Creation of active water-splitting photocatalysts by controlling cocatalysts using atomically precise metal nanoclusters.

With global warming and the depletion of fossil resources, our fossil-fuel-dependent society is expected to shift to one that instead uses hydrogen (H2) as clean and renewable energy. Water-splitting photocatalysts can produce H2 from water using sunlight, which are almost infinite on the earth. However, further improvements are indispensable to enable their practical application. To improve the efficiency of the photocatalytic water-splitting reaction, in addition to improving the semiconductor photocatalyst, it is extremely effective to improve the cocatalysts (loaded metal nanoclusters, NCs) that enable the reaction to proceed on the photocatalysts. We have thus attempted to strictly control metal NCs on photocatalysts by introducing the precise-control techniques of metal NCs established in the metal NC field into research on water-splitting photocatalysts. Specifically, the cocatalysts on the photocatalysts were controlled by adsorbing atomically precise metal NCs on the photocatalysts and then removing the protective ligands by calcination. This work has led to several findings on the electronic/geometrical structures of the loaded metal NCs, the correlation between the types of loaded metal NCs and the water-splitting activity, and the methods for producing high water-splitting activity. We expect that the obtained knowledge will lead to clear design guidelines for the creation of practical water-splitting photocatalysts and thereby contribute to the construction of a hydrogen-energy society.


Yuki Kataoka
Yuki Kataoka Master course student in the Negishi group at Tokyo University of Science. He received a BSc (2019) in Chemistry from Tokyo University of Science. His research interests include the activation of watersplitting photocatalysts by heteroatom doping of cocatalysts and elucidation of the calcination process for loading precise metal NCs on photocatalysts.

Background
In the 21st century, humankind is facing unprecedented serious energy and environmental issues such as the depletion of fossil resources and the destruction of the environment on a global scale. 1 Thus, it is expected to address these problems to create a sustainable society as soon as possible. The energy conversion system shown in Fig. 1 is one of the ultimate systems for constructing such a society. 2 In this system, hydrogen (H 2 ) is produced by a photocatalyst 3 and/or an electrolysis cell, [4][5][6][7][8][9][10][11][12][13][14][15] and the obtained H 2 is converted into electric power by a fuel cell. [16][17][18][19][20][21][22][23][24] With such a system, it is possible to obtain energy from water and sunlight, which are almost infinite on the earth. In addition, when this system is used, the energy medium (H 2 ) can be circulated, preventing the problem of energy depletion. Furthermore, carbon dioxide (CO 2 ), which leads to global warming, is not generated in this system. For constructing this system, modern chemistry is expected to improve the functionality of the photocatalysts, electrolysis cells, and fuel cells.

Water-splitting photocatalysts
For H 2 production, photocatalytic reactions ( Fig. 1) can produce H 2 directly from water and sunlight. [25][26][27][28][29][30][31][32][33][34][35][36][37] Therefore, the photocatalytic water-splitting reaction is considered to be one of the cleanest energy-production reactions for humankind. Among the materials that enable such a reaction to proceed, powder semiconductor photocatalysts have the advantages of a simple system and the ease of increasing the surface area. Because of these advantages, if a large-scale H 2 -production plant using a powder semiconductor photocatalyst is constructed in a desert or uninhabited island where sunlight is abundant, a large amount of H 2 is expected to be generated. 38 However, currently, the solar-to-hydrogen (STH) conversion efficiency is only 1.1%. 39 To enable practical use of a water-splitting photocatalyst, it is indispensable to improve the STH conversion efficiency to approximately 5-10%. 33 When a semiconductor photocatalyst is used, H 2 and oxygen (O 2 ) are produced from water via the following three processes: 25 (i) excitation of electrons from the valence band (VB) to the conduction band (CB) of the photocatalyst by light irradiation, (ii) charge transfer of excited electrons and holes

Shuhei Ozaki
Shuhei Ozaki Master course student in the Negishi group at Tokyo University of Science. He received a BSc (2019) in Chemistry from Tokyo University of Science. In 2019, he worked as a study abroad student at The University of Adelade (Greg Metha's group). His research interests include the activation of visible-light-driven watersplitting photocatalysts.

Masanobu
Kawachi Master course student in the Negishi group at Tokyo University of Science. He received a BSc (2020) in Chemistry from Tokyo University of Science. His research interests include the activation of visible-light-driven water-splitting photocatalysts.

Momoko Hirata
Momoko Hirata Master course student in the Negishi group at Tokyo University of Science. She received a BSc (2020) in Chemistry from Tokyo University of Science. Her research interests include the elucidation of the calcination process for loading precise metal NCs on photocatalysts.

Control of cocatalysts
Chemical reactions occurring on the surface of the cocatalyst involve the adsorption and reaction of reactants and desorption of products. Therefore, the efficiency of the chemical reaction is greatly affected by the adsorption/desorption energy of the reactants/products on the surface of the cocatalyst. Previous studies have revealed that the chemical reaction on the surface of the cocatalyst tends to proceed easily when the Gibbs energy in the process of adsorption/desorption of the reactants/products is Fig. 1 Schematic of the energy conversion system expected for constructing a sustainable society. Note that sunlight also produces wind, biomass, and hydro power in addition to solar power. moderate. For this reason, metal/metal oxide NPs of group 8-11 elements with sizes of several to several tens of nanometers are generally used as cocatalysts. 2 In many cases, the cocatalysts are loaded using the impregnation (Fig. 3A) 35 or photodeposition method (Fig. 3B). 40 These methods enable the cocatalyst to be loaded on the photocatalyst using a very simple procedure. When the photodeposition method is used, it is also possible to preferentially load the cocatalyst on the specific crystal planes of the photocatalysts suitable for the reaction. 41 Through studies using these loading methods, it has been clarified that miniaturization of cocatalysts 42 and control of the electronic structure by alloying 43 are extremely effective in creating highly functional water-splitting photocatalysts.

Contents of this review
We aim to improve the functionality of water-splitting photocatalysts to a practical level (Fig. 4) and thereby contribute to the construction of a hydrogen-energy society. 2 To achieve this, we have attempted to strictly control metal NCs on photocatalysts and thereby clarify the details of the effect of controlling the cocatalyst on the water-splitting activity by introducing the precise-control techniques of metal NCs established in the metal NC field into research on water-splitting photocatalysts. 147,148 This feature article summarizes our previous findings.
In Section 2, we first describe the synthesis method and geometrical structure of metal NCs controlled with atomic precision that are used as precursors of cocatalysts. Then, in Section 3, we describe our research on the functionalization of water-splitting photocatalysts using the metal NCs as precursors of cocatalysts. Specifically, the precise loading of the metal NCs on a photocatalyst (Section 3.1), the electronic/geometrical structure of the loaded metal NCs (Section 3.2), the correlation between the types of loaded metal NCs and the water-splitting activity (Section 3.3), and the functionalization of photocatalysts (Section 3.4) are described. After a brief summary in Section 4, the future outlook is described in Section 5.    93,148 In the method shown in Fig. 5A, metal NCs are prepared by reducing a metal salt or a metal-SR complex with a reducing agent. 162 The metal NCs obtained by this method normally have a distribution in their chemical compositions. Therefore, each metal NC with atomic precision can be obtained by the separation of the mixture using fractionation techniques such as polyacrylamide gel electrophoresis (PAGE), 79,[163][164][165][166][167][168] high-performance liquid chromatography (HPLC), 55,156,[169][170][171][172][173][174][175][176][177][178][179][180][181][182] and thin-layer chromatography (TLC). [183][184][185] As opposed to separating them, when the mixture is exposed to severe conditions, less stable species transform into stable species, thereby allowing stable metal NCs to be size-selectively synthesized (size focusing; Fig. 5B). 153,[186][187][188] In the method shown in Fig. 5C, metal NCs are grown with relatively uniform size by selecting appropriate experimental conditions for slowly reducing the metal ions and aggregating the obtained metal atoms (slow reduction). [189][190][191][192] In the method shown in Fig. 5D, metal NCs with a certain chemical composition are first selectively synthesized, and then the ligands are replaced with ligands with a significantly different structure. This reaction finally results in the selective formation of metal NCs with different chemical compositions (transformation). 193 When using another type of metal salts, metal-SR complexes, or metal NCs as a reactant, it is also possible to selectively synthesize alloy NCs composed of several elements (reconstruction). 194,195 In the method shown in Fig. 5E Fig. 5F). 184,200,201 In these synthetic methods, the method shown in Fig. 5A is superior for systematically isolating a series of metal NCs, and the method shown in Fig. 5B-F is superior for size-selectively synthesizing a specific metal NC. In addition to these methods, there have been several reports on the synthesis of new metal NCs using a combination of several of the methods shown in Fig. 5. 202

Geometrical structures
When research on SR-protected metal NCs began, they were thought to have a geometrical structure where the metal core is covered by SR. 203 44 ] 0 (p-MBA = p-mercaptobenzoic acid) (Fig. 6A). 111,205 Since then, the geometrical structures of many SR-protected metal NCs have been determined by SC-XRD, 153,154 and almost all of the SR-protected metal NCs have been observed to have such ''divide and protect'' structures ( Fig. 6B-D). Several research groups, including those of Zhu, Wang, Jiang, Tsukuda, and Pei et al., have shown that SeR-protected metal NCs (Fig. 6E) 206,207 and CRCR-protected metal NCs (Fig. 6F) 124,208,209 also have such ''divide and protect'' framework structures. 3. Activation of water-splitting photocatalysts using atomically precise metal NCs We aim to functionalize water-splitting photocatalysts that can be used as practical materials. Therefore, we used some of the most advanced photocatalysts. Fig. 7 shows the band structures of several advanced photocatalysts. 28 25 . In our first effort, we used glutathionate (SG)-protected Au 25 NCs (Au 25 (SG) 18 ; Fig. 6B) as a precursor, which are stable in solution and can be size-selectively synthesized. 79,187,223,224 In this experiment, Au 25 (SG) 18 225 In the optical absorption spectrum of the aqueous solution after adsorption, optical absorption was hardly observed, indicating that almost all of the Au 25 (SG) 18 was adsorbed on BaLa 4 Ti 4 O 15 . Glutathionate has two carboxyl groups and one amino group. 226 It can be considered that Au 25 (SG) 18 was efficiently adsorbed on the photocatalyst because the polar functional groups of Au 25 (SG) 18 Fig. 9A-D present the transmission electron microscope (TEM) images, S 2p and Au 4f spectra obtained by X-ray photoelectron spectroscopy (XPS), and optical adsorption/diffuse reflectance spectrum of the photocatalysts before and after calcination (Au 25 (SG) 18 /BaLa 4 Ti 4 O 15 and Au 25 /BaLa 4 Ti 4 O 15 ), respectively, for the photocatalysts containing Au with a weight ratio of 0.1 wt% (0.1 wt% Au). In the TEM image of the sample after calcination (Fig. 9A(c)), only particles with sizes similar to those before calcination were observed, indicating that calcination did not cause the aggregation of NCs. In the S 2p spectrum ( Fig. 9B(b)), a peak was hardly observed, indicating that most of the ligands were removed by calcination. In the Au 4f spectrum of the sample after calcination, the peak positions (83.7 and 87.7 eV; Fig. 9C(b)) were different from those of the sample before calcination (84.7 and 88.5 eV; Fig. 9C(a)) and were located at an energy near those expected for Au(0) (84.0 and 87.7 eV). For Au 25 (SG) 18 , the Au 4f peaks appear on the oxidation side (high-energy side) compared with Au(0) because of the partial charge transfer from Au to S. 225 The shift of the Au 4f peaks after calcination strongly indicates that the ligands were removed by calcination and therefore that Au in the metallic state was loaded on BaLa 4 Ti 4 O 15 . In the diffuse reflectance spectrum of the photocatalyst after calcination (Fig. 9D(c)), the peaks in the visible region, which are characteristic of Au 25 (SG) 18 , were not observed. In the optical absorption spectrum of Au 25 (SR) 18 (Fig. 9D(a)), the peaks in the region of 600-800 nm are attributed to the absorption originating in the central Au 13 core of Au 25 (SR) 18 (Fig. 6B). 112,[227][228][229][230] It can be interpreted that these peaks disappeared because the ligands were removed by calcination, thereby resulting in a drastic change of the structure of the NCs. The slight increase in the particle size of the photocatalysts after calcination (Fig. 9A) can also be explained by this structural change. These results confirmed that the ligands were removed from Au 25 (SG) 18 225 In our later research, it became clear that the precise loading of Au 25 using Au 25 (SG) 18 as a precursor can also be performed in the same way when SrTiO 3 is used as a photocatalyst. 231 Effect of the types of precursor NCs. We also attempted to load Au n NCs (n represents the number of Au atoms) composed of other numbers of constituent atoms on BaLa 4 Ti 4 O 15 . In this experiment, Au n (SG) m NCs (n = 10, 15, 18, 22, 25, 29, 33, or 39; m represents the number of ligands 79 ) were used as a precursor. First, an aqueous solution containing Au n (SG) m NCs (0.1 wt% Au) and BaLa 4 Ti 4 O 15 was stirred at room temperature for 1 h. Then, the adsorption efficiency of the Au n (SG) m NCs was estimated by evaluating the amount of Au contained in the supernatant solution after stirring using inductively coupled   18 aqueous solution and the diffuse reflectance spectrum of (b) Au 25 (SG) 18   respectively. For all the Au n (SG) m NCs, the average particle sizes and particle-size distributions were relatively similar for the Au n (SG) m NCs and Au n (SG) m /BaLa 4 Ti 4 O 15 ( Fig. 10A and B). 232 This finding indicates that no significant change in the particle size (especially the increase of the particle size) of the Au n (SG) m NCs occurs between before and after adsorption. In contrast, different behaviors were observed for Au n /BaLa 4 Ti 4 O 15 depending on the type of Au n (SG) m NCs. According to Fig. 10, the NCs can be broadly classified into the following two groups based on their similar phenomena: (1) n = 10, 15, 18, 25, and 39 and (2) n = 22, 29, and 33. The Au n NCs in group 1 maintained a narrow distribution of their particle sizes even after calcination, although the average particle size slightly increased during calcination ( Fig. 10B and C). In contrast, for the Au n NCs in group 2, the average particle size significantly increased during calcination. These two groups are in good agreement with the classification of the stability of Au n (SG) m NCs against dissociation in aqueous solution: the Au n (SG) m NCs in group 2 dissociate in aqueous solution (release of SG or an Au-SR complex) more quickly than those in group 1. 79 From these results, it can be considered that the stability of Au n (SG) m NCs against dissociation is greatly related to the increase in the particle size of the Au n NCs in group 2. Most likely, part of the Au n (SG) m NCs was dissociated during the stirring process, and the Au-SG complex produced by such dissociation was adsorbed on BaLa 4 Ti 4 O 15 together with the Au n (SG) m NCs. It is presumed that the observed phenomenon occurred because such an Au-SG complex promoted the aggregation of Au n NCs on BaLa 4 Ti 4 O 15 during calcination. These results demonstrate that using Au n NCs with high stability in solution as a precursor is essential to precisely load the Au n NCs on BaLa 4 Ti 4 O 15 (Fig. 11).
Effect of loading amounts. We also investigated the effect of the amount of adsorbed NCs on the particle size of the loaded NCs using Au 25 (SG) 18 as a precursor. Fig. 12 lists the particle size distributions of the loaded particles, which were estimated from TEM images of photocatalysts containing each amount of Au 25 (SG) 18 . When the amount of adsorption of Au 25 (SG) 18 was higher than 0.2 wt% Au, the particle size of the loaded Au NCs increased during calcination. 225 Under these conditions, Au 25 (SG) 18 should be adsorbed on BaLa 4 Ti 4 O 15 at a narrow distance. This appears to induce the aggregation of Au 25 (SG) 18 in the calcination process, leading to the increase of the particle sizes. These results indicate that the loaded amount must be less than 0.2 wt% Au to load Au 25 NCs on the photocatalyst while maintaining the particle size of the precursor NCs. The maximum water-splitting activity of the resulting photocatalysts was observed when the amount of loaded Au was 0.1 wt%, and the water-splitting activity decreased when it was greater than 0.2 wt% Au. 225 The following two factors are considered to be involved in this phenomenon: (1) a decrease in the proportion of surface Au atoms in the cocatalyst with an increase in particle size and (2) a decrease in the amount of light absorption of the photocatalyst with an increase in the amount of cocatalyst. Based on these results, the amount of loaded cocatalyst was fixed to 0.1 wt% Au in the subsequent studies using Au n NCs and their related alloy NCs as cocatalysts.
3.1.2. Use of hydrophobic Au NCs as a precursor. As described above, when hydrophilic Au n NCs are used as  This journal is © The Royal Society of Chemistry 2021 the precursor, the Au n NCs can be easily loaded on the photocatalyst. However, hydrophobic metal NCs easily form crystals; thus, it is possible to determine their geometric structure using SC-XRD. 153,154 Thus, most of the metal NCs synthesized to date have been hydrophobic metal NCs. Therefore, we also attempted to establish a method for loading hydrophobic metal NCs on the photocatalyst at high efficiency. 233 Specifically, we replaced part of the hydrophobic ligands with hydrophilic ligands (ligand exchange; Fig. 13A).
In this experiment, phenylethanethiolate (PET), which has often been used as a hydrophobic ligand, was used as the ligand. First, Au 25 (PET) 18 (Fig. 13B(a)), 112 Au 24 Pd(PET) 18 (Pd = palladium; Fig. 13B(b)), 96 and Au 24 Pt(PET) 18 (Fig. 13B(c)) 96 were synthesized with atomic precision. Then, some of the PET was replaced with hydrophilic p-MBA 111 by a ligand-exchange reaction. The matrix-assisted laser desorption/ionization (MALDI) mass spectra obtained for the sample after ligand exchange only contained peaks attributable to Au 25 (PET) 18Ày (p-MBA) y (y = 7-14; Fig. 14A), Au 24 Pd(PET) 18Ày (p-MBA) y (y = 1-9; Fig. 14B), and Au 24 Pt(PET) 18Ày (p-MBA) y (y = 3-10; Fig. 14C). No intrinsic changes were observed in the diffuse reflectance spectra of the NCs or the TEM images before and after ligand exchange. These results indicate that some of the ligands were replaced in each NC without any change in the framework structure of the NC. Each NC was adsorbed on BaLa 4 Ti 4 O 15 at a high adsorption efficiency of 496% after the ligand exchange ( Fig. 14D-F). In the Au L 3 -edge Fourier-transform extended X-ray absorption fine structure (FT-EXAFS) spectra of the photocatalysts after calcination, almost no peaks attributed to the Au-S bond (B1.8 Å) were observed (Fig. 14G). The coordination numbers of Au (7.1-7.7) of each of the loaded NCs estimated from the Au L 3 -edge FT-EXAFS spectra were between those expected for cuboctahedral-structured Au 13 (5.5) and Au 55 (7.9). In the TEM images of the photocatalysts after calcination, particles with sizes similar to those before calcination were observed. These results confirmed that the ligands were removed from Au 25 (PET) 18Ày (p-MBA) y , Au 24 Pd(PET) 18Ày (p-MBA) y , and Au 24 Pt(PET) 18Ày (p-MBA) y by calcination and, therefore, bare Au 25 , Au 24 15 ) exhibited optical absorption different from those with ligands. This finding indicates that the electronic structures of the loaded metal NCs differed from those of the precursor metal NCs. As later described in Section 3.2.2, the calcination process significantly changed the geometric structure of the metal NCs. The change of the electronic structure of the metal NCs is considered to be mainly caused by the change in the framework structure. The Au L 3 -edge X-ray absorption near-edge structure (XANES) spectra ( Fig. 15) showed that the electron density of the Au 5d orbitals in Au 25 15 , the geometric structure of the loaded clusters was also investigated (Fig. 16). 233 Framework structure of Au 24 M (M = Au, Pd, or Pt). Experiments and DFT calculations by other groups have shown that fine isolated Au n clusters tend to have a planar geometry. [234][235][236][237][238][239] In addition, DFT calculations on Au 10 /metal oxide predicted that Au 10 is most stable on metal oxides when they have a twodimensional structure. 240 Based on these results, Au 25 is presumed to have a relatively planar geometry on BaLa 4 Ti 4 O 15 (Fig. 16A). Indeed, the coordination number of Au estimated from the Au L 3 -edge FT-EXAFS and the particle size observed in TEM and high-resolution (HR)-TEM images of Au 25 /BaLa 4-Ti 4 O 15 strongly support this interpretation. 233 On the other hand, Au L 3 -edge FT-EXAFS, TEM, and HR-TEM measurements suggest that alloy NCs (especially Au 24 Pt NCs) have a more three-dimensional structure than Au 25 NCs. 233 For bare Au n NCs produced in the gas phase, it has been shown that doping with Pd or Pt induces the formation of a threedimensional structure in a smaller size region than for pure Au n . 241,242 Au 24 Pd and Au 24 Pt can be considered to have a more   (Fig. 16B and C). indicate that Pd is located on the surface of the loaded metal NCs and is bound to S (Fig. 16B). 233 In the precursor Au 24 Pd(SR) 18 , Pd was located at the center of the metal core and was not bound to S (Fig. 13B(b)). Therefore, it is inferred that this geometric structure was formed because Pd moved to the surface of the particles during calcination and combined with S that was separated from Au. On the other hand, Pt was found to be located at the interface between Au 24 Pt and BaLa 4 Ti 4 O 15 and bound to several O in BaLa 4 Ti 4 O 15 (Fig. 16C). 233 Pt forms a stronger bond with O than Au (318.4 AE 6.7 kJ mol À1 for Pt-O vs. 223 AE 21.1 kJ mol À1 for Au-O) 243 and can bind with multiple O, thereby causing Pt to be located at the interface    (Fig. 16C). On the other hand, Pd forms a weaker bond with O than Au (145 AE 11.1 kJ mol À1 for Pd-O vs. 223 AE 21.1 kJ mol À1 for Au-O). 243 Thus, it is thermodynamically undesirable for Pd to be located at the interface. Because Pd has a higher surface energy than Au (2.003 J m À2 for the Pd (111) surface vs. 1.506 J m À2 for the Au (111) surface), 244 it is also thermodynamically undesirable for Pd to be located on the surface. Accordingly, the Pd-S bond appears to be formed to suppress the instability of the NCs (Fig. 16B).  (Fig. 16C). Pt can bind more strongly with O than Au. Thus, Au 24 Pt is considered to be more strongly immobilized on BaLa 4 Ti 4 O 15 than Au 25 and Au 24 Pd. In fact, Au 24 Pt did not aggregate on BaLa 4 Ti 4 O 15 as much as Au 25 and Au 24 Pd during the water-splitting reaction. 233 This fact also strongly supports our above interpretation.

Effect of controlling the cocatalysts on the catalytic activity
The effect of the chemical composition of the cocatalyst on the water-splitting activity was successfully elucidated by controlling the chemical composition of the metal NCs with atomic precision.
3.3.1. Effect of the Au NC size. The water-splitting reaction was performed by dispersing 500 mg of photocatalyst in 350 mL of water and irradiating it with ultraviolet light from inside using a high-pressure mercury lamp (400 W). The amount of evolved gas was quantified by gas chromatography (Fig. 17). When the gas accumulates in the reaction cell, the gas evolution is suppressed, and thus the changes in gas evolution over time cannot be accurately tracked. 245 Hence, the gas was flowed through the reaction cell in our studies (Fig. 17).
Under conditions of flowing CO 2 . In our early studies, 225,232 CO 2 was flowed into the reaction cell together with argon (Ar) to accelerate the water-splitting reaction. 246 In these experiments, the evolution of H 2 and O 2 increased continuously over time. The volume ratio of the evolved gas was H 2 : O 2 = 2 : 1. These results indicated that the water-splitting photocatalysis proceeded ideally under the conditions of this gas flow.  Fig. 18, the amount of gas increased continuously with a decreasing number of constituent atoms. 232 This finding indicates that Au n /BaLa 4 Ti 4 O 15 loaded with smaller sized Au n NCs has higher photocatalytic activity for Au n /BaLa 4 Ti 4 O 15 with the same amount (wt%) of Au. Because the increase in activity with a decreasing number of constituent atoms is very gradual, the change in activity between them is interpreted to be mainly caused by the change in the ratio of surface Au atoms, i.e., the change in the number of Au atoms that can react with protons (H + ).
On the other hand, the difference in photocatalytic activity between Au n /BaLa 4 232 This finding suggests that the increase in activity by the ultra-miniaturization of the Au cocatalyst (Au NPs -Au n NCs) (Fig. 18) is mainly caused by the increase in the number of surface Au atoms, which exceeds the decrease in activity per Au atom.   Without flowing CO 2 . Although the CO 2 flow in the abovementioned experiments is an effective means to achieve high activity, the photocatalysts are expected to be dispersed over a large area of water and irradiated with sunlight in practical applications. 38 Therefore, we conducted the same measurements under Ar flow conditions (Fig. 17). 247 The results revealed that the presence of CO 2 affects not only the watersplitting activity but also its dependence on the size of the Au cocatalyst. Fig. 19A 15 . This finding indicates that the size effect of the cocatalyst varies depending on the type of flowing gas and that the water-splitting activity cannot be improved by mere miniaturization of the Au cocatalyst in the absence of a CO 2 flow.
The water-splitting reaction consists of multiple reactions (Fig. 20). Therefore, we investigated how the miniaturization of the Au cocatalyst affects each reaction. First, we examined the HER (Fig. 20A). The results revealed that the miniaturization of Au cocatalysts accelerate the HER. However, as mentioned above, the water-splitting activity was not improved by miniaturization of the Au cocatalyst. Thus, we next investigated the effect of the miniaturization of Au cocatalysts on the reverse reaction. The results indicated that the miniaturization of the Au cocatalyst also accelerates one of the reverse reactions, the O 2 -photoreduction reaction (Fig. 20D). It was thus interpreted that under an Ar flow, the water-splitting activity cannot be enhanced by mere miniaturization of the Au cocatalyst as it causes the acceleration of both reactions, thereby cancelling the acceleration effect. 247

Effect of heteroatom substitution.
We also studied the effect of heteroatom substitutions on the water-splitting activity using Au 25 15 with Pt induces an increase in the watersplitting activity. Such an enhancement of the water-splitting activity by Pt substitution has also been observed in the work of another group, [248][249][250] where larger Au-Pt alloy NPs were used as cocatalysts. Fig. 21 demonstrates that such an effect occurs   To elucidate the reason for this change in the water-splitting activity, we investigated the effect of heteroatom substitutions on each reaction over the photocatalyst (Fig. 20). 233 The results indicated that both heteroatom substitutions accelerated the H 2 evolution (Fig. 22). These substitutions were also observed to accelerate the O 2 photoreduction, especially for the Pd substitution. From these results, it can be interpreted that the water-splitting activity was reduced by Pd substitution because its effect on the acceleration of O 2 photoreduction exceeded its effect on the enhancement of H 2 evolution, whereas the watersplitting activity was increased by Pt substitution because the effect of improving H 2 evolution exceeded the effect of promoting O 2 photoreduction (Fig. 22).
Thus, the Pd and Pt substitutions induce different effects on the reactions that occur on the photocatalyst, respectively. This appears to be largely related to the different substitution positions of both elements. As mentioned above, both Pd and Pt substitutions increase the electron density of Au (Fig. 15). Because both H 2 production and O 2 photoreduction require electrons (Fig. 20), this increase in the electron density of Au appears to cause an increase in the reaction rates of both reactions ( Fig. 22B and C).
However, in Au 24 Pd/BaLa 4 Ti 4 O 15 , Pd, which more easily reduces O 2 than Au, is located on the surface of the alloy NCs (Fig. 22B). Pd was bound to S immediately after the photocatalyst preparation (Fig. 16B). However, after light irradiation, S appears to be reduced to H 2 S and then desorbed from the surface of the metal NCs. Thus, Pd is presumed to be exposed during the water-splitting reaction (Fig. 22B) (Fig. 16C); thus, Pt is not exposed during the water-splitting reaction. In addition, because electron transfer occurs more easily via Pt atoms than via Au atoms at the interface, it is expected that photoexcited electrons are more efficiently transferred from BaLa 4 (Fig. 22C). These facts appear to be largely related to the difference in the heteroatom substitution effect between Pd and Pt substitutions.

Creation of highly active water-splitting photocatalysts
In this way, the establishment of the atomically precise control technique of cocatalysts provided a deeper understanding of (i) the electronic/geometric structure of cocatalysts, (ii) the binding mode at the interface between the cocatalysts and photocatalysts, and (iii) the effect of the number of constituent atoms of the cocatalysts and the alloying on the catalytic activities. We have also attempted to create highly active water-splitting photocatalysts based on this knowledge.
3.4.1. Formation of a chromium oxide shell on cocatalysts for suppressing the reverse reaction. As described in Section 3.3.1, decreasing the particle size of Au cocatalysts accelerates the HER. Substituting one Au atom of the Au cocatalysts with Pd or Pt further accelerates the HER. However, these modifications also accelerate the O 2 -photoreduction reaction. Based on this knowledge, it can be expected that if we could suppress the reverse reaction, a highly active water-splitting photocatalyst could be created based on the characteristic of fine Au n NC and heteroatom-doped Au n NC cocatalysts, namely, high H 2 -evolution activity.
One effective means to suppress the reverse reaction of water splitting is to form a chromium oxide (Cr 2 O 3 ) shell on the surface of the loaded cocatalyst. The Cr 2 O 3 shell is permeable to H + but not to O 2 approaching from the outside. Domen et al. reported that when the cocatalyst surface is covered by a Cr 2 O 3 shell with such characteristics, it is possible to suppress only the progress of the reverse reaction while maintaining the H 2 -evolution activity. Thus, this approach is effective for improving the water-splitting activity. 38,146,[219][220][221][251][252][253][254][255][256] In their research, they used the photodeposition method to form the Cr 2 O 3 shell (Fig. 23A) 233 Thus, when the method reported by Domen et al. is used as is, it is difficult to form the Cr 2 O 3 shell on the surface of the loaded metal NCs while maintaining the number of atoms of the metal NCs. However, research in the field of surface science has revealed that when a metal oxide loaded with metal NPs is heated under an H 2 or O 2 atmosphere, a strong metal-support interaction (SMSI) is induced, thereby leading to the formation of an oxide film on the metal NPs. [257][258][259][260][261][262][263][264] Therefore, we attempted to form a Cr 2 O 3 shell on the surface of the metal NC cocatalyst using such a SMSI effect (Fig. 23B)   In the TEM image of the photocatalyst after calcination (Fig. 24E), particles with an average size of 1.1 AE 0.3 nm were observed. In the HR-TEM image of the photocatalyst after calcination (Fig. 24F), a thin layer with a thickness of approximately 0.7-0.9 nm was observed around particles with high electron density. This indicates that the Au 25 was covered with a Cr 2 O 3 layer during calcination (Fig. 23B). Part of the chromium was oxidized to a highly oxidized state (43+) during calcination. Accordingly, we irradiated the photocatalyst with UV light to reduce the highly oxidized chromium oxide to  (Fig. 19A). On the other hand, Cr 2 O 3 /Au 25 /BaLa 4 Ti 4 O 15 , which had a Cr 2 O 3 shell, evolved H 2 at 3032 mmol h À1 (Fig. 19C). These results demonstrate that forming the Cr 2 O 3 shell enhanced the water-splitting activity of the photocatalyst by approximately 19 times. Cr 2 O 3 itself hardly improved the water-splitting activity. 247 In addition, forming the Cr 2 O 3 shell had little effect on the electronic state of Au 25 . 247 Therefore, it can be considered that the improvement in the water-splitting activity was mainly caused by the suppression of the O 2 -photoreduction reaction due to the Cr 2 O 3 shell formation. Indeed, it was experimentally confirmed that the O 2 -photoreduction reaction was greatly suppressed over Cr 2 O 3 /Au 25 /BaLa 4 Ti 4 O 15 . 247 In this way, we succeeded in creating a highly active water-splitting photocatalyst that takes advantage of Au 25 cocatalysts by forming a Cr 2 O 3 shell on the surface of the Au 25 cocatalysts.
This study also revealed that the aggregation of cocatalysts during light irradiation can be suppressed by forming the Cr 2 O 3 shell. 247 When using our Cr 2 O 3 -shell formation method, Au 25 was covered with a Cr 2 O 3 layer on BaLa 4 Ti 4 O 15 (Fig. 23B). In this case, Au 25 should not easily move around on the surface of the photocatalysts. Indeed, Cr 2 O 3 /Au 25 /BaLa 4 Ti 4 O 15 maintained high water-splitting activity for a long time (Fig. 25A). Unlike for Au 25 /BaLa 4 Ti 4 O 15 (Fig. 25B(a)), the particle size of the Au cocatalyst hardly changed even after 10 h light-irradiation for Cr 2 O 3 /Au 25 /BaLa 4 Ti 4 O 15 (Fig. 25B(b)). These results indicate that the Cr 2 O 3 shell formed using our method improves not only the water-splitting activity but also the stability of the cocatalyst on the photocatalyst surface.   shown in Fig. 23B. 233 Fig. 26A presents TEM and HR-TEM images of the obtained photocatalyst, respectively. In the TEM image, particles of 1.3 AE 0.3 nm in size were observed. This particle size was slightly larger than that of Au 24 Pt-(PET) 18Ày (p-MBA) y /Cr 2 O 3 /BaLa 4 Ti 4 O 15 (1.0 AE 0.2 nm). This indicates that for Au 24 Pt/BaLa 4 Ti 4 O 15 , slight aggregation of the cocatalysts occurred during Cr 2 O 3 shell formation. Based on the particle-size distribution ( Fig. 26A(a)), the main product was estimated to be (Au 24 Pt) 1À3 composed of 1-3 Au 24 Pt units, meaning that B77% of the Au 24 Pt was aggregated to form (Au 24 Pt) 2,3 . 233 p-MBA has a smaller number of hydrophilic functional groups than SG. Furthermore, in Au 24 -Pt(PET) 18Ày (p-MBA) y (y = 3-10), not all the ligands have hydrophilic functional groups. Therefore, Au 24 Pt(PET) 18Ày (p-MBA) y (y = 3-10) should more weakly adsorb to the Cr 2 O 3 surface than Au 25 (SG) 18 . It can be considered that slight aggregation of Au 24 Pt occurred during calcination, most likely due to this weaker adsorption. The HR-TEM images confirmed that the Cr 2 O 3 layer was also formed around (Au 24 Pt) 1À3 (Fig. 26A(b)). Fig. 27 shows the amount of gas evolved over Cr 2 O 3 / (Au 24 Pt) 1À3 /BaLa 4 Ti 4 O 15 . This photocatalyst evolved approximately 20 times more gas than Au 24 Pt/BaLa 4 Ti 4 O 15 (Fig. 27), 233 indicating that Cr 2 O 3 /(Au 24 Pt) 1À3 /BaLa 4 Ti 4 O 15 was even more active than Au 24 Pt/BaLa 4 Ti 4 O 15 . In addition, the total amount of gas increased proportionally with time. After light irradiation, the average particle size slightly increased compared with that of Cr 2 O 3 /(Au 24 Pt) 1À3 /BaLa 4 Ti 4 O 15 before light irradiation (Fig. 26B(a)). However, the increase in particle size (1.3 AE 0.3 nm -1.6 AE 0.6 nm) was far more suppressed than for Au 24 Pt/BaLa 4 Ti 4 O 15 without the Cr 2 O 3 layer (1.1 AE 0.2 nm -2.8 AE 0.9 nm). These results suggest that the combination of Pt substitution and Cr 2 O 3 layer formation makes it possible to create highly active and stable photocatalysts.
In this way, we have succeeded in forming a Cr 2 O 3 layer on (Au 24 Pt) 1À3 and thereby creating a photocatalyst with high activity and stability. Unfortunately, for Au 24 Pt, B77% of the Au 24 Pt aggregated during formation of the Cr 2 O 3 layer. Furthermore, there was a slight increase in the particle size after 10 h of the water-splitting reaction, even for the photocatalyst with a Cr 2 O 3 layer. This increase is most likely because all of the alloy clusters were not necessarily covered with a Cr 2 O 3 layer.   However, if these problems could be addressed, more functional photocatalysts will be successfully created through a combination of decreasing the particle size, alloying, and suppressing the reverse reaction by Cr 2 O 3 -layer formation. 3.4.2. Use of rhodium. In the above series of studies, Au was used as the base element of the cocatalyst NCs. However, a volcano plot of H + adsorption/H 2 desorption 265 predicts that Rh should exhibit higher activity than Au as a cocatalyst for the HER. Therefore, loading an ultrafine cocatalyst composed of Rh and Cr oxides onto a photocatalyst surface is expected to induce higher water-splitting activity. Indeed, Domen et al. reported that photocatalysts loaded with Rh III -Cr III mixed-oxide NPs (Rh 2Àx Cr x O 3 , 10-30 nm) exhibited higher water-splitting activity than those loaded with metal NPs composed of other metal elements. [266][267][268][269] The activity of the photocatalysts is expected to increase with the use of finer Rh 2Àx Cr x O 3 NPs. However, it is difficult to load ultrafine Rh 2Àx Cr x O 3 particles onto photocatalysts using conventional methods ( Fig. 3A and B). Accordingly, we attempted to load fine Rh 2Àx Cr x O 3 NCs on BaLa 4 Ti 4 O 15 using our method (Fig. 28). 270 Unfortunately, there have been no reports on the precise synthesis of Rh 2Àx Cr x O 3 NCs. Therefore, in this experiment, Rh-SG complexes, containing Rh 2 (SG) 2 (Fig. 29A). Then, the Rh-SG complexes were adsorbed on the surface of Cr 2 O 3 /BaLa 4 Ti 4 O 15 (Fig. 29B(a)). Various structural analyses revealed that approximately 6 Rh 2 (SG) 2 complexes aggregated during adsorption. 270 Then, Rh-SG/Cr 2 O 3 / BaLa 4 Ti 4 O 15 was calcined at 300 1C under reduced pressure to remove the ligands from the Rh-SG complexes and form a solid solution of Rh and Cr oxides ( Fig. 29B and C). Finally, a small amount of Cr with a slightly higher oxidation state was reduced to Cr III by UV light irradiation (Fig. 28E) Fig. 29D(a)).
The obtained photocatalyst exhibited an apparent quantum yield of 16% (excitation wavelength = 270 nm; Fig. 30), which is the highest achieved for BaLa 4 Ti 4 O 15 to date. 270 For this photocatalyst, almost no decrease in activity and no increase in particle size were observed even after 10 h of the watersplitting reaction (Fig. 29D(b)). Moreover, it was confirmed that both the reverse reaction (Fig. 20C) and O 2 -photoreduction reaction (Fig. 20D) were well suppressed over this sample, as expected. 270 These results indicate that loading Rh 2Àx Cr x O 3 NCs using our method is very effective for creating highly functional water-splitting photocatalysts.
The method of loading Rh 2Àx Cr x O 3 NCs established in this study can be applied to other photocatalysts in principle.  In addition, Rh 2Àx Cr x O 3 is an effective cocatalyst for various water-splitting photocatalysts. [266][267][268][269]271,272 In the future, it is expected that high quantum yields can be achieved for many water-splitting photocatalysts using this loading method.

Conclusions
We have attempted to create practical water-splitting photocatalysts by combining the most advanced techniques of both metal NCs and water-splitting photocatalysts (Fig. 4). The following findings were obtained regarding the loading of an atomically precise cocatalyst, the electronic/geometric structure of a cocatalyst, the correlation between the types of loaded metal NCs and the water-splitting activity, and the functionalization of a water-splitting photocatalyst.
Loading of atomically precise metal NCs   (iv) The size effect of the cocatalyst changes depending on the type of flowing gas. When using flowing gas not containing CO 2 , the water-splitting activity is not changed by mere miniaturization of the Au cocatalyst. This phenomenon is explained by the acceleration of both the HER and O 2 -reduction reaction due to miniaturization of the Au cocatalyst.

Functionalization
(i) A highly active water-splitting photocatalyst that takes advantage of the high H 2 -production ability of ultrafine Au n NCs can be created by forming a Cr 2 O 3 shell on the surface of Au n NCs. When applying Pt substitution to such cocatalysts, further high water-splitting activity can be obtained.  It is expected that these findings will lead to clear design guidelines for the creation of practical water-splitting photocatalysts.

Outlook
The following efforts are considered necessary to obtain a deeper knowledge of water-splitting photocatalysts and thereby create practical photocatalysts.

Establishment of a synthesis method of atomically precise Rh n NCs
The study on Au n /BaLa 4 Ti 4 O 15 , where the number of constituent atoms of the cocatalyst was controlled with atomic precision, revealed the details of the effect of miniaturization of the Au cocatalyst on each reaction. However, although the number of constituent atoms of the cocatalyst was not controlled with atomic precision, even higher water-splitting activity was achieved for Rh 2Àx

Use of non-noble metal cocatalysts
As shown in this feature article, the use of cocatalysts consisting of noble metals, such as Au, Pt, and Rh, is effective for producing high activity. However, these metals are expensive elements. Since previous studies showed that non-noble elements, such as nickel, cobalt, manganese, etc., 210,222,273 can also be used for the creation of active cocatalysts, research searching for effective non-noble metal cocatalysts is expected to be conducted in future studies. To achieve this, various precise synthesis methods are also expected to be established for non-noble metal NCs. 274 Selective loading of metal NCs on the optimal crystal plane In a semiconductor photocatalyst, there are respective crystal planes which electrons and holes generated by photoexcitation can easily reach. If HER and OER cocatalysts could be selectively loaded on such crystal planes, electrons and holes could be efficiently used in the reaction. 37,275 Thus, techniques for selectively loading metal NCs on specific crystal planes are expected to be developed in the future.
Elucidation of the HER and OER activities of existing metal NCs using electrochemical methods The research field of metal NCs has already realized the precise synthesis of many metal NCs. Among them, there might be metal NCs with electronic structures suitable for HER and OER cocatalysts. Because the HER and OER activity of each metal NC can be estimated by electrochemical measurements, 21 it is expected that electrochemical experiments will be widely conducted for the existing precise metal NCs. These studies could lead to the discovery of novel high-performance cocatalysts.

Elucidation of the geometric structures of loaded metal NCs
To understand the structure-property relationship, it is essential to obtain deeper understanding of the geometrical structure of the loaded metal NCs. Therefore, in the future, it is expected that the geometric structure of the loaded metal NCs will be studied using Cs-corrected TEM 276 and scanning TEM. In addition, operando measurements using X-ray absorption fine structure (XAFS) analysis 277 are also expected to be conducted to obtain a deeper understanding of the geometric structure during the water-splitting reaction.

Structural control of loaded metal NCs
Even if SR-protected metal NCs having no variation in the geometric structure are used as the precursor, there appears to be variation in the geometric structures of the loaded metal NCs. To identify the geometric structures of NCs that create high activity and to selectively load such NCs on the photocatalysts, it is necessary to establish methods to control the geometric structure of the loaded NCs. Previous studies have established various methods to refine the size and structure of SR-protected metal NCs dispersed in solution. 153,158,193,278 In the future, it is expected that methods to refine the size and structure will also be established for the loaded NCs.

Decreasing the size of loaded metal NCs
As shown in Section 3.1.1, decreasing the size of cocatalysts leads to higher activity. Recent studies by another group implied that atomically dispersed metal cocatalysts are also effective for improving water-splitting photocatalysts. [279][280][281] In the future, it is expected that cocatalysts will be controlled in a smaller size region by using smaller ligand-protected metal clusters 114,[282][283][284][285] and thereby the effective size of the cocatalysts will be elucidated in more detail.

Elucidation of the charge-transfer rate at the interface between the photocatalysts and cocatalysts
To obtain a highly active water-splitting photocatalyst, it is also important to increase the rate of carrier transfer and the carrier injection efficiency from the photocatalyst to the cocatalyst. 286 However, currently, there is little information on these properties for photocatalysts loaded with fine metal NCs. In the future, it is expected that deeper understanding of the carrier transfer rate and carrier injection efficiency will be obtained through fluorescence lifetime measurements and transient absorption spectroscopy. [66][67][68][69][70][71][72] Such information would lead to clearer design guidelines for the creation of high-performance cocatalysts that can effectively transfer excited electrons from photocatalysts to cocatalysts.  15 and SrTiO 3 used in our research exhibit high apparent quantum yields and stabilities among ultraviolet light-driven water-splitting photocatalysts (left half part in Fig. 7). However, approximately 40% of solar energy is in the visible-light region. Therefore, effective use of visible light is indispensable for the practical application of water-splitting photocatalysts. 78 Currently, there are only a few semiconductor photocatalysts that can completely split water in one step using visible light, such as GaN:ZnO, g-C 3 N 4 , etc. (right half part in Fig. 7). 78 However, overall water-splitting using visible light can also be achieved by a photocatalytic system that uses a two-step reaction called the Z-scheme, which imitates photosynthesis in plants. 25,26,29,[31][32][33][34] In the future, it is expected that both of these visible-light-responsive water-splitting photocatalysts will be more activated based on the knowledge obtained in previous studies.

Guidance by DFT calculations
As described above, in recent years, it has become possible to control the chemical composition of cocatalysts with atomic precision. For photocatalysts loaded with such atomically precise cocatalysts, DFT calculations could predict highly functional photocatalytic systems. 287 In the future, it is expected that research aimed at the improvement of the functionality of photocatalysts will shift from trial-and-error experiments to prediction by DFT calculations.
We hope that many water-splitting photocatalysts for practical use will be created by overcoming these issues and that we can welcome a society in which energy and environmental problems have been solved as soon as possible.

Conflicts of interest
There are no conflicts to declare.