A review of nanostructured non-titania photocatalysts and hole scavenging agents for CO 2 photoreduction processes

The imperative for the development of sustainable energy technologies to alleviate the heavy reliance on fossil fuels as well as to mitigate the serious environmental issues associated with CO 2 emission has fostered the development of solar fuels through CO 2 photoreduction. The well-documented TiO 2 and modi ﬁ ed TiO 2 -based photocatalysts have been shown to photoreduce CO 2 into hydrocarbons. Meanwhile, there is also an increasing interest in the utilisation of non-titania based materials, namely metal sulphides, oxides, oxynitrides and nitrides, for CO 2 photoreduction. Distinct from other published reviews, we discuss here recent progress made in designing metal sulphide, oxide, oxynitride and nitride photocatalysts for CO 2 photoreduction through morphological changes, aiming at providing a systematic summary of non-titania based materials for CO 2 photoreduction. Furthermore, the introduction of hole scavengers in order to maximise the CO 2 photoreduction e ﬃ ciency is also reviewed.


Introduction
Fossil fuels are currently unrivalled for energy generation, and our existing infrastructure is built to handle fossil fuels for transportation, heating and electricity. 1 Our heavy reliance on fossil fuels results in annual emissions of 32 Gt of CO 2 . 2 This is likely to increase to 36-43 Gt by 2035, subject to policies governing CO 2 emissions and energy use, even with increasing renewable energy sources. 3 To mitigate these environmental issues as well as alleviate our dependence on fossil fuels, harvesting the seemingly innite solar energy and storing it in the form of chemical fuels hold signicant promise to address current and future energy demands. Moreover, the chemical industry and a vast amount of chemical products rely heavily on using fossil fuel feedstock. This further motivates the development of sustainable processes to generate fuels and chemical feedstock from water and CO 2 using solar energy. Such a process is akin to photosynthesis in nature, and therefore, it is referred to as the articial photosynthesis.
Photoelectrocatalytic reduction of CO 2 in aqueous suspensions using semiconducting powders was rst proposed by Inoue et al. in 1979. 4 Later in 1987, the photocatalytic reduction of CO 2 to CH 4 in the presence of H 2 O was proposed by Thampi et al. 5 Since then, an increasing number of studies on the photo(electro)catalytic reduction of CO 2 have been conducted (Fig. 1). Among these studies, almost 50% focused on the materials employed as photocatalysts for conversion of CO 2 under UV and/or visible irradiation. The rest of the studies concentrated mainly on modelling or process development. The use of TiO 2 as a photocatalyst for CO 2 reduction has been extensively studied and has been reviewed elsewhere. [6][7][8][9][10] However, the lack of systematic studies of non-TiO 2 semiconducting materials, namely metal sulphides, oxides, oxynitrides and nitrides, for CO 2 photoreduction (CO 2 PR) has inhibited the development of these photocatalysts compared to titania-based photocatalysts.
Although different photocatalysts (i.e., titania and non-titania based semiconductors) have been proposed in the literature, the overall CO 2 PR conversion remains low especially under sunlight irradiation, making the CO 2 PR system not practical for commercialisation. To further increase the efficiency of CO 2 PR, the introduction of scavenging agents into the CO 2 PR system has been proposed. However, so far, the introduction of hole scavenging agents has not been systematically studied, though studies started in the last century. Therefore, the necessity to systematically scrutinise the recent development of non-TiO 2 photocatalysts and hole scavenging agents for CO 2 PR is of great demand.
There are enormous scientic and technical challenges involved in making even the simplest fuel, H 2 , and even more so for carbon-based fuels by means of CO 2 photoreduction. Similar to other photocatalytic processes, solar-driven photocatalytic conversion of CO 2 in the presence of H 2 O to hydrocarbon fuels uses semiconducting materials to harvest solar energy and provides active sites to allow the photocatalytic conversion process to occur. The basic steps of the photocatalytic process can be summarised as follows: (1) generation of charge carriers (electron-hole pairs) by semiconducting materials upon absorption of photons with appropriate energy from the irradiation of light, (2) separation of charge carriers and their transportation to the surface of the photocatalyst, and (3) chemical redox reactions between the charge carriers and the reactants. CO 2 PR with H 2 O into fuels is illustrated in Fig. 2. TiO 2 was the rst material used for CO 2 PR, 5 and since then it has been widely used because of its abundance, availability, high chemical stability, low cost and non-toxicity. 12 Despite the great effort made in the CO 2 PR using TiO 2 and its derivative materials, the efficiency of the process remains low, 7 mainly attributed to the following factors: (a) Rapid recombination of photogenerated electron-hole pairs; 10 (b) Mild reducing power; The potential of the conduction band electrons is only slightly more negative than the multi-electron reduction potentials of CO 2 , thus providing a very small driving force, whereas the potential of the valence band holes is much more positive than the water oxidation potential. 7 (c) Limited visible light absorption due to the wide bandgap (3.0-3.2 eV) of TiO 2 . 13,14 Strategies including doping, 15,16 coupling with semiconductors, [17][18][19] dye sensitizing, 20,21 surface modication 22,23 etc. have been extensively used to improve TiO 2 photocatalysts and are summarised elsewhere. 9,14,24,25 However, the two most commonly used methods for extending the absorption range to visible light, namely sensitization or doping, do not fully address the optical issue of wide bandgap materials. Sensitizing agents (e.g., dyes or quantum dots) oen degrade when exposed to UV light and photogenerate oxidizing holes in TiO 2 . 7 Dopant atoms, on the other hand, can become the centers of charge recombination. Moreover, the additional energy states associated with foreign atoms are highly localized, resulting in suppressed charge mobility. 27 Hence, while TiO 2 remains a benchmark photocatalyst, there is a lot of interest in developing other materials for CO 2 PR, such as carbon-based semiconductors (e.g., graphene-based composites, 28,29 carbon nanotube composites, 30 g-C 3 N 4 based composites 31-33 and hybrid organic-inorganic materials [34][35][36][37] ) and other inorganic transition or main group metal oxides, sulphides, oxynitrides, and nitrides. Since the use of carbon-based semiconductors for CO 2 PR has been reviewed elsewhere, 30 used for solar-driven reactions. They possess relatively high stability, are low cost and absorb light consisting of photons with energy equal to or greater than their bandgap. 40 This very diverse group of materials includes both narrow and wide bandgap semiconductors; yet many of them offer a more favourable bandgap than TiO 2 . Moreover, many recent CO 2 PR developments follow similar trends to those for photocatalytic water splitting, as both processes share similar constraints on energy bands. [41][42][43] Specically, the quest for new semiconductor materials is focused on the following points: 27 (a) rising the valence band energy to decrease the bandgap, (b) moving the conduction band to more reductive potentials, (c) improving the quantum efficiency of exciton formation whilst suppressing charge recombination and (d) using novel nanoscale morphologies to provide a large surface area with multiple photocatalytically active sites.
To achieve the quest mentioned above, different methods have been proposed previously and are reviewed in the following sections.

Non-TiO 2 materials for CO 2 photoreduction reactions
Although the position of conduction and valence bands is important for photocatalytic properties, the morphology of materials plays a critical role. Furthermore, manipulating the microstructure has also shown to alter the bandgap energy, 44 suppress the charge recombination, 45 enhance the diffusion of electrons towards the surface of photocatalysts, 46 induce quantum connement effects 47 and provide more photocatalytic active sites, thereby enhancing the photocatalytic performance. In this section, nanostructured non-TiO 2 semiconducting materials for CO 2 PR published in the last two decades are reviewed, including metal sulphides, oxides, oxynitrides and nitrides.

Sulphides
Sulphide semiconductors received a lot of attention for CO 2 PR. This was because their valence band, made of 3p orbitals of the sulphur atoms, is located higher than those of their oxide analogues, resulting in the conduction band being more reductive. 42 Many sulphides have a narrow bandgap (e.g., PbS and Bi 2 S 3 ), with the absorption onset in the visible and infrared regions. Amongst sulphide semiconductors, ZnS and CdS were the most studied sulphides for CO 2 PR. ZnS is a wide bandgap semiconductor (E g ¼ 3.66 eV in the bulk); however, it possesses  a strong reducing power of the conduction band (E CB ¼ À1.85 V vs. the NHE at pH 7). 48 Zinc-based materials. The surface area of the photocatalyst is one of the key factors that can signicantly affect the efficiency of the photocatalytic process. Kočí et al. proposed the immobilization of ZnS on montmorillonite, a representative natural clay mineral, which possesses a high surface area and layered structure, to optimise the efficiency of the CO 2 PR (Table 1 entry  1). 49 The study demonstrated that the amount of ZnS loaded affected the degree of agglomeration that consequently inuenced the electronic conguration as well as the efficiency of the ZnS/montmorillonite nanocomposite in the CO 2 PR under UV irradiation (254 nm). A similar approach was demonstrated by Petra et al., in which ZnS was loaded onto large-surface-area SiO 2 (340 m 2 g À1 ) to reduce CO 2 to formate using 2,5-dihydrofuran as the reducing agent. 50 The study revealed that the loading amount of ZnS signicantly affected the yield and the optimal loading was 13% of ZnS into SiO 2 , resulting in 7 mmol g À1 h À1 of HCOOH. Nonetheless, the fabricated samples with coverages above 7% of ZnS on the SiO 2 matrix could suppress the photo-corrosion of ZnS to Zn(0), which is the major disadvantage of sulphides in an aqueous dispersion because the oxidation of lattice S 2À ions leads to elemental sulphur and eventually to sulphate. 51 Meng et al. proposed the co-doping of Cd and Cu into ZnS as one of the most active and optimised design routes for metal sulphide photocatalysts so far. 52 It was found that the doping of Cu could promote the formation of S vacancies and narrow the bandgap energy of ZnS, whereas surface modication of Cudoped ZnS with Cd 2+ enhanced the product selectivity towards HCOOH (99%) under solar light irradiation. Recently, solid solutions of ZnLn 2 S 4 with a ower-like microstructure decorated with a cubic CeO 2 co-catalyst have been shown to exhibit enhanced CH 3 OH production (0.542 mmol g catalyst À1 h À1 ) when compared to pristine CeO 2 and ZnLn 2 S 4 (0.139 and 0.073 mmol g catalyst À1 h À1 , respectively) under visible light irradiation (l $ 420 nm). 53 Cadmium-based materials. CdS (2.4 eV and the absorption onset at 520 nm) is a narrow bandgap metal sulphide photocatalyst. Hence, CdS suffers from rapid recombination of photogenerated electron-hole pairs. In order to enhance the separation of photogenerated electron-hole pairs, surfacephase junctions deduced by the same semiconductors were proposed. Chai et al. fabricated a mixed-phase CdS that is composed of wurtzite and zinc-blende crystalline phases recently (Table 1 entry 2). 54 The fabricated sample exhibited a long photogenerated electron lifetime and efficient charge transfer. The maximum CO and CH 4 evolution rate was 1.61 and 0.31 mmol h À1 g À1 , respectively, and these production rates were maintained even aer 100 h.
The conduction band of CdS is less reductive (E CB ¼ À0.9 V at pH 7 vs. NHE) than that of ZnS. Therefore, CdS is always decorated with noble metals, such as Ag. For instance, Zhu et al. proposed that the loaded Ag could act as an electron trap as well as an active site for CO 2 PR on CdS. 55 The photoproduction of CO was improved by three times when compared with that obtained with bare CdS. Alternatively, CdS can be supported with other wide bandgap semiconductors to enhance its reducing power for CO 2 PR. Kisch et al. found that the coupling of CdS with ZnS strongly enhanced the CO 2 PR activity when compared to SiO 2supported CdS or ZnS samples because CdS and ZnS can absorb light at #530 nm and #330 nm, respectively. 56 The study reported that 5 wt% CdS loaded onto ZnS induced a 40-fold and 16-fold enhancement in the production of HCOOH ($80 mM, l $ 320 nm, 3 h) when compared to unmodied CdS and ZnS, respectively. This strong enhancement was attributed to the electronic semiconductor-support interaction effect that improved the charge separation efficiency of the coupled semiconductor system. A similar observation was also reported by Kočí et al. recently, in which core-shell CdS/ZnS nanoparticles deposited on montmorillonite prepared by a one-pot synthesis exhibited enhanced CO 2 PR activity in water under UV irradiation (l ¼ 365 nm). 57 The increase in the yield was due to the enhanced charge separation of CdS cores by ZnS shells, the increase of surface area and the inhibition of CdS photocorrosion. CO 2 PR performed with CdS coupled with Bi 2 S 3 , having smaller bandgap energy than CdS, was also reported. 59 The Bi 2 S 3 /CdS nanocomposite fabricated with 15 wt% Bi 2 S 3 exhibited the highest methanol production from CO 2 (6.13 mmol g À1 h À1 , Table 1 entry 3), which was at least 50% higher than those obtained with bare Bi 2 S 3 (3.14 mmol g À1 h À1 ) and CdS (2.01 mmol g À1 h À1 ), under visible light irradiation. The enhanced photocatalytic activity suggested that the establishment of a heterojunction between CdS and Bi 2 S 3 could improve charge separation and subsequently prolong the lifetime of photogenerated electron-hole pairs. Moreover, the surface area of the Bi 2 S 3 /CdS nanocomposite, which was 24-27 m 2 g À1 , was slightly higher than those of the bare CdS and Bi 2 S 3 (12 and 21 m 2 g À1 , respectively). Hence, the synergetic effect of surface area and the heterojunction established between these two semiconductors had signicantly improved the overall performance in CO 2 PR. Increasing the specic surface area does not only provide more active sites for the photocatalytic reaction, but also affects the optical properties of the material. For instance, Jin et al. recently proposed that by increasing the length-to-width ratio of Bi 2 S 3 nanoribbons, which increased the bandgap energy of Bi 2 S 3 from 1.22 to 1.38 eV, the CH 3 OH yield obtained was increased from 25.94 to 32.02 mmol g catalyst À1 h À1 under visible light irradiation (l $ 420 nm). 61 However, the coupling of Bi 2 S 3 nanoribbons with CdS was not demonstrated. Hence, it will be interesting to see the performance of Bi 2 S 3 nanoribbons/ CdS nanocomposites in the CO 2 PR. The coupling of CdS with other metal oxides, such as WO 3 , has been demonstrated recently. For instance, Jin et al.
proposed the coupling of WO 3 hollow spheres with CdS to form a hierarchical Z-scheme to increase the CO 2 PR efficiency. 62 The coupling of WO 3 -CdS had greatly enhanced the photoconversion of CO 2 to CH 4 to $1.0 mmol g catalyst À1 h À1 under visible irradiation (l $ 420 nm), whereas pristine WO 3 and CdS only produced trace amounts of CH 4 .
Recently, the synthesis of Zn x Cd 1Àx S solid solutions has attracted extensive attention due to their versatility in tuning the band structures. [63][64][65] Moreover, the introduction of Zn can  Fig. 3), produced 1.28 mmol of CO with a selectivity of 93% aer 4 h. However, pristine CdS and other synthesized Zn x Cd 1Àx S solid solutions produced less than 0.4 mmol of CO under visible light irradiation. The superior performance of ZCS-1 was attributed to the presence of sulphur vacancies that trapped photogenerated electrons, provided CO 2 adsorption sites and facilitated the interaction between the Zn x Cd 1Àx S solid solution and tetra(4carboxyphenyl)porphyrin iron(III) chloride, resulting in efficient interfacial electron transfer for the subsequent photocatalytic reduction reaction.
Copper-based materials. Cu 2Àx S, which have been shown to exhibit localised surface resonance in the near infrared region, and CuS, which has a direct bandgap of 2.0 eV, are nearly ideal for optimal sunlight absorption. 66,67 By carefully controlling the anodization voltage and temperature during the electrochemical anodization of copper foil and copper-coated Kapton substrates, in sodium sulphide electrolyte, copper sulphides with a nanowall nanostructure were obtained (Fig. 4). 60 The sample anodized with 1.5 V at 5 C exhibited the highest methane formation in the CO 2 PR (Table 1 entry 5) under the irradiation of simulated sunlight. At low voltage and temperature, sulphur diffusion was low, leading to a lower concentration of excess sulphur in the sample that yielded Cu 2 S. As a result, less bulk Cu vacancy defects were formed within the sample. Cu 2 S exhibited higher charge mobility than the CuS nanostructured array, which was obtained at high temperature and high voltage.
To engineer the bandgap energy of the photocatalyst that matches the solar spectrum, a solid solution with large and small bandgap semiconductors was proposed. For instance, Arai et al. used the Cu-based sulphide complex Cu 2 ZnSnS 4 with a direct bandgap of 1.5 eV and a large optical absorption coef-cient and obtained a high selectivity of the photoelectrochemical CO 2 reduction reaction (>80%). 68 The Cu-based sulphide complex reported by Liu et al. showcased that the Cubased sulphide complex was able to reduce CO 2 under visible light irradiation in the presence of a Ru co-catalyst. 69 The Ru-Cu x Ag y In z Zn k S m solid solutions induced the formation of methanol in CO 2 PR under visible light irradiation (Table 1 entry  6). Although the study reported that the optimal performance could be obtained through the elemental composition manipulation, the nanostructures of the sulphide complex were not revealed. It is therefore questionable whether the efficiency of these photocatalysts could be further enhanced through the manipulation of their microstructures. Moreover, the stability of metal sulphates in most of the studies has not been demonstrated, and this should be emphasized more in future work.

Oxides
Semiconducting oxides have been widely used as photocatalysts because of their stability and resistance to photocorrosion under irradiation. Hence, oxides have been used for photooxidation and photoreduction reactions. The intrinsic properties of metal oxides play a critical role in determining their feasibility for CO 2 PR. For example, CO 2 PR was observed for p-type NiO covalently linked with a Zn porphyrin light-harvesting sensitizer and rhenium bipyridine system, whereas the CO oxidation reaction was observed when a similar system was coupled with n-type NiO. 70 There are two main groups of metal oxides with a closed-shell electronic conguration that have been at the centre of interest for a CO 2 PR system. The rst group includes octahedrally coordinated d 0 transition metal ions (Ti 4+ , Zr 4+ , Nb 5+ , Ta 5+ , V 5+ , and W 6+ ). Apart from TiO 2 , which is the most prominent member of this group, other binary oxides (e.g., ZrO 2 , Nb 2 O 5 , and Ta 2 O 5 ) have been used in CO 2 PR. A number of more complex oxides referred to as titanates, niobates, tantalates, etc. 71,72 are oen found in a perovskite composite, AMO 3 (A ¼ electropositive cation and M ¼ transition metal; e.g., SrTiO 3 and NaNbO 3 ), or in perovskite-related structures. Since a recent published review has covered the use of perovskite oxide nanomaterials for CO 2 photoreduction, 73 this area will not be further discussed here. The second group includes main group metal oxides in a d 10 conguration with a general formula of M y O z or A x M y O, where M represents Ga, Ge, In, Sn, or Sb. Many of these photocatalytically active binary and ternary oxides initially found application in photocatalytic water splitting, but they have very recently started to be utilised for CO 2 PR. 73 Zinc-based materials. ZnO has been widely used in the photodegradation of organic dyes and chemicals due to its direct and wide bandgap (3.37 eV). 103 Additionally, the bandgap and photocatalytic mechanism of ZnO are similar to those of TiO 2 , and thus, ZnO was also used for CO 2 PR. To compare the CO 2 photoreduction efficiency of ZnO with that of other commonly used wide bandgap semiconductors, Yahaya et al. employed commercially available TiO 2 , ZnO and NiO as photocatalysts for CO 2 photoreduction under 355 nm UV laser irradiation. 74 Among the samples, ZnO and NiO produced high yields of methanol (325 and 388 mmol g À1 h À1 over 1.5 h, respectively, Table 1, entries 7 and 8); whereas TiO 2 had the lowest production yield. In order to enhance the light absorption of commercial ZnO in the UV-vis region, ZnO was calcined to 350 C and the ZnO obtained was immobilised onto a stainless-steel mesh to reduce the agglomeration of the photocatalyst. 104 The maximum conversion of CO 2 achieved was 11.9% (i.e., percentage of CH 4 produced from CO 2 in the presence of a CH 4 reductant). A study revealed that the microstructure of ZnO played a more vital role than doping of ZnO with nitrogen (N-ZnO), even though the latter showed enhanced light absorption from 400 to 650 nm. 75 A uffy mesoporous structured ZnO with a surface area of 29.7 m 2 g À1 exhibited enhanced CO production (0.73 mmol g catalyst À1 h À1 , Doping has been widely used to extend the light absorption of wide bandgap semiconductors to a longer wavelength region by introducing intra-band states above the valence band. However, this approach tends to increase the recombination rate and decrease the charge mobility of the semiconductor, as discussed in Section 1. To avoid these drawbacks, the introduction of foreign cations into the binary semiconductor was considered instead of doping. For example, the ternary Zn 2 GeO 4 semiconductor was used for CO 2 PR under UV-vis irradiation. By fabricating ultralong and ultrathin single crystal Zn 2 GeO 4 nanoribbons, the photocatalytic reduction rate of CO 2 into CH 4 was greatly enhanced to 25.0 mmol g catalyst À1 h À1 when compared to that of the bulk Zn 2 GeO 4 (trace amounts, Table 1, entry 12). 77 The enhanced photocatalytic efficiency was attributed to the superb crystal quality and higher surface area (28.3 m 2 g À1 ) when compared to the bulk Zn 2 GeO 4 (0.75 m 2 g À1 ), resulting in enhanced separation of photogenerated electron-hole pairs and charge mobility. In the following year, the same group proposed the synthesis of the single crystal Zn 2 GeO 4 at 40 C to optimise the surface area. 78 As a result, the surface area of the synthesized Zn 2 GeO 4 nanorods was 33.1 m 2 g À1 which yielded 179 and 35 ppm g catalyst À1 h À1 of CO and CH 4 , respectively.
Further increasing the temperature to 100 C, however, decreased the surface area to 14.8 m 2 g À1 , yielding only 3.2 and 0.4 ppm h À1 of CO and CH 4 , respectively. By reducing the concentration of the Ge-precursor and the solvothermal time employed in the rst study in 2010 (refer to ref. 77) by half, a sheaf-like superstructured Zn 2 GeO 4 was obtained and reported by the same group in 2012 (Fig. 6). 79 Although the CO 2 PR of the superstructured Zn 2 GeO 4 was not reported in this study, the optimised RuO 2 and Pt co-loaded Zn 1.7 GeN 1.8 O nano-sheaves aer nitridation (32.33 m 2 g À1 ) could produce CH 4 with an apparent quantum yield of 0.024% at 450 nm (Table 1 entry 14).  Other nanostructured ternary Zn-based oxides were also proposed by the same group more recently, including ZnGa 2 O nanosheet-scaffolded microspheres 80 and hexagonal nanoplatetextured Zn 2 SnO 4 with micro-octahedron architecture for CO 2 PR application. 81 The unique architecture of the synthesized ZnGa 2 O and Zn 2 SnO 4 signicantly enhanced the separation of photogenerated electron-hole pairs, increased the surface area and extended light absorption. Hence, the methane yield obtained from the CO 2 PR was greatly improved from trace amounts to 69 82 By using in situ FTIR, it was found that the CO 2 amount adsorbed on the surface of Ce-doped ZnFe 2 O 4 was much higher than that on pristine ZnFe 2 O 4 . This phenomenon was attributed to the increase of basicity due to the presence of alkaline CeO 2 and electron density on the surface of the Ce-doped ZnFe 2 O 4 , thereby increasing the number of adsorption bonds between the CO 2 molecules and the surface of the photocatalyst, and activating the O]C bond (Fig. 7). The formation of active b-CO 3 2À and b-HCO 3 À species, which could be readily translated to highly valuable products in the CO 2 photoreduction, was detected. Recently, Xiao et al. discovered that ultrane ZnFe 2 O 4 nanoparticles with a high specic surface area (112.9 m 2 g À1 ) could promote the selectivity of the photoproduction of CH 3 CHO over CH 3 CH 2 OH, and they produced 57.8 and 13.7 mmol g À1 h À1 , respectively, under visible light irradiation (>400 nm). 105 Tungsten-based materials. Among the rst group of metal oxides, WO 3 has the smallest bandgap energy of 2.7 eV and as the edge of its conduction band is located at 0 V vs. NHE at pH 7, it cannot reduce CO 2 . 4 83 The change of the predominantly exposed facets had a signicant effect on the electronic conguration of WO 3 , in which the rectangular sheet-like WO 3 possessed a slightly larger bandgap (2.79 eV) and its conduction band was increased by 0.3 eV. As a result, the conduction band was positioned slightly above the CH 4 /CO 2 potential (À0.24 V), inducing the methane formation from CO 2 at a rate of $0.34 mmol g catalyst À1 h À1 . Chen et al. found that the conduction band of WO 3 was shied to a more negative position (À0.42 V, bandgap energy: 2.79 eV) with a stronger reducing driving force, when the ultrathin ($4-5 nm) single crystal WO 3 was synthesized using a solid-liquid phase arc discharge route in an aqueous solution. 84 The yield obtained from the CO 2 PR was $1.1 mmol g catalyst À1 h À1 under visible light irradiation (l > 420 nm), whereas commercial WO 3 powder produced only trace amounts of methane under the same conditions. A high aspect ratio of ultrathin W 18 O 49 exhibited extended optical properties in the visible and near infrared regions due to the presence of a large amount of oxygen vacancies (Fig. 8). 85 The synthesized   ultrathin W 18 O 49 exhibited the photoreduction of CO 2 to CH 4 at 70 C under visible light irradiation without a co-catalyst (Fig. 8, Table 1 entry 20). The study observed that the selectivity towards CH 4 over other hydrocarbons (e.g., ethanol and acetone) was as high as 95%. The introduction of foreign elements into tungsten oxide, which generated ternary Bi 2 WO 6 , was reported. 86 The Bi 2 WO 6 with predominant {001} facets was proposed to be the most energetically favoured reactive surface for CO 2 dissociation, resulting in 1.1 mmol g catalyst À1 h À1 of methane under visible light irradiation (l > 420 nm), whereas the bulk Bi 2 WO 6 prepared through a solid state reaction produced only trace amounts of methane. Cheng et al. also proposed that the microstructure of Bi 2 WO 6 could enhance CO 2 adsorption. 87 A template-free anion exchange strategy was used to synthesize hollow microspheres of Bi 2 WO 6 ( Fig. 9a and b). The synthesized Bi 2 WO 6 exhibited higher CO 2 adsorption capacity when compared to BiVO 4 and BiPO 4 nanoparticles without hollow structures ( Fig. 9c and d, respectively), leading to high photoconversion of CO 2 into methanol.
Niobate-based materials. Niobates with a perovskite structure have gained some attention because they share many characteristics (i.e., non-toxicity, stability, and indirect wide bandgap) with titanates. Moreover, the conduction band of niobates is slightly more reductive than that of titanates, suggesting that niobates could be a more suitable material for CO 2 PR. A study had shown that the microstructure of NaNbO 3 played an important role in its photocatalytic activity. NaNbO 3 nanowires (653 ppm h À1 g À1 , Fig. 10a) with a smaller bandgap (3.2 eV) and larger surface area (12.0 m 2 g À1 ) exhibited much higher methane formation from CO 2 when compared to the NaNbO 3 bulk (3.2 eV, 1.4 m 2 g À1 , 22 ppm h À1 g À1 , Fig. 10b). 88 The enhanced photocatalytic activity was proposed to be due to the high crystallinity, high aspect ratio and anisotropic effect of the synthesized nanowires. Li et al. also demonstrated that the size and microstructure of photocatalysts play an important role in their photocatalytic activity. 89 KNb 3 O 8 and HNb 3 O 8 nanobelt samples with a higher surface area, which were 28.8 and 39.4 m 2 g À1 , respectively, exhibited a $10 times higher photoproduction rate of methane from CO 2 than the irregularly shaped KNb 3 O 8 and HNb 3 O 8 samples, which were 2.7 and 6.5 m 2 g À1 , respectively (Table 1 entry 24 and 25). A similar observation was reported by Xie et al., in which the nanoplates of SrNb 2 O 6 with an increased surface area revealed improved chemisorption of CO 2 and the separation of photogenerated electron-hole pairs. 107 As a result, more products, such as CO and CH 4 , were obtained from the CO 2 PR compared to the SrNb 2 O 6 nanorods and nanoparticles with a lower surface area. A more recent study revealed that the nanorod-structured SrNb 2 O 6 (1.78 m 2 g À1 , 51.2 mmol g catalyst À1 h À1 ) exhibited a higher photoreduction rate and selectivity towards CO evolution over H 2 (>95%) than the SrNb 2 O 7 nanoakes (3.85 m 2 g À1 , 6 mmol g catalyst À1 h À1 , $39%) and SrNb 2 O 6 nanoparticles even though the latter possessed a higher surface area. 90 This phenomenon was attributed to the separation of the reduction and oxidation sites on the nanorods that decreased the recombination of photogenerated electron-hole pairs. Tantalum-based materials. Tantalum-based semiconductors have been widely used as a photocatalyst for water splitting. Having higher potentials than TiO 2 and above the reduction potential of CO 2 /CH 3  However, the large bandgap energy of Ta 2 O 5 ($3.9 eV) has restricted its light absorption in the visible region. 108 Hence, Sato et al. used N-Ta 2 O 5 to couple a series of ruthenium bipyridine catalysts for the photocatalytic CO 2 reduction to formic acid under visible light irradiation (405 nm). 109 The production rate was found to be $70 mmol g catalyst À1 h À1 in an acetonitrile/ triethanolamine mixture. To enhance the electron transportation and suppress the electron-hole recombination, Ta 2 O 5 was immobilised on reduced graphene and NiO x was used as the co-catalyst (Fig. 11). 91 The highest photoproduction rate of methanol from 3% NiO x - Ta   h À1 aer coupling 1 wt% NiO as the co-catalyst and 1.4 mmol g catalyst À1 h À1 aer the application of reduction-oxidation pretreatment. Tuning the size and crystallinity of InTaO 4 nanoparticles resulted in the bandgap energy range from 2.6 to 3.0 eV and could also enhance the production of methanol from CO 2 . 111 The highest production rate was about 2.7 mmol g catalyst À1 h À1 when 1.0 wt% NiO was added as the co-catalyst.
The methanol generation from InTaO 4 was further enhanced by introducing core-shell Ni/NiO nanoparticles on nitrogen doped InTaO 4 , leading to 160 mmol g catalyst À1 h À1 under the irradiation of light with wavelengths ranging from 390 to 770 nm. 92 KTaO 3 was also used to reduce CO 2 to CO under visible light irradiation. 112 Three samples were synthesized with different bandgaps ranging from 3.5 to 3.7 eV and yielded the highest amount of CO at $62 ppm g catalyst À1 h À1 .
Recently, LaTa 7 O 19 and CaTa 4 O 11 (bandgap energies of 4.1 and 4.5 eV, respectively) were shown to be active for CO 2 PR. CO was produced aer loading with 1 wt% Ag co-catalyst due to the preferable conduction band positions (50 and 70 mmol g catalyst À1 h À1 , respectively, in the presence of 0.1 M NaHCO 3 under the irradiation of a 400 W high-pressure mercury lamp). 93 NaTaO 3 doped with different elements, such as Mg, Ca, Sr, Ba and La, has been proposed as a highly active photocatalyst for CO 2 PR using water as the electron donor in the presence of a Ag cocatalyst under UV irradiation. 94 Among the samples, Ba-doped NaTaO 3 loaded with 1.0 wt% Ag co-catalyst using the liquidphase reduction method exhibited the highest CO production and selectivity from CO 2 ($50 mmol g catalyst À1 h À1 and 56%, respectively). Quaternary tantalates have been developed recently and revealed to be active for CO 2 photoreduction in the presence of water. 113 K 2 RETa 5 O 15 (RE ¼ rare-earth element, namely La, Ce, Pr, Nd, Y, and Sm) was fabricated using the ux method with KCl, which favoured the rod-like morphology, followed by calcination treatment at 1150 C for 6 h. 95 Among the quaternary tantalates, K 2 CeTa 5 O 15 possessed the smallest bandgap energy (2.42 eV, 0.7 mmol g catalyst À1 h À1 ), but K 2 YTa 5 O 15 photoproduced the highest amount of CO (3.86 eV, 91.9 mmol g catalyst À1 h À1 ). The addition of Y was shown to be benecial for capturing CO 2 and subsequently for photoreduction. Meanwhile, the presence of K in the tantalates played an important role in determining the growth orientation of the rod-like structure, thereby affecting the activity in CO 2 photoreduction. Miscellaneous. CeO 2 is a basic metal oxide that can transform inert linear CO 2 to b-HCO 3 À and b-CO 3 2À to reduce the reductive potential of CO 2 . 114 Hence, it has recently attracted a lot of attention. However, it suffers from rapid recombination of photogenerated electron-hole pairs and possesses a large bandgap (3.0-3.4 eV), which restricts the light absorption in the UV range. 115 To improve the performance of CeO 2 for the photocatalytic reduction of CO 2 under visible irradiation, Xiong et al. proposed the coupling of Ag/Ag 3 PO 4 with CeO 2 to construct heterojunctions for enhancing the separation of photogenerated electron-hole pairs and improve light absorption because Ag 3 PO 4 has a narrow bandgap of 2.42 eV. 116 The highest CH 3 OH and C 2 H 5 OH yield obtained was 10.6 and 7.9 mmol g catalyst À1 h À1 , respectively. Zhang discovered that when Ni was loaded on CeO 2 , the nanocomposite exhibited enhanced photo(thermo)catalytic performance and inhibited carbon deposition. 117 Moreover, it is interesting to note that the full light spectrum response from UV to infrared of the Ni metal on CeO 2 decreased the activation energy of C and CH oxidation steps, thus improving the overall photo(thermo)catalytic performance.
MgO was employed to photocatalytically reduce CO 2 into CO with a production rate of $1.6 mmol g À1 h À1 over 6 h in the presence of H 2 as the reductant under UV light (l < 290 nm). 118 Mesoporous Ga 2 O 3 yielded 1.46 and 0.21 mmol g À1 h À1 of CO and CH 4 , respectively, from CO 2 under visible light irradiation. 119 When Ga 2 O 3 was loaded with Ag, the photoproduction rate of CO from CO 2 was 10.5 mmol g À1 h À1 under UV light irradiation (Table 1 entry 33). 96 Iron oxides were proposed as a photoactive centre to induce the photocatalytic reduction of CO 2 . 120 Using electron spin resonance spectroscopy (ESR), the photogenerated electrons from the Fe-O species were efficiently consumed by CO 2 under UV irradiation.
Lamellar BiVO 4 was proposed to exhibit a selective methanol production from CO 2 photoreduction under visible light irradiation. 97 The maximum CH 3 OH production rate was 5.52 mmol h À1 when 0.2 g of BiVO 4 was suspended in 100 mL of NaOH (1.0 M) under full spectrum irradiation of a Xe lamp. The photocatalytic mechanism was proposed according to which the Bi 3+ sites could efficiently receive electrons from the V 3d-block bands of the BiVO 4 to form CO 2 c À radical anions, leading to the formation of methoxyl radicals ($OCH 3 ) and eventually CH 3 OH aer hydrogen abstraction. Wang et al. doped the atomically thin layers of BiVO 4 with different percentages of Co. 121 The Co-doped BiVO 4 exhibited an efficient and stable activity for CO 2 photoreduction to CH 4 . The optimal CH 4 production rate was 23.8 mmol g À1 h À1 , which was three times higher than that of the pristine BiVO 4 , at 60 C with an atmospheric CO 2 concentration ($400 ppm) under a UV lamp (25 W at 254 nm). The enhancement of the production rate of the Co-doped BiVO 4 was suggested to be due to the presence of electron enriched adsorption sites, which was contributed by the Co dopant, activating the CO 2 molecules for further reduction reaction.
Delafossite materials with a general stoichiometry of ABO 2 , in which A is a monovalent metal ion, such as Cu, Ag, and Pt, and B is a trivalent metal ion, such as Al, Ga, and Fe, as the new class of photocatalysts have also been considered for CO 2 PR. 98 CuGaO 2 (bandgap energy $3.7 and weak absorption at 2.6 eV) and the Fe-alloyed CuGa 1Àx Fe x O (1.5 eV) facilitated the photogeneration of CO from CO 2 under the irradiation of a Xe lamp though varied amounts of Fe substituted into CuGaO 2 did not signicantly enhance the CO 2 photoreduction performance ( Table 1 entry 35). 98 Based on the Lewis acidity of CO 2 , alkaline catalysts will benet the adsorption and activation of CO 2 . Layered double hydroxide (LDH) materials usually possess high specic surface areas, which provide numerous active sites for the catalytic reaction. The fabricated CoAl LDH facilitated an enhanced CO 2 photoreduction reaction when compared to P25 due to the surface alkaline OH groups for efficient adsorption of CO 2 at a low concentration. 99 The utilisation of LDH in CO 2 photoreduction has been reviewed previously, and thus, interested readers may refer to the published review articles. 40,122 In summary, metal oxides have shown their ability to promote photocatalytic reduction of CO 2 , as discussed in the previous section. However, most of these photocatalysts only work under UV irradiation due to their large bandgap energies (>3 eV). The relatively large bandgap of metal oxides originates from the valence band maximum, which is formed by O 2p orbitals and is more positive than 3 V. 123 Hence, if metal oxides meet the thermodynamic requirement for CO 2 PR and H 2 O photooxidation, then the bandgap of the metal oxides inevitably becomes larger than 3.0 eV, which is too wide to absorb visible light. 124

Oxynitrides
Tantalum-based materials. The N 2p orbital has a higher potential energy than the O 2p orbital, which indicates that metal oxynitrides as well as metal nitrides could be employed as an efficient photocatalyst. 42,125 For example, in the case of tantalum oxynitride and tantalum nitride, when nitrogenbased N 2p atomic orbitals were introduced into Ta 2 O 5 , new orbitals with a higher bound state energy are generated, resulting in a decrease of bandgap energy. 42,108,126 As a result, the bandgaps of TaON and Ta 3 N 5 (2.5 and 2.1 eV, respectively) are both smaller than that of Ta 2 O 5 (3.9 eV) and thus can effectively absorb visible light and drive the photocatalytic activity. Moreover, the N content plays an important role in determining the bandgap energy in oxynitrides and nitrides. Therefore, Gao et al. proposed a Ca-assisted urea synthesis method to controllably synthesise TaON and Ta 3 N 5 with a tailored N composition. 127 In addition, the initial urea : Ta molar ratio used in the proposed synthesis method was also benecial to control the size and homogeneity of the nal product. Recently, the use of the porous spherical architecture of TaON for CO 2 photoreduction was proposed. 100 The surface area of the porous spherical TaON was $11.12 m 2 g À1 , whereas that of commercial TaON was $7.41 m 2 g À1 . As a result, the CH 3 CHO and C 2 H 5 OH production rates from CO 2 using the porous TaON (0.52 and 2.03 mmol g catalyst À1 h À1 ) were higher when compared to those of the commercial TaON (0.16 and 0.84 mmol g catalyst À1 h À1 ) under visible light irradiation. The enhanced CO 2 PR in the porous TaON was attributed to the increase of surface area. In addition, the reduction of charge transfer distance and enhanced light scattering within the porous spherical structure were also suggested to play roles in enhancing the photocatalytic reduction of CO 2 . The CO 2 PR is a multi-electron process, and a variety of products can be produced using a single semiconducting photocatalyst. The achievement of efficient and selective production of highly valuable fuels is critical for viable CO 2 photoreduction processes. The application of perovskite oxynitrides, such as CaTaO 2 N coupled with the binuclear Ru(II) complex photosensitiser and loaded with the Ag co-catalyst, revealed an enhanced selectivity for HCOOH production (>99%) from CO 2 under visible light irradiation due to the enhanced interfacial electron transfer. 128 A similar approach with the same photosensitiser and co-catalyst coupled with yttrium-tantalum oxynitride (YTON) was recently proposed by the same group. 129 The YTON (2.1 eV) exhibited a smaller bandgap than CaTaO 2 N (2.5 eV), thus extending the light absorption up to 600 nm. Moreover, the selectivity for HCOOH formation from CO 2 was not affected and remained as high as that in their previous study (>99%).
Zinc-based materials. Mesoporous ZnGeON was used as a photocatalyst for CO 2 PR under visible light irradiation (l < 400 nm). 130 The prolonged nitridation time from 1 to 15 h decreased the Zn and O contents, in which Zn was evaporated and O was substituted by N, at 800 C in an NH 3 environment. However, the crystallinity of ZnGeON was enhanced with a slight decrease in the surface area. The ZnGeO nitrided for 10 h (24.4 m 2 g À1 ) exhibited the highest CH 4 production rate of 2.7 ppm g catalyst À1 h À1 , which was higher than those of the and formed the valence band, whereas the conduction band was composed of the Ga 4s and 4p orbitals. 101 As a result, the bandgap energy was reduced to 2.5 eV, resulting from the upliing of the maximum of the valence band and lowering of the minimum of the conduction band. To further enhance the CO 2 PR performance, a ZnGa 2 ON solid solution was modied with ZnAl 2 O 4 that acted as the CO 2 arrester (Fig. 12). 101 The increase of Zn content had also decreased the bandgap energy to 2.3 eV. Beneting from the mesoporous structure, smaller bandgap and enhanced CO 2 adsorption ability, the ZnAl 2 O 4 -modied ZnGa 2 ON showed a methane generation rate of 9.2 mmol g catalyst À1 h À1 from CO 2 , which was 9 times higher than that of the pristine ZnGa 2 ON under visible light irradiation (l $ 420 nm).

Nitrides
Gallium-based materials. Through engineering the nanostructure of the co-catalyst used, selectivity using semiconducting nitrides could be enhanced dramatically, as demonstrated by AlOtaibi et al. 102 The decoration of the nonpolar GaN nanowires with the Rh core and amorphous Cr 2 O 3 shell co-catalyst signicantly increased the production rate of CH 4 from 1.3 (bare GaN) to 3.5 mmol g catalyst À1 h À1 , but the CO production rate decreased from 1130 (bare GaN) to $120 mmol g catalyst À1 h À1 in 24 h (Fig. 12). Due to the effective collection of photogenerated electrons by the Rh core and amorphous Cr 2 O 3 shell co-catalyst, no apparent reductive reaction (e.g., photoreduction of CO 2 to CO) occurred on the surface of GaN without Rh coverage. As a result, the product selectivity towards CH 4 was enhanced in the CO 2 PR under UV-visible light irradiation. In addition, the decoration of the GaN nanowires with Rh/Cr 2 O 3 could suppress the back reaction that formed H 2 O from H 2 and O 2 , and offered adsorption sites for CO 2 .
In summary, both metal oxynitrides and nitrides have shown their capability to photoreduce CO 2 with more favourable optical properties when compared to metal oxides. Unfortunately, these groups of materials have not been extensively explored.

Hole scavengers for CO 2 photoreduction
Various semiconducting materials have been proposed as photocatalysts for CO 2 PR under UV and/or visible irradiation, as discussed in Section 2 and summarised in some other references. [7][8][9][10]103,131 However, the quantum efficiency of the CO 2 photo-conversion into hydrocarbons remained low and could not rely only on the development of photocatalysts. System optimisation plays an important role in optimising the conversion rate and selectivity as well as photocatalyst stability. Therefore, the increase of CO 2 PR efficiency through the introduction of a hole scavenging agent has gained signicant interest. In this section, the use of organic and inorganic hole scavengers is reviewed.

Inorganic hole scavengers
As discussed in Section 2.1, metal sulphides suffer from photocorrosion in an aqueous dispersion due to the oxidation of lattice S 2À ions to elemental S and subsequently to sulphates. 51 Hence, the addition of reducing agents to prevent the oxidation of the lattice S 2À ions by scavenging the photogenerated holes was proposed. Kanemoto et al. achieved a cumulative quantum yield of 72% with irradiation of UV light at 313 nm (i.e., 75.1 and 1.7 mmol g catalyst À1 h À1 of HCOOH and CO, respectively) when NaH 2 PO 2 and Na 2 S (0.35 and 0.24 M, respectively) were added into the system that contained ZnS as the photocatalyst for CO 2 PR (Table 2 entry 1). 132 A systematic study was recently carried out to investigate the effect of Na 2 S as the hole scavenger for ZnS on the CO 2 PR at l ¼ 345 nm. 133 The study elucidated that the photogenerated holes on the surface of ZnS were directly consumed by Na 2 S, whereas photogenerated electrons were pumped into the conduction band simultaneously. In addition, the behaviour of the reaction rate at different pH values resembled that of the solubility of CO 2 , discarding the direct participation of the HCO 3 À and CO 3 2À in the photoreduction process. This observation was supported by a very recent study, in which KHCO 3 was used as the hole scavenger in an aqueous system with ZnS. 134 The study demonstrated that KHCO 3 acted as an effective hole scavenger as well as a buffer to mitigate the pH change induced by the CO 2 saturation. This phenomenon, however, was not observed when only K 2 SO 3 was used as the hole scavenging agent. The optimised solution with 0.1 g of colloidal ZnS, 0.1 M K 2 SO 3 and 0.5 M KHCO 3 achieved 464.2 and 81.3 mmol of HCOOH and CO, respectively, under UV-vis irradiation ( Table 2 entry 2). 134 The selectivity towards HCOOH was reported to be 12.5%, and this could be improved to 95.0% when Cd was added to the colloidal ZnS suspension as the co-catalyst.
Inorganic salts (e.g., NaOH, Na 2 S, etc.) have been reported to have a signicant effect on CO 2 PR. 97,132,134,135 The addition of NaOH had been shown to increase the solubility of CO 2 compared to pure H 2 O because the OH À ions provided by NaOH in aqueous solution reacted with the dissolved CO 2 , and transformed into CO 3 2À and further into HCO 3 À in the CO 2 saturated system. 136 It was suggested that the high the common hole scavengers for CO 2 PR. A previous report proposed that the light energy could be stored within the lightinduced reaction given as The Gibbs free energy of this reaction is +62.8 kJ mol À1 at 25 C. 147 In a system of Cd-loaded ZnS, CO 2 photoproduced formic acid with a quantum efficiency 32.5% in the presence of 1 M 2propanol. 147 Further increasing the Cd concentration resulted in the formation of CO. A study revealed that CdS was capable of photoreducing CO 2 to CO when N,N-dimethylformamide (DMF) containing 1 v/v% water was employed in the system. 148 A similar observation was reported, in which CO was photoproduced when CdS was dispersed in DMF under the irradiation of a 500 W mercury lamp with a 300 nm cut off lter. 140 When DMF was substituted with a low polarity solvent, such as CCl 4 and CH 2 Cl 2 , CO production was dominant, whereas when using a high polar solvent, such as H 2 O, formate was produced. This was because the adsorbability of the CO 2 c À , an intermediate species aer the activation of CO 2 , was strongly dependent on the polarity of the solvent used. For instance, low polarity molecules enabled strong adsorption of CO 2 c À on the Cd sites of CdS through the carbon atom of CO 2 c À , which was not highly solvated in solvents of low polarity, resulting in the formation of CO. When high polar solvents were used, CO 2 c À was stabilised in the system and established only weak interactions with the photocatalyst. As a result, CO 2 c À tended to react with a proton and produced formate.
A recent study suggested that the CO 2 photoreduction process can be greener when glycerol, which is a green solvent derived from vegetable oil, was used as the hole scavenger instead of petroleum-derived solvents. 141 In this study, wurtzite ZnS facilitated the photoproduction of formic acid from CO 2 with an apparent quantum efficiency of 3.2% and 0.9% when glycerol and 2-propanol, respectively, were employed as the hole scavenger.
Cyclohexanol was used as the hole scavenger for the CO 2 photoreduction under UV light irradiation. 149 The optimised sample exhibited the production of cyclohexyl formate and cyclohexanone (178.1 and 170.2 mmol g catalyst À1 , respectively) aer 8 h. The authors elucidated that the production of cyclohexanone was slightly lower than that of cyclohexyl formate because some of the photogenerated holes were consumed by cyclohexanol to form cyclohexyl ether. A recent study demonstrated that a Ru(II)-complex/C 3 N 4 nanocomposite could induce the photocatalytic CO 2 reduction by using a mixture of solvents (N,N-dimethylacetamide and DMA/TEOA). 142 The apparent quantum efficiency achieved was 5.7% at 400 nm ( Table 2 Entry 12). In addition, the product selectivity of the Ru(II)-complex/C 3 N 4 nanocomposite could be enhanced through manipulating the solvent used (Fig. 13). 145 In order to avoid using organic solvents as the medium, the mononuclear Ru(II) complex proposed in a previous study 142 was replaced with a binuclear Ru(II) complex coupled with Ag/C 3 N 4 and was employed as the photocatalyst (Table 2 entry 13). 143 Since no reduction product was obtained in pure water, a hole scavenger (ethylenediaminetetraacetic acid disodium salt dihydrate, EDTA$Na 2 ) was added to promote the photocatalytic CO 2 reduction in water. The main product was HCOOH, and H 2 was produced as a by-product under visible light irradiation (l > 400 nm). Other hole scavenging agents (e.g., potassium oxalate and sodium ascorbate) were shown to be useful for the CO 2 PR. Among the three hole scavenging agents, sodium ascorbate exhibited the best performance with 31.7 mmol and 86% selectivity towards HCOOH. The HCOOH production could be further enhanced to 83.3 mmol with selectivity 97% when K 2 CO 3 (0.1 M) was used as an additive. 144 However, the production of H 2 was reduced by half.
The introduction of organic and inorganic hole scavenging agents has exhibited advantages to enhance the efficiency of CO 2 PR. The presence of hole scavenging agents in the CO 2 PR process is necessary if the oxidation reaction in the CO 2 PR cannot be inhibited by the photocatalyst. Moreover, to avoid carbon contamination and false positive errors for the photogeneration of hydrocarbons in the CO 2 PR process, inorganic hole scavenging agents are preferred.

Conclusions and future directions
To date, signicant achievements have been made in the design and fabrication of photocatalysts and the optimisation of photocatalytic systems. CO 2 PR using metal sulphides, oxides, oxynitrides and nitrides accumulated so far have offered alternative photocatalytic materials other than TiO 2 . Material properties, including the surface area, light harvesting, and charge generation, separation and transportation, have been manipulated through the structural and morphological control during the fabrication processes, leading to enhanced CO 2 PR performance. Amongst the non-titania photocatalysts (metal sulphides, oxides, oxynitrides and nitrides) reviewed here, the ultrathin W 18 O 49 exhibited the highest CH 4 yield (2200 mmol g catalyst À1 h À1 ) from CO 2 under visible light irradiation. The presence of oxygen vacancies was suggested to play an important role in the CO 2 PR. On the other hand, the addition of inorganic salts or organic solvents into an aqueous system has shown to effectively scavenge the photogenerated holes and/or increase CO 2 solubility. Although signicant studies have been carried out on CO 2 PR, some challenges still remain. Firstly, an in-depth understanding of the working mechanism in a CO 2 photoreduction process is still not well understood. Hence, a trial-anderror approach was used when fabricating photocatalysts, attempting to achieve a high CO 2 PR efficiency. Secondly, the insight into the CO 2 PR in the presence of hole scavenging agent(s) is not available. Moreover, due to this lack of knowledge, a rational design to combine state-of-art of photocatalysts with the desired hole scavenging agent(s) for carbon fuel production is difficult to achieve. Therefore, while more effort is required in material advancement, studies of the combined effect of the proposed photocatalyst with a hole scavenger should be encouraged. In addition, further investigation of CO 2 PR at the molecular level through in situ characterisation techniques should be carried out as this is key to boosting the efficiency of CO 2 PR.

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