Introduction
The search for suitable semiconductors as photocatalysts for the splitting of water into hydrogen gas using solar energy is one of the noble missions of material science. An optimal material would combine an ability to dissociate the water molecules, having a bandgap that absorbs light in the visible range and to remain stable in contact with water. Besides, it should be non-toxic, abundant and easily processable into a desired shape. During the past 30 years, various inorganic semiconductors and molecular assemblies have been developed as catalysts for hydrogen production from water under visible light1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12. Semiconductors explored so far are constructed from transition-metal ions with d0 electronic configuration or post-transition-metal ions of d10 configuration, along with group VA or VIA ions as counter-anion components2, 4, 5, 6, 7, 8, 9, 10, 11, 12. For photocatalysis to be chemically productive, precious-metal species11 such as Pt and RuO2 must be used in most cases as extra cocatalysts to promote the transfer of photoinduced charge carriers from the bulk to the surface at which water is converted to hydrogen gas. Metal-based complexes (for example, a complex with four manganese ions in photosystem II (ref. 13) and a di-iron centre in hydrogenases14) are in natural enzymes the active sites photocatalysing the decomposition of water. Synthetic polymer semiconductors such as polyparaphenylene15 have also been used for hydrogen production; however, they are active only in the ultraviolet region and have moderate performance. Here, we show that another simple polymer-like semiconductor, made of only carbon and nitrogen, can function as a metal-free photocatalyst for the extraction of hydrogen from water.
Carbon nitrides can exist in several allotropes with diverse properties, but the graphitic phase is regarded as the most stable under ambient conditions. The first synthesis of a polymeric carbon nitride, melon, was already reported by Berzelius and Liebig in 1834 (ref. 16), and it is therefore one of the oldest synthetic polymers reported. In recent work, we used the thermal polycondensation of common organic monomers to synthesize graphitic carbon nitrides (g-C3N4) with various architectures17, 18. The graphitic planes are constructed from tri-s-triazine units connected by planar amino groups (Fig. 1a). The in-plane organization of tri-s-triazine units and the compression of aromatic planes were found to follow the perfection of condensation (see Supplementary Information, Figs S1,S2), enabling the generation of carbon nitride polymers with adjustable electronic properties, while keeping the free shapeability of a polymer made from liquid precursors. The carbon nitrides used in this study were prepared by heating cyanamide (Aldrich, 99% purity) to temperatures between 673 and 873 K (ramp: 2.2 K min-1) for 4 h. For the material condensed at 823 K, an in-planar repeat period of 0.681 nm (for example, the distance of the nitride pores) in the crystal is evident from the X-ray powder diffraction (XRD) pattern (Fig. 1b), which is smaller than one tri-s-triazine unit (
0.713 nm), presumably owing to the presence of small tilt angularity in the structure. The strongest XRD peak at 27.4°, corresponding to 0.326 nm, is due to the stacking of the conjugated aromatic system, as in graphite (see Supplementary Information, Fig. S2). It should be noted that all of the materials feature some residual amount of hydrogen, which decreases with increasing condensation temperature.
Figure 1: Crystal structure and optical properties of graphitic carbon nitride.
a, Schematic diagram of a perfect graphitic carbon nitride sheet constructed from melem units. b, Experimental XRD pattern of the polymeric carbon nitride, revealing a graphitic structure with an interplanar stacking distance of aromatic units of 0.326 nm. c, Ultraviolet–visible diffuse reflectance spectrum of the polymeric carbon nitride. Inset: Photograph of the photocatalyst.
Full size image (46 KB)Different thermal condensation enables the finer adjustment of the electronic and optical properties, as indicated by the ultraviolet–visible absorption spectrum for carbon nitrides prepared using different condensation temperatures (see Supplementary Information, Fig. S3). The absorption edge is moved towards longer wavelengths, indicating a decreasing bandgap with increasing condensation temperatures. The bandgap of the condensed graphitic carbon nitride is estimated to be 2.7 eV from its ultraviolet–visible spectrum (Fig. 1c), showing an intrinsic semiconductor-like absorption in the blue region of the visible spectrum. This bandgap is sufficiently large to overcome the endothermic character of the water-splitting reaction (requiring 1.23 eV theoretically).
However, in addition to the magnitude of the bandgap, the character of the valence and conduction band and their absolute energies with respect to the reduction and oxidation levels are also important. To analyse the electronic structure, we have carried out density-functional-theory calculations19 for different reaction intermediates along the condensation path in Supplementary Information, Fig. S1. The calculated bandgap decreases from the highest occupied molecular orbital/lowest unoccupied molecular orbital (HOMO–LUMO) gap in the melem molecule of 3.5 eV via 2.6 eV for polymeric melon20 (Fig. 2) down to 2.1 eV for an infinite sheet of a hypothetically, fully condensed g-C3N4. Although the magnitude of the bandgap is underestimated, the trend in our calculations supports our interpretation of the ultraviolet–visible experiments. Figure 2a shows that the band structure of the recently suggested polymeric melon structure20 has a non-isotropic band structure with a direct bandgap at the 
point and only dispersion along the –X direction parallel to the chain. The wavefunction of the valence band in Fig. 2b is a combination of the HOMO levels of the melem monomer, which are derived from nitrogen pz orbitals. The conduction band can similarly be connected to the LUMO of the melem monomer, which consists predominantly of carbon pz orbitals. Photoexcitation consequently leads to a spatial charge separation between the electron in the conduction band and the hole in the valence band. This suggests that the nitrogen atoms would be the preferred oxidation sites for H2O to form O2, whereas the carbon atoms provide the reduction sites for H+ to H2. We have finally calculated the absolute value of the reduction and oxidation levels for H2O to check that they are both located inside the bandgap. The reaction energies were calculated with ab initio thermodynamics21, where the free-energy contribution of the molecules was calculated in the ideal-gas approximation and the solvation energies for H+ and H2O were taken from the literature22, 23. Our value for the reduction level of hydrogen 
eH+/H2=-4.18 eV with respect to the vacuum level corresponds to an absolute value for the normal hydrogen electrode potential (NHE) VNHE=4.18 V. The calculated value for the oxidation level of H2O, hO2/H2O=5.30 eV with respect to the vacuum level, corresponds to the oxidation potential VO2/H2O-VNHE=1.12 V. Both of these values are slightly underestimated compared with experimental values VNHE=4.44 V and VO2/H2O–VNHE=1.23 V (ref. 23). Figure 2a shows that the reduction level for H+ is well positioned in the middle of the bandgap. This ascertains that the reduction process is energetically possible. The oxidation level is located slightly above the top of the valence band, which would permit transfer of holes, but presumably with a low driving force. These calculations provide evidence that carbon nitride has the potential to function as a photocatalyst for hydrogen production.
Figure 2: Electronic structure of polymeric melon.
a, Density-functional-theory band structure for polymeric melon calculated along the chain (–X direction) and perpendicular to the chain (Y– direction). The position of the reduction level for H+ to H2 is indicated by the dashed blue line and the oxidation potential of H2O to O2 is indicated by the red dashed line just above the valence band. b, The Kohn–Sham orbitals for the valence band of polymeric melon. c, The corresponding conduction band. The carbon atoms are grey, nitrogen atoms are blue and the hydrogen atoms are white. The isodensity surfaces are drawn for a charge density of 0.01qe Å-3.
The photocatalysis experiments were carried out with g-C3N4 in powder form (Fig. 1c, inset) to provide sufficient surface area. It is insoluble in water as well as in acid (HCl, pH=0) or base (NaOH, pH=14). The as-prepared g-C3N4 achieved steady H2 production from water containing triethanolamine as a sacrificial electron donor on light illumination (
>420 nm) even in the absence of noble metal catalysts such as Pt, as shown in Fig. 3, curve (i). These results indicate that g-C3N4 functions as a stable 'metal-free' photocatalyst for visible-light-driven H2 production. However, the H2 evolution activity of bare g-C3N4 was fluctuant, and exhibited variation from batch to batch (0.1–4 
mol h-1). Modification with a small amount of Pt solved this problem, as described for other systems in the literature24, improving activity and minimizing the experimental error to lie within 10–15% for different batch samples. This modification violates the 'metal-free' principle, but facilitates electron localization from the conduction band of g-C3N4 to the deposited Pt nanoparticles and simplifies H2 production, here by about a factor of 7. We interpret this finding to be caused by the more favourable hydrogen elimination from Pt–H surface bonds, as compared with the covalent elimination from a hydrogenated g-C3N4 surface. Future work should focus on optimizing reaction conditions and improving surface area/surface functionality to catalyse hydrogen elimination even from the bare g-C3N4 to avoid the use of noble metals, which was however already shown to be possible.
Figure 3: Stable hydrogen evolution from water by g-C3N4.
A typical time course of H2 production from water containing 10 vol% triethanolamine as an electron donor under visible light (of wavelength longer than 420 nm) by (i) unmodified g-C3N4 and (ii) 3.0 wt% Pt-deposited g-C3N4 photocatalyst. The reaction was continued for 72 h, with evacuation every 24 h (dashed line). Unmodified g-C3N4 also photocatalysed steady H2 production from aqueous methanol solution (10 vol%), as shown in Supplementary Information, Fig. S4.
Full size image (35 KB)The H2 production rate increases with increasing Pt content to a plateau at around 2–4%, beyond which it decreases again. A typical time course of H2 production using 3 wt% Pt-deposited g-C3N4 catalyst is shown in Fig. 3, curve (ii). The fact that the synthesized carbon nitride material contains residual hydrogen requires an evaluation of the origin of the hydrogen to exclude the possibility of the catalyst material as the hydrogen source. To evaluate stability, the reaction was allowed to proceed for a total of 72 h with intermittent evacuation every 24 h under visible-light irradiation (>420 nm). Continuous H2 evolution with no noticeable degradation of the carbon nitride was clearly observed from the beginning of the reaction. The total evolution of H2 after 72 h was 770 mol, by far exceeding the amount of surface C6N8 (tri-s-triazine) units (7.6 mol) or the loaded Pt nanoparticles (15 mol). (The number of surface C6N8 (tri-s-triazine) units is estimated to be 7.55
10-5 mol g-1 of g-C3N4. The calculation is based on the specific surface area of g-C3N4(10 m2 g-1), which was determined by the Brunauer–Emmett–Teller method at liquid-nitrogen temperature, and the area of C6N8 units (0.22 nm-2).) To confirm that hydrogen is not coming from g-C3N4, we carried out an extra experiment using Pt/g-C3N4 under ultraviolet light (>300 nm). The system steadily produced 4.5 mmol H2 in 19 h, exceeding the amount of hydrogen (1 mmol) in the catalyst. In most cases where nitrogen-containing materials (for example, (oxy)nitrides) are used as photocatalysts, a low level of N2 evolution is detected in the initial stage of photocatalytic reaction6. This is attributed to the oxidation of N3- species near the material surface to N2. No N2 evolution was observed for the present catalyst even after the extended period of irradiation, which may indicate strong binding of N in the covalent carbon nitride. These results indicate excellent stability of the present material for photocatalytic H2 production. Production of H2 was also observed when other electron donors including methanol, ethanol and ethylenediaminetetraacetic acid were used instead of triethanolamine, although the hydrogen evolution rates were lower (see Supplementary Information, Fig. S5).
To support that reaction proceeds through light absorption within the carbon nitride polymer, we also examined the dependence of the rate of H2 evolution on the wavelength of incident light. As shown in Fig. 4, the trend of H2 evolution rates matches well with that of absorption in the optical spectra. The longest wavelength available for the reaction is 540 nm, which corresponds to the absorption edge of the catalyst. A control experiment showed that indeed no reaction took place in the dark.
Figure 4: Wavelength-dependent hydrogen evolution from water by g-C3N4.
Steady rate of H2 production from water containing 10 vol% methanol as an electron donor by 0.5 wt% Pt-deposited g-C3N4 photocatalyst as a function of wavelength of the incident light. Ultraviolet–visible absorption spectrum of the g-C3N4 catalyst is also shown for comparison.
Full size image (29 KB)It is not the most condensed material (condensed at 873 K) that shows the highest activity, but the more polymeric C3N4 material condensed at 823 K. Obviously, some defect structures are crucial, presumably to localize the electron and hole at specific surface termination sites where they can be transferred to the water molecules.
Note that although hydrogen generation from water is an achievement, it constitutes only the minor half of the problem of water splitting. Water oxidation by non-oxide catalysts remains the real challenge of the community. For inorganic sulphides and (oxy)nitrides, the anion components are less stable and, in some cases, more susceptible to oxidation than water. There are some cases where N2 evolution and sulphur deposition are observed as a result of oxidative decomposition of the photocatalyst25. Indeed, our system in the previous experiments did not eliminate molecular oxygen (presumably again because of the covalent character of the oxygen bonds), and a sacrificial reactant to take up the oxygen was used. The bandgap calculations however indicate that in principle oxygen elimination should be thermodynamically possible, as the HOMO was found to lie below the water oxidation potential.
This stimulates further research for constructing a suitable catalytic active site that efficiently promotes O2 evolution from the g-C3N4 surface. As a first trial, we thus attempted to introduce RuO2, which is well known as a good oxidation catalyst for O2 evolution26, 27, 28, by an impregnation method12, using silver nitrate as a sacrificial electron acceptor.
As shown in Fig. 5, the modified catalyst exhibited visible-light activity (>420 nm) for O2 evolution with no N2 evolution, although the activity for O2 evolution was an order of magnitude lower than that for H2 evolution (Fig. 3). These results indicate that the loaded RuO2 is able to take up the photogenerated holes from g-C3N4, exhibiting functionality as efficient O2 evolution sites. The decrease in activity with reaction time is primarily attributable to the deposition of metallic silver on the catalyst surface, which blocks light absorption and obstructs active sites6, 7. Under ultraviolet irradiation (>300 nm), the amount of O2 evolved for 8 h reaction reached 53 mol (see Supplementary Information, Fig. S6), which is larger than that in the RuO2 catalyst (about 28 mol). (The amount of O2 in the RuO2 (3.0 wt%)-loaded g-C3N4 (0.1 g) is calculated to be about 28 mol. It should be noted that g-C3N4 contains oxygen species as impurities at a rate of about 0.2 wt%.) This fact clearly demonstrates that the observed O2 evolution is no doubt derived from water oxidation. The low activity for O2 evolution can be explained by the band-structure calculation, which showed that the thermodynamic driving force of g-C3N4 for O2 evolution is much smaller than that for H2 evolution (Fig. 3). Although the present preliminary results can be regarded as promising, further optimization for the O2 evolution system is thus needed.
Figure 5: Oxygen evolution from water by g-C3N4.
Time courses of O2 production from water containing 0.01 M silver nitrate as an electron acceptor under visible light (of wavelength longer than 420 nm) by 3.0 wt% RuO2-loaded g-C3N4. La2O3 (0.2 g) was used as a buffer (pH 8–9).
Full size image (25 KB)In summary, we presented here polymeric carbon nitride as a commonly available and simple photocatalyst that is able to generate hydrogen from water even in the absence of noble metals. Although the estimated quantum efficiency of the Pt-modified C3N4 catalyst is still rather low (approximately 0.1% with irradiation of 420–460 nm (Fig. 4)), the present result opens new vistas for the search of energy production schemes, using thermally and oxidation-stable polymeric organic semiconductor structures as the functional material that are cheap and commonly available.
500 nm)
