal activity of g-C₃N₄, yet is not helpful for the NO-removal performance of g-C₃N₄. Density functional theory (DFT) calculations and various characterizations were employed to investigate the effect of Fe(III) on the surface atomic structure of g-C₃N₄ and the removal of different pollutions (RhB, Cr(VI), and NO). The detailed structural information and photocatalytic activity of CNFe_x were then analyzed in detail.

All of the chemicals used in this study were analytical grade and used without further treatment.

g-C₃N₄ was synthesized by heating melamine directly. First, 4 g of melamine was placed in a 30-mL ceramic boat. Then, this ceramic boat was placed in a tube furnace and heated to 520 ℃ with a heating rate of 5 ℃/min. The calcination process was kept at 520 ℃ for 4 h with the protection of Argon gas (99.999%). After 4 h of calcination, the furnace was cooled to room temperature. Subsequently, it was found that the white melamine in the ceramic boat had changed to a yellow solid. This yellow solid was ground into powder and denoted as CN.

For comparison, samples of Fe doped g-C₃N₄ were also synthesized. Typically, 4 g of melamine was dispersed into 100 mL of methanol solution under continuous stirring. Then, a certain amount of Fe(NO₃)₃·9H₂O was also added to this suspension to form a mixture. This mixture was stirred continuously at 60 ℃ until the methanol was completely removed. The obtained solid was transferred in a 30 mL ceramic boat, which was then placed into a furnace for the next calcination. The calcination process was kept at 520 ℃ for 4 hours with the protection of Argon gas (99.999%). Finally, the prepared catalyst was denoted as CNFe_x (x = 0, 0.05, 0.25, 0.5, and 1), where x is the mass percent of doped iron to the total mass of the sample.

Powder X-ray diffraction (XRD) patterns were recorded on a Bruker D8 Advance diffractometer with monochromatized Cu K_a radiation (λ = 1.5418 Å ) to characterize the crystal phase structure. The morphology and structure of the samples were examined by transmission electron microscopy (TEM; JEM 2100). X-ray photoelectron spectroscopy (XPS) with Al K_α X-rays (Thermo Scientific Escalab 250) was applied to investigate the surface compositions of different samples. UV-Vis diffuse reflectance spectra of samples were obtained using an UV-Vis spectrometer (Thermo Scientific Evolution 220) to analyze the light absorption ability and band gap of different samples. The nitrogen adsorption and desorption isotherms were measured using an AUTOSORB-IQ-MP system to measure the surface area of different samples. Fourier-transform infrared spectra (FT-IR) were obtained on a (Nicolet iS50, Thermo) spectrometer using KBr as the reference sample to reveal the existence of chemical bonds. Electron spin resonance (ESR) signals were recorded on an E500-10/12 spectrometer at room temperature to detect defects or free radicals.

To study the oxidation and reduction ability of final samples, we chose RhB, Cr(VI), and NO as model pollutants.

The RhB- and Cr(VI)-removal experiments were conducted in a 100 mL reaction cell with an external cooling water jacket, which ensured that the test temperature was kept at room temperature. Typically, 50 mg of the as-prepared sample was dispersed in 50 mL (10 mg/L) RhB or K₂Cr₂O₇ solution with constant stirring for 1 h in the dark to establish an adsorption-desorption balance. After the adsorption-desorption balance, a 300 W Xe lamp with a 420 nm cutoff filter was chosen as the visible light source. During the visible light irradiation, 5 mL of the suspension was taken out from the reaction cell and to test the concentration of RhB or Cr(VI) at different times. The concentration of RhB was analyzed directly by UV-vis spectroscopy (UV-1800) at λ = 552 nm. The Cr(VI) concentration analysis was conducted using the diphenylcarbazide spectrophotometric method [33].

The photocatalytic NO-removal experiments were performed in a continuous flow cylindrical reactor (V = 0.785 L, R = 5 cm, and H = 10 cm), which was made from glass with a quartz skylight. Before experiments, 50 mg of as-prepared sample was well-dispersed in a glass sample dish (R= 3 cm). In the process of NO-removal, the sample dish was placed in the middle of the reactor. Meanwhile, 600 ppb of NO gas (balanced with air) was introduced into the reactor continuously. The visible light, which was generated from a 300 W Xe lamp, irradiated the sample dish continuously. The concentration of NO was monitored by a NOx analyzer (Thermal 42i).

DFT calculations were carried out with the Cambridge Sequential Total Energy Package (CASTEP) program in Materials Studio 7.0. In the calculation, the Generalized Gradient Approximation (GGA), as well as the Perdew-Burke-Ernzerh of functional (PBE), was deployed. For the sake of accuracy, the cutoff was set to 450 eV and 5 × 5 × 3 k points with the Monkhorst-Pack scheme in the first Brillouin zone was employed. All of the structure models were fully relaxed, with the convergence criteria for geometric optimization and energy calculation set to 1.0 × 10^‒5 eV per atom, 0.05 GPa, 1.0× 10^‒6 eV per atom for maximum force, maximum stress, and the tolerance of energy. We first relaxed a 1 × 1 × 1 super-cell of bulk g-C₃N₄. Then, a single Fe atom was introduced into the triangular pores in the 2 × 1 × 1 and 2 × 2 × 1 optimized super-cells of g-C₃N₄.

To evaluate the photocatalytic oxidation activity of the synthesized materials, RhB was chosen as the model pollutant because the removal of RhB is mainly due to the oxidation of photogenerated holes [34]. Fig. 1(a) and Fig. 1(b) shows that all of the samples could absorb about 5% of RhB when the adsorption-desorption balance was reached. After visible light irradiation for 60 min, the removal efficiencies of RhB over CN, CNFe_0.05, CNFe_0.25, CNFe_0.5, and CNFe₁.₀ were approximately 30%, 61%, 87%, 30%, and 29%, respectively. The degradation rate of CNFe_0.25 is nearly two times greater than that of g-C₃N₄ (Fig. 1(c)).

In addition to the photo-oxidation capacity, the photo-reduction capacities of different samples were also compared by investigating the Cr(VI)-removal behaviors. As shown in Fig. 1(d), after irradiation with visible light for 120 min, Cr(VI) can be removed quickly by the modified g-C₃N₄, while almost no Cr(VI)-removal occurred in the CN system. Among different samples, CNFe_0.5 displayed the best Cr(VI) reduction performance. The Cr(VI)-removal rate of CNFe_0.5 is about 16 times the Cr(VI)-removal rate of CN (Fig. 1(e)). These experiments reveal that coordination doping with Fe (III) could improve the photocatalytic oxidation and reduction performances of g-C₃N₄.

Since both photocatalytic oxidation and reduction performances of g-C₃N₄ could be improved by coordination doping of Fe(III), it is reasonable to consider that the coordination doping of Fe(III) would also improve the NO-removal performance of g-C₃N₄. However, Fig. 1(f) surprisingly shows that the NO-removal rate of g-C₃N₄ dropped with increasing doping amounts of Fe(III). The reasons for this will be discussed later in the text.

XRD was applied to characterize the phase structure of the final samples. From Fig. 2(a), the XRD pattern of g-C₃N₄ shows two distinct peaks at 13.0° and 27.4°. These two peaks correspond to the (100) and (002) crystal planes of g-C₃N₄, respectively[9]. Notably, no new diffraction peaks that relate to the compounds of Fe can be found in the XRD patterns of the CNFe_x samples. However, the existence of iron can be directly proven by XPS. As shown in Fig. 2(b), carbon, nitrogen, and oxygen could be found in the survey spectra of both CN and CNFe_0.5. Additionally, a Fe 2p peak located at 710.2 eV can be found in the high-resolution XPS spectrum of CNFe_0.5, but not in that of CN, indicating that the Fe element exists in the framework of g-C₃N₄ (Fig. 2(c)). These XRD and XPS results suggest that elemental iron was doped into the CNFe_x structure, but not in the form of some compound. If this had been the case, the XRD peak intensities for CNFe_xwould be smaller than that of CN, since the doping of iron would destroy the framework of g-C₃N₄. Compared with the XRD peaks of CN, the peaks of CNFe_x become weaker and broader, confirming the iron was doped into the frame of g-C₃N₄.

To investigate the distribution of elemental iron in the framework of g-C₃N₄, EDS mapping of Fe was measured based on SEM image. As shown in Fig. 3(a), the morphology of CNFe_0.5 is mainly the bulk particles with different sizes. Fig. 3(b) and 3(c) shows that C and N are uniformly distributed in the framework of CNFe_0.5. However, the Fe is mainly distributed at the marginal of the CNFe_0.5 framework (Fig. 3(d)). From the structure diagram of g-C₃N₄, the marginal of g-C₃N₄ consists of many amidogens. Since coordination bonds can be formed between amidogen and Fe(III), the doping of elemental Fe into g-C₃N₄ may be coordination doping. This speculation can be proven by FTIR experiments. From Fig. 3(e), the bands in the range of 1250‒1651 cm^‒1 originated from the stretching vibration signals of the sp³ C-N bonds and typical sp² C=N stretching modes. The peaks at 808 and 3168 cm^-1 can be assigned to the deformation mode of N-H or N=H in amino groups respectively [35]. It is worth noting that these two peaks decrease with increasing Fe-doping amounts. Since amino groups can coordinate with Fe ions and form coordination complexes, the decrease of amino groups is parallel with the EDS mapping of element Fe, further confirming that the doping of Fe in g-C₃N₄ is mainly the coordination doping.

A Brunauer-Emmett-Teller (BET) instrument was applied to measure the specific surface area of the final samples. Fig. 3(f) shows the nitrogen adsorption-desorption isotherms of different samples. The BET surface areas of CN, CNFe_0.05, CNFe_0.25, CNFe_0.5, and CNFe₁.₀ were calculated to be 4.895, 3.404, 5.400, 4.235, and 5.131 m²/g, respectively. Obviously, the change of BET surface areas is very small, suggesting the excellent photocatalytic activity of Fe_0.25-g-C₃N₄ is not caused by the change of BET surface areas.

The photocatalytic activity of a photocatalyst is dependent on its light absorption ability and its band structure. As shown in Fig. 4(a), the UV-vis absorption spectra of CN and CNFe_x are different.

The Fe-doping can improve the light absorption ability of g-C₃N₄ in both the ultraviolet and visible light ranges. Meanwhile, the UV-vis absorption edges of doped samples are red-shifted when they are compared with that of the pure sample. Both the improvement of light absorption and the red-shift effect of the absorption edge are increased with increased doping amounts. From a plot of (ahν)² versus photon energy (hν), the band gaps (E_g) of CN, CNFe_0.05, CNFe_0.25, CNFe_0.5, and CNFe₁ were estimated to be 2.80, 2.78, 2.76, 2.66, and 2.66 eV, respectively (Fig. 4(b)). This observation suggests that the Fe-doping can narrow the band gap of g-C₃N₄ and, thus, leads to improvement of light absorption ability [36].

To find out the origin of the phenomenon that Fe-doping can narrow the band gap of g-C₃N₄, DFT calculations were applied. After the optimization of built models, we found that the Fe-doping made the crystal cell become bigger (Table 1). The calculation results of the band structures show that Fe-doping led to the formation of impurity bands in the bandgap, which narrowed the band gap for more light absorption. From Fig. 4(c) and 4(d), the downward shift of the conduction band (CB) and valance band (VB) implied that the oxidizing ability of photogenerated holes in CNFe_x becomes stronger, while the reducing ability of photogenerated electrons becomes weaker. Referring to the DOS as shown in Fig. 5(a), the top of the VB in both CN and CNFe_x are mainly composed of C 2p. However, the bottom of the CB in CN and CNFe_x are different. In CN, the bottom of the CB is mainly composed of N 2p, while the bottom of the CB in CNFe_x is mainly composed of Fe 2(d). Therefore, in the photocatalytic process of CNFe_x, the electrons are generated from C atoms and excited into Fe atoms. The transfer of photogenerated electrons from C atoms to Fe atoms can also be reflected in the calculation results of the electron density differences. Fig. 5(b) shows the electron density differences of CNFe_x, where the blue color corresponds to a bigger electron density and the red color shows a lower electron density. In the image, it can be seen that the electron density around Fe is smaller than that of C and N. In this case, an electric field (EF) can be formed between N/C and Fe, the direction of which is from Fe to N/C. Under the pressure of this EF, the photogenerated electrons will transfer from C/N atoms to Fe atoms. Based on DFT calculations, Fe-doping not only can narrow the band gap of g-C₃N₄, but can also improve the transfer of photogenerated electrons.

Photoluminescence (PL) spectra were applied to further investigate the separation efficiency of photo-generated electrons and holes, as displayed in Fig. 6(a). A strong emission peak in CN appears at about 452 nm on account of the direct combination of photogenerated electrons and holes [37]. When Fe²⁺ was doped into g-C₃N₄, the intensity of this peak was inhibited significantly, confirming that Fe²⁺-doping can improve the transfer of photo-generated electrons. More interestingly, a new long-wave emission peak at around 442 nm appeared in the PL spectrum of CNFe_0.5. This new peak corresponds to the electron excitation from valance band to the impurity bands of Fe²⁺, which can explain the light absorption of CNFe_0.5 in the range of 450–700 nm.

To analyze the VB/CB potential of CNFe_x, valence-band XPS (VB-XPS) data were examined. From Fig. 6(b), the VB potential of CN and CNFe_0.5 are 1.64 and 1.87 V versus NHE, respectively. Then, according to the formula E_g = E_v – E_c (E_g: band gap, E_v: VB potential, E_c: CB potential), the CB potentials of CN and CNFe_0.5 were calculated to be -1.16 and -0.79 V, respectively. As shown in Fig. 6(c), Fe-doping could shift the VB potential and CB potential negatively, consistent with the results of the DFT calculations. Although the CB position is decrease, the photo-reduction Cr(VI) activity is improved. This is because that the reduction potential is not the only determinant for the Cr(VI) reduction. The amount of actives sites, carrier recombination rate and surface area also affect the reductive reaction of Cr(VI).

As is known, the generation of active species is particularly sensitive to the potential of VB and CB. We therefore measured the active species that may be generated in the photocatalysis process. First, photocurrent experiments indicated that electrons can be generated in both CN and CNFe_x systems under the visible light irradiation. Fig. 7(a) shows that the photocurrent signal of CNFe_0.5 is much stronger than that of CN, indicating CNFe_x can produce more photogenerated electrons and holes than CN. Photogenerated carriers may produce ^•O₂⁻ and ^•OH through the redox reaction. Therefore, the DMPO spin-trapping ESR technique was used to detect ^•O₂⁻ and ^•OH generated in different sample systems. As shown in Fig. 7, no ^•OH and ^•O₂^‒ signals can be detected in either the CN or CNFe_0.5 systems in darkness. Under the light irradiation, six characteristic peaks of DMPO-

^•O₂⁻ can be detected in CN systems (Fig. 7(b)). However, there are no ^•OH signals found in this system (Fig. 7(c)). As for CNFe_x systems, both ^•O₂⁻ and ^•OH can be measured. Meanwhile, the DMPO-^•O₂⁻ signal in the CNFe_0.5 system is much larger than that in the CN system. These phenomena suggest Fe-doping not only enhance the yield of ^•O₂⁻, but also opens the function of

^•OH generation. It is well known that the Fenton reaction (Fe²⁺ + H₂O₂ → Fe³⁺ + ^•OH + OH^‒) between Fe²⁺ and H₂O₂ can produce ^•OH effectively. Therefore, we believe that the Fe²⁺-induced Fenton reaction is the origin of ^•OH in CNFe_x system.

As is known, ^•OH is a very powerful oxidizer. It can oxidize and decompose nearly all organic contaminants. Previous literature also reported that NO can be oxidized into NO₃^‒ by ^•OH. The EPR and VB-XPS showed that Fe-doping not only enhances the yield of ^•OH, but also makes the VB of g-C₃N₄ more positive. Therefore, Fe-doping should improve the NO-removal activity of g-C₃N₄. However, the NO-removal activity of g-C₃N₄ dropped with the increase of Fe-doping amounts. Meanwhile, both the RhB photo-oxidation and Cr(VI) photo-reduction have been improved by the doping of Fe(II), suggesting the decrease of NO-removal activity has nothing to do with the ability to produce high quality carriers. Therefore, we believe that the decrease of NO-removal activity is caused by the decrease of NO adsorption, which can be confirmed by NO-TPD (Fig. 8(a)).

As mentioned earlier, the doping of elemental Fe into g-C₃N₄ can be described as coordination doping, which forms through the coordination between amidogen and Fe(III). Thus, the amino groups are occupied by Fe(III). Larger amounts of Fe(III) doping will induce a smaller surplus of amino groups. Therefore, we suppose that the decrease of NO adsorption over CNFe_x is related to the surface amino group. To confirm this supposition, we prepared two kinds of samples, g-C₃N₄ with fewer surface aminos by deaminization and g-C₃N₄ with surface carbon vacancies, which can expose more surface aminos. Fig. 8(b) shows that the amino group in g-C₃N₄ is significantly decreased after the deaminization. Meanwhile, the calcination under Ar atmosphere makes the amount of amino group in g-C₃N₄ increase. Then, we conducted NO removal experiments with them. However, the results show that the effect of NO removal does not correlate with the number of amino groups (Fig. 8(c)), suggesting that the adsorption sites of NO are not N atoms, but C atoms. This conclusion is reasonable in theory because both the C atom in g-C₃N₄ and the N atom in NO have unpaired electron. Since two unpaired electrons are easier to form covalent bonds with, NO would easily adsorb on the C atom of g-C₃N₄. Therefore, we speculate that C atom may be passivated and, thus, cause the decrease of NO adsorption. To test this speculation, EPR spectrometry was employed to analyze the change of unpaired electrons of the samples. As shown in Fig. 9(a), CN exhibits a strong Lorentzian line that is mainly due to the unpaired electrons on the carbon atoms of the aromatic rings [38]. After the introduction of Fe(III), the intensity of this Lorentzian line decreased about 2/3, suggesting the amount unpaired electrons on the carbon atoms have deceased. In other words, some C atoms in CNFe_x have been passivated after the introduction of Fe(II).

To further prove why Fe-doping causes the decrease of NO adsorption, solid-state nuclear magnetism (NMR) spectra of CN and CNFe_0.5 were studied. As shown in Fig. 9(b), the ¹³C spectra of both samples display two peaks at about 155 and 163.8 ppm, respectively. This phenomenon suggests two kinds of C exist in CN and CNFe_0.5 samples. The peak at around 155 ppm can be assigned to C-N3 (C2), while the peak at 163.8 ppm is ascribed to C-NH_x (C1) [39]. Although both samples contain C1 and C2, the ratios of C1 to C2 in the two samples are different. In CN, the ratio of C1 to C2 is 0.57. However, this ratio in CNFe_0.5 decreased to 0.53, indicating the number of C-NH_x in CNFe_0.5 is fewer than that in CN. This decrease is due to the coordination between Fe and amino groups. Fig. 9(c) shows the ¹H spectra of two samples. The ¹H spectrum of CN shows three peaks at about 8.3, 3.2, and 0.41 ppm, which correspond to N-H, N-H₂, and C-H, respectively. Among these three peaks, N-H₂ possesses the highest intensity, while C-H shows the least intensity. After the doping of Fe, the ¹H spectrum of CNFe_0.5 also shows three peaks at the same positions as that of CN. However, the intensity of these peaks has changed dramatically. After the doping of Fe, the intensity of N-H₂ has become lower, while the intensities of N-H and C-H have become stronger. This phenomenon suggests that Fe-doping makes N-H₂ give one H to the C atom of the triazine ring. It is well established that NO could easily adsorb on the C atom of g-C₃N₄ through free radical bonding. However, a single electron of the C atom was consumed by H after the doping of Fe. Therefore, Fe-doping can decrease the NO adsorption and, thus, inhibit the photocatalytic NO-removal.

In this study, we have demonstrated that the interaction with Fe(III) can be described as coordination doping with amino groups in CNFe_x, which has a great effect on photocatalytic oxidation and reduction. On the one hand, the RhB removal rate was enhanced because more ^•OH was produced by the Fenton reaction. On the other hand, the Cr(VI)-removal efficiency was improved due to high yield of photogenerated electrons and ^•O₂⁻. Furthermore, we studied the mechanism of poor NO removal rate caused by the Fe(III) doping. As a result, we found that the carbon of the triazine ring was the active site of NO removal, but it was passivated by the doping of Fe(III), thus reducing the NO removal activity of g-C₃N₄.