Photocatalytic H₂ production from H₂O splitting is suggested to be a promising strategy for solar–photon energy conversion owing to its green and renewable characteristics [1-3]. In a general artificial photocatalytic procedure, the photocatalytic process involves four procedures, i.e., photon absorption, generation of e^–/h⁺, charge transfer from provider to active sites, and reduction of absorbed H₂O or H₃O⁺ to H₂ [3, 4]. Achieving enhanced photon capture and efficient utilization of solar energy is still challenging [5]. As a result, photon energy needs to exceed the transition energy of photogenerated e^– in the bandgap; the directional charge-transfer and its designated interaction with the absorbed H₂O are required. The former requires significant photon energy to excite photogenerated charges; however, this requirement cannot be met. The second factor dynamically reflects a competition in the utilization of the photogenerated charge, and e^– transfers to H₂O slowly (~μs); however, e^– recombines with h⁺ rapidly (~ps) [6, 7]. Although there are many strategies to be attempted [8, 9], such as the heteroatom doping, heterojunction construction, and plasmonic noble metal loading, it is still difficult to integrate the electricity and optics of semiconductors to simultaneously satisfy the thermodynamic and dynamic requirements of photogenerated carriers.

Taking 2D π-conjugated carbon nitride as a representative, g-C₃N₄ is a typical visible-excitation semiconductor (~2.7 eV), and its CB potential is –1.23 V (vs. NHE at pH = 7) [10, 11]. Generally, g-C₃N₄ has been used as a photocatalyst for reductive photocatalytic reactions, e.g., splitting H₂O into H₂ [12-18], CO₂ [19-23], O₂ [24-26], and/or heavy metal reduction [27-30]. It is known that g-C₃N₄ is usually in the form of the layered bulk agglomeration in thermal polymerization, which results in a small surface area, low-photon absorption, and serious charge–carrier recombination [31]. Thus, noble metal (Pt) is often used as a support for g-C₃N₄ to enhance its photo-induced electron migration and utilization in H₂O splitting into H₂. However, its H₂ evolution efficiency does not match the photocatalyst cost requirements [32, 33]. Essentially, it is necessary to understand that the migration and action mechanism of photogenerated electrons from the atomic orbital structures of g-C₃N₄ and its electronic structure can be consequently adjusted to optimize the utilization of photo-induced electrons and H₂ production performance over g-C₃N₄. It is revealed that g-C₃N₄ is composed of sp²-hybridized C-N bonds with lone electrons on the p_z orbitals, which form π-conjugated structures [34-36]. When the sp²-hybridized structure of g-C₃N₄ deviates from the two-dimensional plane, the π-electron density shifts from the concave to the convex sp²-hybridized structure along with the hybridization intermediate between sp² and sp³ in the g-C₃N₄ layers [36-38]. This causes the electron to directly trend from concave to convex. Moreover, the curved-hollow g-C₃N₄ can increase its surface density, and the undulating surface increases its absorbance and improves the photosensitivity of its intrinsic or composite materials [39]. In view of this, we synthesized hollow-concave carbon nitride (C₃N₄), and CuInS₂, as the typical narrow-bandgap sample, was embedded and grown in the cavity of C₃N₄ by hydrothermal method. Research results confirm that the constructed CuInS₂@C₃N₄ photocatalysts demonstrated increased light-absorption intensity from normal 455 to 480 nm due to the light scattering effect caused by the hollow structure. The good directional migration capability of the charge carriers from CuInS₂ to C₃N₄ has been achieved under the action of the apparent potential difference between the concave to the convex of the sp²-hybridized structure of C₃N₄, which thus promotes the effective spatial separation, and enhances the utilization of photogenerated carriers with the direct S-scheme mechanism. As a result, CuInS₂@C₃N₄ photocatalysts exhibit higher efficiency toward H₂ evolution compared with the bulked g-C₃N₄ supported 1 wt% Pt and g-C₃N₄ loaded 3 wt% Pd.

SiO₂ nanospheres with an average size in the range of 300–350 nm were prepared according to a previous report [40]. C₃N₄ was synthesized by direct heat treatment using SiO₂ nanospheres as a template. Specifically, 2.0 g of melamine powder was evenly mixed with 1.0 g of SiO₂ nanospheres. Finally, the mixture was placed in a crucible (20 mL). The crucible, including precursors and templates, was firstly heated in a muffle furnace at 320 ℃ for 2 h at a linear ramp rate of 10 ℃ min^–1, and then heated at 550 ℃ for 4 h at the same ramp rate. After the temperature was decreased to room temperature, the sample was impregnated into an NH₄HF₂ solution (4 mol·L^–1) for 6 h, and then washed with deionized water. Ultimately, a yellow powder of C₃N₄ was obtained.

In a typical synthesis of CuInS₂@C₃N₄ composites, C₃N₄ (0.49 g), InCl₃·4H₂O (1 mmol), CuCl (1 mmol), and thiourea (3 mmol) were successively added into 50 mL of ethylene glycol, and stirred for 1 h at room temperature; subsequently, the mixture was transferred to a Teflon-lined stainless steel autoclave (100 mL) and heated at 200 ℃ for 12 h. After cooling to room temperature naturally, the product was collected, washed, and dried at 60 ℃ for 4 h to afford CuInS₂@C₃N₄. Besides, the bulk carbon nitride (g-C₃N₄) was prepared by heating 2.0 g of melamine at 550 ℃ for 2 h. Meanwhile, pristine CuInS₂ nanoparticles and CuInS₂/g-C₃N₄ were prepared by the same method as that for CuInS₂@C₃N₄.

The data of the composition, crystal type, and chemical valence of the photocatalytic materials were obtained by XRD-6000 and XPS (XSAM800), and the binding energy for all samples was referred according to C 1s at 284.6 eV in the XPS characterization. The morphology was determined by FESEM (Hitachi, SU-70). The optical property was examined using a UV2600 (Shimadzu) and RF6000 (Shimadzu), and the excitation wavelength of the fluorescence was 325 nm at room temperature. The specific surface area and pore radius distribution were tested on a Micromeritics ASAP 3020. The Mott–Schottky analysis, electrochemical impedance spectroscopy, transient photocurrent response, and linear sweep voltammetry were performed on an electrochemical workstation (CHI 660C Chenhua Instruments) in a Na₂SO₄ solution. Time-resolved photo-luminescence spectra were recorded on an FLS920 fluorescence lifetime spectrophotometer (Edinburgh, Instruments, UK).

Fifty milligrams of the prepared photocatalyst was dispersed into 100 mL of an aqueous solution containing 10 vol% triethanolamine along with Na₂S (0.25 mol L^–1) and Na₂SO₃ (0.2 mol L^–1) solutions. Prior to light exposure, the oxygen in the photocatalytic system was removed by bubbling with N₂. Under irradiation by a 350 W Xe lamp equipped with optical filter (λ ≥ 420 nm), a continuous magnetic stirrer was applied at the bottom of the reactor to contain the photocatalyst particles in suspension throughout the experiment. A portion (0.4 mL) of the product was sampled intermittently with septum (1 mL), and the gas was analyzed using a GC-2014 instrument equipped with a 5 Å molecular sieve column (Shimadzu).

The size of the pristine CuInS₂ is about 100–120 nm (Fig. 1(A)), and C₃N₄ demonstrates the uniform hollow-concave structure with a diameter of 300–350 nm (Fig. 1(B)). Using these hollow concave as nucleation sites, the grown CuInS₂ has a larger size of 250–300 nm than that of the pristine CuInS₂, and finally covers the surface of C₃N₄ (Fig. 1(C)). Element mappings confirm the uniform distributions of C, N, Cu, In, and S for the CuInS₂@C₃N₄ sample in Fig. 1(D). The XRD spectra of the as-prepared g-C₃N₄, C₃N₄, CuInS₂, CuInS₂@C₃N₄ samples are shown in Fig. 2. Both g-C₃N₄ and C₃N₄ demonstrate two characteristic diffraction peaks at 13.1° and 27.4°, which reflect the (100) plane originating from the in-planar packing structure of conjugated tri-s-triazine, and the (002) plane corresponding to the interlayer stacking of the conjugated aromatic structure [23, 24], respectively. However, the intensity of diffraction peak over C₃N₄ is weaker in comparison with that of g-C₃N₄, which is because the electroneutrality of the delocalized tri-s-triazine structure is broken by the apparent potential difference between the concave and convex surfaces in the process of deviating from the plane [41, 42]. In combination with CuInS₂, there are diffraction peaks at 27.9°, 46.4°, and 54.9° over the CuInS₂@C₃N₄ sample corresponding to the (112), (204), and (116) crystal planes of the chalcopyrite-type CuInS₂ (JCPDS 85-1575), respectively. It is noticed that the half peak widths of these peaks are decreased compared with those of the CuInS₂ sample, suggesting that the hollow-concave as nucleation sites was beneficial to the CuInS₂ growth. It is observed that the strong diffraction peak of (002) in C₃N₄ is almost the same with that of (112) in CuInS₂, and the former is overlapped by the latter, while the peak of (100) for C₃N₄ is extremely weak, and does not indicate the presence of C₃N₄ in the composite. Thus, there appear to be no diffraction peaks for C₃N₄ in the XRD pattern of CuInS₂@C₃N₄. The composite and chemical valence states of the photocatalysts were tested by XPS, and the corresponding spectra are shown in Fig. 3. The binding energies in the tests are referred to according to C 1s at 284.6 eV. In Fig. 3(A), there are two Cu 2p signals of the pristine CuInS₂ at 931.7 and 951.5 eV corresponding to the Cu 2p_3/2 and Cu 2p_1/2 of Cu⁺ [43, 44]. However, for the Cu⁺ signal of the CuInS₂@C₃N₄ sample, the binding-energy data are increased to 931.9 and 951.7 eV, respectively. The N 1s spectra of the C₃N₄ sample in Fig. 3(B) are fitted into the signals at 398.2, 399.7, and 401.2 eV, corresponding to sp²-hybridized N (C=N–C), bridging N ((C)₃–N), and N–H in the delocalized tri-s-triazine structures [45, 46], respectively. In combination with CuInS₂, the N 1s binding-energy data of the sp²-hybridized N is slightly decreased to 398.0 eV. This phenomenon may indicate the strong interaction between the composites attributed to the transportation of electrons from CuInS₂ to C₃N₄ and attainment of thermodynamic balance. Moreover, the C 1s, S 2p, and In 3d spectra of CuInS₂@C₃N₄ are shown in Fig. S1, and these binding energies were not been changed before and after combination between C₃N₄ and CuInS₂. The BET area data were calculated to be 3.2, 30, and 120 m²·g^–1 for CuInS₂, CuInS₂@C₃N₄, and C₃N₄, respectively; the corresponding pore-size distributions using the BJH method are 8.8, 9.0, and 6.3 nm. The elaborate data of BET areas and pore sizes, as well as adsorption-desorption isotherms and pore size distributions are listed in Table S1 and Fig. S2. Therefore, evidently, the two physical properties of CuInS₂@C₃N₄ have been enhanced by the addition of C₃N₄.

According to the above characterizations of crystal composite and morphology, the hollow-concave structure of the C₃N₄ obtained by deviating the delocalized tri-s-triazine from the plane will strengthen the photon absorption capabilities and electron transfer performances of its composite. In Fig. 4, it can be observed that the photon absorption edge of g-C₃N₄ is 455 nm; however, its absorption boundary of C₃N₄ shows an obvious red shift to 480 nm along with enhanced photon absorption intensity due to the light scattering effect caused by the hollow structure [47, 48]. Moreover, it is clear that pristine CuInS₂ displays a wide light absorption owing to its intrinsic narrow bandgap. Taking advantage of the hollow structure of C₃N₄ and the narrow bandgap of CuInS₂, the constructed CuInS₂@C₃N₄ shows enhanced absorption from UV to near NIR, suggesting that the formed structure is beneficial for photon absorption and utilization [48]. According to the Kubelka-Munk function vs. light energy in Fig. 4(B), the band-gap energies of C₃N₄ and CuInS₂ are calculated to be 2.50 and 1.20 eV, respectively. In a subsequent test of photoluminescence emission spectrum (Fig. 5(A)), g-C₃N₄ displays the strongest fluorescence intensity in the wavelength range of 450–1000 nm, revealing the serious recombination of photogenerated e^– and h⁺ over g-C₃N₄. For the C₃N₄ sample, it is seen that its fluorescence intensity is much lower, suggesting the good separation and transfer efficiency of the photogenerated charge carriers over C₃N₄ under the effect of the apparent potential difference between its inner and outer hollow surfaces. Furthermore, the CuInS₂@C₃N₄ sample shows a lower signal intensity compared to that of C₃N₄, confirming that the photogenerated charge carriers can be spatially transferred over the CuInS₂@C₃N₄ sample. In combination with the XPS results in Fig. 3, we can preliminarily prove the electron transfer from Cu⁺ of CuInS₂ to N of C₃N₄, which, therefore, is bound to improve the conductivity and charge utilization efficiency of CuInS₂@C₃N₄. In Fig. 5(B), EIS-Nyquist characterization confirms that improved conductivity of CuInS₂@C₃N₄ has been obtained. It is seen that the semicircle plots originate from the charge transportation from CuInS₂@C₃N₄ to the electrolyte. As a result, the CuInS₂@C₃N₄ sample demonstrates smaller electric-resistance in comparison to those of C₃N₄ and g-C₃N₄. Under visible-light illumination, photogenerated charge carriers over samples migrate into their back contact and interact with the electrolyte to yield photocurrent reflected as the i-t spectra (Fig. 5(C)) [49]. Clearly, CuInS₂@C₃N₄ presents the highest anodic photocurrent intensity. Linear sweep voltammetry was used to investigate the interaction barrier between the electron and H₂O [50-52]. In Fig. 5(D), polarization plots of g-C₃N₄, C₃N₄, and CuInS₂@C₃N₄ were obtained at a scan rate of 10 mV/s. The initial potentials at 0.5 mA/cm² are –0.42, –0.36, and 0.016 V for the g-C₃N₄, C₃N₄, and CuInS₂@C₃N₄ samples, respectively. The overpotential at –10 mA/cm² for C₃N₄ is much lower than that for g-C₃N₄, suggesting that the apparent potential difference of the hollow-concave C₃N₄ can lower the interaction barrier between the electron and H₂O. Under the effect of the apparent potential difference, CuInS₂@C₃N₄ presents the lowest overpotential (–0.52 V), which is favorable for H₂ production in photocatalytic H₂O splitting [52, 53].

Accordingly, the photocatalytic activities have been tested by H₂ yield efficiency using a Xe lamp equipped with an optical filter (λ > 420 nm), as shown in Fig. 6(A). The H₂ evolution efficiencies of pristine g-C₃N₄ and CuInS₂ are 67 and 12 μmol·h^–1g^–1. When C₃N₄ was used as the photocatalyst in H₂O splitting, its H₂ evolution efficiency was 107 μmol·h^–1g^–1. Moreover, the H₂ evolution efficiencies are 135 and 373 μmol·h^–1g^–1 over CuInS₂/g-C₃N₄, and CuInS₂@C₃N_4, respectively. The effects of irradiation time on the H₂ production efficiency over the g-C₃N₄, C₃N₄, CuInS₂/g-C₃N₄, and CuInS₂@C₃N₄ photocatalysts are displayed in Fig. 6(B), and the durative H₂ evolution is achieved with no declining trend. However, the H₂ efficiency trend plots vs. irradiation time are clearly distinct. Under visible-light irradiation, the H₂ production efficiency over g-C₃N₄ increases tardily, while high H₂ evolution efficiencies over C₃N₄, CuInS₂/g-C₃N₄, and CuInS₂@C₃N₄ can be obtained. In particular, CuInS₂@C₃N₄ exhibits an excellent evolution rate. Meanwhile, the apparent quantum yield of CuInS₂@C₃N₄ is 4.32% at a wavelength of 420 nm, which is 1.63 times higher than that of CuInS₂/g-C₃N₄ (2.65%). Moreover, we also compared the H₂ evolution efficiency of CuInS₂@C₃N₄ with that of g-C₃N₄ with 1% Pt and 3% Pd prepared according to literature [45, 54]. In Fig. 6(C), it is presented that CuInS₂@C₃N₄ shows higher activity than the 1 wt% Pt/g-C₃N₄ (238 μmol·h^–1g^–1) and 3 wt% Pd/g-C₃N₄ (276 μmol·h^–1g^–1). Thus, the hollow concave structure of C₃N₄ promotes the transmission and utilization efficiency of photogenerated electrons under the action of apparent potential difference, which has great advantages in replacing noble-metal co-catalysts and accelerating the photogenerated electron transmission. The stability test of CuInS₂@C₃N₄ was performed as shown in Figure 6D. After three recycles, CuInS₂@C₃N₄ presented efficient H₂ evolution performance under visible-light irradiation. However, its H₂-evolution efficiency was decreased from 373 to 355 μmol·h^–1g^–1 in the fourth recycle. When sacrificial agents were added into the reaction system again, the performance over CuInS₂@C₃N₄ increased to 375 μmol·h^–1g^–1. Thus, the decline in performance was due to the consumption of sacrificial agents. Therefore, CuInS₂@C₃N₄ demonstrates high stability. In combination with the series of characterizations, the strengthened light absorption from visible to the near-infrared region, effective charge-transfer capability, and the lowered H₂ overpotential over CuInS₂@C₃N₄ ensure a highly efficient H₂ production, which is attributed to the hollow concave structure of C₃N₄ and its structure-induced apparent potential difference.

Kinetically, time-resolved fluorescence spectrum was recorded to probe the separation, transfer, and utilization of photo-induced carriers [55]. In Fig. 7, The decay spectra of CuInS₂, C₃N₄, and CuInS₂@C₃N₄ have been fitted into three different lifetime processes: τ₁, τ₂, and τ₃, to demonstrate the radioactive process, non-radioactive step, and energy transfer [56, 57]. The radioactive step arises from the recombination of e^– and h⁺. The τ₁ values for CuInS₂ (0.0023 ns) and C₃N₄ (3.2 ns) are lower than that for CuInS₂@C₃N₄ (6.2 ns), and the percent of charge carriers decrease from 95% and 49.2% over CuInS₂ and C₃N₄, respectively, to 32.6% over CuInS₂@C₃N₄, revealing that the photogenerated charge carriers have been spatially transferred from CuInS₂ to C₃N₄ over the CuInS₂@C₃N₄ sample. In the subsequent process, the relaxation effect of the partial photo-induced carriers results in the non-radioactive step and energy transfer, which is critical for the utilization of photogenerated electrons and photocatalytic H₂O splitting. It is seen that τ₂ and τ₃ for CuInS₂@C₃N₄ (35 and 12 ns) are longer than those of CuInS₂ (0.0013 and 1.6 ns), and C₃N₄ (20 and 5.7 ns), and the corresponding proportions of photo-induced carriers over CuInS₂@C₃N₄ (67.4%) are clearly higher than those of CuInS₂ (5%) and C₃N₄ (50.8%), suggesting that more photo-induced carriers participate in the photocatalytic H₂O splitting. According to the Kubelka-Munk function vs. light energy in Fig. 4(B) [58], the band-gap energies of C₃N₄ and CuInS₂ are calculated to be 2.50 and 1.20 eV, respectively. In Fig. S3, the Mott–Schottky plots demonstrate that C₃N₄ possesses the n-typical semiconductor feature, and its CB potential is –1.60 V at pH = 7. Thus, the corresponding VB is calculated to be 0.90 V. In Fig. S3(A), the CB and VB potentials of CuInS₂ have been calculated to be –0.20 and 1.00 V, respectively. Based on these results, we propose a suitable photocatalytic mechanism for photocatalytically yielding H₂ over CuInS₂@C₃N₄, as displayed in Scheme 1. Since C₃N₄ and CuInS₂ possess small band gaps, e^– and h⁺ can favorably generate and migrate from the VB to CB of both semiconductors under visible-light irradiation. The XPS results in Fig. 3 reveal the electron transfers from Cu⁺ of CuInS₂ to N of C₃N₄ over CuInS₂@C₃N₄. For the traditional Type-Ⅱ photocatalytic mechanism in accordance with the research of Yu et al. [59], electrons migrate from a higher CB to a lower CB, and the holes migrate from a lower VB to a higher VB. Noticeably, the XPS results reveal that the electron transfer principle does not conform to the Type-Ⅱ photocatalytic mechanism; however, CuInS₂@C₃N₄ follows the S-scheme mechanism of the photocatalytic H₂O splitting into H₂ [60]. The photogenerated e^– tendentiously migrates from the CB of CuInS₂ to the VB of C₃N₄ through the interaction between N and Cu⁺ under the action of the apparent potential difference between the concave and convex of the hollow C₃N₄. Furthermore, the photogenerated h⁺ remains on the CuInS₂ surface, which efficiently completes the spatial separation and directional migration of the photogenerated carriers. Meanwhile, the H₂O dissociative adsorption is advantageously performed on the convex surface of C₃N₄. As a result, e^– smoothly reduces the H₂O molecule to produce H₂ on the convex surface of C₃N₄, while h⁺ oxidizes the sacrificial agent on the surface of CuInS₂.

In summary, utilizing the advantage of the apparent potential difference of hollow-concave carbon nitride, CuInS₂ nanoparticles have been impregnated within C₃N₄ for the photocatalytic H₂O splitting into H₂. The constructed CuInS₂@C₃N₄ exhibits higher photocatalytic performance than the bulk g-C₃N₄ with 1% Pt and g-C₃N₄ with 3% Pd in a visible-light hydrolysis system. Experimental analyses confirm that the high efficiency of the H₂ evolution over CuInS₂@C₃N₄ mainly arises from (1) the increased light-capturing capability owing to the incident light within the hollow C₃N₄ by multiple light scattering, and (2) the directional migration of the charge carrier due to the apparent potential difference of the hollow-concave C₃N₄ caused by deviating the sp²-hybridized structure of its tri-s-triazine component from the two-dimensional plane. The research develops an effective strategy for promoting successive solar energy conversions over carbon nitride, which can also be suitable for other semiconductors.

Study was supported by the National Natural Science Foundation of China (21871155), the K. C. Wong Magna Fund in Ningbo University, Fan 3315 Plan, and Yongjiang Scholar Plan.