Metal Single‐Atom Cocatalyst on Carbon Nitride for the Photocatalytic Hydrogen Evolution Reaction: Effects of Metal Species

Single‐atom (SA) catalysts exhibit high activity in various reactions because there are no inactive internal atoms. Accordingly, SA cocatalysts are also an active research fields regarding photocatalytic hydrogen (H2) evolution which can be generated by abundant water and sunlight. Herein, it is investigated whether 10 transition metal elements can work as an SA on graphitic carbon nitride (g‐C3N4; i.e., gCN), a promising visible‐light‐driven photocatalyst. A method is established to prepare SA‐loaded gCN at high loadings (weight of ≈3 wt.% for Cu, Ni, Pd, Pt, Rh, and Ru) by modulating the photoreduction power. Regarding Au and Ag, SAs are formed with difficulty without aggregation because of the low binding energy between gCN and the SA. An evaluation of the photocatalytic H2‐evolution activity of the prepared metal SA‐loaded gCN reveals that Pd, Pt, and Rh SA‐loaded gCN exhibits relatively high H2‐evolution efficiency per SA. Transient absorption spectroscopy and electrochemical measurements reveal the following: i) Pd SA‐loaded gCN exhibits a particularly suitable electronic structure for proton adsorption and ii) therefore they exhibit the highest H2‐evolution efficiency per SA than other metal SA‐loaded gCN. Finally, the 8.6 times higher H2‐evolution rate per active site of Pd SA is achieved than that of Pd‐nanoparticles cocatalyst.


Introduction
The transition from a society based on fossil fuels to one that is based on clean renewable energy has been discussed in recent years. Use of water-splitting photocatalysts will be one solution because such catalysts can generate hydrogen (H 2 ) from water (H 2 O) and sunlight, both of which are abundant on Earth. [1] Although previous studies on photocatalytic H 2 evolution have achieved 1.1% solar-to-H 2 conversion efficiencies, [2] the efficiency should be improved from upto 5%-10% for practical use. Accordingly, developing visiblelight-driven photocatalysts, that account for ≈43% of solar energy, has been actively investigated in recent years. [1b-e,3] Regarding activation of a photocatalytic reaction, it is important to load metal nanoparticles (NPs) and/or metal nanoclusters (NCs) [4] as cocatalysts onto the surface of the photocatalyst. The role of the cocatalysts is to i) promote charge separation of electrons and holes generated by photoexcitation, as well as ii) work as the active site for the catalytic reaction ( Figure 1A). [5] Thus, modulating the cocatalyst is important for creating highly active photocatalysts. [5a-c] Recently, using a single atom (SA) of metal (with an even smaller size than metal NPs [6] and NCs [7] ) as a cocatalyst has been an active fields of research ( Figure 1B). The catalytic reaction occurs on the surface (i.e., active sites); thus, metal atoms inside NPs cannot normally contribute to the reaction. Because SAs have no inactive internal atoms, using SAs as reaction sites is expected to result in an increase in the reaction efficiency per metal atom. Such SA-loaded catalysts (SACs) with various metal species have been reported since 2011 [8] in photocatalysis/electrocatalysis. [9] In particular, studies using SAC for H 2 -evolution reaction (HER) had revealed that SACs with various types of photocatalysts promotes the HER. [10] Regarding photocatalysts, graphitic carbon nitride (g-C 3 N 4 ; i.e., gCN) [11] has been widely studied as a visible-lightdriven photocatalyst because it i) contains no metal elements and is cheap, as well as highly stable, and ii) has an energy band that is suitable for both absorbing visible light and promoting the HER ( Figure 1A). Typically, the loading metal weight and metal species of a cocatalyst substantially influence the activity. [5b] However, there are no reports of loading SAs onto gCN at high loading rates yet preventing aggregation. In addition, there are few exam-ples that clarify the dependence of the SA metal species on the photocatalytic HER activity. [10d-f,o] Accordingly, there is no unified knowledge on suitable SA metal species for photocatalytic H 2 evolution at present.
In this study, we loaded 10 species of transition metal (M = silver (Ag), gold (Au), cobalt (Co), copper (Cu), molybdenum (Mo), nickel (Ni), palladium (Pd), platinum (Pt), rhodium (Rh), and ruthenium (Ru)) onto gCN as cocatalysts (M SA/gCN) to investigate the metal-species dependence of the photocatalytic HER activity. We achieved to form SA on gCN at high loading weight (≈3.0 wt.%) for Cu, Ni, Pd, Pt, Rh, and Ru by photodeposition when suppressing the reduction power (Figure 2). M SA/gCN (M = Pd, Rh, and Pt) exhibited relatively high HER activity; Pd SA/gCN exhibited the highest HER efficiency per SA. It was revealed that these tendencies mainly originated from the fact that Pd SA exhibits an electronic structure that is suitable for proton (H + ) adsorption.

Characterization of SACs
We synthesized gCN from urea by thermal polymerization. [11b] We confirmed that the obtained gCN exhibits a morphology and . Spectra of Pd foil, PdO, and PdCl 2 ·2NaCl powder are also shown for comparison. In B), we assigned the peaks at ≈1.6 and ≈2.5 Å to Pd-C/N and Pd-Pd bonds, respectively. C) Pd 3d XPS, as well as D) DR spectra, E) PXRD patterns, and F) ultra-high-resolution HAADF-STEM image of Pd SA/gCN(0.5). G) HAADF-STEM images and EDS elemental mappings (Pd-L, N-K, and C-K) of Pd SA/gCN (0.5). In (C), (F), and (G), we set the metal loading weights of SA onto gCN to 0.5 wt.%.
an electronic structure similar to those in the literature by diffuse reflection (DR) spectra and powder X-ray diffraction (PXRD) patterns ( Figure 2E). Regarding the preparation method of M SA/gCN, we used a modified method by Yao et al. in 2018. [10a] Specifically, we prepared M SA/gCN by stirring the gCN and metal precursor (Table S1, Supporting Information) in ultrapure water, and then irradiating with light by using a mercury (Hg) lamp as the light source for 30 min without sacrificial reagent ( Figure 1C). We characterized the prepared M SA/gCN by various analysis methods ( Figure 2) and estimated the loading metal weight by inductively coupled plasma-mass spectrometry (ICP-MS).
As typical results of characterizing M SA/gCN, in the following, we describe the results obtained for Pd SA/gCN, which exhibited the highest H 2 -evolution activity. Figure 2A,B and Figure S1 (Supporting Information) show the results of X-ray absorption fine structure (XAFS) of Pd SA/gCN synthesized with a loading weight of Pd in the range of 0.1-2.9 wt.% [Pd SA/gCN(0.1-2.9)]. The increase of the white line in the Pd Kedge X-ray absorption near-edge structure (XANES) spectra of Pd SA/gCN are in good agreement with palladium oxide (PdO), demonstrating that we loaded Pd as a divalent cation ( Figure 2A). We also observed a peak attributable to Pd 2+ at ≈337.1 eV in the Pd 3d X-ray photoelectron spectroscopy (XPS) spectrum of Pd SA/gCN ( Figure 2C). In the N 1s XPS spectrum of gCN without M SA (bare gCN), we observed peaks at ≈398.0, 399.1, and 400.5 eV which are attributable to C-N-C, N-(C) 3 , and C-N-H groups, respectively. In the N 1s XPS spectrum of gCN after loading Pd SA, we observed these peaks at a slightly higher energy region. Thus, there was complexation between the N atoms of gCN and Pd SA ( Figure S2, Supporting Information). [10b,d] However, there was almost no change in the DR spectrum of gCN before and after loading SA, indicating that SA loading did not induce substantial changes in the electronic structure of gCN, especially in the band gap (BG; Figure 2D). From these results, we interpreted Pd SA/gCN as exhibiting a structure in which Pd(II) ions were weakly coordinated onto the surface of gCN.
Regarding the effect of the weight of Pd loading on the morphology of SA, we observed almost no change in the Pd Kedge XANES spectra of Pd SA/gCN in the range of the loading weight 0.1-2.9 wt.%. Therefore, the Pd SA/gCN obtained in this study maintained the electronic state of Pd 2+ even at a high loading weight of SA (2.9 wt.%). Namely, we interpreted the Pd cocatalyst to be loaded as SAs even at a high loading weight of the cocatalysts without their aggregation. We did not observe peaks originating from the Pd-Pd bond (≈2.7 Å) and we observed only the peak attributable to the Pd-C/N bond (≈1.6 Å) in the Fourier transform-extended X-ray fine structure (FT-EXAFS) spectra of Pd SA/gCN loaded with 0.1-2.9 wt.% Pd ( Figure 2B and Table S2, Supporting Information). [10c,e] In the PXRD pattern of Pd SA/gCN, although we observed diffraction peaks corresponding to the (100) plane and (002) plane of gCN at 13°a nd 27°, respectively, we did not observe peaks attributable to Pd NP ( Figure 2E). [11a] Furthermore, atomic-resolution high-angle scattering dark-field-scanning transmission electron microscopy (HAADF-STEM) [12] images ( Figure 2F and Figure S3, Supporting Information), HAADF-STEM images, and energy dispersive X-ray spectroscopy (EDS) mapping ( Figure 2G) indicate that we loaded Pd SA onto gCN in a monodisperse manner. Thus, Pd SA formed without aggregation on gCN at least up to ≈3.0 wt.% Pd loading.
Typically, when SACs are prepared at a high weight of metal loading, aggregation of metal SAs into metal NPs readily occurs; thus, it is not easy to prepare SA catalysts without aggregation. However, in this study, we loaded SA at high dispersion without aggregation even at high loading weight (≈3.0 wt.%). We assumed that the following two factors play an important role for this fact: i) selection of gCN, which has a high specific surface area, as a support; and ii) suppression of excessive reducing power by not using a sacrificial agent and shorting the light-irradiation time for photodeposition of the cocatalyst. It was presumed that these factors prevented reduction of Pd(II) ions into metal Pd(0) atoms and thereby aggregation of metal Pd (0) atoms. When we loaded the cocatalyst onto gCN by photodeposition and added methanol (MeOH) as a sacrificial agent, we observed Pd NPs with particle size of 2.6 ± 0.8 nm loaded onto gCN at high weight of Pd loading (3.4 wt.% Pd) because reduction of Pd ions proceeded by acceleration of hole consumption (Pd NP/gCN, Figure S4, Supporting Information). This supports the previously described interpretation ii) that weakening the reducing power is significantly important for high loading of SA without aggregation.
We similarly characterized M SA/gCN prepared with other metal species (Ag, Au, Co, Cu, Mo, Ni, Pt, Rh, and Ru). We achieved to form M SA/gCN with Cu, Ni, Pt, Rh, and Ru, as well as with Pd ( Figures S5-S15, Supporting Information). However, when we used Au, Ag, Co, or Mo as the metal species, we observed metallic electronic states in the XANES spectra and the peaks attributable to metal-metal or metal-O-metal bonds in the FT-EXAFS spectra, upon increasing the loading weight of cocatalysts. Therefore, it is difficult to prepare M SA/gCN at high weights of metal loading by using Au, Ag, Co, or Mo as the metal species.
To understand the reasons why the aggregation depended on the metal species, we conducted density functional theory (DFT) calculations on M SA/gCN. Many reports have used a single layer of gCN for calculations; [13] but in a sheet structure, such as gCN, formation of a layered stacking structure changes the surface energy state. The ease of formation for SA might also change. Accordingly, we obtained the most stable (gCN * ) and metastable (gCN ** ) four-layered stacking structures by optimizing from 17 structures of gCN ( Figure S16, Supporting Information). We added the SA onto these two structures (gCN * and gCN ** ) to obtain information on the adsorption sites. Au SA could selectively adsorb onto the vacancy sites ( Figure S17, Supporting Information). The adsorption energy was almost the same at both vacancy sites of gCN * and gCN ** . We also conducted DFT calculations on M SA/gCN with the other metal species. Au and Ag exhibited a smaller adsorption energy compared with the other metals and consequently longer M-C(N) bond lengths ( Figure S18, Supporting Information). Thus, it was revealed that Au and Ag readily detached from the vacancy and thereby aggregate onto gCN. However, aggregation of Co and Mo cannot be explained by M-C(N) bonding energy and lengths. We assumed that forming Co and Mo SA is difficult because, e.g., additional SA tends to form at the same adsorption site and/or the aggregation energy is large.
Thus, we fabricated M SA/gCN at a high loading rate of ≈3.0 wt.% by using Cu, Ni, Pd, Pt, Rh, or Ru. Regarding Au, Ag, Co, and Mo, it was difficult to fabricate M SA/gCN at high weights of metal loading.

Photocatalytic HER Activity
Regarding M SA/gCN (M = Cu, Ni, Pd, Pt, Rh, or Ru), we measured the photocatalytic H 2 -evolution activity under visible-light irradiation (Figure 3). To investigate the effect of SA as a cocatalyst for the HER (reduction side), we evaluated the photocatalytic H 2 -evolution activity in a half reaction with MeOH as a sacrificial agent on the oxidation side. We observed relatively high H 2evolution activity for M SA/gCN by using Pd, Rh, or Pt as SAs ( Figure S19, Supporting Information). This trend is consistent with previous reports on M SA/gCN. [10a-e,h-k] The electron trapping ability and/or H adsorption energy is not thermodynamically favorable for SAs, such as Cu, Ni, and Ru. This is the reason why we obtained only low H 2 -evolution activity for these M SA/gCNs (Cu, Ni, and Ru).
We also investigated the dependence of the metal loading weight on the H 2 -evolution activity by using M SA/gCN (M = Pt, Rh, or Ru), which exhibited relatively high H 2 -evolution activity. We obtained the highest H 2 -evolution activity at SA loading weight of 0.5, 1.0, and 2.0 wt.% for Pd, Rh, and Pt SA/gCN, respectively. This indicates that the optimum loading weight of SAs for obtaining the highest H 2 -evolution activity varies depending on the metal species (Figure 3a). We calculated and compared the activity per atom ( = number of active sites) of each SA for these three types of M SA/gCN (M = Pt, Rh, or Ru), which indicates that Pd SA/gCN exhibited the highest H 2 -evolution activity per SA (Figure 3b). We also compared the H 2 -evolution activity of the obtained Pd SA/gCN with that of Pd NP-loaded gCN (Pd NP/gCN), prepared by the conventional method ( Figure S4, Supporting Information). We set the loading weight of Pd atoms in Pd NP/gCN to 3.4 wt.%, which is the typical loading weight of metal cocatalysts onto gCN. [11b] Pd SA/gCN exhibited a ≈3.1 times higher H 2 -evolution rate than Pd NP/gCN ( Figure S20, Supporting Information). Regarding Pd NP/gCN, the number of surface atoms (Table S3, Supporting Information) can be estimated by using the average particle size of the Pd NP (2.6 ± 0.8 nm) and the density of the metal atoms in bulk Pd (face-centered cubic structure). We thus estimated the number of active sites to be 2.3 times higher in Pd SAs than in Pd NPs (Table S3, Supporting Information). By calculating the H 2 -evolution rate per active site by taking into account the loading weight of Pd atoms (Pd SA : Pd NP = 0.5 : 3.4 wt.%), we estimated the active site of Pd SA/gCN (i.e., Pd SA) to exhibit an 8.6 times higher H 2 -evolution rate than that of Pd NP/gCN (Table S3, Supporting Information). Thus, the Pd SA cocatalyst substantially surpassed the Pd NP cocatalyst prepared by the conventional method in terms of metal utilization efficiency in photocatalytic H 2 evolution.
We also characterized Pd SA/gCN after photocatalytic activity tests by XAFS, TEM, STEM, atomic-resolution HAADF-STEM, EDS, and DR spectroscopy measurements ( Figures S21-S23, Supporting Information). A substantial change was not evident in terms of the morphology and electronic structure of Pd SA/gCN, even after the photocatalytic activity tests. The Pd SA/gCN also showed the high durability for a long-time (12 h) light irradiation for a photocatalytic H 2 -evolution reaction (Figure 3c). We also observed such high stability of the SA catalysts for Pt and Rh SA/gCN ( Figures S24-S27, Supporting Information). The high stability of these M SA/gCN is attributable to immobilization of SA in the vacancies on the gCN plane.

Reaction Mechanism of SA-Loaded gCN
We performed transient absorption (TA) spectroscopy [14] and electrochemical measurements for M SA/gCN to elucidate the metal-species dependence of the difference of the HER activity. In the TA spectra of each sample, excited with a 420 nm laser pulse in a nitrogen (N 2 ) atmosphere ( Figure S28, Supporting Information), we observed absorption over a broad wavelength range of 625-10000 nm (16000-1000 cm −1 ) for all samples base on BG excitation of gCN. Absorption at a wavelength of 2000 nm (5000 cm −1 ) is mainly attributable to shallowly trapped electrons. [15] Thus, we measured the decay curves of the TA intensity at a wavelength of 2000 nm for gCN without cocatalysts (bare gCN) and for each M SA/gCN (M = Pd or Rh) ( Figure S29 and Table S4, Supporting Information). The decay was faster in M SA/gCN than in bare gCN. Thus, the excited electrons in gCN decreased quickly by movement of the excited electrons onto SA; in other words, SA readily captured excited electrons from gCN. [10e] There was no substantial difference in the decay rate of TA between Pd and Rh SA with the same weight of metal loading (TA intensities after 100 μs of excitation were 0.51 and 0.57 times higher for Pd and Rh SA/gCN, respectively, than for bare gCN). Therefore, there was almost no difference in the transfer efficiency of photoexcited electrons from gCN to SA when we used Pd or Rh SA as a cocatalyst ( Figure S30, Supporting Information).
To further understand the carrier dynamics during the watersplitting reaction, we also performed TA spectroscopy under a H 2 O vapor atmosphere (Figure 4 and Table S5, Supporting Information). Regarding bare gCN, there was almost no difference in the TA decay time before and after introducing H 2 O vapor; whereas regarding Pd or Rh SA/gCN, the TA decay was faster in an H 2 O vapor atmosphere than in an N 2 atmosphere. Thus, in a H 2 O vapor atmosphere, the excited electrons transferred to SA were consumed by reduction of H 2 O. Furthermore, the TA decay of Pd SA/gCN was much faster than in Rh SA/gCN (the TA intensities of Pd and Rh SA/gCN after 100 μs of excitation were 0.55 and 0.67 times larger in an H 2 O vapor atmosphere than in an N 2 atmosphere, respectively). Thus, it was inferred that the HER more readily occurred on Pd SA than on Rh SA (Figure 4b,c).
To confirm these interpretations, we conducted electrochemical measurements (Figure 5).
Linear sweep voltammetry (LSV) indicates that the reduction current began at a more positive applied voltage in M SA/gCN (M = Pd and Rh) than in bare gCN ( Figure 5). Thus, M SA had the effect of greatly decreasing the overvoltage for the HER. [10d] Regarding Pd SA/gCN, the reduction current started at a more positive applied voltage compared with Rh SA/gCN. This means that Pd SA loading is more effective than Rh SA loading in terms of increasing the activity of the HER. It is presumed that the decrease in overpotential is caused by the most suitable Gibbs free energy of H-adsorption on Pd SA/gCN. The H-adsorption characteristics may change depending on not only the type of M SA, but also the structure and constituent elements of the support materials. [16] For instance, Pt SA has been shown to be more active with two N-and two C-coordinated Pt SA (Pt-N 2 C 2 ) compared to four N-coordinated Pt SA (Pt-N 4 ). [16a-c] Therefore, it is expected that the activity of SA can be further enhanced by controlling the H-adsorption energy more precisely by doping or alloying the constituent elements of the SA and the support materials. [10f,g,16d,e] In this way, we have clearly revealed that i) SAs can contribute to the capture of excited electrons in gCN, ii) SAs can contribute to a decrease of the overvoltage of the HER, and iii) Pd SA was particularly superior in terms of effect (ii). These results are in good agreement with previous DFT calculations for Pd SA/gCN. [13d] We interpreted these factors as the reasons why Pd SA exhibited the highest photocatalytic H 2 -evolution activity among SAs of various metal species.

Conclusion
We tried to fabricate M SA/gCN by using 10 species of transition metals and established a method of preparing SAs with- www.advancedsciencenews.com www.afm-journal.de out aggregation even at high loading weights of metal cocatalysts, ≈3.0 wt.%, by controlling the photoreduction power for M SA/gCN (M = Cu, Ni, Pd, Pt, Rh, and Ru). Regarding Au and Ag, we interpreted the low binding energy between gCN and SA as imparting difficulties to forming SA without aggregation at high loading weights of metals based on DFT calculations. Measurements of the photocatalytic H 2 -evolution activity of M SA/gCN revealed that among M SA/gCN (M = Cu, Ni, Pd, Pt, Rh, and Ru), Pd SA/gCN exhibits the highest H 2 -evolution efficiency per SA. TA spectroscopy and electrochemical measurements revealed that a M SA cocatalyst contributes to both i) the capture of excited electrons in gCN and ii) the decrease of the overvoltage to the HER, in which the Pd SA cocatalyst is superior, especially in terms of effect (ii). Finally, the H 2 -evolution rate per active site of Pd SA showed 8.6 times higher than that of Pd-NPs cocatalyst.

Experimental Section
Preparation of Pd SA/gCN: Pd SA metal loadings onto gCN were prepared by photodeposition. In this process, gCN (262.5 mg) and an aqueous solution (350 mL) containing an appropriate weight of PdCl 2 ·2NaCl·3H 2 O was added to a quartz cell. After sonication, the suspension was irradiated with a high-pressure Hg lamp (400 W) under atmosphere at 25°C for 30 min. The actual weight of metal loaded onto gCN were determined by ICP-MS of the aqueous solution after mixing. The solid M SA/gCN was collected by centrifugation, washed with water (150 mL), and dried by evaporation.

Measurement of Photocatalytic H 2 -Evolution Activity:
The photocatalytic water-splitting reaction was performed at room temperature with an experimental apparatus built in-house consisting of a 300-W xenon (Xe) lamp (PerkinElmer, Cermax PE300BF) with a long-pass filter (HOYA L42) and a Pyrex cell. The reaction was performed while flowing argon (Ar) gas at the rate of 12 mL min −1 . Before the measurements, the reaction solution containing the prepared photocatalyst (100 mg) and 108 mL of H 2 O/MeOH (10 vol.% methanol was added as a sacrificial agent) was purged with Ar gas for 1 h to ensure complete removal of air from the reaction vessel. The evolved gases were analyzed with a Shimadzu GC-8A gas chromatograph equipped with a thermal conductivity detector and an MS-5A column (Shimadzu, Kyoto, Japan). The H 2 -evolution rate per atom of M SA/gCN was calculated with the following equation: H 2 -evolution rate per atom = H 2 -evolution rate/(loading weight of M/molar weight of M × 6.022 × 1023).

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.