Role of the Support in Gold-Containing Nanoparticles as Heterogeneous Catalysts

In this review, we discuss selected examples from recent literature on the role of the support on directing the nanostructures of Au-based monometallic and bimetallic nanoparticles. The role of support is then discussed in relation to the catalytic properties of Au-based monometallic and bimetallic nanoparticles using different gas phase and liquid phase reactions. The reactions discussed include CO oxidation, aerobic oxidation of monohydric and polyhydric alcohols, selective hydrogenation of alkynes, hydrogenation of nitroaromatics, CO2 hydrogenation, C–C coupling, and methane oxidation. Only studies where the role of support has been explicitly studied in detail have been selected for discussion. However, the role of support is also examined using examples of reactions involving unsupported metal nanoparticles (i.e., colloidal nanoparticles). It is clear that the support functionality can play a crucial role in tuning the catalytic activity that is observed and that advanced theory and characterization add greatly to our understanding of these fascinating catalysts.


INTRODUCTION
Heterogeneous catalysis, where the reactants are in a different phase (typically gas or liquid) from the catalyst (typically solid), plays a central role in the modern-day production of chemicals and fuels. 1 In 2016, the total catalyst market size was reported to be somewhere around $23 billion and is expected to reach $40 billion by 2022, with an annual growth rate of 4.8%. 2 Supported metal catalysts are an important class of catalysts widely used in industry for several reactions including oxidation, (de)hydrogenation, hydrogenation, hydrotreating, deNO x reactions, ammonia synthesis, and Fischer−Tropsch synthesis. In these catalysts, the metal component is often expensive and is used in very small amounts (typically <5%). 1 Smaller metal particles tend to have more active sites exposed compared to larger metallic particles. Hence, metal particle size is a critical structural parameter that determines the catalytic activity of supported metal catalysts. 3 Common catalysts contain very small metal particles, typically in the nm range, which are dispersed onto a high surface area refractory support. Even before the explosion of nanoparticle synthesis methodologies in the recent nanotechnology era, small metal nanoparticles (NPs), dispersed on a solid support material were already being used as catalysts. 4 Colloidal gold NPs are among the earliest nanomaterials to be produced and exploited in a technological application. For instance, gold colloids were used to introduce the dichroic behavior in the now famous Lycurgus Cup, which dates back to the Romans in the fourth century AD. 5 Since then, gold and other metal NPs have been used to a generate range of colors in glassware and windows. Another important milestone in nanotechnology was the synthesis of stable colloidal Au NPs by Faraday in the mid-19th century. 6 Despite its long history of use in metallurgy and glass technology, because gold was considered to be the archetypal unreactive noble metal, it was incorrectly assumed to be a poor candidate for a catalyst material. This perception changed dramatically after two seminal discoveries in the 1980s. Haruta found that supported gold NPs display unparalleled catalytic activity for low temperature CO oxidation. 7 Hutchings predicted and demonstrated that supported gold is the catalyst of choice for producing vinyl chloride monomer via the acetylene hydrochlorination reaction. 8 Since then, gold catalysis has become the subject of received intense attention from both the academic and industrial research communities. 9−11 It has become clear that the identity of the support material and the gold−support interfacial sites generated often play a crucial role in determining the catalytic behavior of supported gold NPs. 10,12 In the field of supported metal catalysts, support materials were often considered to be inert and their primary role was to enhance the stability of the small metal particles via anchoring. In the 1970s, Tauster introduced the term strong metal−support interaction (SMSI) to explain the unexpected H 2 chemisorption properties of noble metal NPs supported on TiO 2 . 13,14 After this report, the potential of the support playing a more active role has been investigated at a fundamental level, mostly using model catalysts. The role of support in practical catalysts is rather more complex and is still not clearly understood in many cases. Supports can play different roles during a catalytic reaction either directly or indirectly. These include providing specific defects sites onto which the metal NPs can be anchored and even stabilized in the case of metastable particles. The support can also enable electron transfer to or from the metal particles and provide additional functionality such as acidity or basicity to the overall supported metal catalyst. All of these factors can dramatically affect the catalytic properties of supported metal materials, and this is especially true for supported Au-based nanoparticle catalysts.
Because of the recent advances in spectroscopic and microscopic methods to characterize supported metal catalysts, a number of articles have now been published on the role of support on the catalytic properties of supported gold-based catalysts, including both monometallic and bimetallic catalysts. In this review, we discuss selected articles in which the role of support on the catalytic properties has been specifically explored. In the first part of this review, we discuss the role of support material in influencing the structural properties of Aubased NPs. Following that, each subsequent section reviews the role of the support on the catalytic properties of Au-based NPs for specific reactions, including CO oxidation, liquid phase alcohol oxidation, hydrogenation/hydrogenolysis, and C−C coupling reactions. The final section of this review briefly discusses the catalytic properties of unsupported gold-based colloids. We have primarily covered selected articles published over the last nine years on this topic, and we conclude the review with a summary and future outlook for this particular line of catalyst research.

THE ROLE OF THE SUPPORT DURING CATALYST SYNTHESIS
The role of support in Au catalysts can be first seen during the catalyst preparation, which has been a challenging task partly due to a relatively low melting temperature of Au (1064°C) compared to that of other precious metals such as Pd (1555°C) and Pt (1768°C). Historically, the discovery that Au-based catalysts can be effective in many industrially important reactions is owed to the successful development of various preparation methods for Au catalysts pioneered by Haruta. 7 The development and the choice of synthesis method are often dictated by the physical and chemical nature of the support materials, including surface area, surface acid/base properties, possible phase transformations, and the presence of defects. Because Au catalysts are well-known to be sensitive to the synthesis methods, clarifying the role played by the support materials for Au catalysts in the preparation process is the first step toward developing an overall understanding of their role in any specific catalytic processes. This section aims to provide a brief overview of common supported catalyst preparation methods for Au catalysts. In some cases, the support materials interact with Au species during the catalyst preparation process to facilitate a high dispersion of Au. They can also serve as anchor sites to stabilize the catalytically active species. These will be mainly discussed in the section 2.1, Preparing Au Catalysts on Oxides and Other Conventional Supports. In other cases, support materials carry specific morphologies and functionalities that need to be preserved or achieved during the catalyst synthesis procedure. These will be discussed in the section 2.2, Preparing Au Catalysts on "Engineered" Nanostructured Support Materials. The following discussion will focus on supported monometallic Au catalysts. Publications on nanoporous Au and Au alloys will be excluded from this section. impregnation method was due to (i) the weak interaction between the AuCl 4 − ion with the oxide surface in an acidic (pH ∼ 1) environment, (ii) particle growth during the calcination step when the chlorine content is high in the catalyst 16 possibly due to a weak bonding between the cationic Au (a soft acid) and the chlorine ion (a hard base), and (iii) chlorine poisoning of the active sites. 16,17 Therefore, alternative methods for preparing Au catalysts were developed to counter these adverse effects, and these are described in this section. An alternative method using sol-immobilization is discussed subsequently in the review as this is a means of fine-tuning the nature of the nanoparticle prior to being supported.
Deposition−precipitation (DP) is one of the main methods for preparing supported Au NPs. Haruta and co-workers 18,19 first succeeded in making active Au catalysts using this method in the early 1990s. In a typical synthesis procedure for Au/TiO 2 , preformed TiO 2 particles are first mixed with an aqueous solution of HAuCl 4 . The pH of the mixture is then adjusted using a base (e.g., NaOH) to a pH between 6 and 10 and aged at 70°C for an hour. The suspension is then washed, dried, and calcined at 400°C for 4 h. Similar methods have been used for depositing Au on other oxide supports such as Fe 2 O 3 , 20 Al 2 O 3 , 17 and MgO. 19 A variant of the DP method utilizing urea instead of a strong base like NaOH for preparing Au catalysts was first attempted by Dekker et al. 21 and then later by Zanella et al. 22 In this method, the hydrolysis of urea (CO(NH 2 ) 2 ) permits gradual and homogeneous control of the OH − concentration and avoids local increases in pH and subsequent precipitation of Au(OH) 3 away from the support. The DP method is also effective on other types of supports, such as activated carbon, 23 nanodiamond, 24 phosphates, 25,26 hydrotalcite type layered double hydroxides, 27,28 g-C 3 N 4 , 29 and hydroxyapatite (HAP). 30 The DP method can even be used to prepare singleatom dispersed Au catalysts. 31 By using UV irradiation instead of a heat treatment, Flytzani-Stephanopoulos and co-workers 32 prepared a Au/TiO 2 material with atomically dispersed Au species that was shown to be active for the low-temperature water−gas shift (LT-WGS) reaction.
During DP synthesis, a key parameter is the surface charge state of the support material, which depends on the pH value and the isoelectric point (IEP) of the oxide. 33,34 For TiO 2 (IEP ∼ 4.5−6.3), the surface will be positively charged (terminated by − OH 2 + ) at pH values lower than the IEP and negatively charged (terminated by O − ) at higher pH values. In conjunction with this, the nature of the Au species generated also depends on the pH value, as well as the concentration of gold and chlorine present, the ionic strength of the solution, and the reaction temperature. According to Moreau et al., 35 the hydrolysis reactions of AuCl 4 − gives rise to a complex equilibrium of different gold chloro-hydroxy species at a given pH value ( Figure  1a). At pH < 2, AuCl 4 − is the dominant Au containing species; at pH > 8, the dominant species will be Au(OH) 4 − ; for pH values between 2 and 8, the major species present are charged AuCl x (OH) 4−x − anions or neutral AuCl x (OH) 3-x (H 2 O) species. Therefore, in the DP process, electrostatic interaction between the oxide surface and Au species occurs at lower pH levels, which explains a higher Au uptake under those conditions. However, the Cl content remains high, which could result in large Au particles during the calcination step. The optimum pH for TiO 2 usually lies around 6−8, where electrostatic adsorption should not take place. Using X-ray absorption spectroscopy, Louis et al. 36 proposed that in a DP method utilizing NaOH, Au species were grafted onto the OH − groups associated with the TiO 2 support, forming Ti−O−Au(OH) 3 − metal complexes. This explains the relatively low Au uptake (i.e., 60% at pH = 6) 37 from the Au precursor solution, which is one of the main limitations of the DP method. The DP method utilizing urea allows a higher Au loading be achieved (i.e., 8%) through the precipitation of nitrogen-containing amorphous compounds or the adsorption of an ammino-hydroxo-aquo cationic gold complex. 38,39 At pH values >9, the Au loading is limited by the increasing solubility of the Au hydroxide species. 38 Another significant limitation of the DP method is that it does not work well for oxides having a low IEP, such as SiO 2 (IEP < 2−4). 34 Anion adsorption (AA) methods have also been attempted by many researchers. In an AA process, the surface of the support is tuned to be positively charged, on which the negatively charged gold chloro-hydroxy species can gradually electrostatically adsorb. The AA process usually takes a long time to complete (typically ∼16 h). This AA method was used by Zenella et al., 22 and Au particles smaller than 4 nm supported on TiO 2 were achieved. Lessard et al. 40 used this approach and made active gold catalysts for low-temperature water−gas shift reactions on La 2 O 3 and La 2 O 2 SO 4 . Because of the high IEPs of these supports, the pH of the mixture was tuned so that the main species adsorbed was Cl-free (i.e., Au(OH) 4 − ). The Au uptake and highest Au loading attainable depends on the nature of the support. By carefully tuning the HAuCl 4 concentration and the pH of the solution, Pitchon, Petit, and coworkers 41,42 achieved 100% Au uptake and a 2% final loading on an Al 2 O 3 support. Furthermore, no gold was lost during the filtration and washing steps. It was also suggested that there is some kind of anion exchange process taking place between the Au species with the surface hydroxyl groups associated with the Al 2 O 3 support. A similar method was later used by the same group to disperse Au onto layered double hydroxides. 43 Nguyen et al. 44 reported a better thermal stability of the Au/γ-Al 2 O 3 catalysts prepared by the AA method compared to those prepared by conventional deposition precipitation (DP).
Single-atom Au catalysts can be also prepared using the AA route. Wang et al. 45,46 reported atomically dispersed gold on ZnZrO x , while Qiao et al. prepared single-Au atom catalysts on Co 3 O 4 , 47 CeO 2 , 48 and FeO x . 49 The metal loading of these catalysts were usually kept at a very low level of around 0.05 at %. For a SiO 2 support that has a low IEP, cationic adsorption is one possible strategy to prepare active Au catalysts. Trichlorobis(ethylenediamine)gold(III) (Au(en) 2 Cl 3 ) can serve as the cationic precursor, and Au/SiO 2 catalysts have been prepared by Zanella et al. 36 and later by Dai and Overbury 50−52 using this cationic absorption approach. As mentioned earlier, in the DP urea or ammonia process, the cationic adsorption of amino-hydroxo-aquo cationic Au complexes were thought to be taking place. 39,53 Preparing Au/SiO 2 with the more readily available HAuCl 4 precursor can be done by adjusting the surface functionality of the SiO 2 . Amine functional groups can provide positively charged aminium ions in an acidic solution, and therefore AuCl 4 − anions can be electrostatically adsorbed. In a 2009 study by Liu et al., 54 commercial SiO 2 supports were refluxed with APTES (H 2 N(CH 2 ) 3 Si(OEt) 3 ) in ethanol for 24 h, so that the amine functional group can be grafted onto the support before adding HAuCl 4 . This effectively shifts the IEP of the SiO 2 to a higher value. 55 In a more recent study, branched polyethylenimine was used to functionalize the SiO 2 to anchor glutathioneprotected Au clusters. 56 It should also be noted that Au catalysts on silica supports can also be prepared using a double-support strategy, which can in some sense be considered as functionalizing the SiO 2 surface with another oxide. For instance, Au catalysts on TiO 2 , 57 CoO x , 58 and FeO x -modified 59 SiO 2 supports have been reported.
Functionalizing activated carbon surfaces by acid washing has been studied by Willock and co-workers. 60 It was found that washing the carbon support with nitric or hydrochloric acid can almost exclusively generate surface hydroxyl groups, which can better assist the nucleation of Au particles compared to a carbon surface covered with ketone groups.
Another way of achieving high dispersion of Au and an intimate contact with the support is to prepare both components cooperatively. The most common approach is the coprecipitation (CP) method, developed by Haruta and co-workers 18,61 for Au catalysts. They reported that certain oxide supports (e.g., α-Fe 2 O 3 , Co 3 O 4 , and NiO) can be precipitated out from the solution largely simultaneously with Au, therefore ensuring good mixing. Haruta's original method involves quickly pouring an aqueous solution of HAuCl 4 and the corresponding metal nitrate precursors into an aqueous solution of sodium carbonate (Na 2 CO 3 ). The resultant precipitates (usually hydroxides) will then be subjected to washing, vacuum drying, and calcination, typically at 400°C to form the final catalyst. In this CP process, through a quick mixing of the acidified precursor solution into the basic Na 2 CO 3 solution (pH = 8), the AuCl 4 − will undergo the hydrolysis process described earlier and release Cl − , which can be further removed during the washing step. The final Au/ metal oxide catalyst is generated during the calcination step. As confirmed by XRD and TEM characterization, at Au loadings of 5−10 at % with respect to the support transition metal, Au NPs below 10 nm in size are generally formed. 18 Andreeva et al. 62 and Hutchings and co-workers 63,64 later reported slightly different CP methods for preparing such metal oxide supported Au catalysts. The modification comes from varying the sequence of mixing the acid and base precursors as compared to Haruta's original method. 18,61 In this case, the Na 2 CO 3 solution was added gradually into an aqueous solution of metal nitrates and HAuCl 4 until a pH of 8−9 was attained, followed by the usual washing, drying, and calcination steps. The resultant catalysts were shown to be active for low-temperature water−gas shift, 62 CO oxidation at ambient temperature, 63 and the direct synthesis of hydrogen peroxide. 64 With the advances of aberration-corrected scanning transmission electron microscopy (AC-STEM), Herzing et al. 65 reported that Au subnanometer clusters and isolated atomic Au species also exist in the catalysts prepared by the CP method. Furthermore, it was proposed that the subnanometer clusters might be responsible for the high activities observed in CO oxidation reactions rather than the Au NPs. Using cyanide leaching and in situ electron microscopy, Allard et al. 66 demonstrated that a significant amount of atomically dispersed Au can be trapped inside the CP generated support materials, which can subsequently diffuse outward to the surface during subsequent heat treatment (Figure 2). This is not too surprising considering the nature of the CP method. He et al. 67 later studied Au on Fe 2 O 3 e prepared by the above two CP methods and found that a larger fraction of Au can be trapped into the support when the acidic and basic solutions are mixed quickly (i.e., via Haruta's original method). The dynamic evolution of Au species during the heat treatment (e.g., via diffusion and aggregation of Au on the surface and via outward diffusion of trapped internal Au species) determines the final population distribution of Au species on the oxide surface as well as the catalytic activity of the material in the CO oxidation reaction.
Kudo et al. 68 further modified Haruta's CP method by adding HAuCl 4 approximately 1.5 h after mixing the metal nitrates with Na 2 CO 3 . They found that this procedure significantly increased the available Au sites for CO adsorption, probably because the support and Au no longer precipitate simultaneously, meaning that much less Au will be trapped inside the support. The method then becomes very similar to a deposition−precipitation method. Another modification of the conventional CP method was reported later by Zhang et al. 69 Here, an electrochemical approach was used to monitor the concentration of Cl − ions in the solution mixture after coprecipitation, which is thought to affect the precipitation of the support and gold hydroxides. The most active catalysts can then be reliably reproduced when the [Cl − ] concentration lies in the 1−3 ppm range and the Au NPs are thought to be mainly sitting on the edge of nanosized Fe 2 O 3 particles after calcination.
One recent example of Au catalysts prepared by coprecipitation is the Au/α-MoC catalysts reported by Ma and coworkers. 70 The catalysts were prepared by mixing aqueous solutions of (NH 4 ) 6 Mo 7 O 24 ·4H 2 O and HAuCl 4 , followed by washing, drying, and a 500°C calcination step. AC-STEM characterization of their materials confirmed the formation of epitaxial Au rafts, 1−2 nm in diameter and 2−4 atomic layers thick, grown on an α-MoC support, which are highly active for the water−gas shift reaction even at room temperature.
Finally, heat treatment is usually needed when preparing Au catalysts in order to convert the precursors of Au and support oxides into their active forms. One of the key roles of the support is to stabilize the Au species and maintain their high dispersion. Recently, de Jongh and co-workers 71 showed that on a TiO 2 support, Au particle agglomeration can be accelerated by the presence of water and/or the presence of Cl − , and it is more pronounced in an oxidizing atmosphere. In contrast, Au on nonreducible supports such as SiO 2 and Al 2 O 3 are remarkably stable in a nonoxidizing atmosphere. In another study by Zhang and co-workers, 72 2−5 nm Au particles epitaxially supported on MgGa 2 O 4 spinel were shown to retain their original particle size even after heating above the melting temperature of bulk gold (1064°C), demonstrating the potential efficacy of particle stabilization effects from the support lattice.

Preparing Au Catalysts on "Engineered" Nanostructured Support Materials
In many cases when synthesizing Au catalysts, efficiently and homogeneously dispersing the Au is the main concern during the preparation, especially when commercial metal oxide support materials are used. In other cases, however, creating and/or maintaining the special nanostructure of the support material can be equally important. Support materials with welldesigned nanostructures and architectures can not only bring additional functionalities to the Au catalysts, but it also can serve as model catalysts for mechanistic investigations. 73 In this section, we will discuss the preparation of Au catalysts on a variety of support materials having specially designed nanostructures. These include (i) ordered porous materials, such as zeolites, mesoporous silica, and metal−organic frameworks (MOFs), (ii) Au catalysts with iron oxide heterostructures that allow magnetic separation, and (iii) Au catalysts with yolk−shell or core−shell nanostructures and so-called "inverse" oxide/ metal catalysts.
2.2.1. Au on Mesoporous Supports. Mesoporous silica, zeolites, and other materials with controlled pore structures are widely used as catalysts and support materials. 74 Preparing Au catalysts on such mesoporous materials is potentially attractive for at least two reasons: First, it is harder for Au species in the mesoporous materials to migrate and agglomerate due to the geometric constraints imposed by the structure, thus resulting in Au catalysts having superior stability compared to those prepared on conventional high surface area metal oxide supports. For example, Dayte and co-workers 75 showed that the sintering of Au particles is dependent on pore size, pore wall thickness (which determines the pore wall strength), and pore connectivity. For instance, Au supported on materials with twodimensional pore structures and lower connectivity (e.g., SBA-15) showed better stability compared to those supported on materials with 3-D pore structures and high interconnectivity (e.g., SBA-12), in which Au can migrate more easily between pores and even to the outer surface ( Figure 3).
Second, it is often desirable to combine some catalytic functionality imparted by the mesoporous support with that of gold to generate "multifunctional" catalysts. One good example is Au supported on titanium-silica 1 (TS-1), which itself is an important catalyst for the selective oxidation or epoxidation of olefins. The combination of TS-1 with Au can create a bifunctional catalyst, whereby olefins can be epoxidized over TS-1 with H 2 O 2 formed in situ from H 2 and O 2 produced over Au. 76−78 For mesoporous silica, the challenges in preparation are similar to those described for the regular silica supports which were discussed earlier. Mesoporous silica materials can also be deliberately decorated with functional groups, such as amine, 79−83 thiol, 84,85 and pyrrolidone, 86,87 to increase the interaction of the support with Au. Functionalizing the preformed Au particles/clusters before deposition on the mesoporous silica support is another approach which has been tried. For example, Tsukuda et al. 88 have reported an active alcohol oxidation catalyst in which Au particles supported on mesoporous silica were prepared using triphenylphosphine functionalized Au 11 clusters. Secondary support layers/particles comprised of, for example, TiO 2 , 89−92 CeO 2 , 93−95 and BaSO 4 , 96 within the mesoporous silica structures have also been commonly used to enhance Au attachment. Conventional preparation methods for dispersing Au on those materials can be directly applied when such a double-support configuration is employed.
To ensure most of the Au is located within the pore structure of the support, it is often desirable to have a one-pot synthesis of the mesoporous support and Au NPs. Overbury et al. 97 developed a coassembly technique using a bifunctional organosilane ligand in which the amine group complexes with Au(III) and the siloxane group interacts with the silica matrix during a sol−gel template synthesis. Uniform 2−5 nm Au NPs with up to 7 wt % loading were prepared within the pores of an MCM-41 material by this method. A similar approach using thiolcontaining organosilane ligands was subsequently reported by Wu et al. 81,98 and Chen et al. 99, 100 Budroni and Corma 101 have reported preparing a Au−organic−silica catalyst that was synthesized from Au NPs capped with both 1-dodecanethiol (DT) and 3-mercaptopropyltrimethoxysilane and dispersed in ethanol containing tetraethyl orthosilicate (TEOS), followed by the hydrolysis of the TEOS.
In addition to mesoporous silica, other mesoporous oxides have been also explored as support materials for Au catalysts such as mesoporous titania, 102−104 mesoporous ceria, 105−108 and mesoporous iron oxide. 109 More traditional preparation methods, such as incipient wetness, deposition precipitation, 110 ion exchange, 111,112 chemical vapor deposition, 113 or solimmobilization, can be directly used to make Au catalysts on these nonsilica-based mesoporous supports. 114 Au single-atom catalysts have also been successfully prepared using mesoporous supports. For example, Flytzani-Stephanopoulos and co-workers 115 reported making low loading (i.e., 0.25 at %) Au catalysts on mesoporous MCM-41 and KLTLzeolite materials doped with alkali ions (i.e., Na + and K + ) using simple incipient wetness impregnation methods. It was found that the alkali ions are responsible for stabilizing the atomically dispersed Au in the form of Au−O(OH) x , which is found to be highly active in the low-temperature water−gas shift reaction.
Porous coordination polymers, especially metal−organicframeworks (MOF), 112 are emerging as important catalysts and/or catalyst support materials. Hermes et al. 116 have reported Au/MOF-5 materials prepared by a CVD method using a (CH 3 )Au(PMe 3 ) precursor. Haruta and co-workers 117 later prepared Au catalysts on several porous coordination polymers with a solid grinding method using Me 2 Au(acac), which is also a convenient source of Au in CVD-based preparation methods. 113 A typical preparation procedure involves first grinding the coordination polymers with Me 2 Au-(acac), a slightly volatile organogold complex, using an agate mortar and pestle, in air for 20 min without the need of organic solvent, and then a heat treatment at relatively low temperature (e.g., 120°C). The resultant catalysts were found to be active for benzyl alcohol oxidation. Additional work has been reported more recently using a similar method for producing ultrafine Au NPs within MOF structures. 118,119 Impregnation in organic media using Me 2 Au(acac) is also a possible route to make Au catalysts supported in MOFs, as reported by Epron and coworkers. 120 Providing that the MOF is properly functionalized, e.g., amino-functionalized MIL-53(NH 2 ) 121 or UiO-66-(NH 2 ), 122 HAuCl 4 can be used as a precursor in an in situ reduction type method using NaBH 4 as the reductant in the solution. Dai et al. 123 have recently reported a charge modulation approach, in which charged species were created within the framework of a pyridine-functionalized conjugated organic network (P-CON) by introducing fluoride anions, which serves as a stabilization center for Au NPs.
2.2.2. Au on Magnetic Supports. Magnetically recyclable catalysts are of considerable interest to the research community as they potentially enable separation steps in various reaction process and can serve as a bridge between homogeneous and heterogeneous catalysis. 124,125 For Au catalysts, there are a reasonable number of publications that focus specifically on preparing composite materials between Au NPs and magnetic NPs (e.g., iron oxide), the latter of which serves the dual purpose of acting as a "support" for the gold and providing the magnetic functionality to the catalyst. 126 Sun and co-workers 127 reported the synthesis of Au/Fe 3 O 4 "dumbbell" NPs, a few nanometers in dimension on each side, using a bottom-up approach. The synthesis involves the decomposition of Fe(CO) 5 on the surface of preformed Au NPs followed by oxidation in a 1-octadecene solvent. Additional control over the size, structure, and chemical nature of such composite materials are also possible by regulating the reaction conditions. By depositing such particles onto more conventional catalyst supports, such as metal oxide or carbon, highly active and magnetically recoverable catalysts can be made ( Figure  4). 128,129 It has also been reported that the magnetic properties of the iron oxide can be enhanced due to an interfacial interaction with the metal particle. 130 A slightly different preparation method was reported by Lin et al., 131 whereby decomposition of an iron−oleate complex was carried out at high temperature (i.e., 310°C) to produce very similar dumbbell morphologies.
Spivey and co-workers 132 reported making particles having an Fe 3 O 4 core of ∼5 nm diameter coated with a thin (i.e., 0.5 nm) Au shell, which were themselves supported on TiO 2 . These were prepared by reducing Au precursors in the presence of iron oxide NPs. Meng et al. 133 have also reported a one-pot solvothermal synthesis method for making Au−Fe 3 O 4 nanocomposites, in which the resultant iron oxide particles are usually in the 10−100 nm size range. Kong et al. 134 reported a one-pot hydrothermal synthesis method, which incorporates magnetic γ-Fe 2 O 3 and Au particles into the wall of mesoporous silica materials. The resultant material is both catalytically active and magnetically recoverable. Magnetic materials other than iron oxides have also been explored, including Ni/SiO 2 @Au microspheres, 135 139 demonstrated that a porous TiO x overlayer can be intentionally introduced to pin down Au NPs through a wet chemistry approach. Such a layer not only discourages Au from sintering but also enhances CO oxidation through a strong metal−support interaction (SMSI).
The encapsulation of Au NPs can also be achieved through a colloidal synthesis approach, forming a Au-core/porous shell structure. Gao et al. 140 reported small (<3 nm) and highly stable Au NPs encapsulated by SiO 2 , prepared from ultrasmall gold hydroxide NPs, followed by SiO 2 coating within reverse micelles, then finally converting Au hydroxide to Au through thermal annealing. Amine functionality can also be introduced during the SiO 2 coating procedure in order to achieve a better affinity for the metal. 141 Bai et al. 142 have reported a method for preparing mesoporous aluminosilicate encapsulated Au@SiO 2 multilayer core−shell catalysts. Au core−shell catalysts with TiO 2 143 and titanium silicate 142 shells have also been reported recently.
Another type of Au catalyst with an encapsulating support was first introduced by Schuẗh and co-workers in 2006. 144 As shown in Figure 5, the synthesis begins with Au NPs, upon which a SiO 2 shell was then deposited, followed by deposition of another thin layer of porous ZrO 2 with a typical pore size of 3−4 nm. The SiO 2 then can be etched away, leaving Au encapsulated by a hollow ZrO 2 shell, in a so-called "yolk−shell" configuration. Compared to the "core−shell" nanostructure, the "yolk−shell" architecture obviously allows more Au surface sites to be exposed for participation in catalytic reactions. The shell not only serves as the support but can also provide mass-transfer controls between the reactant molecules with the gold. The immediate benefit is a significantly improved thermal stability: the catalyst retains the same activity in CO oxidation even after a high-temperature calcination treatment at 800°C. The same group later reported a carbon-based yolk−shell structure 145 and demonstrated the possibility of using cyanide leaching as a method for fine-tuning the size of the Au particles after the yolk−shell structure had been synthesized. 146,147 Zaera and co-workers 148,149 have prepared Au@TiO 2 yolk− shell structures using a similar approach. To demonstrate that Au@SiO 2 yolk−shell structures are also possible, Song and coworkers 150 used a much larger Au particle (∼120 nm) as the original core, followed by SiO 2 coating, and then employed cyanide to leach away some of the internal Au material. Zhang and co-workers 151 used polystyrene-co-poly(4-vinylpyridine) microspheres as both templates to fabricate a porous silica shell via a sol−gel process which acts as a scaffold to immobilize the Au NPs. A yolk−shell structure was then formed after calcination, during which the polymer microsphere is burnt away. Au-based yolk−shell catalysts incorporating TiO 2 , 152 CeO 2 , 153,154 and ZrO 2 155 shells have now also been reported. Finally, because the interface between Au and the support are often considered catalytically important, 156 an inverse catalyst concept has also been investigated for Au catalysts, whereby the original support is now created in the Au NPs. Dumesic and coworkers 157 demonstrated such a case in 2015, in which MoO x moieties are dispersed onto Au NPs using a controlled surface reaction approach. The interface between the MoO x species and the Au are considered to be the active site for the reverse water− gas shift reaction in this catalyst. In this case, the concept of what is actually acting as the support material is getting blurred.
In this section, we have surveyed the plethora of preparation methods for synthesizing Au catalysts on different types of support materials. One challenge in determining the full role of the support during the synthesis is that advanced techniques for characterizing the material, such as synchrotron-based X-ray absorption spectroscopy (XAS) and AC-STEM, are not always readily available to the materials chemist. Given the fact that Au catalysis is very sensitive to nanostructure, there is still much scope for understanding the details of these synthesis methods and the crucial role played by the support in determining the final catalyst nanostructure. Understanding the impact of the support on Au dispersion, the choice of Au precursors and the required treatments (e.g., heat treatment, leaching, surface functionalization) is an important goal in order to reveal the precise role of the support in the actual catalytic reactions, a topic which will be covered in more detail in subsequent sections of this review article.

CO Oxidation and Related Reactions
The CO oxidation reaction is one of the most widely studied processes in the area of Au catalysis. It also is important for our understanding of a range of related reactions, such as preferential oxidation of CO in the presence of H 2 (PROX) and the water− gas shift (WGS) reaction, which will only touch on in this review. The resilience of bulk gold to oxidation has led to much discussion on the mechanisms for oxygen adsorption and activation over supported NPs. Clearly, O 2 must be activated for CO to be transformed to CO 2 , but the observed high activity of gold NPs is in stark contrast to the well-known inert behavior of single-crystal surfaces of Au. 158 The field of CO oxidation by supported Au particles has been the subject of a number of reviews. 159 These have highlighted the major aspects of this chemistry, including the support influence on reactivity, discussion of the active oxidation state of gold, the importance of water in the oxidation reaction, and particle size and morphology effects. These continue to be hot topics of discussion and are also now enhanced by new approaches to catalyst synthesis, new insights from characterization methods and kinetics, and the advent of well-defined single-metal site catalysts. The area has also benefitted from close interactions between specialists in materials synthesis, characterization, 160 catalytic testing, surface science, and applied theoretical chemistry. 161 Theoretical studies have shown that isolated Au particles are capable of oxidizing CO. 162 However, in the most widely accepted mechanism for practical rates of oxidation, oxygen Chemical Reviews pubs.acs.org/CR Review adsorption is believed to occur on the support or at the metal− support interface. In particular, oxygen adsorption is thought to be favored by the presence of oxygen vacancies that would be expected on supports that are semiconductor materials (e.g., TiO 2 and ZnO) as a consequence of the Schottky junction at the metal/support interface. 163 Oxide supports have been broadly divided into "active" semiconducting, which can sustain oxygen vacancy creation by reduction of metal centers (i.e., reducible supports), and wide band gap materials, which tend to be irreducible and are thought to be "inactive" in oxidation reactions. 164 The influence of the point of zero charge on the Au/support effects are also reported to have important implications for CO oxidation catalysis. 165 For large Au NPs, (i) CO reversibly adsorbs at low temperatures (T < 150 K), 166 or is in dynamic equilibrium at sufficiently high partial pressures, which is also the case at around room temperature and above, 167 (ii) CO oxidation takes place readily only when oxygen is provided in atomic form, 168−170 and (iii) dissociative adsorption of oxygen is strongly hindered by a high dissociation barrier due to a weak coupling to the Au substrate due to the filled dstates, 165,171,172 and (iv) adsorbed CO may be directly oxidized to CO 2 or transformed into carbonates or formates, which then decompose to CO 2 and water or surface hydroxyl groups. 159 Which route dominates depends on catalyst composition, preparation method, and reaction conditions. Oxide supports with irreducible character, such as Al 2 O 3 , SiO 2 , and MgO, generally show a low ability to adsorb or store oxygen and yet CO oxidation is observed for Au NPs on these supports. 173 This indicates that oxygen adsorption and dissociation must be possible on the gold clusters themselves, although DFT calculations suggest there is only a weak interaction 162 except for particular facets of small NPs. For Scheme 1. MvK Scheme for Oxidation of CO at the Au Nanoparticle Interface with an Oxide Support a a Note: the square symbol is used to indicate a lattice vacancy. Chemical Reviews pubs.acs.org/CR Review these irreducible supports, activity for CO oxidation critically depends on the diameter of the gold particles, and usually only very small particles (<2 nm) yield highly active catalysts. 164,174,175 The increasing activity with smaller particle size was explained by an enhanced dissociative adsorption of oxygen on small gold particles due to a higher density of reactive defect sites (edge, kink, or step sites) 176,177 or a gradual change in the electronic structure at decreasing size. 174,178,179 On the other side are catalysts which are supported on reducible transition metal oxides, such as Fe 2 O 3 , CeO 2 , NiO x , CoO x , and TiO 2 . These exhibit a superior activity for oxidation reactions, with activities higher than those of irreducible supports by up to 1 order of magnitude. On these catalysts, the size of the gold particles also seems to play a more secondary role, with even large Au particles up to 30 nm in size demonstrating activity, 164 although in some cases deactivation has been noted as particles sinter and so grow in size. 179,169 This observation may also be explained by the metal loading, where the metal covers enough area of the support to mitigate its role providing atomic oxygen. 169 One widely discussed explanation for the higher CO oxidation activity of Au NPs supported on reducible supports is the supply of oxygen from the lattice of the support in a Mars van Krevelen (MvK) process. Scheme 1 shows one reaction sequence that includes a MvK style supply of oxygen. First CO adsorbs close to the interface between metal particle and support. Then lattice oxygen forms a new bond with carbon, forming an oxygen anion vacancy. The vacancy can be filled in a second step through the dissociation of molecular oxygen at the defect site. When lattice oxygen (O 2− ) is removed during this oxidation route, the two electrons from the ion can be accommodated by the reduction of support cations (e.g., Ti 4+ τo Ti 3+ ) which is not possible on an irreducible support. This also means that the activation of O 2 must take place at the Au/ oxide interface as molecular oxygen dissociates to heal the anion defects that are formed. On the practical assessment of oxidation catalysts, Kipnis et al. has analyzed work on CO oxidation and preferential oxidation of CO in the presence of H 2 (PROX) to emphasize the importance of the exothermic nature of CO oxidation reactions. 180 By measuring the temperature at the inlet, middle and outlet of the catalyst bed in a plug flow reactor, the temperature gradient can be used to highlight the exotherm effect. Figure 6 shows this measurement for a PROX reaction in which a 1:1 mixture of CO was used in a large excess of H 2 . Not all of the CO is consumed as the oxygen will also react with H 2 . At steady state, there is a 20°C difference between the inlet and the outlet temperatures as the exotherm has caused light-off at the front of the bed and the oxidation reaction becomes diffusion limited there. The rest of the catalyst has no oxygen supply, and so oxidation cannot take place in these regions. This makes it difficult to compare catalysts under high CO conversion conditions, and reliable kinetic parameters are only obtained from experiments at low CO conversion. Many researchers simply compare catalysts based on light-off curves generated by monitoring CO conversion as the temperature of the catalyst bed is increased.
3.1.1. Irreducible Oxides As Supports. One of the features worthy of discussion for Au catalysts supported on reducible supports is their oxygen storage capacity. This is usually thought of as the ability of the catalyst to use up lattice oxygen from the support for CO oxidation, which can be replenished by O 2 . However, irreducible supports can also show some oxygen storage ability. Gleaves and co-workers have used a temporal analysis of products (TAP) reactor to study CO oxidation over a Au/SiO 2 catalyst prepared by magnetron sputtering. 181 TEM analysis shows that materials prepared in this way have a mean particle size of around 3.2 nm. In a TAP experiment, the catalyst bed is held under vacuum and exposed to short (10 s of ms duration) pulses of reactant gases with the outlet of the reactor monitored using a mass spectrometer. The pulses contain very few molecules (around 10 −9 mol per pulse) so that gas molecule collisions are minimized and the temporal profile of gases exiting the reactor can be deconvoluted into surface reaction kinetics and a Knudsen diffusion component. In Figure 7, a titration experiment using the TAP reactor approach is shown. Here, the reactor bed is first exposed to a flow of O 2 before the catalyst is placed under vacuum. The TAP reactor is then used to pulse CO. In the first few pulses, all CO is converted into CO 2 so that the oxygen required can only be provided by that stored on the catalyst. By integrating the pulse areas, the amount of oxygen stored can be estimated. Analysis of the dependence of the oxygen reservoir formed in the pretreatment of the catalyst on the pretreatment pressure showed that oxygen is stored both on the Au surface and subsurface regions of the NPs. This study was also able to estimate the heat of adsorption of CO on a fully reduced Au/SiO 2 sample to be −24.4 ± 3.7 kJ mol −1 .
Schuẗh and co-workers have prepared Au catalysts supported on Mg(OH) 2 and MgO prepared from the dehydration of the hydroxide. 182 Au was deposited using a colloidal method at a loading of 0.7 wt % producing particles with a mean diameter of 3.1 nm. It was found that this produced negligible numbers of particles below 1 nm based on HAADF-STEM images. Both supports were produced from a mesoporous MgO starting point, and this turned out to be key because the high CO oxidation ability of these materials was not reproduced using conventional magnesium oxides or hydroxides. The catalysts were found to be active at temperatures as low as −85°C (below the sublimation temperature of CO 2 , although the partial pressure of CO 2 is too low for solid to form over the bed of the reactor). Figure 8 shows the interesting temperature dependence observed for CO conversion. At very low temperatures and low space velocity (80 000 mL h −1 g cat −1 ), 100% conversion of CO is obtained, and this performance is maintained up to 30°C, at which point the conversion dips to 73% at around 90°C before returning to full conversion at 230°C and above. At higher space velocity (400 000 mL h −1 g cat −1 ), the CO conversion is seen to rise from 35% at −85°C until it eventually reaches 100% conversion at around −60°C and the U-shaped behavior at Chemical Reviews pubs.acs.org/CR Review higher temperatures is more pronounced. This work also used the flow reactor to titrate available oxygen by flowing CO over a preoxidized catalyst. A reservoir of active oxygen was identified, which was large enough to rule out oxygen storage by Au alone. This suggests that the mesoporous support was able to provide a reservoir of oxygen to supply the reaction even at very low temperatures. The authors also pointed out that the high activity of these materials is achieved despite their containing negligible amounts of Au particles below 1 nm. This they attribute to the support's ability to supply active oxygen into the reaction even at very low temperatures.
For the Au/MgO(001) with an irreducible support, Duan and Henkelman have used DFT calculations to look at O 2 adsorption and activation. 183 They found that, even though the MvK process is not possible, the support still has an influence on oxygen activation. The Au work function is reduced on the MgO surface due to the interfacial dipole so that charge transfer to the O 2 2π* orbital becomes easier and makes the dissociation of O 2 more facile than for the Au surface away from the interface. Even so, the dissociation of O 2 was found to be kinetically hindered so that the work suggests a mechanism in which O 2 scission only occurs as CO 2 is formed. This is similar to the Auonly mechanism originally proposed by Norskov. 184 3.1.2. Reducible Oxides as Supports. Perhaps the reducible support that has been most widely studied for CO oxidation and related reactions is titania. TiO 2 is a semiconducting oxide which has a number of polymorphs that differ in their chemical and electronic properties including their band gap. 185 The usual forms used as catalyst supports are rutile and anatase, with the common Degussa P25 titania consisting of a mix of the two (80% anatase, 20% rutile). 186 The World Gold Council (WGC) has also produced a reference standard for Au/ TiO 2 using P25 as the support material and depositing Au by the DP method with NaOH as the precipitating agent.
Chenakin and Kruse have synthesized pure anatase TiO 2 using a titanyl oxalate complex as the titanium source. 187 Gold catalysts (GC) were produced at a 1.5 wt % loading of gold using a deposition precipitation approach with NaOH (GC1) or urea (CO(NH 2 ) 2 ) (GC2 and GC3) as precipitating agents. The choice of urea as a precipitating agent avoids the introduction of Na + ions into the catalyst material and other aspects of the synthesis for GC2 and GC3 were also designed to exclude Na + . Calcination of the materials was used to control mean particle size with GC3 (2.7 nm) and GC2 (2.8 nm) containing particles near to that of the WGC reference material, while the Au particles on GC1 (4.6 nm) were considerably larger. The measured CO oxidation rates (2% CO, 2% O 2 in Ar; GHSV 30 000 mL g cat −1 h −1 ) followed an inverse power law in relation to particle size with an exponent of −2.2. This was interpreted as confirming that the periphery of the particle, and so the interface between nanoparticle and support was the important factor leading to CO oxidation activity, in line with earlier work. 188 Even so, the WGC catalyst did not lie on the same trend line, showing 20% higher activity than either GC2 or GC3. The same work also reported detailed XPS analysis of the WGC catalyst ( Figure 9). The Au 4f signal shows the expected characteristic shape for Au in the metallic state but shifted slightly to lower energies. This the authors interpreted as indicating negative charging of the NPs to produce Au δ-; correspondingly, the Ti 2p spectrum shows a Ti 3+ feature suggesting that the oxide has been reduced through electron transfer to both Au particles and Ti cations when surface anion vacancies are formed. Water and hydroxyl groups are also evident from the O 1s spectra. A direct correlation was obtained between the fraction of Ti 3+ found by XPS for the four catalyst samples and their observed reaction  rates for CO oxidation. So that, besides the expected Au particle size effect, the availability of surface defects and the presence of Au δwere found to be important factors in determining catalytic activity.
DFT+U calculations have also shown that the vacancy creation energy for oxygen defects on the anatase (101) surface are affected by the presence of a Au cluster ( Figure 10). 189 The O vacancy defect formation energy refers to the removal of a surface oxygen anion to form 1 / 2 O 2 (g), i.e.,

TiO
TiO (vac) 1/2O (g) Calculations for the internal energy change for this process, based on electronic structure calculations, have been available for several years. It is now becoming more common to estimate the free energy for the reaction based on the calculated frequencies to estimate the contribution of vibrational states to the entropy and standard statistical mechanics of the ideal gas to estimate translational and rotational contributions. 190,191 The resulting plot with temperature for the case of oxygen defect creation on Au/TiO 2 (101) anatase shows that the interface O atoms (those in Ti−O−Au bonds) have similar defect formation energies to the clean surface. For perimeter sites (Ti−O−Ti oxygens near to the Au cluster), very small clusters (Au 3 ) have little effect, but inclusion of a Au 10 cluster significantly reduces the defect formation energy, with a negative defect formation free energy occurring above around 500 K. In the same work, it is also shown that this results in an easy pathway by which CO adsorbed an Au particle can be oxidized at the interface with the TiO 2 support. 192 This concept of forming TiO 2 with a semireduced surface has been used to rationalize the improved performance of Fe-doped TiO 2 as a support compared to a reference pure TiO 2 material. 193 The titania support was prepared using titanium-(IV) isopropoxide in a sol−gel synthesis approach with iron(III) nitrate introduced into the sol−gel to provide the desired level of doping. Au particles were added using the deposition precipitation route with urea as the reducing agent. An Fe doping level of 1 wt % was found to result in an increased Ti 3+ content according to XPS and gave a conversion of 88% for CO oxidation at room temperature compared to 61% for a TiO 2 catalyst prepared without any Fe-doping.
The role of the Au nanoparticle in facilitating the removal of lattice oxygen has been studied using temporal analysis of products (TAP) studies and DFT calculations by Widmann and co-workers. 194 Au on P25, the most widely used commercial titania, was used in this study. As P25 TiO 2 consists of ∼80% anatase, their simulation studies consider anatase (101) as the model surface. They found that, at temperatures above −20°C, the MvK mechanism is preferred. However, below that temperature, the activation energy required to extract a lattice oxygen from the surface is too great and reaction with physisorbed molecular oxygen becomes the preferred pathway. This means that, in most practical situations, a reducible support is required to observe high activity for the CO oxidation reaction as the most efficient catalysis takes place via the MvK route. CO adsorption on the metal component, rather than support, is also preferred at these higher temperatures and the barriers to CO diffusion across the surface to the metal/oxide interface sites are comparatively small (around 0.5 eV). Also using the TAP approach, Widmann and Behm have shown that the oxygen storage capacity for Au/TiO 2 shows a temperature dependence, increasing with reaction temperatures between 80 and 400°C . 195 They argue that oxygen is supplied from a highly stable active oxygen species, O act , only formed at the perimeter of Au  Chemical Reviews pubs.acs.org/CR Review particles. 196 The reaction of O act with CO is activated, but the active oxygen species itself is formed in a barrierless process. The TiO 2 lattice around the perimeter of the Au NPs can act as a store for O act , which can also be replenished by adsorbed O 2 under flow reactor conditions. In the TAP reactor, at higher temperatures, lattice diffusion of oxygen effectively extends the area around each particle that can be considered part of the perimeter and a greater region of the support can be used to supply O act . In these TAP experiments, alternating CO and O 2 pulses were studied. They found that CO 2 is only evolved during the CO pulses, so that direct CO oxidation is preferred over any surface intermediate route under the conditions encountered in the TAP reactor. The discussion on the relative importance of particle size effects, low coordination Au atoms and the particle interface with the oxide in the CO oxidation reaction catalyzed by Au/ TiO 2 has also been taken up by Lu and co-workers. 197 They prepared Au/TiO 2 using Degussa P25 as support and deposition−precipitation with urea to form Au NPs from the usual HAuCl 4 ·4H 2 O precursor. Catalyst samples were washed to remove excess chloride once the material had been prepared. Catalysts were calcined at 523 K in 10% O 2 in He, with the time of calcination used to control particle size. HRTEM was used to confirm that materials with different particle size distributions had been prepared, the samples were classified as small (av size 2.9 ± 0.6 nm), medium (av size 5.0 ± 0.8 nm), and large (av size 10.2 ± 1.6 nm). Light-off curves demonstrated the expected dependence on particle size, with the samples containing the smallest particles showing the lowest light-off temperature. The next step was to add an additional overcoat of TiO 2 to the samples using atomic layer deposition (ALD) of titanium isopropoxide (TTIP); different timings of exposure to the TTIP vapor allowed control of the thickness of the overcoat. The group had already shown that the addition of TiO 2 using ALD to Au catalysts prepared with irreducible supports (Au/Al 2 O 3 and Au/SiO 2 ) could improve the performance of these materials for CO oxidation. 198 When used with Au/TiO 2 , HRTEM showed that the overcoat layer following the longest exposure to TTIP (50 cycles) was around 1.5 nm thick and that the ALD process did not detectably affect the Au particle size distribution. XPS also indicated that the Au NPs remained in the metallic state. CO-DRIFTS measurements showed that bands associated with low-coordination Au sites had lower intensity for the coated Au/ TiO 2 samples than for the materials without coating applied. This indicated that the overcoating oxide covers low coordination sites or that the ALD process causes changes in nanoparticle morphology to reduce their number. For the CO oxidation reaction, samples with 10 or 20 cycles of ALD showed small reductions in the light-off temperature but the reduction for the 50 cycle material, with the thickest coating, was significant, shifting to higher temperatures by around 40°C. The activation energies also increased with ALD coating thickness with a measured activation energy for the uncoated small particle Au/TiO 2 case of 26.5 ± 0.7 kJ mol −1 and 35.4 ± 0.9 kJ mol −1 for the 50 cycle ALD coated catalyst. Using the CO-DRIFTS data, it was found that the number of low coordination sites on the 50 cycle ALD coated small Au/TiO 2 particles was more than 2 orders of magnitude lower than that on the uncoated sample and yet the CO conversion at 298 K was only halved by the addition of the coating. The small particle Au/ TiO 2 with this thick coating also showed higher conversions compared with the uncoated medium and large Au particle catalysts. They concluded that the coating provides additional Au···TiO 2 interface sites, a factor which is able to outweigh the effect of the reduction in available low coordination Au.
Haruta has suggested that the perimeter region between Au NPs and their oxide supports is generally where the active sites for CO oxidation are located. 156 Using Au supported on a wide range of oxide materials, prepared by both coprecipitation and deposition−precipitation methods, the group considered the dependence of the temperature at which CO conversion reaches 50% (T 1/2 ) and the heat of formation of the oxides per O atom (−ΔH f 0 ). The idea was to test if this quantity is useful as an indicator of how easily oxygen vacancies will form. 199 Figure 11 shows the full range of oxides considered which span oxides of Pt group metals, first row transition metal oxides, and irreducible oxides. Figure 11a illustrates a volcano type dependency with Au/Co 3 O 4 prepared by coprecipitation giving the lowest T 1/2 Chemical Reviews pubs.acs.org/CR Review value at around 220 K. In fact, Co 3 O 4 alone when prepared as nanorods has been shown to give notable low temperature CO oxidation activity. 200 Figure 11b highlights the change in T 1/2 between the bare oxide and the Au/MO x catalyst. The value for TiO 2 is lowered by nearly 600 K, which was proposed to be due to the creation of oxygen vacancies in the perimeter region around the Au clusters. Neurock and Yates have also looked at the oxidation of CO over Au/TiO 2 at very low temperatures (110−130 K), mainly using IR spectroscopic methods combined with DFT calculations to unravel the mechanism for CO oxidation. 202 At these low temperatures, oxygen migration from the lattice is suppressed due to the barrier associated with oxygen defect creation. 194 Accordingly, a model was developed in which O 2 is activated at an Au−Ti 4+ dual site which provides a favorable adsorption energy site (DFT calculated E ads (O 2 ) = −1.01 eV). 203 In the DFT model of Au on rutile TiO 2 (110), this species can oxidize CO adsorbed at the Au/TiO 2 interface with a calculated barrier of only 0.1 eV. In this mechanism, the perimeter region of the catalyst is important but only to provide adsorption sites for gas phase O 2 rather than supplying lattice oxygen into the reaction. Figure 12 shows how low temperature (120 K) adsorption of CO acting as an IR probe molecule can distinguish sites on the Au/TiO 2 surface with distinct bands for CO adsorbed at exposed Ti 4+ cations on the support as well as at cationic (Au δ+ ) and metallic (Au 0 ) Au sites. Indeed, DFT calculations by Boronat et al. 204 have suggested that the 2126 cm −1 feature is associated with adsorption of CO at Au in Au− O−Ti bridge structures. By pretreating catalysts in oxidizing (O 2 ) or reducing (CO) atmospheres, the relative proportions of Au δ+ and Au 0 can be controlled. 201 At the low temperatures employed in these experiments, it was found that reduced catalysts are more effective at CO oxidation than preoxidized materials containing Au δ+ . This is rationalized from DFT calculations which show two effects: First, low coordinated Au atoms can strongly chemisorb oxygen and are effectively locked out of taking part in CO oxidation. Second, cationic Au centers are less effective at stabilizing the surface OC···O 2 transition state than are metallic Au species because the required electron transfer from Au to O 2 lessened. The calculated barrier for this transition state located at the interface between a Au particle and TiO 2 for Au δ+ was found to be 0.32 eV, around three times that found for metallic Au. Behm and co-workers have used cyanide leaching to consider the relative importance of cationic and metallic Au in Au/TiO 2 catalysts. 205 Cyanide leaching can be used to preferentially remove metallic Au and so leave behind a higher proportion of cationic species than found in the as-synthesized materials. They also found that Au 0 NPs are more effective for CO oxidation than is Au δ+ and used DRIFTS measurements to suggest that cationic Au can be reduced by CO under reaction conditions.
One of the most active catalysts identified so far for CO oxidation is Au supported on iron oxide. This catalyst also highlights the importance of preparation method in synthesizing the catalyst and the use of aberration corrected STEM in understanding the link between structure and activity. Hutchings and co-workers have shown that samples prepared by coprecipitation at a loading of 2.9% Au/FeO x had activities that were highly dependent on the catalyst drying procedure. Materials dried under static air in a tube furnace and heated up to 120°C before being dried at that temperature gave virtually no activity for CO oxidation at 25°C (0.5% vol CO, flow rate 66 000 h −1 ). In contrast, material prepared from the same precursor but introduced into a preheated GC oven with flowing air at 120°C produced catalysts which showed complete conversion of CO at 25°C using the same reaction conditions. Conventional XRD, XPS, and high resolution electron microscopy showed the particle size distribution to be the same in both samples and showed that they contained the same proportions of metallic and cationic gold. 206 It was only when a higher resolution microscopy technique (namely HAADF-STEM), capable of imaging single Au atoms, became available that all Au species in these catalysts samples could be resolved. 65 By taking an active Au/FeO x catalyst and then calcining it at different temperatures, the particle distribution could be altered and at the same time the FeO x support was transformed into lower surface area Fe 2 O 3 . Calcining at progressively higher temperatures led to a loss of CO oxidation activity, so that the 100% conversion found for the as-prepared and dried sample essentially became <1% conversion for a material calcined at 600°C . Figure 13 summarizes the resulting changes in particle size/ species distributions based on a statistical analysis of HAADF-STEM imaging data. The Au species observed were classified as single atoms, subnanometer monolayer clusters, subnanometer bilayer clusters, and NPs greater than 1 nm is size. This revealed a correlation between catalyst activity and the number of 0.5 nm Au bilayer clusters present. Clearly, in these materials, Au particle size and shape are important factors in the activity for CO oxidation, but the range of particles present in the distribution produced by precipitation methods makes definitive identification of the most active form of Au difficult. Detailed electron microscopy studies have also been used to show how atomically dispersed species, subnanometer clusters and Au NPs are affected by heat treatment and cyanide leaching of metallic Au from samples. 66 Colloidal deposition (also referred to as solimmobilization) offers much narrower particle distributions, and Schuẗh and co-workers have used this approach to study Au/FeO x . 207 By preparing materials with a mean particle size of Chemical Reviews pubs.acs.org/CR Review 2.1 ± 0.54 nm and working with aberration corrected HAADF-STEM, they were able to exclude the presence of the very small particles observed on in the earlier study. Even so, these materials were highly active for CO oxidation, giving TOF values of 0.5−1 s −1 , compared with the 3.5 s −1 estimated in the earlier study based only on the concentration of subnm Au bilayer clusters. Schuẗh and co-workers, have also compared Au NPs supported on iron oxides (FeO x ) and hydroxides (FeOH) utilizing Au NPs with a uniform size around 2 nm prepared by deposition−precipitation (DP) and colloidal deposition (CD) methods. 208 The oxide materials were produced by calcination of the hydroxide (400°C for 2 h). Electron microscopy showed that all four catalyst materials contained particles around 2 ± 0.6 nm. XRD and BET characterization also showed that the calcination step led to the formation of α-Fe 2 O 3 for the FeO x materials with around one-third the surface area of the parent hydroxide. Opposite trends for CO oxidation were observed on the oxide and hydroxide catalysts produced by DP or CD methods, with the order of activity: DP (Au/FeOH < Au/FeO x ) and CD (Au/FeOH > Au/FeO x ). In this work, XANES was used to analyze the relative amounts of Au δ+ and Au 0 species present in catalyst samples using features present on the Au L-III absorption edge. Both FeO x and FeOH supported Au catalysts produced by the DP method showed Au δ+ features after the drying stage of the preparation (Figure 14a), but for calcined (300°C for 30 min) and used materials a typical Au 0 absorption edge was observed. Materials produced by the CD approach showed only Au 0 for the dried material (Figure 14b). This would be expected because the CD particles have been formed and reduced in the colloidal suspension. For the DP prepared samples, ionic Au species are deposited which are then reduced to form particles during calcination. In situ measurement of the XANES spectra for DP-Au/FeO x and DP-Au/FeOH confirmed that the proportion of Au δ+ remained below 10% throughout the CO oxidation reaction (1% CO/16% O 2 in He, 25°C). Even so, H 2 -TPR showed that the addition of Au NPs by the DP method gave rise to low temperature reduction features centered at 128°C for DP-Au/FeOH and 232°C for DP-Au/FeO x , with XRD showing the formation of Fe 3 O 4 above these temperatures, indicating that Au NPs are prompting the reduction of the iron oxide and hydroxide supports. For the CD prepared catalysts, a similar effect was observed, but the reduction temperatures were higher (188°C for CD-Au/FeOH and 257°C for DP-Au/ FeO x ), suggesting that the interaction between the support and NPs is not so strong when using CD deposition. Cui et al. have also worked with Au/FeO x and Au/FeOH materials to study the effect of the strength of nanoparticle/oxide interaction, as measured by TPR, on the CO oxidation activity. 209 They found that the strength of this interaction could be controlled by the pH used in the deposition−precipitation of the Au precursors and conclude that a strong interaction favors high CO conversions and imparts good catalyst stability.
DFT calculations using the PBE functional and DFT+U approach show that O 2 will dissociate at the interface between a Au 10 cluster and the clean stoichiometric (0001) surface of α-Fe 2 O 3 more readily than on an isolated Au particle. 210 In later calculations, it was also shown that the Au 10 particle could become oxidized at the perimeter where the interface with the α-Fe 2 O 3 can stabilize the oxidation of Au atoms, but this oxidation does not extend to the central region of the cluster. 211 The presence of the cluster lowered the defect formation energy for   Haruta has also considered the polymorph of the iron oxide support, producing an Au catalyst supported on mesoporous maghemite, γ-Fe 2 O 3 , nanoflakes 212 using a deposition− precipitation method. 213 The γ-Fe 2 O 3 appears to be a support, giving Au catalyst with higher CO oxidation activity than α-Fe 2 O 3 due to it being more easily reduced. 214 The nanoflake synthesis produces catalysts with high surface area (up to 193 m 2 g −1 ) capable of over 90% CO conversion and a specific activity of 8.41 mol CO g Au −1 h −1 at room temperature. The particle size range from electron microscopy was determined to be 2−5 nm, which, it was suggested, provides an optimal surface coverage of particle perimeter sites for catalysis.
Ceria, CeO 2 , is another oxide widely used in catalysis for its oxygen storage capacity, which is related to the ease of reduction of Ce 4+ to Ce 3+ , making it possible to capitalize on the material's redox abilities. 215 Hence Au/CeO 2 applied to CO oxidation is another widely studied catalyst. DRIFTS measurements have been used to identify surface species and, using the characteristic bands for key intermediates/side products, several studies have linked operando spectroscopic measurements to catalyst deactivation. For example, Chen et al. studied Au/CeO 2 with Au loadings of 0.18%, 0.97%, and 5.7%, and TEM analysis confirmed that the mean particle size increases with metal loading (0.18%-Au/CeO 2 : 1.7 ± 0.6 nm), (0.97%-Au/CeO 2 : 2.6 ± 0.6 nm), and (5.7%-Au/CeO 2 : 3.7 ± 0.6 nm). 216 Kinetic measurements in a 1% CO, 1% O 2 gas mix in N 2 over a temperature range of 300−360 K allowed these researchers to estimate the activation energies for CO oxidation for the three materials. The 0.97%-Au/CeO 2 and 5.7%-Au/CeO 2 gave similar results with estimated activation energies of 28.4 ± 1.6 and 25.7 ± 1.3 kJ mol −1 , respectively, whereas the activation energy measured for the lowest Au loading (0.18%-Au/CeO 2 ) was found to be substantially lower at 18.7 ± 3.2 kJ mol −1 , in line with the expected higher activity per Au site for the smallest Au NPs. In the room temperature DRIFTS spectra obtained from these materials, bands for bidentate carbonate (CO 3 2− ), monodentate carbonate, bridged carbonate, bidentate bicarbonate (CO 3 H − ), and carbonite (CO 2 2− ) could be detected upon the adsorption of CO. Furthermore, bands due to hydroxyl groups on the oxide surface could be classified into terminal and bridging configurations. In addition, CO(a) at Au 0 and Au δ+ sites were detected, with the ratio of the two decreasing with increasing Au loading, implying that the positively charged Au centers are located at the edge of the Au NPs. By studying the DRIFTS peak as a function of time during CO oxidation (1% CO, 1% O 2 in Ar), Chen et al. concluded that the oxidation activity for the direct oxidation of CO(a) does not exhibit a strong particle size dependence. By way of contrast, the production of CO 2 via the decomposition of carbonates, bicarbonates, and formate species contributed to CO conversion over the larger particles studied, but these same species inhibited oxidation of CO(a) over the smaller particles contained in the 0.18%-Au/CeO 2 catalysts.
El-Moemen et al. have used a similar spectroscopic approach to probe the effect of the reactant mix on the catalytic activity of 4.5 wt % Au/CeO 2 catalysts for CO oxidation. 217 They compared catalysts prereduced in H 2 or CO with those heat treated in N 2 or preoxidized at 400°C. They find that the activation energy measured for CO oxidation is independent of the pretreatment procedure (E act = 32 ± 2 kJ mol −1 ). Even so, the pretreatment does influence the initial catalytic activity and the degree of deactivation during a 17 h test (80°C, gas mix: 1% CO, 1% O 2 in N 2 ), with preoxidized samples giving higher rates that prereduced, but the prereduced catalysts showing the greatest stability under reaction conditions. Nanoparticle sizes were not influenced by the pretreatment steps, but the surface species observed using DRIFTS experiments were affected. On the preoxidized materials signals for carbonate were detected, but these were not seen on the prereduced material. However, after around 10 min on-stream, the surface composition according to DRIFTS reached an equilibrium state that was essentially identical for all catalysts. XANES measurements have also been used by this group to show that Au rapidly reaches a metallic state during CO oxidation. 218 Changing the ratio of CO to O 2 in the reaction mixture to an O 2 -rich flow resulted in an abrupt increase in reaction rate, indicating a positive reaction order with respect to O 2 . Switching to a CO-rich mixture in a reaction initiated at a 1:1 ratio of CO to O 2 , a gradual decrease in CO conversion was observed, indicating that the oxygen reservoir on the catalyst surface was sufficient to maintain the reaction under these conditions.
The preparation method used to produce the ceria support itself can be tailored to generate cubic, polyhedral, or rod-shaped nanocrystallites of the oxide. 219 The nanorod and nanocube forms show higher oxygen storage ability than the nanopolyhedra form due to the differing expression of {100}, and {110}-type facets on these structures. The nanorod form of CeO 2 has a high proportion of {110} facets which are also particularly suited to providing the high dispersion of Au required for single-atom catalysts to be produced. 220 Gold supported on nanorod ceria has been studied by Guo et al., who have used a deposition−precipitation approach to prepare single-Au atom catalysts and then subsequently obtained Au clusters (<2 nm in size) supported on the same material by hydrogen reduction of the corresponding single-atom samples. 221 In addition, larger Au particles (3−4 nm in size) were prepared by colloidal deposition of Au onto the ceria nanorod support. XANES was used to show that the single-atom catalysts contained largely cationic Au species while the clusters and particles showed spectra typical of metallic particles. EXAFS data taken under reaction conditions also demonstrated that the three different forms of Au were stable under reaction conditions (25°C, 1% CO, 16% O 2 in He). Testing the CO oxidation activity of the three types of catalyst showed that the CO oxidation rate per unit mass of Au was higher with the 3−4 nm particles than for the metallic clusters and that the single-atom cation Au catalyst showed a much lower ability to oxidize CO. A 600°C oxidative treatment of CeO 2 nanocubes at 600°C can produce {111} nanofacets on the {100} faces of the cubes. 222 This significantly reduces the light-off temperature for CO oxidation. DRIFTS operando measurements showed that Au δ--CO species appear on these materials upon exposure to CO and can be correlated with CO oxidation activity. 223 The nanofaceting appears to stabilize the Au particles and give rise to a higher proportion of Au δ--CO signal in the DRIFT spectra.
Chemical Reviews pubs.acs.org/CR Review DFT ab initio molecular dynamics has also been used to study a Au 20 particle supported on a CeO 2 (111) surface. 224 The authors chose to site the particle over an oxygen defect in the surface to represent a partially reduced ceria surface. The cluster was then modeled using molecular dynamics both with and without the inclusion of adsorbed CO (Figure 15a). By plotting a distribution of the Au atom distances from the center of mass of the cluster (Figure 15b), the atomic shells of the optimized cluster can be clearly seen as separated peaks at around 2, 3.5, and 5 Å. The motion of the atoms in this MD run for the cluster without adsorbates is sufficient to smear out these peaks and gave a continuous distribution, showing that the cluster stayed connected together throughout the simulation. However, when CO was adsorbed, a single-atom can be seen to have broken away from the cluster to a distance of over 6 Å from its center of mass. Snapshots from the trajectory shows that this breakaway Au atom carries a CO molecule with it, and analysis of the electronic states shows this to be a Au + -CO complex. The electrons lost to form the Au + cation were found to be accommodated by reduced (Ce 3+ ) cations in the supporting ceria. 225 The study went on to show that oxidation even at the periphery of the cluster is limited by the strong adsorption energy of the CO 2 − species that is formed. In contrast, the use of lattice oxygen to oxidize the CO in the Au + -CO complex is much more facile as the electrons generated when lattice oxygen is used to oxidize CO are accommodated by reducing the Au + cation, which then rejoins the cluster. A similar process has been suggested based on calculations for Au/TiO 2 along with microkinetic modeling, implying that this process is preferred at low temperatures (<130°C) under oxidizing conditions but that at higher temperatures the perimeter oxidation mechanism discussed above takes over. 226 For Au on manganese oxide supports, Schuẗh and co-workers have found that the preparation method for the support material can be crucial for observing high activity. 227 Prior to this work, the highest activity for gold supported on manganese oxide for CO oxidation was 0.098 mol CO h −1 g Au −1 . 228 In an earlier study, MnO 2 and Mn 2 O 3 were compared as supports with the conclusion that the redox ability of the Mn 2 O 3 phase led to facile formation of active oxygen species on the surface of the catalyst and consequently a performance in CO oxidation as compared to Au/MnO 2 . Schuẗh and co-workers prepared Au/ MnO 2 using techniques designed to produce nanoscale metal oxide particles as supports. Samples of α-MnO 2 nanowires (NW) were prepared using a hydrothermal method, while highly ordered mesoporous β-MnO 2 (meso) was produced by a nanocasting method from the aqueous manganese nitrate using SBA-15 as a hard template. In each case, PVA protected Au NPs were deposited onto the support from an aqueous colloidal solution. These catalysts exhibited CO conversion at very low temperatures. For example, the Au/α-MnO 2 (NW) showed a CO conversion of 37% at −90°C flow rate; 80 000 mL h −1 g cat −1 ). The materials also showed some oxygen storage capacity, as CO could be used to titrate active oxygen species from the catalyst surface in the absence of gas phase O 2 , as shown in Figure 16.
3.1.3. Non-metal-Oxide Materials as Supports. While metal oxide materials remain the most studied support materials for CO oxidation and related reactions, reports have appeared using alternative materials as supports. Haruta and co-workers have produced a Au@ZIF-8 by simple solid grinding of the Au source ((CH 3 ) 2 Au(acac), acac = acetylacetonate) with the metal organic framework (MOF) material. 118 ZIF-8 was chosen for its high thermal stability (over 500°C) and large pore volume (11.6 Å pore diameter) to accommodate the Au NPs that were expected to form. No features for Au in the XRD of the resulting material were seen, but UV−vis adsorption provided some evidence that Au NPs were present. The resulting catalysts were tested for CO oxidation producing light-off curves with 50% conversion between 175 and 225°C depending on the catalyst loading (0.5−5.0 wt % Au).
3.1.4. Ligand Protected Au NPs. Supported Au catalysts produced by conventional impregnation or deposition precipitation routes will always produce a broad range of metal particle sizes and may also result in a mix of metallic and cationic gold. Specific Au colloids in solution with protective surface ligands are one approach that has been tried to produce uniform singlesize Au NPs in sufficient quantities for catalysis. For example, Au    18 supported on CeO 2 nanorods. 231 They found that a heat treatment step is required to remove the SR ligands from the nanoparticle in order to observe catalytic activity, as the ligands tend to block CO adsorption. IR spectroscopy was used with CO as a probe molecule to identify a significantly greater role for Au δ+ in the CO oxidation reaction than was found for "clean" NPs deposited by more conventional deposition or impregnation methods. Removal of the thiolate ligands was determined to commence at the perimeter of the nanocluster and so the activity of Au n (SR) m after even partial removal of SR ligands emphasizes the importance of perimeter sites in CO oxidation. 232 Li et al. have carried out CO oxidation experiments at 80°C and temperature-programmed oxidation experiments to show that Au clusters stabilized with SR ligands would be expected to retain their ligands under the reaction conditions. 233 Indeed, a Au 144 (SR) 60 material pretreated at 300°C exhibited 100% CO conversion at 40°C, whereas a catalyst pretreated at 80°C only achieved this level of conversion at 60°C. This difference in performance was attributed to the thiolate ligands covering some of the catalytically active sites for the material pretreated at the lower temperature. For the ligand protected Au 144 (SR) 60 / CeO 2 catalyst, an activation step was required before the catalyst showed a high level of activity. This involved holding the catalyst at 80°C for of 10 h under the oxidative conditions (1.67% CO, 3.37% O 2 in 94.96% He), and over this time period CO conversion was observed to increase from 4.5% to 75.8% and eventually reached 94.7% after 24 h.
The stabilization of Au NPs deposited on TiO 2 using a colloidal deposition method with octadece-9-nyl amine as the surfactant has been studied by Dai and co-workers. 138 They monitored the stability of the catalysts once the ligand shell is removed. During conventional preparation methods involving washing steps and calcination in air, the Au particles are known to migrate and sinter. In this work, an alternative approach was developed, which is illustrated schematically in Figure 17. In this process, the deposited colloidal particles were first heated to 500°C in flowing N 2 , which leads to a carbonaceous layer from the decomposition products the surfactant molecules which form over the surface of the NPs. It was claimed that the formation of this layer in the inert atmosphere improves the interaction of the NPs with the support surface, possibly by reducing the oxide locally in the vicinity of the nanoparticle as the hydrocarbon chain of the surfactant is partially decomposed. The improved nanoparticle−support interaction effectively suppressed particle growth in the final calcination step. The choice of surfactant was also noted to be important with functionality in the aliphatic part of the molecule, providing cross-linking during the heat treatment. The authors used TEM, Raman spectroscopy and XRD to show that the particle size is not adversely affected by the treatment under conditions of low particle dispersion. Interestingly, 6 nm colloidal Au particles were found to have a multiply twinned icosahedral morphology which was maintained upon formation of the carbonaceous coating. Both the supported colloid and heat-treated materials exhibited a low capacity for CO adsorption. However, the materials become active for CO oxidation once a further conventional calcination step was performed to remove the carbonaceous coatings. The materials also exhibited improved thermal stability even when this second calcination step was carried out to remove the protective carbon overlayers from the Au NPs.
3.1.5. Particle Anchoring and Encapsulation. As has been previously discussed, the particle size effect for Au oxidation catalysis has been a matter of much discussion, with Chemical Reviews pubs.acs.org/CR Review most researchers now agreeing that sizes below 5 nm are required to observe high CO oxidation activity. This roughly coincides with the length scale at which a band structure picture of the electronic states of the metal breaks down and a discrete molecular orbital picture becomes more appropriate. 234 There are many applications where the catalyst needs to be robust at high temperatures, at which point, for conventionally supported Au catalysts, sintering becomes a significant route to deactivation. 235 For example, in the car cold start problem, for which the activity for CO oxidation at low temperatures makes supported Au catalysts an attractive option, the catalyst will necessarily be required to spend time in the hot exhaust stream once the engine reaches its normal operating temperature. This highlights the need to devise novel approaches to firmly anchor Au NPs to the support to improve their sintering resistance and hinder coalescence. Several strategies to make support structures that hinder sintering have been developed over recent years. Even so, the deactivation of catalysts may be caused by factors other than particle agglomeration, such as surface poisoning by reaction intermediates such as carbonates or by water. 217 The mobility of Au particles on conventional oxide supports has been highlighted by environmental TEM studies, where the ability to record the dynamic behavior of catalyst particles while exposed to low pressure reactant gases (100s of Pa) even at elevated temperature allow the visualization of particles in conditions approximating to their catalytically active state. Au NPs supported on the oxygen terminated surface of CeO 2 have been seen to reversibly migrate and rotate even at room temperature, indicating that the interaction with the surface atoms in general is weak. However, the particles may be interacting with surface defects because they can return to their original positions/orientations rather than making further migrations on the 10 s of time scale. 236 The use of microporous, mesoporous, or hierarchical materials to isolate Au NPs deposited in their channels is an attractive one. For example, Scurrell and co-workers have shown that Au deposited by incipient wetness from a KAu(CN) 2 solution on to zeolite-Y shows activity for CO oxidation. 231 However, on reduction to the metallic form, the initially highly dispersed Au was found to form large particles and the catalytic activity was lost. Ion exchange of zeolite-Y with transition metal cations such as Cr 3+ , Fe 3+ , Co 2+ , or Ni 2+ was found to improve the thermal stability of the 4% Au/NaY material, with the effect inversely dependent on the reduction potentials of the exchanged metals. XRD analysis of the Au/NaY material without transition metal cation exchange shows Au particles of around 20 nm in size are obtained on reduction, whereas AuCr/ NaY has an average particle size of around 5 nm and correspondingly improved CO oxidation activity. AuFe/NaY showed stable conversions for CO oxidation at 200°C with the amount of deactivation seen to decrease with increased levels of Fe introduced by ion exchange. Even so, the level of conversion did lower from 55% for catalysts containing 2 wt % Fe to 28% at a level of 6 wt % Fe. The authors proposed that the transition metal cations can act as anchor sites for the anionic Au species deposited during the incipient wetness procedure. This then lowers the level of migration displayed by the metallic Au NPs when the catalyst is in operation.
The silica walls of the pores in mesoporous materials also provide insufficient anchoring and Au particles can be observed to migrate easily at temperatures as low as 550°C and even make transitions between neighboring channels within the material at this temperature. 237 Accordingly researchers have looked to modify mesoporous materials with additional anchor sites that can limit the migration of the NPs.
Kucěrováet al. have explored a grafting approach with a Ti(OiPr) 4 precursor to produce a mesoporous SBA-15 material coated with a TiO x surface layer. Earlier work had shown that impregnation of titanium(IV) isopropoxide into SBA-15 to form TiO 2 in situ could improve the dispersion of Au within the mesoporous material. 90 They found that the inclusion of the TiO x layer improved the catalyst activity for CO oxidation by a factor of ∼10 compared to Au/SBA-15 without this modification. IR spectroscopy of adsorbed CO was used to infer that Au deposited on the TiO x modified SBA-15 was largely associated with the TiO x coated areas of the support. For reactions carried out at 30°C catalyst deactivation was observed but could be assigned to water inhibition and was reversible by a 400°C calcination treatment of the used catalyst.
Peŕez-Cabero et al. have selected UVM-7 as a bimodal mesoporous silica for producing supported Au catalysts using nanodomains of transition metal oxides as Au anchoring points. 238 For example, the group have produced UVM-7 with TiO x nanodomains introduced into the mesopores by either (i) a one-pot cohydrolysis/condensation route in which the silica and titania regions of material are prepared together, or (ii) wet impregnation of a pure silica UVM-7 with TiO(acac) 2 . 239 At the pH value used (7−8) for the deposition−precipitation of Au, the dominant solution phase species is anionic Au(OH) 4 − and, as TiO 2 has a point of zero charge at ∼6, the TiO x domains deposited in the UVM-7 material provides anchor points for the deposition of Au. Calcination of these materials at 800°C produced little growth in the Au NPs which maintained an average size of ∼3.5 nm. Materials produced in this way show high activity for CO removal from a simulated vehicle exhaust stream, even after 10 reaction cycles working at temperatures up to 600°C in each cycle. 240 TiO x modified mesoporous silica SBA-15 has been used as a substrate for supported Au catalysts by Behm and co-workers. 241 They point out that to prepare a reference Au/SBA-15 via the usual deposition precipitation route from AuHCl 4 ·3H 2 O is difficult as the isoelectric point for silica is quite low (∼2), so that the negatively charged Au species produced in solution will not have a high affinity for the surface at the deposition pH. Instead, they turned to using the cationic complex [Au-(ethylendiamine) 2 ]Cl 3 (formed from [Au(en) 2 ]Cl 3 ) as the Au precursor. This was found to interact strongly with the negatively charged SBA-15 surface at pH values above 2. The SBA-15-TiO x samples were prepared using a grafting approach from a titanium isopropoxide (Ti(OiPr) 4 ) precursor. 242 Catalytic testing at reaction temperatures of 30, 80, and 180°C showed that these materials give activity for CO oxidation, which increases with Ti loading and reaction temperature. The deactivation behavior with time-on-stream was not affected by the Ti loading with reaction rates declining by 80% or more after 1000 min even at a reaction temperature of 30°C. However, a subsequent 400°C heat treatment was found to completely restore the CO oxidation activity, pointing to a catalyst poisoning effect rather than particle sintering.
Creating core−shell nanostructures is another strategy that has been explored to stabilize Au NPs. 244 In a two-step synthesis process, Au clusters capped with 11-mercaptoundecanoic acid (MUA) were formed and then dissolved in THF. This was then mixed with a THF solution of titanium butoxide and hydrolyzed to produce Au particles coated with a porous TiO 2 shell (denoted Au@TiO 2 ). These Au@TiO 2 particles were then Chemical Reviews pubs.acs.org/CR Review immobilized onto silane functionalized alumina to produce the final heterogeneous catalyst. A reference catalyst formed by impregnation of MUA-Au onto Degussa P25 TiO 2 was also produced. Both materials had a nominal Au loading of 1 wt % and were calcined at 400, 600, and 700°C. TEM was employed to show that for both catalyst variants the material prior to calcination contained uniform NPs with a mean diameter of 3 nm. The reference Au/TiO 2 catalyst showed an increase of particle size even when calcined at 400°C, with a distribution of particle sizes between 5 and 15 nm being measured. For the Au@TiO 2 /Si-Al 2 O 3 material, the Au particles maintained the original 3 nm average diameter even after calcination at 600°C. It was also suggested that the activity of these materials was improved by the highly defective TiO 2 coating. Indeed, at 30°C the Au@TiO 2 /Si-Al 2 O 3 catalyst calcined at 400°C showed a CO conversion of around 85%, while the comparable Au/TiO 2 catalyst gave only 60% conversion. Light-off curves were also used to demonstrate that only calcination at 700°C of these modified materials led to significant deactivation. Another approach to encapsulation of Au NPs is by the synthesis of yolk−shell structures. Here Au colloids are coated with a silica shell using a method developed to produce uniform silica spheres on the 100 nm length scale. The resulting Au@ SiO 2 spheres are then coated by the final support material (e.g., ZrO 2 ), and the silica is then removed by treatment with strong base (NaOH) to produce the Au yolk confined within the porous zirconia shell, as shown in Figure 5. 144 This synthesis approach leads almost exclusively to one Au particle per shell and produces catalysts for CO oxidation that are resistant to sintering. The size of Au NPs and level of defects can also be controlled using post synthesis Au leaching with cyanide 245 and calcination/quenching procedures. 146 Detailed TEM analysis also revealed that the Au particles produced by the yolk−shell synthesis contain more twinning and stacking fault defects (Figure 18a,b) than those found for conventional colloidal synthesis methods. 243 The catalytic activity measured using light-off curves could also be correlated with the measured density of these defects, with higher defect densities giving lower light-off temperatures.
3.1.6. Effect of Water. The role of water in the oxidation of CO by supported Au NPs has been a key point of discussion for several years. Haruta 246 has outlined the main ways that water may influence the low temperature CO oxidation reaction, namely: (i) water may act to maintain the cationic state of Au required by some mechanisms, (ii) water, or OH − generated from water, may be directly involved in the CO oxidation as a cocatalyst or reagent (Scheme 2), (iii) water may help to activate oxygen, or (iv) water may serve to activate intermediates or remove spectators such as carbonates. Very recently, Haruta has also reported that the influence of water on the CO oxidation rate is particularly profound for the single-atom catalyst Au 1 / CeO 2 . 246 Ojeda et al. 247 showed that the yield of CO 2 depends strongly on the partial pressure of water and proposed that water can act as a cocatalyst for CO oxidation via a mechanism in which it helps to activate coadsorbed O 2 . In this mechanism CO, O 2 , and H 2 O all adsorb reversibly to the Au or Au/oxide interface (steps 1−3). Water and oxygen can then react by proton transfer to form a surface bound hydroperoxy species (step 4). Even though kinetic fitting of equilibrium constants suggested that the hydroperoxy is a minority species, it was proposed that its oxidizing power will still facilitate the ready conversion of CO to CO 2 , leaving a second surface hydroxyl species (step 5). The combination of surface hydroxyls to form water and a surface oxygen atom regenerates water (closing the cycle for the cocatalyst, step 6). The resulting isolated surface oxygen atom can then readily oxidize a further CO molecule (step 7).
The effect of water on the reaction kinetics for CO oxidation has been observed for a number of catalysts. Schuẗh and coworkers using manganese supported Au catalysts found that the measured activation energy for CO oxidation was lower when around 3 ppm water was present in the feed gas than when the gases were carefully dried ahead of the catalyst bed. Figure 19 shows that under dry conditions the measured activation energy f o r A u / M n O 2 ( n an o wi r e -N W ) ca t a l y s t a n d A u / MnO 2 (mesoporous) catalysts are both around 41 kJ mol −1 . In the presence of water, the activation energy for the Au/ MnO 2 (NW) material reduces by 13 kJ mol −1 to only 28 kJ

Scheme 2. Proposed Mechanism for Water as Co-catalyst in CO Oxidation
Chemical Reviews pubs.acs.org/CR Review mol −1 . In situ Raman spectroscopy was used to look for evidence of superoxide or peroxide species on the catalyst surface, but only bands due to metal−oxygen modes were observed, and the researchers concluded that dioxygen anion species were not significant for this class of catalysts. For 2 wt % Au/ZrO 2 , heat treatments (from 85 to 550°C) have been used to control the hydroxylation level of the supporting oxide material and the CO oxidation activity has been systematically studied as a function of surface hydroxyl concentration. 248 The degree of surface hydroxylation and the classification of the OH − groups into monodentate and bidentate was carried out using FTIR spectroscopy. CO oxidation activity was then tested using either humid or dry air as the oxygen source. The highest CO conversions were found for the catalysts that had undergone the lowest temperature heat treatment and so contained the highest surface OH − coverage. On the basis of DFT calculations, the OH − group is thought to increase the charge transfer between the Au NPs and the support, leading to a positively charged Au species to which CO can bind more strongly. The transfer of O from a coadsorbed O 2 molecule is then suggested to proceed with a calculated barrier 0.2 eV lower than for the Au cluster on a clean stoichiometric oxygen terminated zirconia surface.
Saavedra et al. have studied the effect of water on a Au/TiO 2 catalyst measuring a kinetic isotope effect (KIE) of k H /k D = 1.8 for CO oxidation when using 700 Pa of H 2 O/D 2 O in the feed stream of their reactor (Reaction conditions: 20°C, 1% CO, 20% O 2 , space velocity (SV) = 36 l g cat −1 min −1 ). 249 The observation of a KIE suggests that water or surface hydroxyl groups must take part in a kinetically important step in the reaction. In supporting DFT calculations, a Au 10 cluster was generated on a rutile TiO 2 (110) surface and the exposed oxide hydroxylated. The effect of water was then considered by adding one or two water molecules to the model as required. The adsorption of water to perimeter site OH groups (E ads = −1.80 eV) was found to be significantly more favorable than adsorption of water to either the Au cluster itself (E ads = −0.85 eV) or regions of the hydroxylated surface without the Au cluster (E ads = −0.36 eV). Further calculations suggested that coadsorbed water at the Au/TiO 2 interface can protonate adsorbed O 2 to produce an *OOH entity capable of oxidizing CO (following step 5). However, the DFT energies and barriers suggest that *CO and *OOH interact to produce *COOH and *O rather than CO 2 directly and a surface hydroxyl. This CO assisted scission of the *O−OH bond has a calculated barrier of only 0.1 eV, compared with barriers for CO oxidation by Au−O or Au− OH of 0.65 and 0.40 eV, respectively. Experimentally, they also found that for catalysts carefully dried in N 2 , the IR signals for molecular water decreased significantly but surface hydroxyls remain unaffected. Catalysts treated in this way showed up to an order of magnitude reduction in CO oxidation turnover frequency (TOF), supporting the idea that molecular oxygen activated by molecularly adsorbed water was the main source of oxidizing agent in these reactions. Surface hydroxyls and lattice oxygen are thought to be less significant when water is available as a cocatalyst.
Lin and co-workers reasoned that the hydrophilicity of oxide surfaces makes the direct observation of the hydroperoxy intermediate produced by the interaction of water with adsorbed O 2 very difficult to detect by IR spectroscopy. 250 This is the case partly because of overlapping bands from surface hydroxyl groups and water itself and partly because the behavior of the reaction as a function of water concentration shows a surface poisoning effect as the water concentration is increased. They decided that Au supported on hexagonal boron nitride should mitigate both factors, as the support has no spectroscopically competing surface species and is hydrophobic and so should not suffer from surface poisoning. Indeed, they found that Au/BN shows increasing CO conversion with H 2 O content with no maximum right up to a partial pressure of water of 3000 kPa. The rate of conversion of CO is low compared to oxide supported catalysts (3.5 mmol CO g Au −1 min −1 ), but they were able to demonstrate that the IR signature of the proposed OOH oxidant was present using DRIFTS. In the same work, oxygen labeled water H 2 18 O was used in the reaction alongside C 16 O and 16 O 2 . DRIFTS measurements then found that the CO 2 produced is a mix of C 16 O 2 and C 18 O 16 O, demonstrating that the OH − produced after proton transfer from water to adsorbed O 2 can also take part in the oxidation of CO. Fu et al. have used DFT calculations to look at Au on a hexagonal BN support 251 and found that there is electron transfer between the BN surface and the Au 9 cluster they employed. This leads to a weakening of the Au−CO bond compared to an isolated cluster, which, they suggest, explains the experimentally observed blue-shift for CO adsorbed to Au/BN reported by Lin and co-workers. 250 In the DFT calculations, it was also found that water activates O 2 to make HO 2 and COOH in a gold-only mechanism similar to that suggested for CO oxidation on Au/TiO 2 in the presence of water. 249

Alcohol Oxidation
Manufacturing of bulk and fine chemicals involves functional group transformations and oxidation of alcohol to carbonyl compounds is one of the key transformations. 252 Typically, these transformations are carried out using stoichiometric oxidants such as chromates, permanganates, or iodates which can result in the generation of highly hazardous waste in addition to the desired products. Hence, it is preferable to use molecular oxygen for this oxidation reaction using a catalyst. 252 However, catalysts need to be able to activate molecular oxygen. Several catalysts, including supported noble metal catalysts, supported transition metal catalysts, and metal oxides, have been reported to date. 253  Within the scope of this review, in this section, we discuss the role of the support during the aerobic oxidation of alcohols over supported gold nanoparticles. This section has been divided into two subsections based on the influence of the basicity or acidity of the support on the activity (conversion) and selectivity of the reaction. The acidity and basicity of the supports played a key role in the activity and selectivity for the oxidation of the alcohols.
3.2.1. Role of Support Property on the Activity of the Catalyst. Many researchers have worked extensively on the mechanism of the aerobic oxidation of alcohols over supported Au NPs and the widely accepted mechanistic steps involved are (a) step 1, deprotonation of alcohol to form an alcoholate species, (b) step 2, β-hydride elimination of the alcoholate species to form a carbonyl species, and (c) step 3, elimination of the adsorbed H s as H 2 O by oxygen. 253 Among these steps, the β-hydride elimination step is the ratedetermining step, and this occurs mostly on the metallic site or at the metal−support interfacial site. 12 When basic supports are used to support Au NPs, the basicity of the support enhances the rate of deprotonation of the alcohol to produce the alcoholate species (step 1 in Figure 20). On the basis of this strategy, several active catalysts with basic supports have been reported for the aerobic oxidation of alcohols. Kaneda and co-workers reported     257 In this catalyst, they further report that the synergistic effect between the Au NPs, and the support is because of the Cr 3+ /Cr 6+ redox cycle at the Au− support interface. As a result of this, O 2 activation takes place at the Au−Cr HT interface, which is accompanied by electron transfer from the support to the Au nanoparticle. 258 Santra et al. reported Bi doped CeO 2 supported Au NPs as very active catalyst for the oxidation of alcohols. The doping of Bi 3+ into the CeO 2 matrix increased the surface oxygen vacancy concentration of the support material and increases the basicity of the material. Consequently, Au NPs supported on Bi doped CeO 2 have been found to be an effective catalyst for the aerobic alcohol oxidation reaction. 259 Santra et al. also increased the surface O vacancy of the CeO 2 support by introducing SnO 2 , which in turn resulted in the creation of Au δ± V 0 -Ce 3+ sites (where V 0 = oxygen vacancy), which are responsible for higher activity of this supported Au catalysts. 260 All the examples discussed in the previous section concerned monohydric alcohols. Recently, oxidation of polyhydric alcohols such as glycerol, glucose, and hydroxymethylfurfural has received considerable attention because of the importance in the valorization of biorenewable feedstocks. 261 Aerobic oxidation of these polyhydric alcohols is more challenging compared to that of monohydric alcohols, hence these oxidation reactions are typically carried out in the presence of a base (NaOH or KOH) to increase the rate of the reaction. Among all the polyhydric alcohols, glycerol is the most widely studied substrate because it is formed in large quantities as a byproduct during the biodiesel production by the transesterification of oils and fats. Hence, glycerol has been identified as one of the 12 biobased sustainable molecules that could potentially replace conventional feedstock molecules that are used in the production of chemicals and fuels. 262 Oxidation of glycerol results in several products, including glyceraldehyde, glyceric acid, dihydroxy acetone, and many more ( Figure 21). Villa et al. prepared Au/H-mordenite and AuPt/H-mordenite catalysts and used them for the base-free aerobic oxidation of glycerol by exploiting the acidity of the H-mordenite support. 263 These are the earliest examples of the base-free oxidation of glycerol using supported metal catalysts, however, the role of acidity in enhancing the rate of the reaction has not been reported clearly. Brett et al. reported the use of bimetallic AuPd and AuPt NPs supported on Mg (OH) 2 as effective catalysts for the base-free aerobic oxidation of glycerol. They utilized the basicity of Mg (OH) 2 to enhance the rate of the reaction. 264 Following these two initial examples, many Au NPs based catalysts have been developed for the base-free oxidation of glycerol by utilizing the acid−base properties of the support. 265−267 A major problem associated with using solid base materials, such as MgO, as catalyst support is the deactivation of this catalyst during the reaction. The catalyst deactivation mainly occurs because of the leaching of the basic metal oxide upon reaction with the acidic products. Other reversible modes of deactivation are the phase transformation of MgO to Mg(OH) 2 and/or MgCO 3 after the reaction. 268 Some other supports, e.g., hydrotalcites are found to vacancies are formed and the combination of these two sites have been found to be crucial for the oxidation of secondary alcohol group in glycerol ( Figure 22). They maximized the population density of this site by preparing Au/ZnO catalyst via the deposition precipitation method followed by a high temperature thermal treatment in the presence of air. The support material plays similar roles during in the selective oxidation of other polyhydric alcohols such as ethylene glycol, propane diols, glucose, and hydroxymethyl furfural. 269 In summary, during selective aerobic oxidation reactions, the support plays an active role via one or more of the following routes (a) by providing acidic or basic sites to generate the alcoholoate species, (b) providing electrons to generate either the Au δor Au 0 species, and/or (c) generating oxygen vacancies that participate in the mechanism of the reaction.
3.2.2. Role of Support Property in Tuning the Selectivity. In the aerobic oxidation of benzyl alcohol, the desired product is benzaldehyde, however, other products such as toluene and benzoic acid can decrease the selectivity of this reaction. Toluene is formed by the disproportionation of two molecules of benzyl alcohol to form an equimolar mixture of toluene and benzaldehyde. In parallel, benzaldehyde is also produced by the oxidative dehydrogenation of benzyl alcohol. The benzaldehyde can be further oxidized to produce benzoic acid ( Figure 23). Supported monometallic Au NPs do not catalyze the disproportionation reaction, and hence do not produce toluene. However, the activity of supported monometallic Au NPs is very low, and the addition of Pd to form bimetallic AuPd NPs increases the activity substantially. Pd, however, catalyzes the disproportionation reaction, resulting in the production of toluene. To study to the role of support, Sankar et al. prepared five supported AuPd catalysts using different supports such as TiO 2 , Nb 2 O 5 , C, MgO, and MgO via the sol-immobilization methodology to produce nearly uniform bimetallic NPs and tested them for the oxidation of benzyl alcohol. Bimetallic AuPd NPs supported on a basic support, e.g., ZnO or MgO, did not produce any toluene (Table 2). 270 Figure 23. Schematic representation of the reaction pathways for benzyl alcohol oxidation using supported AuPd catalysts. Through careful analysis, they found that the active site for the oxidative dehydrogenation reaction is the (bi)metallic site, whereas the active site for the disproportionation reaction is the metal−support interfacial site. They tuned the metal−support interfacial site by changing the support, and through careful kinetic studies they concluded that the NPs supported on the more acidic support (TiO 2 , Nb 2 O 5 , activated carbon) catalyze the disproportionation reaction in addition to the oxidative dehydrogenation reaction, whereas the NPs supported on the more basic supports catalyze the oxidative dehydrogenation reaction exclusively resulting in >99% selectivity to benzaldehyde. Using detailed kinetic and in situ spectroscopic studies, they attributed this change in selectivity to different modes of adsorption of benzyl alcohol over supported Au-based catalysts. The basicity of the support promotes the adsorption of benzyl alcohol via the alcoholic oxygen atom which is crucial for the substrate to undergo the dehydrogenation pathway. Both the oxidative dehydrogenation and the disproportionation reactions do not occur without the metal, hence the cooperative role of both the metal and support is essential for these two reactions to occur. 270,271 In another example, the acidity as basicity has been reported to play a crucial role in tuning the selectivity during the oxidation of glycerol using supported Au catalysts. Yuan et al. prepared a series of Au supported on MgO-Al 2 O 3 catalysts with a systematic variation of the Mg/Al ratio to tune the acid−base properties of the support and the catalyst. All of these catalysts were tested for the base-free aerobic oxidation of glycerol at 80°C. The Au catalyst comprising the most acidic support (Al 2 O 3 (0.1)) resulted in the highest selectivity for dihydroxy acetone (DHA). Increasing the surface basicity and lowering the acidity, by increasing the MgO content, of the support, resulted in the selectivity to glyceric acid (GLA) being steadily increased at the expense of DHA selectivity (Figure 24). 272 Prati and co-workers studied the role of the support surface acidity on the catalytic activity and selectivity of AuPt NPs for glycerol oxidation. They found that a basic support resulted in a catalyst that was more active but less selective for the desired C3 products, producing more formic acid. However, AuPt NPs supported on acidic supports were less active but more selective toward C 3 products. 273 To sum up, the properties of the support material (acidity or basicity) play a key role in the resultant activity and/or selectivity of the supported metal catalysts. Basicity promotes both activity as well as selectivity, however these materials have stability issues because of support dissolution. Hence the solution is to improve the stability of the basic support materials or to improve the selectivity of the acidic support materials.

Hydrogenation Reactions
Hydrogenations are a very important class of reactions that are widely used for the production of bulk chemicals and the synthesis of fine chemicals. 274,275 In general, direct hydrogenation using H 2 as the reducing agent is preferred as it is the cleanest approach, and for this reaction to be viable under reasonable reaction conditions, a catalyst is required. Mostly transition metal catalysts are used to activate hydrogen, although lately nonmetal catalysts are being reported 276 and especially frustrated Lewis pairs are finding growing interest. 277 Traditionally, group VIII−X metals, e.g., ruthenium, palladium, and nickel, are the catalysts of choice in hydrogenation reactions. Gold was considered to be catalytically inactive. The publication of "Hydrogenation Over Supported Gold Catalyst" by Bond et al. 278 in 1973 and the demonstration that gold can be a superior catalyst when applied as small NPs by Haruta et al. 7 in the late 1980s were turning points for gold catalysis. Nowadays, gold is a promising catalyst with versatile application also in hydrogenation reactions. 279−281 In this section, we discuss selected hydrogenation reactions in which an influence of the support material for gold-containing catalysts have been reported.
3.3.1. Selective Hydrogenation of Nitro Groups. Hydrogenation of nitro groups is an important transformation as the resulting amines are central intermediates in the chemical industry for bulk and fine chemicals alike. 282 In particular, functionalized amines are of interest creating the challenging task of selectively hydrogenating the nitro group in the presence of hydrogenation-prone functionalities such as aromatic rings, carbon−carbon double bonds, or carbonyl bonds. In contrast to the selective reduction of nitro compounds with stochiometric amounts of reducing agents that lead to large amounts of waste, a catalytic process using H 2 is favorable. 283 However, established hydrogenation catalysts such a supported platinum catalysts usually do not differentiate between functional groups especially at longer reaction times. 284−286 Thus, chemoselective catalysts for the hydrogenation of functionalized nitro compounds are needed.
Corma and Serna in 2006 showed that supported gold catalysts could be exceptional for the selective hydrogenation of nitro compounds. 287 In particular, the use of Au/TiO 2 led to excellent selectivity of functionalized anilines at over 95% at full or close to full conversion. This highly selective nitro group reduction was attributed to an energetically and geometrically preferential adsorption of the substrate at the interface of the NP and the TiO 2 support. 288 In contrast, Au/SiO 2 is not a selective catalyst in this reaction, and because no preferential adsorption of the substrate was found on that catalyst, they concluded that the support plays an important role in the hydrogenation of functionalized nitroaromatics by favorably activating the nitro group.
The importance of the preferential adsorption of functionalized nitroaromatics with respect to the nature of the support for the metal catalyst was underlined in further studies on AuPt bimetallic catalysts for 3-nitrostyrene hydrogenation using both experimental and computational aproaches. 289,290 Serna et al. reported that the addition of very small amounts of Pt (0.01%) to form a gold platinum bimetallic catalyst increased the TOF, significantly maintaining a very high 3-vinylaniline selectivity of 93%. 289 However, an increase to 0.05% Pt decreased the   290 They showed that the addition of one Pt atom to a small gold cluster is sufficient to make the dissociative chemisorption of hydrogen barrierless, thereby explaining the strong increase in TOF with the very low Pt levels. With regard to the substrate adsorption, they could show that the adsorption modes are completely different on Au 13 /TiO 2 to Pt 13 /TiO 2 (Figures 25 and 26). The nitrostyrene molecule preferentially adsorbs on the top of the platinum NP, activating both the nitro group as well as the double bond, explaining the loss in selectivity. In contrast, the aforementioned favorable adsorption of the nitro group at the metal−support interface for Au/TiO 2 catalysts is retained when the Au/Pt ratio is high enough to keep Pt atoms isolated and away from exactly that interface. Another suitable support for gold catalysts in the selective reduction of nitro compounds that aids the catalysis is Al 2 O 3 . Shimizu et al. reported a highly active Au/Al 2 O 3 catalyst, exhibiting very good selectivity in the reduction of various functionalized nitroaromatics as well as providing a higher intrinsic activity compared to a Au/TiO 2 catalyst. 291 They attributed this to the acid and base properties of the support. In the reduction of 4-nitrostyrene, the use of supported gold NPs with a similar mean size resulted in a peak TOF with amphoteric Al 2 O 3 as the support, while a strong basic support material MgO and an acidic support material SiO 2 both showed very low TOFs. As gold NPs supported on carbon gave an even lower reaction rate by 2 orders of magnitude, it can be concluded that both the acidic and basic properties of the support have an influence on the observed activity. Using FTIR spectroscopy, they proposed a mechanism ( Figure 27) that supports the observed structure−activity relationships. Heterolytic cleavage of H 2 was proposed to occur at the gold−support interface, with the hydride located at the low coordination gold atom and the proton located at the oxygen close to the Lewis acid site. These polar hydrogen species preferentially react with the polar nitro group of the substrate enabling the chemoselectivity and preferential adsorption of the nitro group on Au/Al 2 O 3 . Such a heterolytic dissociation is in line with previous studies of H 2 activation. 292,293 Through DFT studies Whittaker et al. showed that H 2 activation predominantly occurs across the metal− support interface for a Au/TiO 2 catalyst, resulting in a proton adsorbed on a titania surface hydroxyl and a formal hydride adsorbed on the gold. 293 In a more recent study by Tan et al., the weak acidity and basicity provided by a zinc aluminum hydrotalcite as a support for gold nanoclusters was described to be beneficial for the selective hydrogenation of 3-nitrostyrene. 294 It was shown that the support was preferentially adsorbing the nitro group of the substrate and not the carbon−carbon double bond. However, the ZnAl-hydrotalcite without any gold present was not active in the hydrogenation reaction. Thus, they considered that the small gold nanocluster (1.7−2.6 nm mean size) might be activating the hydrogen on the low coordination sites and because of the Chemical Reviews pubs.acs.org/CR Review favorable adsorption of the substrate on the support the reduction could happen at the interface, resulting in the high selectivity of 3-vinylaniline. The same group investigated the influence of the divalent metal ions of hydrotalcites in 2018. 295 The previous work, Au clusters supported on ZnAl-hydrotalcite was the most promising catalyst with regards to selectivity and activity. When the more basic MgAl-hydrotalcite was used as support the rate dropped considerably. Although the NiAlhydrotalcite support led to an almost doubled activity, the selectivity to the desired product decreased substantially because of overhydrogenation and was only 9.5% after 16.5 h reaction time compared to 98.1% for the Au 25 -ZnAl-HT-300 catalyst. The high reactivity and overhydrogenation was attributed to the metallic nickel present in the support due to the easy reduction of nickel in close proximity of gold. Hartfelder et al. compared Au/Al 2 O 3 with Au/TiO 2 in the hydrogenation of nitrobenzene. 296 The authors showed that in the alumina-supported catalyst the gold NPs with sizes smaller than 2 nm are responsible for the observed catalytic activity, as these have the potential to activate the hydrogen. For the titaniasupported catalysts, the activity was much less dependent on the particle size, which was related to the fact that titania facilitates hydrogen dissociation. As no other nitroaromatics with additional functional groups were tested, no comments on the selective nature of the catalyst can be made.
In 2017, Anandkumar et al. 297 showed that the reducibility of the support is important for the hydrogenation of nitroaromatics when reducing agents such as sodium borohydride were used. Partially oxidized gold species (Au δ+ ) were found to be beneficial for the catalytic performance of ceria-supported gold catalysts. Through the redox cycle of Ce 4+ and Ce 3+ , CeO 2 is able to stabilize the Au δ+ on the support. Additionally, an improved adsorption of the reactants on the catalyst surface make CeO 2 an excellent support. Both, the substrate 4-nitrophenol and the sodium borohydride need to adsorb on the catalyst surface. Through the hydrolysis of BH 4 − to B(OH) 4 − , hydrogen is produced and a Au−H bond is formed. Surface hydrogen can react with the 4-nitrophenolate ions to yield the product 4-aminophenolNPs ( Figure 28).
3.3.2. Selective Hydrogenation of Alkynes and Alkadienes. Selective hydrogenation of acetylene to ethene is both academically challenging and industrially important. 298−300 Cracking of naphtha results in a stream of ethene with acetylene impurities and selective hydrogenation is used to increase the purity of ethene feedstock, which is used for the production of polyethylene. Complete hydrogenation to ethane and oligomerization reaction, resulting in butadiene and higher hydrocarbons that often poison the catalyst reduce the selectivity to ethene. Most of the alkyne selective partial hydrogenation studies report the tuning of process parameters and feedstock composition to control the selectivity. 301,302 Typically, Pd-based catalysts are used for this selective hydrogenation reaction; however Cu, Ni, Au, and Ag have also been reported to be active for this reaction. 298−300,303 Supported Au monometallic NPs have been reported to have excellent alkene selectivity. 302, 304 Au has also been used as a second metal to improve the catalytic properties such as activity, selectivity, and stability of Pd-based catalysts. 298,305,306 Mitsudome et al. designed a selective Au@CeO 2 catalyst with a gold core and a CeO 2 shell, resulting in a high number of interfacial sites for the semihydrogenation of alkynes to alkenes ( Figure 29). 307 When comparing this catalyst to a Au/CeO 2 catalyst with gold NPs of similar size but without the core−shell structure, the core−shell structured catalyst gives 100% alkene yield even at high conversion, while the alkene yield was only 84% for the Au/CeO 2 catalyst and was further decreased with increasing reaction time. The authors attribute this outstanding selectivity of the Au@CeO 2 to heterolytic hydrogen dissociation occurring on the Au−ceria interfacial sites, as polar hydrogen species, especially hydrides, react more favorably with alkynes compared with alkenes, owing to the electrophilicity of alkynes. In contrast, the exposed Au sites on the unmodified Au/CeO 2 catalyst induce homolytic cleavage of H 2 into nonpolar hydrogen species, resulting in unselective hydrogenation of alkynes. Gluhoi et al. 308 and Peng et al. 309 reported Au/CeO 2 to be a selective catalyst for the partial hydrogenation of acetylene to ethene. In both studies, the thermal pretreatment of the catalyst was varied; calcination resulted in Ce 4+ and Au 0 , whereas reduction resulted in Ce 3+ and Au 0 species. The calcined sample proved more active for catalytic acetylene hydrogenation, which was ascribed to the reduced amounts of oxygen vacancies. The selectivity for both catalysts was similar.
Other impurities in alkene feedstock obtained by cracking are alkadienes. Similar to alkynes, they need to be selectively hydrogenated to the corresponding alkenes before further processing especially in the polymer production to avoid lowgrade end products. 310 It was shown that supported gold catalysts are active for the selective 1,3-butadiene hydrogenation, however, the metal oxide supports had little or no influence on the activity of the catalyst. 311 Masoud et al. investigated Au/TiO 2 and Au/SiO 2 for the selective hydrogenation of 1,3-butadiene in the presence of propene. 312 Au/TiO 2 had a much higher initial activity, however, the catalyst deactivated faster under these reaction condition. It has been proposed that the deactivation is due to the deposition of carbonaceous species, facilitated by the surface properties of the TiO 2 support. They further proposed that the active Au sites on the small clusters close to the TiO 2 support are the most susceptible sites for deactivation. In contrast, on the silica support, only limited coke formation was observed, making Au/SiO 2 a stable catalyst for a very long reaction time. This is an example where the role of support is not to influence the activity of the catalyst but its stability.
In contrast to metal oxides, carbon materials as supports seem to have a greater influence on the 1,3-butadiene hydrogenation as reported by Castillejos et al. in 2015. 313 Au NPs supported on nitrogen-doped graphene oxides (GOs) and carbon nanotubes   Chemical Reviews pubs.acs.org/CR Review (CNTs) were investigated for the selective hydrogenation of 1,3-butadiene. Interestingly, the specific activities obtained with the different catalysts could neither be directly correlated to the average gold particle sizes nor the amount of smaller Au particles. They attributed this to hydrogen spillover onto the support, leading to an increased availability of the reactive hydrogen species. Hydrogen spillover, however, was not directly related to the amount of surface nitrogen groups. In the GOsupported catalysts, a continuous network of oxygen groups was proposed to facilitate the diffusion of H atoms, resulting in several strong adsorption sites. On the CNT-supported catalysts, the H atoms were suggested to spill over through defect sites where this species could also be stored. They also investigated amine-functionalized GOs and CNTs as supports for 1,3-butadiene hydrogenation. 314 Interestingly, with reuse of the functionalized GO materials, the catalytic activity was enhanced although the size of the gold NP was not changed. The authors proposed that this might be due to intercalation of the substrate between the layers during restacking of the support material. 3.3.3. Hydrogenation of CO and CO 2 . The hydrogenation of CO and CO 2 to produce methanol are reactions of great interest because methanol is not only widely used as solvent and starting material but is considered an alternative fuel and can be used for conventional energy storage for fuel cells. 315 Additionally, formic acid is a promising molecule for hydrogen storage, making the hydrogenation of CO 2 to yield formic acid, an important transformation that is increasingly studied. 316 Hartadi et al. reported the use of TiO 2 , Al 2 O 3 , ZrO 2 , and ZnO as supports for gold NPs with very similar mean particle sizes in the hydrogenation of CO 2 . 309 They observed the production of methanol as well as CO due to the reverse water gas shift (RWGS) reaction. The best results with regard to methanol selectivity were obtained using a Au/ZnO catalyst. Because the formation rates of methanol are in the same order of magnitude for all the catalysts evaluated, the authors attributed this to a lower activity of the Au/ZnO in the RWGS reaction. Additionally, ZnO was the only support that showed any activity for methanol production as well as in the RWGS reaction without Au present. They proposed that one reason for the improved selectivity at lower activity for Au/ZnO could be due to strong adsorption of the CO 2 on the oxide support material. However, this cannot be the only reason as Au/Al 2 O 3 shows a much lower selectivity and activity for methanol formation. Another explanation is, in an analogous manner to Cu/ZnO catalysts, in the reducing environment a (surface) reduction of the support and the formation of Zn surface species on the Au NPs can occur, which leads to stronger binding of the intermediates and a decrease in reaction barriers which ultimately results in a higher activity.
Au/ZnO was also investigated as a promising catalyst for methanol synthesis from synthesis gas as well as CO 2 -containing synthesis gas by Strunk et al. 317 In this study, a correlation between oxygen vacancies in the ZnO support material, obtained by N 2 O reactive frontal chromatography, and the catalytic activity in methanol synthesis in both CO 2 -free and CO 2 -containing synthesis gas was found. Through a comparison of the N 2 O consumption with TEM analysis, it was concluded that the oxygen vacancies are mainly located in the perimeter of the gold NPs. These oxygen vacancies at the Au−ZnO interface were identified as the active sites for methanol synthesis.
Filonenko et al. investigated the liquid phase hydrogenation of CO 2 to formates over different gold catalysts in the presence of base. 318 While colloidal gold NPs were inactive, Au supported on metal oxides resulted in active catalysts for this reaction. However, the nature of support material is important, as Al 2 O 3 Chemical Reviews pubs.acs.org/CR Review led to double the TONs as compared with TiO 2 as support.
They also found that a high temperature pretreatment of the Au/Al 2 O 3 had a negative influence on the catalytic activity. It was suggested to be a result of a change in the degree of hydration of the alumina, which can influence its ability to dissociate H 2 heterolytically. They proposed a mechanism for CO 2 hydrogenation using Au/Al 2 O 3 ( Figure 30). Starting with the heterolytic H 2 dissociation at the gold and support interface into a metal hydride and surface hydroxyl, the CO 2 is then adsorbed on the support, forming a bicarbonate species through reaction with the surface hydroxyl, which is subsequently attacked by the metal hydride, leading to an adsorbed formate species.
In 2017, Liu et al. studied organic−inorganic hybrid silica materials in the liquid phase hydrogenation of CO 2 to formate. 319 It was found that through modifying the support with a Schiff base, a highly active catalyst could be obtained in the presence of a NEt 3 additive in a polar solvent. Using in situ DRIFTS as well as theoretical calculations, it was found that the CO 2 forms a weak carbamate zwitterionic intermediate at the Schiff base and gold interface. This activation of CO 2 is important for the formation of formate. Subsequently, in the proposed mechanism (Figure 31), H 2 is dissociated at the lowcoordinated sites of the gold NPs and the resulting H species hydrogenate the carbamate zwitterion intermediate at the gold− Schiff base interface. They also suggest that through electron donation of the nitrogen groups the electron-rich gold surface could produce a more negative hydride, which increases the reactivity of the nucleophilic attack at the carbon atom of the CO 2 molecule and therefore, is beneficial for the CO 2 hydrogenation.
All in all, the influence of the support on the catalysis of goldcontaining NPs in selective hydrogenation reactions can be versatile. For example, reduceable supports can stabilize catalytically active, partially oxidized Au species. Depending on the acid and base properties of the support, the heterolytic cleavage of hydrogen can be facilitated on the Au−support interface. Furthermore, the right support can direct the adsorption of bifunctional substrates to give high selectivities of the desired products.

C−C Coupling Reactions
A number of reviews prior to 2015 exist for the application of supported Au NPs toward C−C coupling reactions. 12,320,321 More recent reviews have focused on the C−C coupling with mono-and bimetallic gold NPs in both homogeneously and heterogeneously catalyzed applications 230,230,322−326 although without necessarily focusing on the role of the support. Here, we consider the specific role of the support for C−C coupling reactions, where examples exist for Glaser homocoupling (sp− sp), Ullmann homocoupling (sp 2 −sp 2 ), Sonogashira crosscoupling (sp−sp 2 /sp 3 ), Heck cross-coupling (sp 2 −sp 2 ), Suzuki−Miyaura and Hiyama cross-coupling (sp 2 −sp 2 /sp 3 ), as well as oxidative and reductive coupling, and three-reagent A3 crosscoupling of aldehydes, alkynes, and amines. The supporting material for the gold-based nanocatalysts used in these reactions falls into three categories: porous frameworks, photoactive oxides, and the more traditional low porosity carbons and metal oxides that are generally thought to be inactive during these reactions.
3.4.1. Porous Frameworks as Supports. Several groups have considered mesoporous or hierarchically structured silica to support gold nanocatalysts, with the advantage that these materials are easily separable from the reactants and products. Liu et al. supported AuCl on MCM-41, a hierarchical silica, identifying Au(I) active sites for oxidative cross-coupling of propargylic acetates and arylboronic acids, noting reasonable yields (71%) in only 15 min, with recyclability over seven cycles but with slower kinetics being observed after three reaction cycles. 327 Sadeghzadeh et al. used a MOF-like mesoporous silica KCC-1, with phosphine linkers, to stabilize Au(III) complexes; arylation of benzoxazole with allylarene gave an 87% yield in 7 h. Similar activity was noted for a Au(III) homogeneous catalyst, although the separability of the supported catalyst makes the heterogeneous form beneficial. 328 Mesoporous silica has also been used when coupled with additional materials: Movahed et al. used graphene coated with mesoporous silica, with pores of Chemical Reviews pubs.acs.org/CR Review ∼2.5 nm, to support small Au NPs. The Au NPs were ∼2 nm in size and predominantly Au(0) according to XPS analysis. This catalyst was successfully applied to Suzuki−Miyaura crosscoupling for phenylboronic acid and phenyl iodides/bromides, with a 93% yield being observed with the phenyl iodide reaction after 6 h at 100°C. The catalyst was reused six times, maintaining a 90% yield, and hot-filtration experiments showed the catalyst was working as a heterogeneous catalyst rather than as a precursor for homogeneous catalyst. Furthermore, the composite catalyst was applied to A3 coupling, with yields of 77−96% with only 4.9% leaching of the metal after six cycles, although no aggregation was noted. 329 Vilhanováet al. considered a simpler silica support with amino functionalization to support Au NPs with a size of 0.8 ± 0.2 nm. When applied to Glaser-type alkyne coupling, 10 different products were successfully isolated with yields >84%; the catalysts were separable and recyclable up to five times with no changes in particle size. XPS analysis showed high quantities of Au(III), which was deemed to be part of the reaction cycle ( Figure  32). 330 Polymeric-type supports offer an alternative form of porous framework to support Au nanocatalysts, typically for larger NPs of >10 nm size. For example, Elhage et al. used nonsilanized glass wool, composed of fibers, to support Au NPs in a size range of 23 ± 10 nm. Here, photoreactivity was induced through exposure to green light to enable sp 3 −sp 3 coupling of benzyl bromides with 80% yield in 7 h. Reusability in this case was not promising, with yields decreasing to 66% in 24 h during a second application; postapplication characterization showed larger NPs of 35 ± 12 nm had been formed. 331 Poupart et al. formed supports of polystyrene-block-poly(D,L-lactide), with 1−2 μm pores, that 100 nm Au NPs were then deposited on; these proved excellent for the homocoupling of benzeneboronic acid, giving full conversion and 90% yield in 2 h. 332 Liang et al. considered hydrophobic polymers of polydimethylsiloxane sponge to support Au NPs and applied them to the crossdehydrogenative coupling of tertiary amines. A 92% yield was observed for the supported NPs, compared to 26% for unsupported NPs. The performance was observed and reusability performed over 20 cycles, with the 98% yield decreasing to 80% in 24 h. Finally, as an illustration of the potential for these porous supports, the catalyst was applied in a flow-reactor with 90% yield in 24 h. 333 Thomas et al. considered a more complex composite porous system, made of self-organized strontium ion cross-linked alginate/carboxymethyl cellulose composites, then with Au NPs and graphene oxide deposited on this support. The material, which had pores of up to 100 μm, gave 98% yield in 4 h for Suzuki−Miyaura coupling and recyclability was demonstrated for up to six cycles; high yields of 85% (4 h) were also observed in the absence of graphene oxide. 334 Slightly smaller pores were observed in the melamine polymer cross-linked Fe 3 O 4 supports developed by Pourjavadi et al., being only 5−10 nm, then with small Au NPs of 6 nm in size deposited. This system was also applied to Suzuki−Miyaura coupling for aryl boronic acids and aryl halides, giving 92% yield (1.5 h), using H 2 O as a solvent. These catalysts offer great novelty as the magnetic core allows facile catalyst separation. However, 7% activity was lost over five reaction cycles, and the reaction rate deteriorated due to loss of catalyst. 335 A magnetic-core catalyst was also created by Li et al., who supported PdAu NPs on Fe 3 O 4 @MgAl-LDH core@shell materials. This material had   336 Maya et al. used bentonite as a support for 4 nm Au NPs. This support is highlighted as being inexpensive and with a readily modifiable surface; when applied for oxidative C−C coupling, the material gave up to 90% yield (12 h) in optimized conditions. The supported NPs, which had protected ligands, were compared to activated and acidified supports, as well as bare NPs, and yields were poorer. A range of ketones were investigated which, using optimized conditions, gave >80% yield, except for linear alcohols, with reusability over five reaction cycles. 337 Kaur et al. have used Amberlite XAD-4 as a support for 3−8 nm Au NPs, which when applied for microwave-assisted Hiyama cross-coupling of aryl halides with phenyltrimethoxysilane, yields of up to 90% were achieved in times of only 10 min, compared to a conventional reaction time 6−12 h. In this case, the activity did, however, decrease slightly over five reaction cycles. 338 3.4.2. Photoactive Supports. In contrast to porous frameworks, photoactive supports participate directly in the observed catalytic reactivity, although the reported mechanisms are contrasting. Photoactivity here refers to materials that have a chemical or physical response to illumination, rather than a direct involvement in catalysis. Clearly, there is potential for many semiconducting materials to be considered in this context, but there are few reported investigations. Crabbe et al. recently synthesized KNb 3 O 8 supports, with a band gap of 3.7 eV, on to which they deposited Au NPs of 17.9 ± 10.1 nm. Under UVA light, Ullmann coupling of iodobenzene was achieved with 65− 98% yield (2 h); this was repeated three times without significant loss of reactivity. Biaryl formation from aryl halides was also investigated, but in this case, the conversion decreased with increasing electronegativity of the halide species (Br, 20%; Cl, 0%). The mechanism for reaction was identified as a hole/ electron pair forming in the photocatalyst (Figure 33), with the electron transferring to the Au nanoparticle where the reaction occurred. 339 A similar idea was proposed by Han et al. for AuPd NPs supported on TiO 2 (P25) photoactive supports. The NPs were 3 nm in size, and so, under visible light, a plasmonic response was observed from the Au and "hot electrons" could transfer to the Pd, where the catalysis occurred; these electrons were postulated as being replaced by photoexcited charge carriers in the conduction band of the TiO 2 . When applied to the Suzuki− Miyaura cross-coupling of iodobenzene and phenylboronic acid, 98% biphenyl yield was achieved, compared to trace yield over Au Ni; in the absence of light, this dropped to 17% yield and 0%  Chemical Reviews pubs.acs.org/CR Review in the absence of the AuPd NPs. No change in activity or particle size was observed after four cycles. When the catalyst was investigated with a wider range of aryl halides, yields again decreased with increasing electronegativity for Br (70%) and Cl (14%). 340 In contrast to these studies, Lanterna et al. report an alternative mechanism when supporting Au NPs (6 nm) on TiO 2 supports for the reductive dimerization of 4-nitrobenzyl bromide, and in this case an 83% yield was achieved in 5 h, with the reaction requiring photostimulation at 532 nm. At this wavelength, this is postulated as resulting from plasmonic hot electrons being injected into the conduction band of the TiO 2 ; these electrons then are transferred to the aryl halides from the support (Figure 34b). However, reuse experiments showed that yields decreased to ∼55% after only three cycles. 341 3.4.3. Carbon and Metal Oxides As Supports. 3.4.3.1. Carbon. On activated carbon, Parmentier et al. deposited atomically dispersed AuCl x nanocatalysts, and compared these to sol-immobilized NPs of 6.6 ± 3 nm in size for the homocoupling of phenylboronic acid to form biphenyl, the AuCl x nanocatalyst was shown to be active with modest yields of 25% in 6 h; in contrast, the sol-immobilized Au NPs were inactive (Figure 35). While the cationic Au(I) and Au(III)  Chemical Reviews pubs.acs.org/CR Review were therefore deemed reactive, it was noted that strong surface attraction of the active Au species (to the supporting carbon materials) is necessary to prevent reduction/sintering of NPs; such sintering was observed in the samples after catalytic testing (30−40 nm). 342 Strong support interactions were also considered by Candu et al. 336 and Primo et al., 337 who both prepared Au "platelets" with (111) facets over defective H-doped graphene. In both cases, the NPs were 3 nm deep and had surface areas of ∼20 nm. Candu et al. considered these materials for Ullmann homocoupling of iodobenzene, where good turnover numbers (TON) and selectivity to diphenyls were observed. 343 Primo et al. made a comparison with 5 nm Au NPs on graphene, and a 5-fold improvement in TON was observed with 83% selectivity, although this was observed only at low conversion (0.3%). 344 Graphene oxide was used to support Au NPs of 3.5 nm by Mondal et al. 338 Again, a strong support−nanoparticle interaction was formed between the cationic Au and the anionic oxygen of graphene oxide. When applied to Suzuki−Miyaura cross-coupling of aryl halides and phenylboronic acid, 93% yield was observed at 70°C. Only the combination of the support and nanoparticle was active, and hot filtration experiments showed catalyst was stable throughout, although yield did decrease by ∼10% over five cycles. 345 3.4.3.2. Metal Oxides. For metal oxide supports, a stronger surface−NP interaction can occur than for the more inert carbon materials, and this can be applied in the coupling reactions. DFT calculations were used by Boronat et al. to investigate whether Au(I) or Au(III) was the active species for Sonogashira coupling reactions on Au/CeO 2 , the most selective of supports considered in experimental work ( Figure 36). The small particles were found to spread the charge so that oxidized Au δ+ was at a particle surface, or particle−support interface, and active for deprotonation, while Au(0) was on the inside of the NPs. Interestingly, experimental testing showed homocoupling for biphenyl preferred Au(0), whereas homocoupling of alkynes was more active in NPs with mixed metal oxidation states, similar to as observed in DFT. 346 Further reducible oxide supports (CeO 2 , TiO 2 , ZrO 2 ) were tested by Liu et al. as Au nanoparticle supports, with the oxides themselves supported on SBA-15. The metal oxide quantity was 21−34% by mass, with Au NPs 1.3−2.3 nm in size. Initially, XPS measurements showed all NPs comprised some Au(III) but, after reduction, a dominance of Au(0) was observed. When tested for the basefree homocoupling of phenylboronic acid, all the reduced samples proved more active than the as-deposited catalysts. Conversion (>74%) and productivity (>94%) were high; recyclability was tested with the reduced CeO 2 support, and here the yield decreased to 73% in five cycles due to leaching and aggregation, with nanoparticle sizes increasing from 1.3 to 2.3 nm. 347 Albadi et al. also considered monoxide CuO and ZnO supports for Au NPs applied in Suzuki−Miyaura cross-coupling; a 90% yield was achieved in 4.5 h for coupling of aryl halides with arylboronic acid, and the catalyst was recyclable in testing over three runs. 348 Gholinejad et al. also considered mixed metal oxides of CuFe 2 O 4 , supported on silica, to support Au NPs for Sonogashira coupling. The support was proposed to aid nanoparticle separation, and for the coupling of iodobenzene and phenylacetylene, this catalyst gave 95% yields and demonstrated recyclability over four reaction cycles; furthermore, the magnetic nature of the support again made it easily separable from the reaction mixture. 349 Several investigations have considered the use of rigorous size-selected, thiolate-protected Au NPs deposited on metal oxide supports, where here the support acts to simply stabilize the NPs. Li et al. tested small [Au 25 (SR) 18 ] − [TOA] + NPs supported on CeO 2 for Ullmann heterocoupling of 4-methyliodobenzene with 4-nitro-iodobenzene, where TOA = tetraoctylammonium. The reactivity of the NPs is strongly affected by the type of thiolate ligand (SR) encapsulating the Au nanoparticle. Maximum conversion (91%) was observed for R = 1-naphthalenethiolate, with a selectivity of 82%; however, in recycling, this decreased to 50% and 15%, respectively, by the third reaction cycle, postulated to be due to ligand removal. 350 In other similar work, Li et al. deposited Au 38 (SR) 24 NPs (R = C 2 H 4 Ph) on TiO 2 , CeO 2 , and SiO 2 , and the activity on CeO 2 and SiO 2 was similar to the unsupported NPs for A3 coupling, being >98% in all cases. It was concluded after further analysis that both cationic "surface" and neutral "core" Au atoms play synergistic roles in the reactivity, which is aided by ligand and support interactions. 351 Li et al. went further and considered the reactivity of doping Au 25 (SR) 18 NPs (R = CH 2 CH 2 Ph) with Cu, Ag, and Pt, all supported on TiO 2 , all for application in the Sinogashira cross-coupling of p-iodoanisole and phenylacetylene. Overall, the pure Au nanoparticle has the highest selectivity, with the Pt-doped catalyst being equivalent; the Cu-doped catalyst in contrast preferred Ullmann homocoupling.
In conclusion, the authors determined that the electronic effect (change in energy levels) from nanoparticle dopants is crucial in reactivity and is determined by the shell species (element) of the nanoparticle rather than being related to the choice of support. 352

Reactions Catalyzed by Unsupported Au-Based NPs
It has been assumed that most of the catalytic activity of the supported NPs is due to the influence of the support, for example, by the creation of interfacial sites between the NP and the support or the changing in nanoparticle morphology that the support can induce on the NPs. Hence, the activity of unsupported colloidal NPs has not been considered in many studies to date. However, for the completeness of this review, we consider this topic should be addressed as there are examples where unsupported colloidal NPs can give superior performance.
Colloidal NPs have been studied to show their intrinsic catalytic activity and to elucidate the catalytic mechanism for hydrogenation, oxidation, coupling reactions, etc. 353,354 These NPs mimic the metal surface activation and catalysis at the nanoscale typically observed in heterogeneous catalysis but also are dispersible in a range of solvents (unlike solid catalysts) and can be used as homogeneous catalysts. Hence, these nanoparticle catalysts can bridge both the homogeneous and heterogeneous catalysis communities, and these catalysts are sometimes referred to as semiheterogeneous. 355 Nanosized gold colloids display interesting optical properties together with high catalytic activities. They are intensely colored due to the surface plasmon resonance, with the particular color of the colloid depending not only on the size and shape of the NPs but also on the properties of the solution such as the dielectric constant of the dispersion medium. 353 Au and Au−Pd NPs have been shown to exhibit catalytic activities without the presence of a support for oxidation reactions for substrates as benzyl alcohol, 356 glycerol, 357 glucose, 165,358 styrene, and CO, 359,360 oxidative polymerization of aniline and pyrrole, 361,362 hydrogenation of cinnamaldehyde and benzalacetone, 363,364 and Chemical Reviews pubs.acs.org/CR Review epoxidation and allylic oxidation. 365 Monodispersed PVP stabilized Au NPs were synthesized and investigated by Tsukuda and co-workers for aerobic oxidation of benzylic alcohols and homocoupling of phenylboronic acid. 356,366 The catalyst was prepared by rapid reduction using NaBH 4 into an aqueous solution the [AuCl 4 ] − −PVP complex, which yielded NPs with 1.3 nm mean diameter. The catalyst was treated with Na 2 SO 3 to increase the particle size to study the effect of particle size on catalytic activity. Recently, colloidal NPs were also shown to oxidize methane to methanol in aqueous solutions at mild temperatures. 367,368 Compared to the same NPs supported on titanium oxide, much higher activity and selectivity was obtained with much lower consumption of H 2 O 2 . 369,370 Previously, homogeneous Au and Pd have also been investigated for low temperature methane oxidation. 371,372 However, without strong oxidants such as selenic acid or trifluoroacetic acid, the catalyst was not effective in activating methane to produce oxygenated products. Oxidation in aqueous conditions with homogeneous chloroauric acid led to precipitation of Au(0) and deactivation of the catalyst. 373 Polymer protected bimetallic Au−Pd NPs were Chemical Reviews pubs.acs.org/CR Review shown to possess much higher activity, with no deactivation observed for up to 4 h. 367 This size effect on catalytic activity was further investigated under the influence of identity and concentration of the stabilizing ligand used to prepare the colloids. The role of the ligand was also studied in the enhancement or suppression of the catalytic activity while avoiding particle aggregation. 374 It has been considered that strong Au−ligand interactions could decrease the particle size and avoid particle aggregation but could also decrease the catalytic activity considerably by partially or completely blocking access of the reactant molecules to the metal surface. 353,374 A compromise between the activity and reusability must be established with respect to the type of stabilizing ligand used and its concentration. For example, the effect of the type of stabilizer used to prepare colloidal catalysts for glycerol oxidation has been investigated by Prati and coworkers. 357 An increase in the catalytic activity was observed with decrease in the particle size when catalysts were prepared using PVA, tetrakishydroxypropylphosphonium chloride (THPC), and citrate as protecting agents ( Figure 37). The catalysts showed similar selectivity, but citrate-stabilized Au NPs, with mean particle size of 9.8 nm, showed the least activity (TOF 160 h −1 ) compared to particles stabilized by PVA (715 h −1 ) and THPC (2478 h −1 ), which had much lower particle sizes of 2.5 and 2 nm, respectively. On aging, PVA and citrate showed much less agglomeration compared to THPC stabilized, which indicated that electrostatic stabilizers were much less effective as a protective agent compared to steric stabilizers. The bulky steric stabilizers prevented agglomeration and hence were active over longer use periods but were found to be much less active compared with THPC, probably again due to its steric congestion and limiting active site accessibility. Stabilizer-free NPs have been synthesized, but due to their instability in solution over an extended period of time, were supported onto TiO 2 and tested for glycerol oxidation. Again, a decrease in activity was observed with increase in particle size and also a difference in selectivity was observed with PVP, PVA and stabilizer free catalysts. 375 PVP-stabilized gold NPs were shown to display enhanced catalytic activity compared to the colloids prepared by PAA for aerobic oxidation, which was attributed due to electron donation properties observed for PVP. 376 Similarly, the metal:polymer ratio has been varied to probe the effect of polymer concentration on particle size and catalytic activity. Particle size was confirmed using electron microscopy, and the highest polymer to Au ratio of 50:1 gave small size particles with very narrow particle size distribution in the range of 1.8−2.8 nm. 364 On the other hand, a much broader particle size distribution from 2 to 12 nm was obtained with a polymer to Au ratio of 3:1. Particles prepared with 50 times the polymer concentration showed the best activity toward the reduction of nitrophenol compared to the Au catalyst with bigger sizes. Similarly, the mean diameter of monodispersed Pd NPs was controlled from 1.7 to 3.0 nm by changing the amount of PVP added during synthesis. 377 Along with the catalytic activity and selectivity, aging, and reuse of unsupported NPs has also been studied. De Vos et al. studied the stability of the Au colloids for oxidation of 1,2propanediol. 378 It was observed that significant level of activity (>50%) was maintained even after extended periods of aging (24 days). Another point noted was the recovery and reuse of the colloidal catalysts after consecutive experiments. This was addressed by using a poly(dimethylsiloxane) (PDMS) membrane to recover the sol by nanofiltration, by which over 99% of metallic colloid could be retained. 379 These studies indicate the intrinsic activity of the unsupported NPs, and supporting material does not necessarily improve catalytic activity or selectivity. Colloidal catalysts can be used for a range of reactions, with stability and reuse possible for properly designed processes. However, more research is now required to understand and improve on efficiency of these colloidal catalysts as they can help to understand the role of the support and the metal−support interface in heterogeneous catalysis.

FUTURE PERSPECTIVES
Since the use of gold NPs or highly dispersed gold catalysts was highlighted as effective catalysts in the 1980s, the field of gold catalysis has grown markedly and now gold and gold-containing NPs are the subject of extensive studies for a whole range of reactions in both heterogeneous and homogeneous catalysis. For supported gold and gold-containing NPs, the support is often crucial in securing the required catalytic function and this has been the focus of this review. In some cases, improved catalysis is observed in the absence of the support, but such examples are few and far between at present. However, we consider the study of unsupported colloids will provide important insights into how supported catalysts function, and so we expect such studies to continue in the future. Where there is perhaps the greatest scope for future improvement is in the field of deliberately designing supports with specific functionality. There have been some attempts at this; specifically, with the fine-tuning of acid/base properties of surfaces. However, we consider there is immense scope for further studies in this area. One approach will be to design the surfaces of the supports so they contain specific sites that can anchor gold nanoparticles. As gold is a soft metal, such an approach should be focused on adding in soft functionalities as the soft−soft interaction will necessarily lead to enhanced stability and interaction with the support. With the significant advances in materials characterization methods that we have witnessed and exploited in the past decade, coupled with advanced theoretical studies, it is clear that we will be able to probe the interaction between the support and the nanoparticle in ever more detail. Use of in situ characterization methods will be most useful, as it is known that NPs can be fluxional under reaction conditions. This we consider will be a key area for advances in future studies. However, the major problem that needs to be addressed with gold catalysis is the need to ensure that the Au NPs, clusters, or cations that constitute the active site of the catalyst, for there are examples where each of these structures is the active center and do not agglomerate as this inevitably leads to deactivation. In some cases, this may be holding up the further commercialization of Au catalysts. In this respect, attention to the design of the support so that it can inherently stabilize the active gold center is the area for the most research attention at this time. It is here that we consider that theory can help in the selection and design of novel support structures, and we envisage this being a very active research area in the future.

ASSOCIATED CONTENT Special Issue Paper
This paper is an additional review for Chem. Rev. 2020, volume 120, issue 2, "Nanoparticles in Catalysis".