Controlled colloidal metal nanoparticles and nanoclusters: recent applications as cocatalysts for improving photocatalytic water-splitting activity Journal of Materials Chemistry A REVIEW

In recent years, research on the use of metal nanoparticles (NPs) and nanoclusters (NCs) synthesized by liquid-phase reduction in water-splitting photocatalysts has been actively conducted. Water-splitting photocatalysts have been attracting attention because they can produce hydrogen (H 2 ), which is attractive as a next-generation energy source, from solar energy and water. However, further improvement of water-splitting photocatalysts is required for their practical use in society. Recent studies have demonstrated that the active sites (cocatalysts) of water-splitting photocatalysts can be controlled using the advanced NP/NC syntheses and structural modulation techniques established in the ﬁ elds of colloid, NP, and NC chemistry and thereby highly active water-splitting photocatalysts can be developed. If such research progresses further, it is expected that a transition to a new society using H 2 as the main energy source will become possible. However, such applied research has just started and examples of such research are currently limited. The purpose of this review is to introduce the importance of controlled colloidal NPs/NCs in research on water-splitting photocatalysis to readers by summarizing the existing research. We hope that this review will raise interest in the application of metal NPs/NCs in water-splitting photocatalysis and that a society actively addressing energy and environmental problems will become a reality as soon as possible.


Metal nanoparticles and nanoclusters
The nanometer scale is the smallest on which the function of a material appears. Therefore, if a material can be controlled at the nanometer scale, almost all the properties of the material can be regulated. Furthermore, it is possible to produce new physical/chemical properties and functions different from those of the bulk material by controlling the atomic array structure (crystal structure) at the nanometer scale. Because of the importance of and interest in producing nanomaterials with intended structure and function, technology to control nanomaterials (i.e., nanotechnology) is being promoted as a national policy in many countries.
Metal nanoparticles (NPs) (Fig. 1), 1,2 which are aggregates of metal atoms, play a central role in such nanotechnology research. Research on metal NPs was initiated by Faraday in the 1840s in his investigation of metal colloids. 3 Since the 1980s, the research on metal colloids has progressed substantially and the expression "metal NPs" has been coined. [4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23] In the last two decades, research on metal NPs has increased explosively, and the techniques to synthesize metal NPs have advanced dramatically. In recent years, it has become possible to control not only the size of metal NPs but also their geometrical structure. Metal NPs composed of coinage metals such as gold (Au), silver (Ag), and copper (Cu) exhibit localized surface plasmon resonance absorption in the visible region ( Fig. 1), [23][24][25][26][27] Kosuke Wakamatsu. Master course student in the Negishi group at Tokyo University of Science. He  which makes these NPs even more attractive. Because of these properties, metal NPs can be used in surface-enhanced Raman spectroscopy. 28 Therefore, studies on the use of metal NPs to detect trace amounts of molecules are being conducted for applications. 29 In addition, techniques to control the arrangement of metal NPs have been established, 30 which has opened the way to apply metal NPs in the eld of electronic devices. Furthermore, metal NPs are also expected to be useful in biotechnology applications such as drug delivery systems and diagnosis (Fig. 2). 31 Metal nanoclusters (NCs), as shown in Fig. 1, are smaller in size than metal NPs and are also important materials in nanotechnology. 32,33 There is no clear denition of the boundary between metal NPs and metal NCs. However, when the particle size is less than 2 nm, the materials are generally called metal NCs (Fig. 1). Such ultrane metal NCs possess electronic and geometrical structures that are different from those of both the corresponding bulk metal and metal NPs, which leads to the appearance of new physical/chemical properties and functions (Fig. 1). [34][35][36][37][38][39][40][41][42][43][44][45] Furthermore, because the physical/chemical properties and functions of NCs strongly depend on the number of constituent atoms, if the number of constituent atoms in metal NCs can be controlled, numerous functions can be realized by one metal element. When multiple elements are used in NCs, it is possible to access further various functions. [46][47][48][49][50][51][52][53][54][55][56] Gas-phase experiments played a leading role in research on metal NCs in the 1980s and 90s. [57][58][59][60][61][62][63][64][65][66][67][68][69] Also, the synthesis of metal NCs composed of gold, palladium (Pd), and platinum (Pt) started at that time. [70][71][72][73][74][75][76][77][78] However, it was only aer 2000 that research on these NCs began to increase explosively. 79 In 2005, the rst precise synthesis method for thiolate (SR)-protected metal NCs was established. 80 Since then, numerous noble metal and alloy NCs have been precisely synthesized using SR, phosphine, and alkyne ligands.  Since 2007, it has become possible to determine the geometrical structure of NCs by single-crystal X-ray diffraction (XRD) analysis. [110][111][112][113][114] Thus, at present, inorganic chemists precisely synthesize SR-protected metal NCs as organic chemists synthesize organic molecules. It is possible to obtain a deep understanding of the structureproperty relationship of metal NCs with precisely determined geometrical structure.  The use of these well-dened NCs as catalysts, chemical sensors, photosensitizers, and solar cell components is currently being studied (Fig. 2). 34,79,[115][116][117][118][119][120][121][122] In this way, the syntheses and applications are being actively investigated for metal NPs and metal NCs at present. Multiple reviews have been published on the recent research development of such metal NPs and NCs. 34,79,[115][116][117][118][119][120][121][122] Readers hoping to obtain comprehensive knowledge of the synthesis techniques of metal NPs/NCs, their geometrical structures, and potential applications should refer to these reviews.

Application of metal NPs/NCs as cocatalysts in watersplitting photocatalysts
This review outlines the research on the use of metal NPs/NCs in the eld of water-splitting photocatalysis, in which many groups have begun to work in recent years (Fig. 2). With the   depletion of fossil resources and the aggravation of climate change, the transition from a society that depends on fossil fuels to one that uses clean and renewable energy is needed. Hydrogen (H 2 ) releases a large amount of energy upon combustion. In addition, the combustion of H 2 generates only water, which does not pollute the environment. Thus, H 2 has great potential as a new energy source that solves energy and environmental problems (Fig. 3).
Most H 2 is currently produced by steam reforming of fossil resources. However, this method releases carbon dioxide as a byproduct and consumes fossil resources. Therefore, if H 2 continues to be produced by this method, it will not provide a solution for both energy and environmental problems. The water-splitting photocatalytic reaction 123 has been proposed as a clean and renewable H 2 production method (Fig. 3). 124 Using this reaction, it is possible to produce H 2 from sunlight and water, which are available in almost unlimited quantities on the earth. This is different from the case of the water-splitting electrocatalytic reaction [125][126][127][128] in which electric power is consumed to proceed the reaction (thus, the combination with the solar cells is indispensable in this case 129,130 ). However, although water-splitting photocatalysts have been attracting attention for many years, further improvement is still required to realize their practical use.
Water-splitting semiconductor photocatalysts are oen composed of semiconductor photocatalysts and metal NP/NC cocatalysts that work as reaction sites (Fig. 4). 131 The cocatalyst plays a role in promoting the photocatalytic reaction, and it has been shown that control of the cocatalyst particle size and dispersion improves photocatalytic activity. However, it is difficult to control the particle size and chemical composition of cocatalysts loaded using conventional photodeposition and impregnation methods because the metal NPs/NCs are grown on the surface of the photocatalyst in these methods (Fig. 5A). 132,133 To overcome these problems and produce highly active water-splitting photocatalysts suitable for practical use, it is essential to introduce new techniques for the preparation of water-splitting photocatalysts.
As described in Section 1.1, techniques to control the particle size distribution of metal NPs fabricated by liquid-phase synthesis have already been established. When such controlled metal NPs are adsorbed on a photocatalyst and the surface protective organic molecules (ligand, polymer, etc.) are removed, it is possible to obtain controlled metal NP-loaded photocatalysts with high watersplitting activity (Fig. 5B). 132,133 When precisely controlled metal NCs are used as the precursor, it is possible to regulate the Fig. 4 Schematics of photocatalytic reactions. (A) One-step photoexcitation system for overall water splitting, (B) Z-scheme system for overall water splitting, and half reactions using a sacrificial reagent; (C) H 2 evolution reaction (HER) and (D) O 2 evolution reaction (OER). VB and CB mean the valence and conduction band, respectively. cocatalyst on a photocatalyst with atomic precision. [134][135][136] For such precise metal NCs, detailed information about electronic and geometrical structures can be obtained by various high-resolution measurements and theoretical calculations, as shown in the previous studies on the supported metal NPs/NCs. 118,137 Therefore, loading the cocatalyst controlled with atomic accuracy would lead to the clear design guidelines for high activation of water-splitting photocatalysts. If appropriate metal NCs are designed and fabricated based on the knowledge obtained in this way, the activity of water-splitting photocatalysts is expected to be further enhanced.

Purpose of this review
As described in Section 1.2, water-splitting photocatalysts can be activated using cocatalysts fabricated by precise synthesis and structure control techniques (nanotechnology) established in the elds of colloid, NP, and NC chemistry. Such highly active water-splitting photocatalysts could lead to the resolution of energy and environmental problems. However, such applied research has only recently started, and at present the examples are limited. The purpose of this review is to introduce the importance of using controlled colloidal metal NPs/NCs in research on water-splitting photocatalysis by summarizing the existing research. We hope that this review will increase interest in the application of metal NPs/NCs in water-splitting photocatalysts. Ultimately, it is anticipated that the progress of this research will lead to a society in which energy and environmental problems have been solved.

Contents of this review
This review is constituted as follows. First, Section 2 provides an overview of water-splitting photocatalysts. Then, Section 3 outlines the methods used to load cocatalysts. Section 4 presents examples of research using metal NPs/NCs fabricated by liquid-phase synthesis as cocatalysts in water-splitting photocatalysts. In this section, we also present examples of studies using metal NPs/NCs for H 2 or O 2 production from water (the half reactions of water splitting), although these reactions do not realize complete decomposition of water. Section 5 shows the importance of elucidating the interface between metal NPs/ NCs and substrates, and introduces appropriate methods for such a purpose. Section 6 summarizes this review, and Section 7 briey describes the prospects of metal NPs/NCs in water splitting.
It should be noted that there are many studies using NPs/ NCs not as active sites but as light absorption sites in watersplitting photocatalysis. [138][139][140][141][142][143][144] A number of reviews on the use of NPs/NCs as photosensitizers have already been published, 37,115 so this review does not cover such research.

Reaction mechanism
As described in Section 1.2, a water-splitting photocatalyst generally consists of a semiconductor photocatalyst and cocatalyst. When the photocatalyst is irradiated with light, electrons are excited from the valence band (VB) to the conduction band (CB) in the semiconductor photocatalyst. The excited electrons in the CB and holes remaining in the VB move to the photocatalyst surface and/or cocatalyst, where the electrons reduce water and the holes oxidize water (Fig. 4). 131,132 Theoretically, if the bottom edge of the CB of the semiconductor photocatalyst is more negative than the reduction potential of water (0 V vs. normal hydrogen electrode; NHE), the reduction reaction of water proceeds and H 2 is produced; this is called the H 2 evolution reaction (HER). In addition, if the top edge of the VB of the semiconductor photocatalyst is more positive than the oxidation potential of water (1.23 V vs. NHE), the oxidation reaction of water proceeds and O 2 is produced, which is called the O 2 evolution reaction (OER) (Fig. 6). [145][146][147] However, a semiconductor that satises these conditions is not always capable of completely decomposing water (i.e., overall water splitting), which is largely related to the following factors: (i) the high activation energy of water splitting makes it difficult for this reaction to proceed; (ii) the reaction is deactivated by recombination of electrons and holes; and (iii) the produced H 2 and O 2 can undergo the reverse reaction. Therefore, it is important to suppress these negative factors when designing a cocatalyst.

Reaction system
The recent major systems used for the water-splitting reaction are one-step photoexcitation systems 148 and two-step photoexcitation systems (i.e., Z-scheme systems). 131,132,147,[149][150][151][152] In addition to these systems, much research has been conducted on the half reactions of water splitting (HER, OER) using sacricial agents. Here, these reactions are outlined.
2.2.1. One-step photoexcitation systems. In one-step photoexcitation systems, water is completely decomposed by a single semiconductor photocatalyst (Fig. 4A). These one-step photoexcitation systems for water-splitting photocatalysis originated from the Honda-Fujishima effect 123 reported in 1972. Early research mainly focused on the development of photocatalysts only photoexcited by ultraviolet (UV) light. [153][154][155] These studies produced many UV-active water-splitting photocatalysts with high quantum yields.
On the other hand, to realize the practical application of water-splitting photocatalysis, it is necessary to use visible light, which accounts for about 40% of solar energy. As described in Section 2.1, to completely decompose water, the positions of the CB and VB of a photocatalyst must be at appropriate energies to enable the HER and OER, respectively. Furthermore, to suppress recombination of excited electrons and holes, the photocatalyst needs to have high crystallinity and large specic surface area. Because of these requirements, only a limited number of photocatalysts for one-step photoexcitation systems that can completely decompose water with visible light have been reported. 151 2.2.2. Z-scheme systems. There are few semiconductor photocatalysts that satisfy the following three factors: (i) absorption of visible light; (ii) suitable energy levels for overall water splitting; and (iii) appropriate properties including high stability and non-toxicity. However, if a semiconductor that can catalyze the HER is combined with another that catalyzes the OER and a redox couple that transfers charges (electrons and holes) between them, overall water splitting is possible. Such a system, which imitates photosynthesis in plants, is called a Zscheme system (Fig. 4B). 132,147 Z-scheme systems include semiconductor photocatalysts that proceed a half reaction of water splitting (HER or OER). Therefore, compared with the case for one-step photoexcitation systems, there are many photocatalysts available, and it is possible to use a longer wavelength of sunlight. Additionally, because the HER and OER occur on different photocatalysts, it is possible to suppress the reverse reaction between the generated H 2 and O 2 by using a two-cell reaction tube with an ion-exchange membrane. Also, an operation to separate the generated gas is unnecessary. However, in Zscheme systems, the reverse reaction involving a redox couple occurs, which is different from the case for one-step photoexcitation systems. Furthermore, because two photons are required for one reaction in overall water splitting, Z-scheme systems have the disadvantage of lower theoretical conversion efficiency than that of one-step photoexcitation systems.
2.2.3. Half reactions using sacricial reagents. As described above, overall water splitting by a semiconductor photocatalyst is generally difficult to achieve. Therefore, the photocatalytic ability of a semiconductor to drive each halfreaction (HER, OER) is oen evaluated using an electron/hole trapping agent, which is called a sacricial agent. For example, when light is irradiated on water containing a hole trapping agent (methanol, ethanol, ascorbic acid, or triethanolamine (TEOA)) that is more easily oxidized than water, the holes generated by photoexcitation are preferentially used to oxidize the hole trapping agent. Therefore, the reduction of protons (H + ) by photoexcited electrons is likely to occur and H 2 is easily produced. These measurements can evaluate the HER ability of a cocatalyst without being affected by OER activity (Fig. 4C). Similarly, when the activity of the water-splitting reaction is measured in water containing an electron trapping agent (such as Ag + or Fe 3+ ), water oxidation is likely to occur, so the OER activity of various photocatalysts can be easily evaluated (Fig. 4D). Measurement of the activity of such a half reaction is very important for determining whether a photocatalyst satises some of the requirements for water splitting and for exploring photocatalysts that could be used in Z-scheme systems.

Semiconductor photocatalyst materials
Semiconductors used as photocatalysts are mainly metal oxides, metal nitrides, and metal suldes (Fig. 6). 131,151 Also, metal-free graphitic carbon nitride (g-C 3 N 4 ) can be used as a watersplitting photocatalyst. 148,156,157 Metal oxides generally have high stability but do not readily absorb visible light because their VB mainly consists of oxygen 2p orbitals and is located to the positive side of the redox potential of the OER. To overcome these problems, the following strategies have been used to modify photocatalysts: (i) metal-ion doping (

Cocatalyst materials
The cocatalyst promotes the transfer of electrons and holes generated by photoexcitation and also plays a role in lowering the activation energies of the HER and OER (Fig. 4). Each reaction on the cocatalyst involves the adsorption of the substrate, reaction of the substrate, and desorption of the products. The progress of each reaction is strongly related to the adsorption/desorption energies of the substrate/product on the cocatalyst surface. In particular, the chemical reaction on the cocatalyst surface proceeds easily when the Gibbs energy of formation is moderate for the state where the substrates and products are adsorbed on the cocatalyst. This is because the reaction does not occur without the substrate being adsorbed, but the reaction is inhibited when the adsorption is too strong. Generally, metal or metal oxide NPs composed of group 8-11 elements are used as the cocatalyst because of their suitable binding affinity with substrates. These NPs oen have diameters of several to several tens of nanometers. 133,170 Additionally, it has recently been reported that MoS 2 and WS 2 171 are highly active HER cocatalysts. 170 An important point is that metal/ metal oxides NPs/NCs that are highly active in the HER and/or OER but also highly active towards the reverse reactions cannot be used as cocatalysts. Metal/metal oxides NPs/NCs are generally active in reverse reactions, so when they are used as cocatalysts, formation of core/shell structure, alloy state, or a solid solution is necessary to suppress the reverse reactions on the cocatalyst surface. 147

Towards practical use
For the practical use of a water-splitting photocatalyst, the conversion efficiency of solar energy to H 2 (STH) needs to be at least 10%. However, the current maximum STH for water splitting using semiconductor ne particles is about 1.1%. 151,172 Therefore, further improvement of STH is required for the practical use of water-splitting photocatalysts. Among such efforts, the development of visible light-responsive photocatalysts and highly functional cocatalysts are particularly important research subjects. To develop more advanced cocatalysts, it is necessary for the density of active sites to be increased, and the substrate adsorption energy to be optimized by the modulation of the cocatalyst electronic structure. Renement of the cocatalyst is an effective strategy to achieve both of these characteristics and alloying is also effective for the latter. In addition, techniques for loading ne cocatalyst particles on photocatalysts with high dispersity and improving cocatalyst durability are also important for practical application of water-splitting photocatalysts.

Conventional methods
In previous studies, cocatalysts have been loaded on photocatalysts by impregnation and photodeposition methods. These loading methods are widely used because of their simplicity.
The impregnation method ( Fig. 5A(a)) was rst used for loading the cocatalysts on the photocatalysts by Domen et al. in 1980. 172 Since then, this method has been used as an effective method for loading the cocatalysts. In this method, the photocatalyst and precursor metal salt of the cocatalyst are thoroughly mixed using a mortar and pestle and then calcined to deposit ne cocatalyst particles on the photocatalyst surface. In the photodeposition method ( Fig. 5A(b)), a photocatalyst is dispersed in a solution containing a precursor metal salt of the cocatalyst and then light is irradiated on the photocatalyst. The cocatalyst is deposited on the photocatalyst surface through the reduction or oxidation of the metal salt by the electrons or holes generated via photoexcitation. 173 Using the photodeposition method, it is possible to preferentially load the cocatalyst particles on a specic crystal plane of the photocatalyst, which become the reaction sites on the photocatalyst surface. 174 In the catalyst systems fabricated using these two methods, the metal NP/NCs necessarily have a size distribution because the metal atoms are aggregated on the photocatalyst (Fig. 5A). 175 Also, it is not possible to apply methods such as size separation and size convergence to these loaded metal NPs/NCs, unlike metal NPs/NCs dispersed in a solution. Therefore, it is difficult to load cocatalysts with a uniform size on photocatalysts using these conventional methods.

Recent method using colloidal metal NPs/NCs: liquidphase adsorption
In recent years, it became possible to synthesize various types of metal NPs/NCs with controlled size and chemical composition in the liquid phase. Colloidal metal NPs/NCs of the same size can be adsorbed on a photocatalyst and then the protective organic molecules are removed, leading to the production of controlled metal NP/NC-loaded photocatalysts. This approach is oen called liquid-phase adsorption.
In the liquid-phase adsorption method, rst, metal NPs/NCs are adsorbed on the photocatalyst by mixing the metal NPs/NCs and photocatalyst in a solvent. Because metal oxide photocatalysts generally have numerous surface hydroxyl groups (-OH) in water, metal NPs/NCs with protective organic molecules with dissociative functional groups (such as -CO 2 H, -SO 3 H, or -NH 3 groups) can be adsorbed on the photocatalyst  surface at a high adsorption rate via hydrogen bond formation between the protective organic molecules and photocatalyst surface. 176 Even when the protective organic molecules do not have dissociative functional groups, if they contain phenyl groups, the metal NPs/NCs can be adsorbed on the photocatalyst at a relatively high adsorption rate via dipole-dipole interactions. 177 Aer adsorption, the protective organic molecules are removed from the metal NPs/NCs by calcination, photocatalytic oxidation/reduction by light irradiation, or ozone oxidation. This approach has attracted much attention in recent years as a novel method to load photocatalysts with cocatalysts with controlled particle size and chemical composition (Fig. 5B).

Application in overall water splitting
This section introduces research using metal NPs/NCs in the overall water-splitting reaction. The photocatalysts used in these studies are summarized in Fig. 7. The materials and conditions used in these studies are summarized in Table 1. 4.1.1. Use of colloidal metal NPs. Domen et al. 176 started to use metal NPs synthesized by the liquid-phase reduction method for overall water splitting in 2009. They used monodispersed colloidal rhodium (Rh) NPs as a precursor. The sizecontrolled Rh NPs (1.9 AE 0.6 nm) were loaded on a GaN:ZnO solid solution, 160 which is a visible light-responsive photocatalyst ( Fig. 8 and Table 1). The Rh NPs with a size of 1.7 AE 0.3 nm were synthesized using sodium 3-mercapto-1propanesulfonate as a ligand ( Fig. 8A(a)). This ligand is suitable for the synthesis of small NPs and is easily removed from the NPs by calcination aer adsorption on a photocatalyst. Because the sulfonate groups in the ligands of the Rh NPs were not adsorbed at a high rate on the photocatalyst surface, they changed the functional group of the ligands from sulfonate groups to sulfonic acid by cation exchange aer NP synthesis ( Fig. 8A(b)). This conversion promoted the formation of hydrogen bonds and acid-base interactions between the Rh NPs and photocatalyst surface. Thereby, the Rh NPs were adsorbed on the photocatalyst at a high rate (Fig. 8A(c)). Then, the ligands were removed from the Rh NPs by calcination at 673 K for 30 min under reduced pressure, leaving the Rh NPs loaded on the photocatalyst (Fig. 8A(d)). The particle size of the loaded Rh NPs (1.9 AE 0.6 nm) was about one quarter of that of Rh NPs loaded by the conventional photodeposition method. This indicates that the loading method used in this study is suitable for loading ne NPs on the photocatalyst surface. The researchers formed a Cr 2 O 3 shell on the surface of the Rh NPs to suppress the light-induced O 2 reduction reaction (ORR), which is a reverse reaction of water splitting that occurred on the surface of the loaded Rh NPs without a Cr 2 O 3 shell. The Cr 2 O 3 shell was permeable to H + but not O 2 approaching from the outside. [178][179][180] When the cocatalyst surface was protected by a Cr 2 O 3 shell with such characteristics, it was possible to suppress the progress of the reverse reaction while maintaining the H 2 production ability of the catalyst, thereby leading to high water-splitting activity. The team formed Cr 2 O 3 shells on the Rh NPs loaded on photocatalysts prepared by both methods (Fig. 8B). The photocatalyst with Rh NPs loaded by the liquidphase adsorption method exhibited water-splitting activity that was about three times higher than that of the photocatalyst with Rh NPs (7.6 nm) loaded by conventional photodeposition (Fig. 8C). This indicates that miniaturization of the metal core of the cocatalyst is an effective approach to improve watersplitting activity. The group also reported in this paper that cocatalyst loading by liquid-phase adsorption is also effective for loading NPs composed of other noble metals. When the cocatalyst NPs were composed of Pd or Pt, decreased particle size also enhanced the activity of the NP-loaded GaN:ZnO solidsolution photocatalyst.
In 2010, the same group further developed such research. In this study, various noble metals (Rh, Pd, and Pt) and metal oxides (NiO x , RuO 2 , and Rh 2 O 3 ) were used as HER cocatalysts for GaN:ZnO photocatalysts. 181 Both conventional methods (photodeposition and impregnation methods; Fig. 5A) and liquid-phase adsorption (Fig. 5B) were used to load NPs on the photocatalyst. A Cr 2 O 3 shell was also formed on the cocatalyst surface. The highest activity was obtained when Rh was used as the core metal and the cocatalysts were loaded by liquid-phase adsorption. In the liquid-phase adsorption method, Rh cocatalysts with ne particle size can be loaded with high dispersion. Additionally, it is presumed that Rh cocatalysts loaded by liquid-phase adsorption have more metallic properties than those loaded by conventional methods. The photocatalyst with Rh NPs loaded by liquidphase adsorption was considered to show high watersplitting activity because of these factors. This study also revealed that the morphology of the Cr 2 O 3 shell depended on the valence state of the core Rh NPs and pH of the solution during the photodeposition of the Cr 2 O 3 layer. Based on this knowledge, they formed a Cr 2 O 3 shell with appropriate thickness on metallic Rh NPs at an appropriate pH (3.0-7.5) (Fig. 9), which yielded an efficient water-splitting photocatalyst.
In 2013, to further functionalize the Rh NP cocatalyst, Teranishi and colleagues investigated the correlation between the particle size of Rh NPs and water-splitting activity. 182 In this study, Rh NPs were synthesized by polyol reduction using polyvinylpyrrolidone (PVP, M w : 10 000 or 40 000) as a protective polymer. The effects of the pH and temperature of the reaction solution (ethylene glycol solution) on the particle size of the resulting PVP-protected Rh NPs (PVP-Rh NPs) were investigated. The nucleation rate of PVP-Rh NPs was controlled by the pH. The results obtained in this experiment were in good agreement with those of a theoretical calculation reported by Goia and co-workers. 183 The reaction temperature also strongly affected the nucleation rate of PVP-Rh NPs. Based on these ndings, they selected the appropriate pH and reaction temperature conditions and fabricated size-controlled PVP-Rh NPs (1.6 AE 0.3, 2.7 AE 0.3, and 5.1 AE 0.5 nm). Each sample of PVP-Rh NPs was stirred in ethanol with a GaN:ZnO photocatalyst to induce cocatalyst adsorption. Then, PVP was removed from the NPs by calcination at 673 K (Fig. 10A). A transmission electron microscope (TEM) image of the calcined sample revealed that although slight aggregation occurred on  the GaN:ZnO photocatalyst, the Rh NPs still maintained their high monodispersity (1.5 AE 0.3, 3.8 AE 0.8, and 6.6 AE 1.1 nm). Then, a Cr 2 O 3 layer was formed on the Rh NPs by photodeposition. Measurements of the water-splitting activity of the series of photocatalysts revealed that their water-splitting activity increased with decreasing cocatalyst size (Fig. 10B). They attributed this phenomenon to the following effects caused by the miniaturization of Rh NPs: (i) increased relative surface area of smaller Rh NPs; (ii) improved charge separation; and (iii) increased active sites for the HER.
The above examples demonstrated that control of the HER cocatalyst is an effective approach to improve the water-splitting activity of photocatalysts. In 2010, Domen et al. 184 showed that co-loading of an OER cocatalyst with an HER cocatalyst led to even higher water-splitting activity. In this study, GaN:ZnO was used as the photocatalyst, Cr 2 O 3 /Rh (shell/core) NPs were employed as the HER cocatalyst, and Mn 3 O 4 NPs as the OER cocatalyst (Fig. 11A). First, MnO NPs (9.2 AE 0.4 nm) were synthesized by a liquid-phase reduction method as a precursor of the OER cocatalyst. The MnO NPs were adsorbed on the GaN:ZnO photocatalyst and then calcined at 673 K to form Mn 3 O 4 NPs. Next, Cr 2 O 3 /Rh NPs were loaded on the GaN:ZnO photocatalyst as an HER cocatalyst. TEM measurements conrmed that the two types of cocatalysts were loaded on the photocatalyst without covering each other (Fig. 11B)   photocatalyst loaded with only the HER cocatalyst (Cr 2 O 3 /Rh NPs) (Fig. 11C); and (iii) water-splitting activity depended on the loading amount of OER cocatalyst (Mn 3 O 4 NPs). It was noted that such a co-loading method of both HER and OER cocatalysts is also applicable to cases where NPs composed of different metals are used as cocatalysts.
Thus, control of the cocatalyst is effective for both the HER and OER. However, in the early stages of research on cocatalysts, it was unclear whether the HER or OER was the ratelimiting step in the water-splitting reaction. In 2014, Domen and colleagues conducted a study to determine the rate-limiting step of water splitting. 185 In this study, GaN:ZnO was again used as the photocatalyst, along with Mn 3 O 4 NPs, RuO 2 NPs, or IrO 2 NPs as the OER cocatalyst and Cr 2 O 3 /Rh NPs as the HER cocatalyst (Fig. 12A). Evaluation of the water-splitting activity of each photocatalyst revealed that their activities were almost independent of the type of OER cocatalyst (Fig. 12B). Also, the amount of loaded OER cocatalyst to achieve high activity was much smaller than that of the HER cocatalyst. These results suggested that the HER was the rate-limiting step of water splitting by the photocatalyst used in this study. Based on these ndings, they also optimized the HER cocatalyst. Specically, GaN:ZnO photocatalysts co-loaded with RuO 2 NPs and Cr 2 O 3 / Rh NPs were fabricated by both liquid-phase adsorption of Rh NPs (3.3 AE 0.8 nm) and photodeposition of Rh NPs. They measured the water-splitting activities of both types of Cr 2 O 3 /Rh NPs + RuO 2 NPs/GaN:ZnO photocatalysts. The photocatalyst with Cr 2 O 3 /Rh NP cocatalyst loaded using liquid-phase adsorption showed higher water-splitting activity than that with the cocatalyst NPs loaded using the photodeposition method. These results indicate that controlling the HER cocatalyst is very important for enhancing the water-splitting activity of the GaN:ZnO photocatalyst.
When SrTiO 3 was used as the photocatalyst instead of GaN:ZnO solid solution, control of the OER cocatalyst also proved effective to improve water-splitting activity. In 2018, Teranishi et al. 186 synthesized Co x Mn 3Àx O 4 NPs (Co/(Co + Mn) ¼ 0-40 mol%) by doping Co into Mn 3 O 4 NPs using a liquid-phase reduction method (Fig. 13A). The obtained Co x Mn 3Àx O 4 NPs were loaded on an SrTiO 3 photocatalyst as OER cocatalysts aer Cr 2 O 3 /Rh NPs were loaded as HER cocatalysts (Fig. 13B). TEM analysis conrmed that the particle size ($9 nm) of the Co x -Mn 3Àx O 4 NPs was maintained aer loading on SrTiO 3 . The water-splitting activity of the photocatalyst increased with the amount of Co of Co x Mn 3Àx O 4 . The photocatalyst loaded with Co x Mn 3Àx O 4 NPs with a Co/(Co + Mn) ratio of 40 mol% showed water-splitting activity that was 1.8 times higher than that of the catalyst loaded with Mn 3 O 4 NPs ( Fig. 13B and C). To investigate the origin of such activity enhancement, they examined the electrocatalytic activity of Co x Mn 3Àx O 4 -loaded BiVO 4 photocatalysts in the OER. The results revealed that the efficiency of hole transfer from the photocatalyst to the cocatalyst was not increased by Co doping. Based on these results, they attributed the improvement of water-splitting activity to the increased OER activity on the cocatalyst surface induced by Co doping.
4.1.2. Use of atomically precise colloidal metal NCs. As shown in Section 4.1.1, using controlled colloidal metal NPs as a cocatalyst precursor can lead to photocatalysts with improved water-splitting activity. Among the metal NPs/NCs synthesized in the liquid phase, SR-protected metal NCs (metal ¼ Au, Ag, Cu, etc.) can be synthesized with a particle size of about 1 nm.  If such ultrane metal NCs are used as cocatalysts, it is expected that highly active water-splitting photocatalysts will be obtained based on the drastic increase in the surface atomic ratio of such small cocatalyst particles. In addition, it recently became possible to substitute some of the metal atoms of SRprotected metal NCs with a different element. [46][47][48][49][50][51][52][53][54][55][56] If these alloys NCs are used as cocatalyst precursors, it is expected that the effect of the chemical composition of the cocatalyst on the water-splitting activity of photocatalysts could be claried at the atomic level, thereby providing clear design guidelines for the development of highly active photocatalysts.
Since 2013, Negishi's group has published several papers on the use of atomically precise metal NCs as precursors of HER cocatalysts. First, in 2013, they succeeded in loading Au 25 (Fig. 14A), and that most of the ligands were removed by calcination. The water-splitting activity of Au 25 (Fig. 14B). These results indicate that the use of the ne Au NC cocatalyst is indeed effective for enhancing the water-splitting activity of the photocatalyst. In another paper, Negishi's group reported that such an improvement in water-splitting activity caused by the ultra-miniaturization of the Au cocatalyst also occurred when SrTiO 3 was used as a photocatalyst. 188 This result demonstrates that loading the ne Au NC cocatalyst (more widely metal NCs 120 ) is also possible for the other photocatalysts when using SG as a ligand of NCs.
In 2015, Negishi's group synthesized a series of Au n (SG) m NCs (n ¼ 10, 15, 18, 25, and 39) with controlled numbers of constituent atoms with atomic precision and used the NCs as cocatalyst precursors. 189 They investigated the correlation between cocatalyst size and water-splitting activity. In this study, when Au n (SG) m NCs (n ¼ 22, 29, and 33) were used as precursors, the aggregation of Au NCs on the photocatalyst surface occurred during adsorption or calcination. It has been revealed that Au n (SG) m NCs (n ¼ 22, 29, and 33) are metastable species which are kinetically trapped during the NC formation and decompose (release of SG or Au-SG oligomers) in aqueous solution in a shorter time when compared with Au n (SG) m NCs (n ¼ 10, 15, 18, 25, and 39). 81 It was that for Au n (SG) m NCs (n ¼ 22, 29, and 33), part of the Au n (SG) m NCs dissociate during the stirring process and Au-SG oligomers, formed from the dissociated products, adsorb conjunctively with the Au n (SG) m NCs. The existence of such Au-SG oligomers was considered to promote cohesion of clusters on BaLa 4 Ti 4 O 15 during calcination. These results indicate that it is difficult to load Au n (SG) m NCs that are unstable in solution on the BaLa 4 Ti 4 O 15 photocatalyst while maintaining the number of constituent atoms of NCs. This means that it is essential to use metal NCs that are stable in solution as a cocatalyst precursor to achieve precise loading of metal NCs (Fig. 15A). Then, the correlation between cocatalyst size and water-splitting activity was investigated by evaluating the water-splitting activities of Au n NCs/BaLa 4 Ti 4 O 15 photocatalysts prepared using Au n (SG) m NCs (n ¼ 10, 15, 18, 25, and 39), which are stable in solution, 81 as cocatalyst precursors. It was found that the water-splitting activity increased as the size of the Au n NC cocatalyst decreased (Fig. 15B). Because the increase in activity with cocatalyst size decrease was modest, it was considered that the change in activity between the catalysts was mainly caused by the change in the proportion of Au NC surface atoms with miniaturization (i.e., Au atoms that react with hydrogen). Conversely, the difference in water-splitting activity between Au n NCs/BaLa 4 Ti 4 O 15 and Au NPs/BaLa 4 Ti 4 -O 15 could not be fully explained by the difference in the number of surface atoms (Fig. 15B). From various analyses, it was found that the activity per Au atom on the surface of Au 10 NCs/ BaLa 4 Ti 4 O 15 was only about 15-25% of that of Au NPs/BaLa 4 -Ti 4 O 15 . Therefore, it was concluded that the main factor behind the improved activity caused by the ultra-miniaturization of the HER cocatalyst (Fig. 15B) was the increase of the number of surface Au atoms with an efficiency exceeding the decrease in activity per Au atom.
In the same paper, Negishi et al. 189 also examined the importance of ligand removal in such composite systems. Multiple studies have reported that for composites composed of Au n (SG) m NCs and semiconductor photocatalysts, electron transfer occurs even when the ligands are present. 139 Indeed, Au 25 (SG) 18 NC/BaLa 4 Ti 4 O 15 that had not been calcined showed higher water-splitting activity than that of BaLa 4 Ti 4 O 15 without the NC cocatalysts (Fig. 15C). This indicates that electron transfer occurs between the Au 25 (SG) 18 NC and BaLa 4 Ti 4 O 15 without removing the ligands, and thus Au 25 (SG) 18 NC also function as a cocatalyst. However, the water-splitting activity of Au 25 (SG) 18 (Fig. 15C). This means that the presence of the ligands lowers the efficiency of the electron transfer between Au 25 NC and BaLa 4 Ti 4 O 15 or decreases the activity of individual Au atoms. These results claried that ligand removal by calcination is very important to obtain composite photocatalysts with high water-splitting activity.
In this way, the use of the ne metal NCs as a cocatalyst is effective to improve the water-splitting activity of photocatalysts. However, the ORR, which is a reverse reaction of water splitting, proceeds simultaneously with the HER on the Au 25 NC surface. 188 Therefore, to more effectively utilize the high surface area unique to ne metal NCs and thereby obtain highly active photocatalysts, it is necessary to form a shell that suppresses the ORR on the surface of Au 25   Surface science research has revealed that when a metal oxide supporting metal NPs is heated under H 2 or O 2 atmosphere, a strong metal-support interaction (SMSI) is induced and thereby an oxide lm is formed on the metal NPs. [191][192][193][194] Kurashige et al. attempted to use such an SMSI effect to form a Cr 2 O 3 shell on the Au 25 NC surface (Fig. 16A) (Fig. 16A). Then, Au 25 (SG) 18 was adsorbed on Cr 2 O 3 /BaLa 4 Ti 4 O 15 , and both ligand removal and surface protection by the SMSI effect were achieved by calcination. A TEM image of the photocatalyst aer calcination showed the formation of a thin lm with a thickness of about 0.7-0.9 nm around particles with a diameter of about 1 nm (Fig. 16B). This indicates that the Au 25 NC were covered with a chromium oxide layer during calcination (Fig. 16A) (Fig. 16C). This indicates that the formation of the Cr 2 O 3 shell is effective at suppressing the ORR on the surface of the Au NC cocatalyst. The formation of such a Cr 2 O 3 shell also suppressed aggregation of the cocatalyst during light irradiation. This indicates that the Cr 2 O 3 shell formed by the method established by Kurashige and colleagues improves not only the water-splitting activity but also the stability of the cocatalyst on the photocatalyst surface.
It is expected that photocatalysts with higher water-splitting activity could be realized by heteroatom doping of ne Au NC cocatalysts. In previous studies on the effect of heteroatom doping on photocatalyst activity, experiments have been conducted using photocatalysts with distributions of the particle size and doping ratio (chemical composition) of the cocatalysts. [195][196][197][198][199][200][201] To obtain a deep understanding of the effect of heteroatom doping on photocatalytic activity and thereby establish clear design guidelines for photocatalyst activation, it is essential to study composite photocatalysts on which the cocatalysts have strictly controlled chemical composition. Recently, Kurashige et al. 202 attempted to use Au 24 Pd and Au 24 Pt NCs, in which one Au atom of the Au 25 NC is substituted with Pd or Pt, as an HER cocatalyst. Au 24 Pd(SR) 18 and Au 24 Pt(SR) 18 NCs can be precisely synthesized only when a hydrophobic ligand is used. 202 However, the metal NCs protected by hydrophobic ligands could not strongly interact with the hydrophilic surface of BaLa 4 Ti 4 O 15 , which led to poor adsorption on the photocatalyst. Then, they replaced some of the ligands of Au 24 -Pd(SR) 18 and Au 24 Pt(SR) 18 NCs with a hydrophilic ligand, which allowed them to adsorb on BaLa 4 Ti 4 O 15 at a high adsorption rate (Fig. 17A). The ligands on the NC surface were removed by calcination (Fig. 17A).
Investigation of the photocatalyst loaded with cocatalyst NCs controlled with atomic precision revealed the following three points about the heteroatom doping of the Au cocatalyst: (i) the Pd atom was located on the surface of the metal NC and the Pt atom was located at the interface between the metal NC and photocatalyst (Fig. 17B); (ii) Pd doping induced a decrease of water-splitting activity and Pt doping caused water-splitting activity to increase (Fig. 17C); (iii) these opposite effects of Pd and Pt heteroatom doping are strongly related to the location of the doped heteroatom (Fig. 17B). The results also showed that combining Pt doping and surface protection of the cocatalyst with a Cr 2 O 3 shell increased the activity and stability of the photocatalysts to a greater extent than only Pt doping (Fig. 17D).
In the above series of research, Au was used as the base element of the NC cocatalyst. It has been predicted that Rh has higher catalytic activity than Au with respect to the HER based on volcano plots for hydrogen adsorption and desorption. 203 Therefore, it is expected that a highly active water-splitting photocatalyst could be obtained by loading Rh and Cr oxide NPs/NCs as an HER cocatalyst on the photocatalyst. Indeed, Domen and co-workers reported that a photocatalyst loaded with Rh(III)-Cr(III) mixed oxide NPs (Rh 2Àx Cr x O 3 ; particle size ¼ 10-30 nm) showed higher water-splitting activity than that of photocatalysts loaded with metal NPs composed of other elements. [204][205][206][207] If ultrane Rh 2Àx Cr x O 3 NCs could be loaded on  Fig. 16A. Unfortunately, the precise synthesis of Rh 2Àx Cr x O 3 NCs has not been reported. Therefore, in this experiment, a complex containing Rh 2 (SG) 2 as the main component was used as a precursor. First, Rh 2 (SG) 2 was adsorbed on BaLa 4 -Ti 4 O 15 covered with a Cr 2 O 3 lm, and then the resulting photocatalyst was calcined. These operations resulted in loading of Rh 2Àx Cr x O 3 NCs on the photocatalyst (Fig. 18A). Various structural analyses showed that about six Rh 2 (SG) 2 complexes aggregated during adsorption and that the Rh and Cr 2 O 3 lms formed solid solutions during calcination (Fig. 18B). Monodisperse Rh 2Àx Cr x O 3 NCs with a particle size of 1.3 AE 0.3 nm were loaded on BaLa 4 Ti 4 O 15 by this method (Fig. 18C). The obtained photocatalyst exhibited an apparent quantum yield of 16% under 270 nm excitation, which is the highest achieved for BaLa 4 Ti 4 O 15 to date (Fig. 18D). These results indicate that loading Rh 2Àx Cr x O 3 NCs by this method is an effective approach to improve the water-splitting activity of the BaLa 4 Ti 4 O 15 photocatalyst. In principle, this method is applicable to the other photocatalysts. Rh 2Àx Cr x O 3 has already proved a useful cocatalyst in many water-splitting photocatalysts. [204][205][206][207]209,210 In the future, it is expected that high quantum yields can be achieved for many water-splitting photocatalysts using this technique.
In the above research, the Cr 2 O 3 shell on the cocatalyst surface suppressed the ORR, which is one of the reverse reactions of water splitting. When Pt NPs were used as a cocatalyst, the hydrogen oxidation reaction (HOR), which is also a reverse reaction of water splitting, proceeds simultaneously with the HER. In 2013, Wang and colleagues showed that platinum oxide (PtO) NCs could suppress the HOR. 211 In their study, UV-responsive TiO 2 was used as the photocatalyst and PtO or Pt NCs were used as the HER cocatalyst. To load PtO NCs on the photocatalyst, TiO 2 {001} nanosheets and poly(methacrylic acid) were dispersed in an aqueous solution containing chloroplatinic acid. Fine PtO NCs were loaded on TiO 2 by injecting an aqueous solution of NaBH 4 into the vigorously stirring reaction solution to form PtO NCs/TiO 2 . 212 The polymer in PtO NCs/TiO 2 was removed by washing the sample with ethanol several times. Pt NCs were loaded on TiO 2 to form Pt NCs/TiO 2 using the same mixing method without polymer. Scanning TEM (STEM) images revealed that the loaded PtO NCs (Fig. 19A) and Pt NCs had particle diameters of 1.0 AE 0.3 and 2.0 AE 0.5 nm, respectively. The HER and HOR activities of the obtained photocatalysts were evaluated (Fig. 19B). It was revealed that PtO NCs/TiO 2 showed high HER activity and suppressed HOR activity. Similar HOR suppression was not observed for Pt NCs/TiO 2 (Fig. 19B). To clarify the reasons for this difference, they conducted density functional theory (DFT) calculations using Pt 8 O 8 /TiO 2 and Pt 12 /TiO 2 as models (Fig. 19C)  ( Fig. 19D). In this study, because PtO NCs were synthesized and loaded on TiO 2 in a single reaction system, it is difficult to judge whether size-controlled PtO NCs were synthesized and subsequently loaded on TiO 2 or the size of PtO NCs loaded on TiO 2 was controlled by the coexistence of polymer. In addition, the number of constituent atoms of the PtO and Pt NCs was not controlled with atomic precision. However, this study is important because it revealed a different approach to inhibit reverse reactions of water splitting.
In this way, the use of metal NCs with controlled ne size as a cocatalyst can induce an increase in water-splitting activity because of their large surface area. Such precisely controlled cocatalysts make it possible to obtain a deep understanding of the effects of cocatalyst size and heteroatom doping on photocatalyst activity, as well as the origins of these effects. The ndings obtained from these studies are expected to lead to clear design guidelines for the development of highly active and stable water-splitting photocatalysts. Cl 2 $H 2 O) on the surface of an anionic nanosheet by electrostatic interaction and then conducting the hydrogen reduction on the resulting sample (intercalation method). 213 Pt NCs/KCa 2 Nb 3 O 10 obtained by this method showed higher water-splitting activity than Pt NCs/KCa 2 Nb 3 O 10 prepared by the impregnation method and RuO 2 NPs/KCa 2 Nb 3 O 10 fabricated by the intercalation method. It was interpreted that the Pt NCs were loaded monodispersely between the nanosheets because it was difficult for the formed Pt NCs to aggregate between the nanosheets due to steric hindrance. In this way, the control of the particle size of the cocatalyst by a method other than the liquid-phase adsorption method has also been progressing in recent years.

Application in H 2 evolution
In addition to the studies on overall water splitting described in Section 4.1, many reports on the improvement of the activity of each half reaction of water splitting (HER and OER) by controlling the cocatalyst have appeared recently. In the following subsections, we focus on research of HER using controlled colloidal metal NPs/NCs and introduce some typical research examples. The photocatalysts used in these studies are also summarized in Fig. 7. The materials and conditions used in the experiments are summarized in Table 2  face. Therefore, the adsorption of NPs on a substrate differs depending on the crystal face, and thereby the activity of the metal NPs also varies depending on the exposed crystal face. In 2013, Lu et al. 214 showed that controlling such crystal faces indeed led to changes in photocatalytic activity. In their study, UV-responsive TiO 2 was used as a photocatalyst. First, cubicshaped Pt{100} NPs (4.7 AE 1.6 nm), hexagonal-shaped Pt{100/ 111} NPs (6.0 AE 1.5 nm), and tetrahedral-shaped Pt{111} NPs (6.5 AE 2.0 nm) were synthesized (Fig. 20A). Each type of Pt NPs was adsorbed on TiO 2 by stirring the Pt NPs with TiO 2 in water. In a series of photocatalytic HER activity measurements, the photocatalyst using Pt{111} NPs as the cocatalyst showed the highest HER activity (Fig. 20B and C). Pt{111} NPs have a higher percentage of Pt atoms at the corners and edges than is the case for Pt{100} NPs and Pt{100/111} NPs. Therefore, the apparent activation energy of Pt{111} NPs for H 2 generation is low (Fig. 20D). In addition, the Pt{111} NPs contain many active sites. It was considered that these reasons led to the rapid progress of the HER on the Pt{111} NP surface. Model calculations showed that it was easier to form H 2 molecules via bonding of each hydrogen atom on the Pt{111} NP cocatalyst surface than was the case on the Pt{100} NPs and Pt{100/111} NPs.
In 2014, Lu's group also conducted similar research using Au NPs. 215 In this experiment, rst, cubic-shaped Au{100} NPs, truncated cubic-shaped Au{100/111} NPs, and octahedralshaped Au{111} NPs were synthesized by liquid-phase reduction. Each type of Au NPs was adsorbed on TiO 2 by stirring the Au NPs and TiO 2 in water. A series of photocatalytic HER activity measurements revealed that the photocatalyst using Au{111} NPs as a cocatalyst (denoted as Au{111} NPs/TiO 2 ) possessed the highest HER activity of the samples. Photoluminescence measurements and model calculations indicated that Au{111} NPs/TiO 2 showed high HER activity because of the same reasons as Pt{111} NPs/TiO 2 .
As shown in the above examples, to control the cocatalyst, it is very important to select an appropriate element, rene the particle size, improve the dispersibility, form an alloy, form a shell to suppress reverse reactions of water splitting, and control shell thickness. In 2014, Hensen et al. 216 showed that the valence state of the Rh NP surface had also a large effect on the HER activity of a complex system composed of Rh NPs and visible light-responsive g-C 3 N 4 photocatalyst. They synthesized PVP-Rh NPs with sizes of 4.1 AE 0.8, 5.9 AE 1.1, 7.8 AE 1.4, and 9.1 AE 1.7 nm and adsorbed them on g-C 3 N 4 to form PVP-Rh NPs/g-C 3 N 4 (Fig. 21A). Then, PVP on the Rh NP surface was removed by ozone oxidation. In general, removal of the protective ligand induces an increase in HER activity. However, the Rh NPs/g-C 3 N 4 samples obtained aer ozone oxidation showed lower HER activity than PVP-Rh NPs/g-C 3 N 4 (Fig. 21B). When the samples aer ozone oxidation were calcined under H 2 ow, their activity increased. These results indicated that the activity decreased when the surface of the Rh NP cocatalyst was oxidized by ozone, whereas the activity increased when the Rh NPs were reduced by calcination under owing H 2 . Thus, it was claried that the high metallicity of the cocatalyst surface is important to improve the HER activity of Rh NPs/g-C 3 N 4 photocatalysts.
Recently, Hojamberdiev et al. 217 also studied the HER activity of g-C 3 N 4 . In their study, Ni(OH) 2 , which is a relatively inexpensive HER cocatalyst, and halloysite (Al 2 Si 2 O 5 (OH) 4 ) as a hole trapping agent were co-loaded on g-C 3 N 4 (Fig. 22A). The HER activity of the obtained photocatalyst was measured in an aqueous solution containing 10 vol% methanol as a sacricial agent. The HER activity of the catalyst loaded with 1 wt% Ni(OH) 2 was about 40 times higher than that of g-C 3 N 4 alone. This indicates that co-loading Ni(OH) 2 and halloysite is an effective approach to improve the HER activity of g-C 3 N 4 . Ni(OH) 2 loaded on a photocatalyst promoted charge separation by capturing photoexcited electrons, and the negative charges on the halloysite surface captured holes generated by photoexcitation. This behavior led to the extremely high HER activity of the catalyst co-loaded with Ni(OH) 2 and halloysite (Fig. 22B). A model calculation was performed to evaluate the adsorption affinity of water and methanol molecules on the catalyst surface. The results showed that the combination of g-C 3 N 4 , halloysite, and Ni(OH) 2 promoted the adsorption of water and methanol on the cocatalyst surface.
In the above research on HER cocatalysts, mono-metal NPs were used. In 2016, Yao and colleagues reported a study using alloy NPs as a cocatalyst. 218 In this study, visible light-responsive CdS was used as the photocatalyst and Pt-Pd alloy NPs were used as the cocatalyst. The loaded Pt-Pd NPs were Pt-Pd nanocubes with an {100} crystal plane (Fig. 23A) and Pt-Pd nano-octahedra with the {111} crystal plane. Both types of NPs were synthesized with multiple compositions (Pt : Pd ¼ 1 : 1, 2 : 1, 3 : 1, and 1 : 2). The HER activity of the obtained photocatalysts was measured in aqueous solutions containing ammonium sulte as a sacricial agent. The catalyst loaded with Pt-Pd NPs with a Pd : Pt ratio of 2 : 1 showed the highest activity. This nding was attributed to the optimized hydrogen adsorption/desorption energy on the cocatalyst surface with a Pt : Pd ratio of 2 : 1. In addition, comparison of cocatalysts with different crystal planes revealed that the Pt-Pd nanocubes  with the {100} crystal plane had higher activity than that of the Pt-Pd nano-octahedra with the {111} crystal plane (Fig. 23B). The {100} crystal plane has a lower atomic density than that of the {111} crystal plane. They interpreted that the Pt-Pd nanocubes showed higher activity than the Pt-Pd nano-octahedra because the electron transfer from the photocatalyst to the cocatalyst occurred more efficiently on the {100} crystal plane than on the {111} crystal plane due to the lower atomic density of the former than the latter.
The shape and crystal plane of the photocatalyst also affect HER activity. In 2011, Sun et al. 219 reported that the HER activity of TiO 2 was improved when its morphology was modied to form cylindrical TiO 2 nanotubes (TNT) (Fig. 24A(a)). In this study, CuO NPs were used as an HER cocatalyst. CuO NPs were loaded on TiO 2 and TNT by adsorption and calcination (denoted as A-C; this is the same as liquid-phase adsorption) using Cu(NO 3 ) 2 as a precursor to form CuO NPs(A-C)/TiO 2 and CuO NPs(A-C)/TNT, respectively (Fig. 24A(b)). For comparison, CuO NPs were also loaded on TiO 2 and TNT by wet impregnation (WI) to give CuO NPs(WI)/TiO 2 and CuO NPs(WI)/TNT, respectively (Fig. 24A(c)). The HER activity of this series of photocatalysts was measured in an aqueous solution containing methanol as a sacricial agent. The results revealed that the HER activity of the photocatalysts increased in the following order: CuO NPs(A-C)/TNT > CuO NPs(WI)/TNT > CuO NPs(A-C)/ TiO 2 > CuO NPs(WI)/TiO 2 (Fig. 24B). CuO NPs(A-C)/TNT showed the highest activity, which was higher than that of the corresponding photocatalyst using noble metal Pt/Ni NPs as a cocatalyst. Because TNTs are cylindrical, they have a high specic surface area (Fig. 24C). In addition, adsorption and    calcination loaded cocatalyst NPs with higher dispersibility than that in the case of wet impregnation, and NP aggregation was less likely to occur in the former samples than in the latter. The high HER activity of CuO NPs(A-C)/TNT was ascribed to these factors.

4.2.2.
Use of atomically precise colloidal metal NCs. In 2015, Orlov et al. 220 used metal NCs instead of metal NPs as HER cocatalysts. First, they synthesized ne PVP-Pt NCs by liquidphase reduction. The obtained PVP-Pt NCs were adsorbed on CdS as a photocatalyst. Calcination of PVP-Pt NCs/CdS at 400 C under reduced pressure for 24 h and then 400 C under owing hydrogen for 2 h gave Pt NCs/CdS. TEM measurements revealed that the loaded Pt NCs had an average particle size of only about 0.9 nm (Fig. 25A). The HER of the obtained Pt NCs/CdS photocatalyst was measured in an aqueous solution containing Na 2 S and Na 2 SO 3 as sacricial agents. The HER activity of Pt NCs/CdS was 17 times higher than that of CdS without a cocatalyst (Fig. 25B). Also, Pt NCs/CdS showed higher HER activity than CdS with Pt NPs with a size of about 5 nm as a cocatalyst. These results indicated that Pt NCs are effective as HER cocatalysts and that miniaturization of cocatalysts can induce activation. Using a model in which a Pt 38 NC was adsorbed on the CdS (1 010) plane, they investigated the interface structure and electronic interaction between the Pt NC and CdS in this composite system by a DFT calculation (Fig. 25C). The results revealed that the electronic structure and local surface potential changed on both the Pt NC and CdS surfaces because of the strong geometrical and electronic interactions between them. They stated that understanding the structural and electronic interactions between the NCs and photocatalyst is important for elucidating the reaction mechanism to optimize hydrogen production.
In the above study, although the cocatalyst ($0.9 nm) was in the size range of NCs, it is not clear whether the number of constituent atoms was controlled with atomic precision. In recent years, there have been several studies using NCs with atomic precision as cocatalyst precursors. In 2017, Yang et al. 221 studied the HER activity of photocatalyst systems using [Ag 25 223 respectively. Each type of NC was adsorbed on g-C 3 N 4 by stirring for 12 h in a toluene/ dichloromethane mixture with g-C 3 N 4 . The samples were calcined at 150 C under owing Ar for 2 h to yield Ag 25 NC/g-C 3 N 4 and PtAg 24 NC/g-C 3 N 4 . Both types of NCs had a particle size of about 1 nm. The removal of the ligands was conrmed by photoelectron spectroscopy. X-ray photoelectron spectroscopy and X-ray absorption ne structure analyses conrmed that Pt and Ag were in a zero-valent oxidation state in the NCs loaded on the photocatalyst. The photocatalysts were dispersed in an aqueous solution containing TEOA as a sacricial agent and irradiated with visible light (>420 nm) to induce the HER. It was found that PtAg 24 NCs/g-C 3 N 4 generated four times as much hydrogen as Ag 25 /g-C 3 N 4 . The HER activity of PtAg 24 NC/g-C 3 N 4 was 330 times higher than that of g-C 3 N 4 without a cocatalyst ( Fig. 26A and B). Open-circuit voltage decay (OCVD) measurements revealed that PtAg 24 NC/g-C 3 N 4 had a longer carrier lifetime than that of Ag 25 NC/g-C 3 N 4 (Fig. 26C). The high HER activity of PtAg 24 NC/g-C 3 N 4 was ascribed to effective trapping of the photoexcited electrons by Pt, which suppressed the recombination of electrons and holes (Fig. 26D). Thus, electron transfer was promoted by substituting one Ag atom of Ag 25 with Pt, thereby improving the HER activity of the resulting NCloaded photocatalyst.
In 2018, Fang et al. 224 also succeeded in improving the HER activity of g-C 3 N 4 using precisely controlled metal NCs as a cocatalyst. They used Au 25 (Cys) 18 with L-cysteine (Cys) as a ligand as a cocatalyst. First, Au 25 (Cys) 18 (0.18, 0.49, or 0.96 wt%) and g-C 3 N 4 were stirred in ethanol for 1 h to adsorb Au 25 (Cys) 18 on g-C 3 N 4 . The obtained series of photocatalysts was dispersed in aqueous solutions containing TEOA as a sacricial agent and then the HER was induced by irradiating the aqueous solution with visible light (>420 nm). The photocatalyst with 0.96 wt% Au 25 (Cys) 18 exhibited the highest HER activity of the samples (Fig. 27A and B). Photoelectrochemical and photoluminescence measurements showed that the photocatalyst loaded with 0.96 wt% Au 25 (Cys) 18 promoted charge separation to the greatest extent of the catalyst series. Based on these results, they concluded that the adsorption of Au 25 (Cys) 18 not only provided HER active sites, but also induced effective interfacial charge transfer between the cocatalyst and g-C 3 N 4 ( Fig. 27C and D).
Thus, the precisely controlled NCs function as an effective HER cocatalysts. In addition, as described in Section 4.2, when precisely controlled NCs are used as a HER cocatalyst, it is easy to analyze the structure of the cocatalyst and clarify the origin of the improved activity. To utilize the advantages of such precise NCs, it is necessary to suppress aggregation of the cocatalyst on the photocatalyst surface. Section 4.2 revealed that the formation of a Cr 2 O 3 shell suppressed both cocatalyst aggregation and the reverse reaction of water splitting. In 2019, Shi et al. 225 revealed that encapsulating precisely controlled NCs with metal-organic frameworks (MOFs) can also suppress the aggregation of NCs. In this study, Au 25 was used as a precisely controlled NC and ZIF-8 was used as the MOF. First, Au 25 (SG) 18 ($1.2 nm) was encapsulated into ZIF-8 by a coordinationassisted self-assembly strategy. Next, TiO 2 crystals were grown on the ZIF-8 surface by a hydrothermal method. Finally, a rhenium (Re) complex was adsorbed on the surface of the TiO 2 crystals to form ZIF-8/Au 25 (SG) 18 NC/TiO 2 -ReP with an Au content of 0.5 wt% (Fig. 28A). In this photocatalyst, Au 25 (SG) 18 was used as an HER cocatalyst and the Re complex was used as a cocatalyst for the CO 2 reduction reaction. TEM measurements conrmed that during the preparation of ZIF-8/Au 25 (SG) 18 NC/ TiO 2 -ReP, the aggregation of Au 25 (SG) 18 NC hardly occurred. The obtained photocatalyst was dispersed in an aqueous solution containing TEOA as a sacricial agent and irradiated with visible light (>420 nm) to induce the HER and CO 2 reduction reaction. The HER activity of ZIF-8/Au 25 (SG) 18 NC/TiO 2 -ReP was about twice that of the case using Au NPs ($10 nm) as the HER cocatalyst. The turnover frequency of ZIF-8/Au 25 (SG) 18 NC/TiO 2 -ReP was 87 h À1 (based on an Au content of 0.5 wt%) and its apparent quantum yield at 500 nm was 2.06%. No decrease in activity was observed even when the light irradiation was continued for 7.5 h. In addition, high-angle annular dark eld-STEM-energy-dispersive X-ray spectroscopy measurements indicated that the monodispersity of the Au 25 (SG) 18 NC in ZIF-8/ Au 25 (SG) 18 NC/TiO 2 -ReP was maintained aer the photocatalytic reaction (Fig. 28B). These results demonstrated that encapsulating the cocatalyst with an MOF is also an effective way to produce stable and highly active photocatalysts. 4.2.3. Use of other controlled NPs/NCs. In this way, using the liquid-phase adsorption method makes it easy to control the particle size of the HER cocatalyst. On the other hand, there are several examples in which the particle size of the HER cocatalyst was successfully controlled by the conventional method. For example, in 2008, Mallouk et al. directly deposited Rh(OH) 3 NPs on the layers of Dion-Jacobson-type perovskite HCa 2 Nb 3 O 10 or KCa 2 Nb 3 O 10 , and then calcinated the sample to uniformly load Rh 2 O 3 NPs between the layers. 226 Fine Rh 2 O 3 NPs were considered to be loaded in a highly dispersed state between the layers of KCa 2 Nb 3 O 10 because Rh 3+ in Rh(OH) 3 NCs was immobilized  promoted the charge separation in the composite of Co 3 O 4 NPs/ g-C 3 N 4 and showed an apparent quantum yield of 1.1% (at 420 nm) for oxygen production from aqueous silver nitrate solution ( Fig. 29 and Table 3 228 In this study, WO 3 , RuO 2 $nH 2 O NPs, and IO 3 À /I À were used as the OER photocatalyst in a Z-scheme system (Fig. 4B), cocatalyst, and electron mediator, respectively. Because the reduction reaction of IO 3 À is a six-electron reaction, it is rate-limiting in the water-splitting reaction of Z-scheme systems. Therefore, it is necessary to load PtO x or RuO 2 as a cocatalyst on the OER photocatalyst to accelerate this rate-limiting reaction. However, because the conventional loading method includes calcination, it is difficult to use this loading method for thermally unstable photocatalysts. Therefore, they attempted to load the cocatalyst on the photocatalyst (RuO 2 $nH 2 O/WO 3 ) simply by stirring the cocatalyst precursor (RuCl 3 ) and photocatalyst (WO 3 ) together in solution. They also prepared a sample that was calcined aer adsorption for comparison (RuO 2 /WO 3 ). These two photocatalysts were subjected to electrochemical measurements under light irradiation to evaluate their IO 3À reduction activity. RuO 2 $nH 2 O/WO 3 showed higher activity than RuO 2 /WO 3 for the reduction of IO 3À . RuO 2 $nH 2 O/WO 3 also showed higher activity in the photocatalytic OER than RuO 2 /WO 3 . These results were attributed to RuO 2 $nH 2 O loaded by just adsorption accelerating the IO 3 À reduction reaction more than RuO 2 loaded by a calcination process. Thus, the cocatalyst of the OER photocatalyst affects the reduction reaction rate of the electron mediator. This means that control of the cocatalyst of OER photocatalysts is also effective for enhancing the functionality of Z-scheme-type photocatalysts. This paper further showed that such a method is also effective for enhancing the activities of H 2 WO 4 and Ta 3 N 5 , which are thermally unstable OER photocatalysts. In another recent paper, Maeda et al. 229 loaded RuO 2 on a nitrogen/ uorine-co-doped rutile TiO 2 (R-TiO 2 :N,F) photocatalyst by the same loading method. This demonstrates that the adsorption method reported by Abe's group is highly versatile.

Information on interface between NPs/NCs and photocatalysts
In order to improve the functionality of water-splitting photocatalysts on the basis of the design guidelines, it is essential to  Thus, spectroscopy, SPM, and DFT calculations are powerful tools for elucidating the phenomenon occurred at the interface of NPs/NCs-photocatalysis. In addition to these methods, electrochemical measurements are also useful for predicting the appropriate cocatalysts 121,122,244 because the overpotential obtained by electrochemical measurements are strongly related to the activation energy of each reaction. At present, these measurements and analyses have not been necessarily conducted for the photocatalysts described in Section 4. However, these measurements and analyses should be conducted also for the studies using the advanced water-splitting photocatalysts (Section 4). The results would provide a deeper understanding of their interfaces and enable us to predict the appropriate elements, size, and geometric structure of the cocatalysts suitable for obtaining highly active photocatalysts.

Summary
This review summarized recent studies on the application of metal NPs/NCs in water-splitting photocatalysis. Through this summary, the followings points regarding cocatalyst control were claried.
(i) The use of precise synthesis and structure control techniques (nanotechnology) established in the elds of colloid, NP, and NC chemistry are effective for controlling the cocatalysts of water-splitting photocatalysts to obtain highly active watersplitting photocatalysts. At present, the kinds of ligands of the NPs/NCs and the photocatalysts used for liquid-phase adsorption method are limited to several species because this type of study just started in recent years. However, considering the mechanism (Section 3.2), other ligand-protected metal NPs/ NCs 245 and photocatalysts 131 seem to be also applicable to the liquid-phase adsorption method.
(ii) To control a cocatalyst, it is very important to select appropriate elements, rene the particle size, improve dispersibility, form an alloy, form a shell with controlled thickness that suppresses the reverse reaction, expose a suitable crystal face, and control the charge state.
(iii) The formation of a shell that suppresses the reverse reaction is effective to enhance both the activity and stability of the cocatalyst.
(iv) Although the above-mentioned approaches provide an overall guidance to enhance cocatalyst activity, the most effective means to achieve high activation differs slightly depending on the combination of photocatalyst and cocatalyst.
(v) In one-step photoexcitation systems for overall water splitting, control of both HER and OER cocatalysts is effective for enhancing the overall activity of water-splitting photocatalysts.
(vi) For Z-scheme catalysts, reduction of the mediator may be rate-determining. Thus, control of the cocatalyst that promotes mediator reduction is an effective approach to increase the rate of this reaction.
(vii) Using precisely controlled NC cocatalysts provides a deeper understanding of geometrical structure and electronic state of the cocatalyst, and thereby greater knowledge of the main factors inuencing photocatalyst activation.
These ndings are expected to be useful for researchers in the eld of water-splitting photocatalysis, as well as those interesting in the applications of controlled NPs/NCs.

Outlook
As mentioned above, it is necessary to improve the STH of a water-splitting photocatalyst to 10% or more to enable its practical use. Thus, the following studies are considered important in the future.
(i) Close collaboration between photocatalyst chemists and NP/NC chemists. To enhance the activity of water-splitting photocatalysts, it is necessary to improve both the semiconductor photocatalyst and cocatalyst. Regarding semiconductor photocatalysts, photocatalysis researchers would continue to make substantial improvements. On the other hand, the recent studies revealed that the use of precise synthesis and structure control techniques established in the elds of colloid, NP, and NC chemistry is very effective at raising cocatalyst activity. Therefore, researchers in the eld of NPs/ NCs should also conduct thorough research on enhancing the activity of cocatalysts. Effective approaches to maximize activation depend on the specic combination of photocatalyst and cocatalyst. If NCs/NPs researchers continue to conduct cocatalyst research using commercial photocatalysts, the results would not lead to the high activation of the most advanced photocatalysts. To develop highly active water-splitting photocatalysts for practical use, researchers in both elds need to collaborate more closely in future research.
(ii) Selective loading of cocatalysts on the optimal crystal plane. The crystal planes of semiconductor photocatalysts show different transport rates of photogenerated electrons and holes. If HER and OER cocatalysts could be selectively loaded on crystal planes with rapid electron and hole transport, electrons and holes could be efficiently used in the relevant reactions. In fact, research on photocatalysts in which a cocatalyst was loaded by photodeposition demonstrated that such selective cocatalyst deposition is an effective approach to increase activation. 246 In the case of photodeposition, the HER cocatalyst is formed by reduction of metal ions and the OER cocatalyst is formed by oxidation of metal ions. Therefore, each cocatalyst is inevitably loaded on a crystal plane where electrons and holes can easily reach the active sites. Even for colloidal NPs/NCs, if such crystal plane-selective loading can be achieved, higher activity should be obtained compared with that provided by the present method. One of the methods seems to adsorb the colloidal metal NPs/NCs on the photocatalyst aer a particular crystal-plane which is undesired site are selectively covered with a surfactant.
(iii) Control of the geometrical structure of loaded NPs/NCs. The geometrical structure of loaded NPs/NCs appears to show some variation. In the case of NPs, the size distribution occurs at the synthesis stage. In the case of NCs, there is no variation in size (chemical composition) and geometrical structure at the synthesis stage, but the NCs aer loading sometimes seem to have varied geometrical structure. To identify the geometrical structures of NPs/NCs that lead to high activity and selectively load NPs/NCs with such geometrical structures as cocatalysts, it is necessary to establish a new method for controlling the geometrical structure of the loaded NPs/NCs. For NPs/NCs dispersed in solution, post-treatment methods for size 247 and structure convergence 55 have been reported. In the future, it is expected that post-treatment to achieve size and structure convergence will also be established for loaded NPs/NCs, which would realize the loading of NPs/NCs with controlled geometrical structure.
(iv) Elucidation of the calcination process. Protective organic molecules on colloidal metal NPs/NCs are oen removed by calcination. However, the presence or absence of protective organic molecules is generally determined only by X-ray photoelectron spectroscopy and/or X-ray absorption ne structure measurements. In the case of NCs, although NCs are synthesized while controlling the chemical composition at the atomic/molecular level, ambiguity remains about the elimination of the protective ligands. For ultrane NCs, the remaining protective ligands should strongly affect the electronic/ geometrical structure of the NCs. To obtain reliable design guidelines for improving the activity of cocatalysts, it is necessary to determine whether or not the protective ligands remain on the NCs at the atomic/molecular level. Thus, it is necessary to gain a deep understanding of the calcination mechanism and to further improve the structural analysis techniques for loaded NCs. If the calcination mechanism can be claried at the molecular level, it will not be necessary to consider the protective ligands remaining aer calcination and will also be possible to conduct calcination at an appropriate temperature to suppress cocatalyst aggregation on the photocatalyst. This understanding is also necessary for better control of the geometrical structure of loaded NCs.
(v) Utilization of various metal NPs/NCs. In recent years, it has become possible to synthesize various monodisperse NPs/ NCs with controlled size and structure. Inorganic chemists can now synthesize metal NCs as precisely as organic chemists synthesize organic molecules, and the types of synthesized NCs are steadily increasing. However, the types of NPs/NCs used in the study of water-splitting photocatalysts are limited. In the future, more of the existing NPs/NCs should be used in watersplitting photocatalysis research. This would expand knowledge of the major factors affecting cocatalyst activity and increase the possibility of producing photocatalysts with high water-splitting activity.
(vi) Theoretical calculations of real systems. For understanding the main factors inuencing the activity, theoretical research using the actual geometrical structures is required. At present, deep understanding of the chemical composition and geometrical structure of loaded NPs/NCs and their loading sites on photocatalysts have not been obtained experimentally. In the future, it is expected that the structural analysis techniques for loaded NCs will be improved and thereby enable theoretical calculations to be carried out using actual geometrical structures obtained by such experiments. If such a progress would be achieved, it became possible to predict appropriate photocatalyst systems by DFT calculations.
We hope that by overcoming these current limitations of water-splitting photocatalysts, a society able to solve energy and environmental problems will emerge as soon as possible.

Conflicts of interest
There are no conicts to declare.