Introduction

Sunlight is considered to be a desirable future energy source, with the benefits of clean, cost-free, and inexhaustible supply1,2. Hydrogen generation from water splitting by photocatalysts under solar light irradiation is a promising pathway to store solar energy as chemicals1,2,3,4,5,6. In recent decades, semiconductor photocatalysts have been widely explored to develop more efficient and less expensive photocatalysts toward solar-driven water splitting7,8,9. Among these studied catalysts, graphitic carbon nitride (C3N4)-based catalysts have attracted tremendous attentions due to their excellent physicochemical stability, visible-light response, metal free as well as its appealing electronic band structure (Eg ≈ 2.7 eV) with band positions straddle the redox potentials of water splitting theoretically8,9,10,11,12. However, the fast recombination of photo-generated electron–hole pairs restricts extensive application13. Using noble metals as co-catalysts is an effective strategy to enhance photocatalytic efficiency of semiconductors, for which noble metals not only serve as an electron-capturing acceptors to promote the separation of photoinduced charge carriers but also act as active sites for H2 evolution reaction14,15. However, high-cost and low reserves impede their practical applications. Single-atom noble metal catalysts can maximize the atom efficiency of noble metal and have high numbers of active sites for catalysis owing to their high ratio of under-coordinated metal atoms16,17,18. Importantly, the physical and chemical properties (such as bandgap and redox properties) of the catalyst will be significantly changed with the particle size of the metal being reduced to cluster or atomic scale due to the size effect19, which results in the varied adsorption states of reactants and reduced reaction barrier for improving activity/selectivity of catalysts compared with nanoparticle catalysts20,21,22,23,24,25. Despite their high utilization efficiencies, single-atom catalysts need further optimization to mitigate serious aggregation or coarsening during the preparation and catalytic reaction process26,27.

An appropriate support or ligand that strongly interacts with the noble metal atoms is required to stabilize atomic noble metals18,28,29,30,31,32,33,34. The pyridinic nitrogen atoms from bulk C3N4 can fix the isolated metal atoms leading to catalysts which exhibit excellent catalytic activity and product selectivity15,32,33,34. Generally, compared to Pt-based catalysts, Pd-based catalysts for water splitting are less attractive due to the fact that metallic Pd has more positive chemical potential and stronger adsorption of hydrogen. Recently, a Pd atom deposited on C3N4 was proved to narrow the bandgap of pristine C3N4 (ca. 2.7 eV) to 0.2 eV by DFT calculations34, which indicated a promising method for obtaining long-wave solar-light responsive catalyst by loading atomic Pd or Pt on C3N4. Wang et al.9 developed a reduced bandgap and promoted charge separation of polymeric photocatalyst of C3N4 by O doping, resulting from the unique structure with both oxygen linkers and nitrogen linkers. Gao et al.35 reported that the binding free energy of hydrogen atom on C3N4 is very sensitive to mechanical strain. They replaced a bridging carbon atom in C3N4 with an isoelectronic silicon atom to induce mechanical strain, and then attained a zero of binding free energy of hydrogen on C3N4, which is favorable for the H2 evolution35. Very recently, Cao et al.15 found that both interlayers intercalated and surface-anchored Pd atoms presented in an atomic Pd/C3N4 layer material. This unique structure provided the driving force for the directional charge-transfer in both vertical and in-plane transportation via Pd atoms and the surface Pd atoms served as reactive sites. However, the surface Pd atoms with various oxidation states strongly indicated a complex state of Pd atoms on C3N4, which disturb the analysis of the role of Pd in water splitting. Additionally, the superior performance and reaction route of atomic Pd/C3N4 in comparison with benchmark catalysts based on different Pd sizes (i.e., nanoparticles and clusters) have not been revealed yet.

In this work, we investigate the catalytic properties of atomic Pd on bulk C3N4 catalyst and its roles in catalyzing water splitting reactions in comparison with benchmark catalysts with different Pd particle sizes. Especially, a distinctive reaction route of photocatalytic H2 evolution is clarified in our study that the pyridine N in the C3N4 cavity adjoining Pd-loaded cavity rather than Pd atom is the active site of atomic Pd/C3N4. This finding presents insights into the roles of pyridine N in the cavities of C3N4 with decorated heteroatoms in H2 evolution process and may give inspiration for designing and investigating atomic catalysts for water splitting on which the reaction mechanism may be distinct from those on nanoparticle or cluster catalysts.

Results

Identification of single-atom catalyst

The bulk C3N4 powder was prepared via thermal condensation of urea36. The Pd/C3N4 catalysts were synthesized using a facile liquid-phase adsorption–deposition method, denoted xPd/C3N4 in which x stands for the Pd loading amount in weight (e.g., 0.1 Pd/C3N4 means that the Pd loading amount is 0.1 wt%). The textural properties of catalysts are shown in Supplementary Tables 1 and 2. As depicted by the X-ray diffraction (XRD) results (Supplementary Figure 1), all catalysts display similar patterns with a typical C3N4 structure and no signals assignable to palladium species can be found. The high angle annular dark field (HAADF) scanning transmission electron microscopy (STEM) graphs of the prepared Pd/C3N4 photocatalysts are shown in Fig. 1a–d. As for 3 Pd/C3N4, most of Pd species exist as spherical nanoparticles with average size of 2–3 nm and the lattice fringes with d spacing of 0.23 nm, which are in good agreement with Pd (111) (JCPD 65-2867). As for 0.1 Pd/C3N4, no nanoparticles or large clusters can be observed (Fig. 1c); instead, the high-resolution STEM image in Fig. 1d suggests the presence of isolated Pd, as indicated by the white arrows.

Fig. 1
figure 1

Microstructure and H2 evolution performance of catalysts. a HAADF-STEM image and b HRTEM image of 3 Pd/C3N4. In b, the dotted circle refers to Pd nanoparticle; the inset shows the enlarged area from white box region and the crystalline interplanar spacing of Pd nanoparticle is marked by white lines, showing a typical d spacing of 0.23 nm ascribed to Pd (111). HAADF-STEM images of 0.1 Pd/C3N4 c without and d with detected Pd. The white arrows in d denote the single Pd atoms. Scale bars: a, c 5 nm; b, d 2 nm. e EXAFS spectrum of 0.1 Pd/C3N4. f Photocatalytic H2 evolution performances measured by per gram weight of catalysts. g H2 evolution measured by per gram of Pd. h Cycling test of H2 evolution activity over 0.1 Pd/C3N4

EXAFS spectrum of the 0.1 Pd/C3N4 catalyst is depicted in Fig. 1e and the related results are summarized in Supplementary Table 3. No peaks ascribed to bulk Pd phase (~2.7 Å) can be detected37,38. Moreover, a single sharp peak at about 1.6 Å (not phase-corrected) appears in the EXAFS spectrum of the 0.1 Pd/C3N4, which is ascribed to the Pd–C/N bond33. This result corroborated the atomic distribution of Pd in 0.1 Pd/C3N4. Generally, EXAFS cannot distinguish the coordinated C and N atoms, because they give similar scattering parameters due to their neighboring positions in the periodic table of elements. Several previous works reported the preferential coordination of Pd atom to nitride rather than carbon when palladium is loaded on nitrogen-doped carbon39,40. To consolidate the coordinated station of Pd with N atoms, we employed the density-functional theory (DFT) method to optimize the most stable adsorption sites for atomic Pd on C3N4, and the results turned out to be consistently well with the EXAFS results (discussed later).

Photocatalytic H2 evolution performance

Considering that the UV light only accounts for about 5% of the total solar energy, which is far less than the visible-light (~43% of solar spectrum)41. For this reason, we evaluated the photocatalytic H2 evolution under visible light irradiation. It is seen from Fig. 1f that the pristine C3N4 shows rather poor H2 production (1.4 µmol g−1 h−1). When introducing 0.05 wt% of Pd on C3N4, the H2 evolution is dramatically enhanced to 435.3 µmol g−1 h−1. With increasing the Pd loading amount, the H2 evolution rate firstly increase and then decrease after a threshold. The optimal H2 evolution rate is up to 728 µmol g−1 h−1 on 0.1 Pd/C3N4 catalyst which is remarkably higher than the reported activity of Pd catalyst (326 µmol g−1 h−1)42. Besides, we also calculated the activities on per milligram Pd, these corresponding activities of xPd/C3N4 catalysts monotonically increase with decreasing the Pd loading amounts. The 0.05 Pd/C3N4 and 0.1 Pd/C3N4 catalysts have similarly much higher turn over frequency (TOF). These results indicate that the single Pd atoms interacting with C3N4 are the most active sites on xPd/C3N4 catalyst. CO pulse chemisorption measurements were also introduced to access the metal dispersion. However, different from conventional Pd catalysts, atomic 0.1 Pd/C3N4 exhibited no CO uptake, which was further verified by Fourier transform infrared spectroscopy (FTIR, Supplementary Figure 2). This phenomenon suggests that isolated Pd-based catalyst behaves differently from those conventional Pd-based catalysts, in accordance with the recent work by Pérez-Ramírez33. From Fig. 1g, one can see that the 0.1 Pd/C3N4 catalyst after running four cycles does not exhibit apparent deactivation, suggesting the good stability during H2 evolution reaction under visible light irradiation. which has been further confirmed by the HAADF-STEM images of the used sample (only isolated Pd can be found) in Supplementary Figure 3b.

Experimental characterization of catalysts

In an attempt to unravel the mechanism behind the aforementioned change between palladium content and H2 evolution performance, we resorted to femtosecond time-resolved transient absorption (fs-TA) spectroscopy, a robust tool for tracking the time-resolved photoexcited electron dynamics of nanosystems32,43,44. The samples under investigation were 0.1 Pd/C3N4, 0.5 Pd/C3N4, and 3 Pd/C3N4, apart from the bare C3N4 as a reference. In the fs-TA measurements, a pump-probe scheme with an ultraviolet pump and a white-light continuum probe was adopted. The center wavelength of the pump was chosen at 400 nm, which is effective for promoting electrons from the valence band to the conduction band of C3N4. Since the 450–650 nm probe was found to produce essentially the same fs-TA kinetics for each sample, we show here a representative set of data taken at 520 nm (Fig. 2a). All of the fs-TA signals manifested as negatively valued photoinduced bleach and their recovery can be well described by a bi-exponential function: τ1 = 0.14 ± 0.01 ps (66%) and τ2 = 8.5 ± 0.5 ps (34%) for C3N4, τ1 = 0.30 ± 0.05 ps (47%), and τ2 = 13.7 ± 1.4 ps (53%) for 0.1 Pd/C3N4, τ1 = 0.28 ± 0.03 ps (63%) and τ2 = 7.8 ± 0.7 ps (37%) for 0.5 Pd/C3N4, τ1 = 0.29 ± 0.03 ps (58%) and τ2 = 5.3 ± 0.5 ps (42%) for 3 Pd/C3N4. On average, the mean recovery lifetimes (〈τ〉) are 2.98 ± 0.18, 7.40 ± 0.77, 3.06 ± 0.28, and 2.39 ± 0.23 ps for C3N4, 0.1Pd/C3N4, 0.5Pd/C3N4, and 3Pd/C3N4, respectively, as plotted in Fig. 2b. Remarkably, 0.1Pd/C3N4 stands out with a longest mean recovery lifetime, which is roughly 2.5-fold of that observed in bare C3N4 (i.e. 7.4 ps for 0.1Pd/C3N4 vs. 2.98 ps for C3N4), echoing well to the volcano-shaped relationship in Fig. 1f. Additionally, both photocurrent response (PC) and electrochemical impedance spectra (EIS) measurements (Supplementary Figure 4) were performed. 0.1Pd/C3N4 exhibits the highest photocurrent intensity and smallest interfacial charge transfer impedance, suggesting that 0.1Pd/C3N4 possesses more efficient charge separation and faster charge transfer than other catalysts, which are consistent with the fs-TA results. Similar to our previous investigation on the single atomic Pt/C3N4 system32, in this particular case of 0.1Pd/C3N4 the addition of isolated single Pd atoms into the C3N4 network may turned out to induce a pronounced intrinsic change of the near band-edge electron trap states of C3N4 (Fig. 2c) in such a way that the longer-lived photogenerated electrons can have more opportunities to participate in the H+ reduction, leading to its best performance of photocatalytic H2 evolution among others. Cao et al.15 clarified that the interlayer intercalated isolated Pd atoms could provide a vertical channel for directional charge transfer from the bulk to the surface via DFT calculation, which is also benefit for the efficient charge separation.

Fig. 2
figure 2

Charge-transfer characterization of fs-TA. a Representative fs-TA kinetics recorded on C3N4, 0.1Pd/C3N4, 0.5Pd/C3N4, and 3Pd/C3N4. Pump: 400 nm; probe: 520 nm. b The mean recovery life time as a function of the Pd loading amount. The gray dashed arrow denotes the Pd loading amount of the catalyst which stands out with the longest mean recovery lifetime. The error bars are determined from the chi-square values obtained in the fit. c A plausible mechanism underlying the involved photophysical processes

X-ray photoelectron spectroscopy (XPS) were employed to confirm the material structure and chemical state, and the related results are depicted in Supplementary Figure 5. Etching with an Ar+ beam was conducted to remove surface layers to obtain access to the bulk composition of the material. Like the results from Cao et al.15 part of Pd atoms are incorporated into the subsurface of C3N4 were found after Ar+ etching. Consequently, the presence of interlayer intercalated isolated Pd may benefit the charge transfer in bulk C3N4. Moreover, the Pd 3d profiles (Supplementary Figure 5a) of 3Pd/C3N4 and 0.5Pd/C3N4 can be deconvolved into two pairs of doublets with 3d5/2 positions at 335.8 and 337.9 eV, 3d3/2 positions at 341.1 and 343.2 eV, which are assigned to metallic Pd0 and oxidized Pdσ+, respectively39. Compared with 3Pd/C3N4 and 0.5Pd/C3N4, few signals for Pd0 existed in atomic 0.1Pd/C3N4 before or after Ar+ etching (also verified by the DFT results in Supplementary Table 4), which indicates that the electronic structure of atomic 0.1Pd/C3N4 is different from those catalysts with high loading amount of Pd. It should be noted that both metallic Pd0 and oxidized Pdσ+ were detected on the surface of single-atom Pd/C3N4 without Ar+ etching by Cao et al.15 which is different from the results in our study. This distinction may contribute to the different preparation approach, leading to different surface properties of materials. The UV-vis spectra in supplementary Figure 6 indicates that grafting single atomic Pd to the structure of C3N4 has little effect on the light harvest of C3N4.

Catalytic mechanism

We further conducted DFT calculations in order to explore the possible photocatalytic mechanism of water splitting on xPd/C3N4. For comparison, we choose the models of pristine C3N4 and C3N4 loaded with isolated Pd atom (Pd1/C3N4) and Pd6 cluster (Pd6/C3N4) to investigate the size effect of Pd on the photocatalytic H2 evolution performance. The charge densities of highest occupied molecular orbital/lowest unoccupied molecular orbital (HOMO/LUMO) and the alignments of HOMO/LUMO of C3N4, Pd1/C3N4, and Pd6/C3N4 are shown in Fig. 3 to compare their separation of electron and hole excited by light. The coordinates of Pd1/C3N4 and Pd6/C3N4 are shown in Supplementary Data 1 and 2. The geometrical structures of hydrogen adsorption on Pd1/C3N4 and Pd6/C3N4 and Pd (111) are depicted in Fig. 4 to reveal the active sites for photocatalytic H2 evolution, which may be influenced by size-effect of loaded Pd.

Fig. 3
figure 3

Charge densities and band alignments of HOMO and LUMO. ac The charge densities of HOMO/LUMO of C3N4, Pd1/C3N4, and Pd6/C3N4, respectively. The brown, gray, and olive spheres refer to carbon, nitrogen, and palladium atoms, respectively. The isovalue of the charge density is 0.01 e(Å3)−1. d The potentials vs. NHE of HOMO/LUMO of C3N4 (red line), Pd1/C3N4 (blue line), and Pd6/C3N4 (magenta line). The dash lines refer to the highest occupied orbital bellow the fermi level of C3N4 in Pd1/C3N4 and Pd6/C3N4

Fig. 4
figure 4

Geometrical structures of hydrogen adsorption on Pd1/C3N4, Pd6/C3N4 and Pd (111) surface. Geometrical structures and adsorption energies of hydrogen adsorption on the a nitrogen and b palladium atoms in Pd1/C3N4. Geometrical structures and adsorption energies of hydrogen adsorption on the c nitrogen atoms and d palladium cluster in Pd6/C3N4. e Geometrical structure and adsorption energy of hydrogen adsorption on the palladium in Pd (111). The brown, gray, olive, and red spheres refer to the carbon, nitrogen, palladium, and hydrogen atoms, respectively. The adsorption sites of hydrogen are marked with red dash cycles

The optimized structures of catalysts are shown in Supplementary Figure 8. Our results reveal that the cavity of C3N4 can stabilize the atomic Pd and Pd6 cluster, similar to previous study34. After loading of Pd1 or Pd6, the band gap will decrease from 2.79 eV (pristine C3N4) to 0.58 eV (Pd1/C3N4) and 0.30 eV (Pd6/C3N4). The decrease of band gap is caused by the respectively electron states of Pd1 and Pd6 fill the band gap of C3N4 to improve the potential of valence band in Pd1/C3N4 and Pd6/C3N4 (details in Supplementary Figure 9). The distribution of charge density of HOMO and LUMO is studied since it is closely related to the activation behavior of photocatalyst. For pristine C3N4 (Fig. 3a), the HOMO mainly consists of charge densities from the edge N atoms in cavity of C3N4, and the LUMO consist of charge densities from the edge C atoms in cavities and the center graphitic N atoms in the C3N4 sheet. When C3N4 is loaded with Pd1, the HOMO mainly consists of majority of charge densities from isolated Pd atom and very few charge densities from the C and N atoms interacting with Pd, and the LUMO mainly consists of charge densities from C and N atoms not interacting with Pd atom, within which some atoms are far from isolated Pd (Fig. 3b), indicating a strong spatial separation of electrons and holes. Similar to isolated Pd1, the HOMO of Pd6/C3N4 consists of majority of charge densities from Pd6 cluster and the LUMO consists of charge densities from supporting C3N4 (Fig. 3c). Wang et al.9 also reported the similar effect of O on the spatially separated HOMO and LUMO of C3N4. Therefore, the atomic-scale Pd loading on C3N4 will promote the separation of photogenerated electrons and holes while pristine C3N4 shows poor separation of electron–hole pairs when compared with the co-catalyst loaded with Pd.

The potentials of HOMO and LUMO versus NHE of pristine C3N4, Pd1/C3N4, and Pd6/C3N4 were also calculated and the results are shown in Fig. 3d. The potentials of HOMO and LUMO of CN are −1.17 and 1.62 eV, respectively, matching the previous study well45. With loading of Pd1 and Pd6, the potentials of HOMO decrease dramatically from 1.62 to −0.47 eV and −0.60 eV, respectively. The potentials of LUMO of Pd1/C3N4 and Pd6/C3N4 increase slightly from −1.17 to −1.09 eV and −0.90 eV, respectively. The potentials of calculated HOMO of supporting C3N4 without interaction with Pd atom are closed to the HOMO of pristine C3N4. The potential difference leads to the transfer of holes from the HOMO with high potential to the HOMO with low potential. The potentials of Pd1 and Pd6 are much smaller than that of the supporting C3N4, indicating that the photogenerated holes of C3N4 transfer to Pd1 and Pd6 and the electrons remain in C3N4. Considering the narrower bandgap of C3N4 after loaded with Pd, the visible triggered H2 evolution reactions may mainly happen on the atomic Pd-loaded C3N4 over Pd1/C3N4 catalyst.

Based on the study of optical activation behavior of Pd1/C3N4, and Pd6/C3N4, we found that the electrons will accumulate in supporting C3N4 and holes will accumulate in Pd atoms. Then, we turn to the study of hydrogen adsorption on Pd1/C3N4 and Pd6/C3N4. There are many adsorption sites that we took into consideration, such as Pd1 and Pd6, hole of C3N4 with or without loading of Pd1 and Pd6 (details in Supplementary Figure 10). Inspired by the above conclusion that the supporting C3N4 of the co-catalysts is the accumulation site of electrons, which is the most possible reactive site for hydrogen evolution, so the hydrogen adsorption energies on the supporting C3N4 and the loaded Pd1, Pd6, and Pd (111) (represent the large Pd particle) are studied and the most stable structures of hydrogen adsorption are summarized in Fig. 4. As shown in Fig. 4a, the most stable adsorption site of hydrogen on Pd1/C3N4 is pyridine N in the C3N4 cavity adjoining Pd-loaded cavity rather than the isolated Pd atom (−0.43 vs. 0.17 eV). It is worth to note that the adsorption of hydrogen on the cavity of C3N4 without loading of Pd1 is much stronger than that with loading of Pd1 (Supplementary Figure 10). In Pd1/C3N4, the cavity of C3N4 adjoining Pd1-loaded cavity is the ground-stable adsorption site of hydrogen and accumulation site of activated electrons simultaneously. This is the reason why Pd1/C3N4 possesses the excellent photocatalytic H2 evolution performance. For Pd6/C3N4 (in Fig. 4b), the size increase of atomic-scale Pd will enhance the adsorption ability of hydrogen, so the adsorption of hydrogen on Pd6 is stronger than that on supporting C3N4 (−0.58 vs. −0.44 eV). For large Pd particle, the stable Pd (111) are used to simulate its surface behavior and adsorption of hydrogen will further be enhanced when comparing with Pd6 cluster (−0.73 vs. −0.58 eV). When the size of atomic-scale Pd increases, the hydrogen atom prefers to be adsorbed on Pd clusters instead of supporting C3N4, which is the accumulation site of activated electrons. Therefore, the large size of atomic-scale Pd is harmful to the improvement of photocatalytic H2 evolution performance of co-catalyst. Our work interprets a distinct active site for H2 evolution route with those reported routes on noble metals for H2 evolution (photoinduced electrons transfer from semiconductor to metal and then reduce H+ to hydrogen molecules)7,32,46,47,48,49,50. These interesting results reveal that the size effect of noble metal could affect the active sites for H2 evolution and present new insights into the roles of pyridine N in the cavities of C3N4 with decorated heteroatoms in H2 evolution process, which may give inspiration for designing atomic catalysts for water splitting.

Discussion

In summary, single-atom Pd is successfully prepared and stably located in the six-fold cavity of C3N4. The distinctly different element constitutes of energy bands of the 0.1Pd/C3N4 give intrinsic character of catalyst for the charge separation. The separation of photoinduced charges on 0.1Pd/C3N4 is more like a semiconductor/semiconductor composite catalyst (C3N4 unites near atomic Pd is composited with those C3N4 units far away from atomic Pd): photoinduced electron–hole pairs from C3N4 units will transfer to the C3N4 units interacting with atomic Pd and be separated; most importantly, C3N4 units interacting with atomic Pd can utilize the long wave visible light to generate hot electrons; on the C3N4 units near atomic Pd, H ions adsorb on pyridine N atoms in the C3N4 cavities adjoining Pd-loaded cavity and are reduced by electrons and then desorb as H2 molecules. The H2 evolution on C3N4 can be greatly boosted by loading atomic Pd resulting from the appropriate sites for hydrogen adsorption, as well as accumulation of photoinduced electrons. This work provides new fundamental insights into the roles of pyridine N atoms in the cavities of C3N4 with decorated heteroatoms in H2 evolution process and extends the photocatalytic H2 evolution mechanism of C3N4 materials, which is expected to give inspiration for investigating atomic catalysts for water splitting on which the reaction mechanism may be distinct with those on nanoparticle or cluster-based catalysts, as well as designing high-performance catalysts for H2 evolution.

Methods

Catalyst preparation

C3N4 was prepared via thermal condensation of urea. In detail, urea was placed in a closed alumina crucible, then it was heated to 600 °C in a muffle with a heating rate of 2 °C/min and held in this temperature for 4 h in air. After cooling it to room temperature, the as-obtained yellowish product was ground into fine powder with an agate mortar. The Pd/C3N4 were synthesized using a facile liquid-phase adsorption method as follows: The as-prepared C3N4 (0.5 g) was firstly suspended in 50 ml ultrapure water and sonicated for 60 min, then to gain various Pd loading content hybrid Pd/C3N4 (0.05, 0.1, 0.5, 3 wt%), a certain volume of H2PdCl4 (0.01 M) solution was added into the C3N4 aqueous dispersion and then sonicated mixed again for 60 min. Subsequently, in order to adjust the pH up to 9, appropriate Na2CO3 (0.1 M) was added into the mixed solution dropwise and slowly, after stirring for 18 h, the resultant product was washed with a large amount of ultrapure water for several times and further dried in an 60 ℃ oven overnight. The final obtained samples were labeled as 0.05Pd/C3N4, 0.1Pd/C3N4, 0.5Pd/C3N4, and 3Pd/C3N4, respectively. All as-prepared catalysts were reduced under H2 atmosphere at 473 K for 1 h before characterization.

Characterization

The X-ray powder diffraction (XRD) patterns were performed on a D/max-RB X-ray diffractometer (Rigaku, Japan) with a Cu Kα radiation source (λ = 1.54065 Å, 40 kV, 100 mA), operating at a scanning angle range of 10–80° (5 °C/min). The nitrogen adsorption and desorption isotherms were conducted on a Micrometrics ASAP 2020 instrument. The Brunauer–Emmett–Teller (BET) surface area was obtained by a multipoint BET method using adsorption data in the relative pressure (P/P0) range of 0.05−0.25. Transmission electron microscopy (TEM) and high-angle annular dark-field STEM (HAADF-STEM) were used to identify the morphological and structure features. The HAADF-STEM images were achieved by a FEI Tecnai G2 F30 HR-TEM/STEM microscope (Japan) and operating at an accelerating voltage of 200 kV. XPS were measured on a Physical Electronics PHI5802 X-ray spectrometer with Mg Kα X-ray (Kα = 1253.6 eV) as excitation source and using the C 1s peak of adventitious carbon (284.8 eV) for calibration. The metal loading concentrations were determined by ICP-AES on a Shimadzu Corporation-ICP-7500 instrument. UV–visible diffuse reflectance spectra were recorded using a Shimadzu UV–vis S-4100 spectrophotometer equipped with an integrated sphere, using BaSO4 powder as reflectance material over the wavelength range of 200–800 nm. The photocurrents and EIS measurements were recorded on an electrochemical analyzer (CHI 750D Instruments) with a standard three-electrode system using the samples coated on fluorine-doped tin oxide (FTO) glass as the working electrodes, a Pt foil as the counter electrode, and Ag/AgCl (saturated KCl) as a reference electrode. The light was provided by a 300 W Xe arc lamp. The working electrodes were prepared as follows: 10 mg of photocatalyst was dispersed in a mixture of 200 μl of isopropanol and 20 μl of Nafion to make a slurry. A 30 μl portion of the slurry was then coated onto a 1 cm−2 FTO conductive glass and dried to form working electrode. The ultrafast TA measurements were performed under ambient conditions, on a Helios pump-probe system (Ultrafast Systems LLC) in combination with an amplified femtosecond laser system (Coherent). The 400-nm pump pulses (~100 nJ/pulse at the sample) were delivered by an optical parametric amplifier (TOPAS-800-fs), which was excited by a Ti:sapphire regenerative amplifier (Legend Elite-1K-HE; center wavelength 800 nm, pulse duration 35 fs, pulse energy 3 mJ, repetition rate 1 kHz) seeded with a mode-locked Ti:sapphire laser system (Micra 5) and pumped with a 1-kHz Nd:YLF laser (Evolution 30). The stable white-light continuum (WLC) probe pulses (450–650 nm for this work) were generated by focusing the 800-nm beam (split from the regenerative amplifier, ~400 nJ/pulse) onto a sapphire crystal plate. A reference beam split from the WLC was used to correct the pulse-to-pulse fluctuation of the WLC. The time delays between the pump and probe pulses were varied by a motorized optical delay line. The instrument response function (IRF) was determined to be ~100 fs by cross-correlating the pump and probe pulses at the sample. A mechanical chopper operating at 500 Hz was used to modulate the pump pulses such that the TA spectra with and without the pump pulses can be recorded alternately. The temporal and spectral profiles (chirp corrected) of the pump-induced differential transmission of the WLC probe light (i.e., absorbance change, ΔA) were visualized by an optical fiber-coupled multichannel spectrometer (with a CMOS sensor) and further processed by the Surface Xplorer software equipped with the Helios system. The samples well dispersed in pure ethylene glycol were contained in a 0.7 mL sealed quartz cuvette under a continuous magnetic stirring condition ensuring that the photoexcited volume of the sample was kept fresh during the TA measurements. Fourier transform infrared spectroscopic (FTIR, Thermo Scientific Nicolet IS50, USA) of CO adsorption was performed to determine the initial states of Pd species. The spectra of CO adsorption at 298 K were recorded after the sample was exposed to CO (1 vol%) balanced by N2 (100 ml min−1) for more than 0.5 h and then purged with a N2 flow (100 ml min−1) for 40 min to remove the weak adsorbed CO. The spectra of CO adsorption were recorded each 10 min until reaching saturated CO adsorption state.

Photocatalytic performance and durability test

Visible-light-driven H2 evolution were conducted in a gas-closed circulation system equipped with a vacuum line. In detail, 100 mg sample powder was suspended in 100 ml 20 vol% triethanolamine (TEOA) aqueous solution and vacuumed for 30 min to remove dissolved oxygen in the mixed aqueous solutions before evaluation. Visible light radiation was provided by a 300 W Xe lamp (Ceaulight) with a UV-CUT filter (λ > 400 nm). The relative focused intensity is about 180 mW cm−2. To keep the catalyst powders suspended in the mixed solutions uniformly, continuous stirring was applied during the reaction of water splitting driven by visible light. The generated H2 was determined by an online gas chromatography (Ceaulight, GC7920, 5 Å molecular sieve columns, N2 carrier) equipped with a thermal-conductivity detector (TCD). The photocatalytic H2 generation durability was characterized by performing four cycle runs with each 4 h under similar conditions.

Computational methods and details

All density-functional theory (DFT) calculations in this study were implemented by the Vienna Ab initio Simulation Package (VASP)51,52. The projected augmented wave (PAW) potential53,54 was employed to describe the electron–ion interactions. When optimizing the geometric structures, the generalized gradient approximation (GGA) method55 with the Perdew−Burke−Ernzerhof (PBE) functional56 and vdW correction based on DFT-D3 method57 was applied to describe the exchange-correlation energy. The cutoff energy for plane wave basis was set to 400 eV, and the convergence criteria of energy is 10–5 eV. Due to the poor results of band gap energies for semiconductors calculated by the PBE method, the hybrid functional (HSE06)58 was applied to the electronic structure calculations (Supplementary Figure 7). To obtain the correct band gap energies, the cutoff energy was increased to 520 eV. The optimized lattice parameter of C3N4 unit cell was 7.13*7.13*16 Å, which consistent with previous result59.

The 3*3 supercell of C3N4 was built to the stable adsorption structures of Pd1/C3N4 and Pd6/C3N4 and relative hydrogen adsorption morphologies. The vacuum slab perpendicular to the C3N4 surface was set to the 15 Å to avert the interaction between the periodic layers. The Brillouin zone is sampled with a single gamma-centered Monkhost–Pack mesh. All the structures were fully relaxed until the force on each atom was less than 0.01 eV Å−1.