In the fields of renewable energy development and environmental remediation, the conversion of solar energy into clean chemical fuels by semiconductor photocatalysts has been a popular strategy worldwide [1-7]. Various kinds of semiconductors, including oxides, sulfides, and nitrides, have been investigated as active photocatalysts in the past decades [8-13]. However, the greatest restriction on their practical application is currently the low energy conversion efficiency of photocatalysis. One possible solution is to develop highly efficient photocatalytic materials that can trigger hydrogen production in the visible-light region.

Recently, intense interest has been focused on the metal-free semiconductor graphitic carbon nitride (g-C₃N₄) [14, 15]. This material exhibits potentially desirable photocatalytic activity because of its band gap of roughly 2.7 eV, allowing visible absorption, and the distinct redox ability of the photoinduced electrons and holes [16, 17]. Moreover, polymeric g-C₃N₄ possesses extended π-conjugated layer networks, and the convenient thermal polymerization synthesis allows for its large-scale preparation and structural modification [18, 19]. Nevertheless, g-C₃N₄ still shows low activity for photocatalytic H₂ evolution, suffering from the limited ability to utilize sunlight below 450 nm and high recombination rate of photoinduced electron-hole pairs [20, 21]. To date, various modification strategies have been explored to optimize this promising material for better performance [22-31]. Among them, band gap engineering through element doping has proven an effective approach to improve the photocatalytic activity of g-C₃N₄. Introducing impurities at the atomic level has been demonstrated to extend the light absorption region, modify the band structure, and generate electron/hole traps for charge separation [32-37]. In particular, Li, Na, and K have been recognized as effective doping additives into the inner structure of g-C₃N₄ [38-41]. However, few comparison studies have investigated the differences among g-C₃N₄ photocatalysts doped with various alkali metals, which has limited our understanding of their effects.

Herein, we present a comparison study of g-C₃N₄ doped with different alkali metal ions (Li⁺, Na⁺, and K⁺) for photocatalytic H₂ evolution. Detailed analysis of optical absorption, intrinsic band structure, and charge transfer are carried out to investigate the positive effects of doping metal ions in g-C₃N₄ matrices for visible-light photocatalytic activity.

Pristine g-C₃N₄ was prepared in the conventional manner by heating melamine precursor at 550 ℃ for 2 h in air with a heating rate of 5 ℃ min^-1, and denoted as CN. To synthesize alkali metal-doped g-C₃N₄, the obtained g-C₃N₄ powder (500 mg) was dispersed in 15 mL deionized water containing 15 mg MOH (where M represents Li, Na, or K). Then the mixture was stirred for 1 h and heated to 100 ℃ to remove water. The uniformly mixed solid was further ground and calcined at 600 ℃ for 1 h under N₂ atmosphere in a tube furnace, followed by washing with diluted HCl solution to remove the excess alkali metal ions and washing with deionized water several times until pH = 7. The resulting products were labeled as Li-CN, Na-CN, and K-CN, respectively. For comparison, the as-prepared g-C₃N₄ was also calcined without adding any MOH, and the as-obtained sample was denoted as C-CN.

The doping amounts of alkali metal ions in the products were measured by inductively coupled plasma atomic emission spectrometry (ICP-AES, Optima 4300 DV, Perkin Elmer). The X-ray diffraction (XRD) patterns were recorded on a D/Max-RB X-ray diffractometer with Cu-K_α radiation at a slow-scan rate of 0.005° (2θ) s^-1. Scanning electron microscopy (SEM) and elemental mapping images were obtained on a scanning electron microscope (FESEM, JSM 7500F). UV-visible diffuse reflectance spectra were investigated on a UV-vis spectrophotometer (UV2600, Shimadzu) using a BaSO₄ reference. Nitrogen adsorption-desorption isotherms were analyzed by a BET analyzer (Micromeritics ASAP 3020, USA). Chemical states and valence band spectra were collected by X-ray photoelectron spectroscopy (XPS, VG ESCALAB 210). An Al-K_α excitation source was employed and the C 1s peaks (284.8 eV) were referenced for all binding energies. Photoluminescence (PL) emission spectra were examined on a Hitachi F-7000 fluorescence spectrophotometer under the excitation wavelength of 350 nm. Both the excitation and emission slits were 5.0 nm in width and the voltage was 400 V. Time-resolved photoluminescence (TRPL) decay curves were measured on a fluorescence lifetime spectrophotometer (FLS920, Edinburgh Instruments, UK) excited at 360 nm.

Photocatalytic H₂ production reactions were performed at room temperature in a three-neck vessel, of which the three openings were sealed with rubber plugs and tubes. A 350 W Xe lamp with a UV cutoff filter (λ > 400 nm) was used as the light source. Typically, 50 mg photocatalyst powder was dispersed in 80 mL aqueous solution containing 10 vol% triethanolamine as scavenger. All samples were loaded with 1 wt% Pt by photodeposition for 30 min. The reaction system was evacuated for 30 min before light irradiation to ensure anaerobic conditions. The amount of evolved H₂ was detected using a GC-14C gas chromatograph (Shimadzu, Japan, TCD) with N₂ carrier gas and a 5 Å molecular sieve column.

Electrochemical impedance spectra (EIS) and photocurrent responses were tested on a CHI 660D electrochemical workstation (Chenhua Instruments, Shanghai, China) with a traditional three-electrode system. This system included the prepared sample as the working electrode, a Pt wire as the counter electrode, and a Ag/AgCl electrode as the reference electrode. A Na₂SO₄ aqueous solution (0.5 mol·L^-1) was used as the electrolyte and a low-power LED (3 W, 420 nm) was used as the visible-light source. The potential was set as 0.5 V for photocurrent measurement. The working electrode was prepared by grinding 50 mg photocatalyst, 20 mg polyethylene glycol (PEG, M_w= 20000), together with 2 mL ethanol to form a slurry. The obtained slurry was coated onto a 2 cm × 1.2 cm fluorine-doped tin oxide (FTO) glass substrate and then annealed at 450 ℃ for 30 min in a muffle furnace.

Fig. 1(a) displays the XRD patterns of g-C₃N₄ and modified g-C₃N₄. All g-C₃N₄ samples show two diffraction peaks at ca. 13.1° and 27.4°, which are typical of g-C₃N₄ graphitic stacking layers, and are similar for each sample. Specifically, the former peak corresponds to an in-plane structure repeating with the distance of the in-plane cavity between three tri-s-triazine units, i.e. 0.68 nm, while the latter (strong) peak is assigned to the interplanar stacking of g-C₃N₄ [42-44]. These peak intensities are weakened for the metal ion-doped g-C₃N₄, indicating a decreased crystallinity degree from Li-CN to K-CN. This implies that the in-plane structure may be partially fragmented into smaller crystallites in the doped samples. Such distortion of g-C₃N₄ usually leads to the formation of defects and band-structure changes. The slow-scan XRD patterns were then measured in the range of 25°-30° and are shown in Fig. 1(b). It is clearly seen that after heating at the higher temperature of 600 ℃, the dominant (002) peak of g-C₃N₄ shifts to the higher angle of 27.6°, which reflects a higher degree of stacking order and a narrower interlayer distance, resulting from the layer compaction of g-C₃N₄ [45, 46]. Meanwhile, in the XRD patterns of metal-doped g-C₃N₄, the (002) diffraction peak gradually shifts back toward a lower 2θ value, from 27.6° to 27.4°, with the successive introduction of Li, Na, and K. This indicates that successively larger interlayer distances are created by the interstitial doping of those three ions into the structure [47]. Clearly, larger atom sizes induce larger interlayer distances. It is probable that the metal ions are incorporated into nano-sized cavities, where they coordinate with the electron-donating N-containing functional groups of g-C₃N₄ [36]. The elemental mapping images of the samples, shown in Fig. 2, reveal the existence of C and N in all cases, as well as Na or K in the corresponding samples; however, the Li content in Li-CN is beyond the detection limit of this method. It can be seen that the metal ion species are uniformly distributed within the g-C₃N₄ matrix. The doping concentrations of Li, Na, and K measured by ICP-AES are 0.118, 0.052, and 0.059 mol% for Li-CN, Na-CN, and K-CN, respectively.

The BET specific surface areas and pore structures were also investigated to clarify the structural variation among the samples. From the nitrogen adsorption-desorption isotherms in Fig. 3, it can be seen that all the g-C₃N₄ photocatalysts show a type Ⅳ isotherm and H3 hysteresis loops, indicating the formation of slit-like mesopores from aggregated g-C₃N₄ plates. The data listed in Table 1 reveal that post-calcination g-C₃N₄ has a much higher specific surface area and larger pore volume than pristine g-C₃N₄, suggesting that high temperature increases the porosity of g-C₃N₄ [46]. However, the BET specific surface areas of Li-CN, Na-CN, and K-CN are 15, 28, and 21 m²g^-1, respectively, i.e. they are reduced compared with that of C-CN. This is possibly attributed to the treatment of the g-C₃N₄ surface with basic reagents (MOH) after heat treatment, resulting in the destruction and collapse of the pore structure [48]. Note that Na-CN shows the highest specific surface area among the metal-doped samples. Generally, a larger S_BET provides more active sites and facilitates light absorption for enhanced photocatalytic activity [49, 50].

UV-vis diffuse reflectance spectra of M-CN and pristine g-C₃N₄ were collected to investigate their light absorption properties. The corresponding band gaps were obtained by plotting (αhν)² against photon energy (hν) (Fig. 4(b)) and are listed in Table 1. Fig. 4(a) shows a steep rise in absorption intensity from 452 nm to the ultraviolet region for pristine g-C₃N₄, which represents a relatively narrow band gap of ~2.74 eV. In comparison, the absorption edge of C-CN is red-shifted and the band gap is calculated to be even smaller, 2.52 eV. The narrowed band gap results from the small-size effect of the decreased particle sizes as well as the increased degree of condensation for high-temperature-treated g-C₃N₄ [45]. An additional absorption shoulder also appears near 500 nm, which can be assigned to the n-π^* transitions occurring in a π-conjugated system [51]. This is another effect of the high-temperature treatment: the increased layer condensation causes structural distortions, which increases the number of lone-pair electrons on the N-containing groups at the edges of localized regions of g-C₃N₄. These N-containing groups with high electron density are basic in nature and behave as active sites to accommodate metal ions by electrostatic interaction or complexation [51-53]. Hence, the post-treatment at high temperature facilitates the metal doping process. Moreover, the absorption edges of the alkali metal-doped samples exhibit even greater shifts to longer wavelengths and the corresponding band gaps decrease to 2.48, 2.42, and 2.39 eV for Li-CN, Na-CN, and K-CN, respectively. This result demonstrates that alkali metal ions were successfully doped into the g-C₃N₄ matrix with the aid of high-temperature treatment. The resulting decreased band gaps should endow the M-CN catalysts with a better ability to harvest visible light, thus increasing their photocatalytic activities.

The chemical states of C and N and the presence of dopant elements in the g-C₃N₄ samples were analyzed by XPS, as shown in Fig. 5. The two dominant peaks at 284.8 and 287.9 eV in the C 1s region correspond to typical C-C graphitic carbon and sp²-hybridized C atoms (N=C-N) in the g-C₃N₄ aromatic structure, respectively. The high-resolution N 1s XPS of all g-C₃N₄ samples can be fitted to three characteristic peaks located at 398.3, 399.5, and 400.8 eV. They are assigned to sp²-hybridized N atoms (C=N-C), N-(C)₃ in the tri-s-triazine rings, and terminal amino functional groups (NH₂ and NH), respectively [54, 55]. All the C 1s and N 1s spectra are in accordance with the typical chemical states and bonding forms of C and N in g-C₃N₄. Note that no obvious shift of the C 1s and N 1s binding energies is observed for the C-CN sample in comparison with pristine g-C₃N₄. However, the alkali metal-treated g-C₃N₄ show a positive shift, by 0.2-0.4 eV, of the corresponding C 1s and N 1s binding energies. Such a shift reveals the formation of an interaction, presumably chemical coordination, between the doped ions and g-C₃N₄ aromatic rings.

The Li 1s, Na 1s, and K 2p peaks appearing at 55.3, 1071.4, and 292.6 eV are attributed to metal-N bonding, and all are lower in energy than that of the relevant alkali salts with M-O bonding (56.0, 1072.1 and 292.9 eV, respectively) [39, 41, 48]. This result suggests that during high-temperature calcination, the triazine units in g-C₃N₄ became partially disordered, and the nitrogen atoms with increased lone-pair electron density (see above) had a strong coordination affinity toward the metal ions. This caused the formation of strong (high-energy) N-bridging covalent bonds between the N-containing groups and metal ions [47, 56]. In this way, the metal doping changed the electron density around the N atoms, subsequently altering the electronic structure and band gap of g-C₃N₄.

The valence band spectra were thus measured for all samples to determine the relative valence band positions. As shown in Fig. 5(f), the energies of the valence bands gradually increased from CN, through C-CN, to metal-doped g-C₃N₄. Combined with the band gaps calculated from the UV-vis spectra, the valence band energies allowed us to obtain the electronic structures and relative band positions of the prepared photocatalysts, as depicted in Fig. 6. Both post-calcinated and doped g-C₃N₄ show narrower band gaps, more positive valence bands, and less negative conduction band positions than pristine g-C₃N₄, allowing for more efficient visible-light utilization and charge excitation, while still providing enough energy for water reduction. Consequently, these materials should show improved visible-light photocatalytic performance for hydrogen production.

The photocatalytic hydrogen production performances of the as-prepared g-C₃N₄ samples were evaluated under visible-light (λ > 400 nm) irradiation. In the absence of photocatalysts or light irradiation, no product was detected in the system. From Fig. 7, it can be seen that all the modified g-C₃N₄ catalysts show enhanced photocatalytic activities in comparison with pristine g-C₃N₄ (5.0 μmol/h). Importantly, the introduction of alkali metal ions into g-C₃N₄ significantly raises the H₂ production rates, exceeding that of post-calcinated g-C₃N₄. Especially, Na-CN has the highest photocatalytic activity (18.7 μmol h^-1), which is more than three times and two times that of pristine g-C₃N₄ and C-CN, respectively.

As mentioned above, the modification of g-C₃N₄ led to increased specific surface areas and broadened visible-light absorption, both of which contribute to the enhanced photocatalytic activities. To obtain an in-depth understanding of the photocatalytic process, the charge transfer dynamics was further investigated by PL, TRPL analysis, and photoelectrochemical measurements.

The PL spectrum of pristine g-C₃N₄ shown in Fig. 8(a) displays a strong characteristic peak near 445 nm, which can be assigned to its band-to-band emission. All the modified g-C₃N₄ photocatalysts have similar but highly quenched PL peaks at about 465 nm, indicating successful suppression of the recombination of photogenerated charge carriers. Particularly, the sharply decreased PL intensity for C-CN can be ascribed to the structural distortion. The inset of Fig. 8(a) shows that the intensity was further decreased after doping with metal ions from Li-CN to K-CN. This suppression of PL by metal doping can be attributed to two causes: the presence of slight defects and the introduction of impurity levels between the gaps, both of which benefit charge transfer and reduce the recombination rate of photoinduced electron-hole pairs [57, 58]. This tuning of the electronic structures occurs to different degrees for different metal ions, so that the PL intensities vary among Li⁺-, Na⁺-, and K⁺-doped g-C₃N₄.

The lifetimes of charge carriers were calculated by analyzing the TRPL spectra (Fig. 8(b)), which were fitted with bi-exponential decay kinetics. The results are listed in Table 2. As can be seen, both the lifetime components (τ₁ and τ₂), representing radiative and non-radiative processes, respectively, are shortened after the high-temperature and doping treatments, consistently with the above-mentioned suppression of PL. This further confirms that the recombination of charge carriers is successfully retarded in modified g-C₃N₄ because of the distortion of the microstructures and the alteration of the electronic structures, which generate defects and impurity levels for trapping charge carriers [57, 59].

Photoelectrochemical measurements, including transient photocurrent responses and electrochemical impedance spectra, were further conducted to determine the charge transfer efficiency. As shown in Fig. 9(a), all the photocatalysts show prominent photocurrent responses and maintain relatively good stability through repeated cycling runs, suggesting their ability to efficiently transfer photogenerated electrons to an external circuit under visible-light irradiation [60]. The photocurrent intensity of C-CN is slightly higher than that of CN because of the broader light utilization and increased photogenerated charge separation of the former. Meanwhile, the three M-CN samples show considerably increased current intensities, among which Na-CN exhibits the highest value, nearly 7 times that of CN. These results agree well with the photocatalytic activities.

Similarly, the EIS spectra also provide information about the recombination/separation of electrons and holes [61-63]. In Fig. 9(b), the smaller arc radii for C-CN and M-CN relative to CN represent better conductivity or more efficient electron transfer under visible light, which is consistent with the above results. It is noted that the sample with the best performance in retarding charge recombination, according to the PL and TRPL spectra, is K-CN, while the sample with the best performance in promoting charge transfer, based on the photoelectrochemical measurements, is Na-CN. This difference implies that the utilization of photogenerated charge carriers should be more effective for Na-doped g-C₃N₄ during the water splitting process. The K-doped sample undergoes the most favorable change in the electronic structure of g-C₃N₄ among all samples, and thus exhibits the lowest recombination rate of photogenerated charge carriers. However, the conduction band of K-CN is also shifted to the least negative energy among all samples. As a result, under visible-light irradiation, the electrons in the conduction band of K-CN show the reduced ability to reduce protons for H₂ evolution.

Therefore, based on the above discussion, it can be deduced that the improved visible-light photocatalytic hydrogen evolution by metal ion-doped g-C₃N₄ originates from two causes: the optimization of the microstructures for a high specific surface area, and the optimization of the electronic structures for fast charge transfer and effective charge utilization.

g-C₃N₄ samples doped with alkali metal ions (Li⁺, Na⁺, and K⁺) are successfully synthesized by high-temperature assisted calcination. The doped photocatalysts exhibit larger specific surface areas, narrower band gaps, and more efficient charge-carrier transfer, all of which greatly enhance their photocatalytic hydrogen production performances under visible light. Among the metal ion-doped g-C₃N₄ photocatalysts, the Na-doped g-C₃N₄ shows the highest photocatalytic activity, because the electronic structure and redox potentials are tailored optimally for the effective utilization of photogenerated electrons during the water reduction process. This work provides a protocol for the optimization of the doping method to design efficient photocatalysts.