Perspective: Photocatalytic reduction of CO₂ to solar fuels over semiconductors

To date, the depletion of finite fossil fuels possesses numerous challenges to the survival of humans, and the pursuit for new energy sources is urgent. Simultaneously, the global warming caused by increasing emissions of the greenhouse gas “CO₂,” mainly from the combustion of fossil fuels, induces serious environmental crises on the earth such as the polar ice caps melting and fast rising sea levels.^1–3 Therefore, as a sustainable alternative to the depletion of fossil fuels as well as a means to reduce CO₂ emissions, reduction of CO₂ to useful fuels by water is receiving extensive attention.^4–23 Due to the high stability of CO₂ (ΔG_f⁰ = −394.4 kJ·mol⁻¹) caused by the highly oxidized state of carbon and stable C=O bonds, the transformation of CO₂ to reduced products requires a significant energy input (ΔG > 0).²⁴ Inspired by photosynthesis of plants, photocatalytic reduction of CO₂ is gaining increasing interest and considered as one of the most promising strategies to produce renewable fuels as it is driven by photo-excited charge carriers. The first work of CO₂ photoreduction, reported by Inoue et al., showed that the photocatalytic reduction of CO₂ in an aqueous suspension of a semiconductor can form formaldehyde, formic acid, methanol, and methane.²⁵ Following this breakthrough, numerous researchers have focused on the photocatalytic mechanism of CO₂ conversion over a wide variety of semiconductors.^{8,9,17,22,26–34} However, the reaction efficiency and selectivity for the production of hydrocarbon fuels from CO₂ and H₂O are still unsatisfactory. A comprehensive understanding of the photoreduction mechanism, including the adsorption and activation of CO₂ molecules, and subsequent processes of reduction remains elusive.^35,36

There are numerous informative perspectives and reviews introducing current research and recent advances of CO₂ photoreduction. These perspectives also highlight design strategies of photocatalysts as well as the primary challenges in the field.^{3,7,24,35,37–41} Different from those perspectives and reviews, this work mainly focuses on understanding the adsorption and reaction mechanism of CO₂ as well as the modification of photocatalysts from theoretical perspectives. The adsorption of CO₂ is an important step during the reduction process. The high antibonding orbital of gas-phased CO₂ determines its need to adsorb on the surface first in order to be activated before accomplishing photocatalytic reduction.⁴² The adsorption and activation of CO₂ are somewhat complex because it involves photogenerated electrons.^10,36 It is therefore important to know how the photogenerated electrons participate in the adsorption and activation of CO₂ molecules. In this perspective, the role of the photogenerated electrons in the adsorption and activation of CO₂ molecules is elucidated. Additionally, the dominating factors affecting the adsorption and reaction pathways, such as the solvation effect and surface defects, have been addressed. The solvation effect is a significant factor in CO₂ reduction as most of the experiments occur in aqueous solution.^8,43–45 The water environment can change the adsorption configuration and reaction energies of CO₂ photoreduction and is therefore important when discussing the reaction.⁴⁶ Surface defects such as oxygen vacancies are common during the preparation of photocatalysts and create distinctive electronic structures. This unique defective system can greatly alter the adsorption behavior of CO₂ and induce different reaction pathways.⁴⁷ Upon chemical adsorption, CO₂ is reduced by surface charge carriers. The reaction involves multiple proton-coupled electron transfers and can lead to the formation of many different products depending on the chosen reaction pathways, which makes the process rather complex.^7,35 Therefore, the reaction mechanism and pathways of CO₂ reduction along with their dependent factors will be discussed in detail. The modification of photocatalysts is a substantial and important issue within the literature. It is acknowledged that incorporation of co-catalysts can improve the performance of semiconductor photocatalysts.²⁴ It is well known that the use of co-catalysts provides additional active sites; however, many other structural and electronic changes are evident which enhance the activity of photocatalysts. Thus, the beneficial impact of co-catalyst loading on the photocatalytic system will be fully elucidated. The discussions in this perspective will improve the existing and relevant understanding of CO₂ photoreduction and should help future efforts to further improve the overall efficiency and selectivity of the reaction. To conclude this work, a brief perspective on the challenges and possibilities within this field is presented.

CO₂ adsorption has been widely studied on various photocatalysts, in particular, on titanium dioxide, because it is highly stable under photocatalytic conditions, inexpensive to use, and able to perform a variety of photocatalytic reactions such as water splitting.⁴⁹ Figure 1 illustrates the adsorption of CO₂ on a TiO₂ surface with different adsorption configurations in the presence of one photogenerated electron. The configuration of A is a linear physical adsorption of CO₂, with one oxygen of CO₂ weakly bonding with the Ti site. The CO₂ molecule can also have chemical bonding with the surface and act as an electron donor and acceptor simultaneously to form configuration B. The lone pair of electrons on the oxygen of CO₂ can promote the adsorption of CO₂ on the Ti sites and form the adsorption structure C. The carbon atom can also gain electrons from oxygen on the TiO₂ surface, forming a carbonate CO₃⁻ species (D). The most stable configuration of CO₂ adsorption is the linear physical adsorption configuration (A) with O weakly interacting with the adsorption site.¹⁰ However, it is proposed that multiple reaction pathways of CO₂ photoreduction start with the formation of a CO₂^⋅− radical, which was detected by infrared spectroscopy on the surface of TiO₂ under UV illumination. This was also verified on a MgO surface by electron paramagnetic resonance.^33,50–52 Theoretically, CO₂^⋅− was also found to be activated by CO₂ in the presence of a photogenerated electron on the TiO₂ surface.⁵³ Moreover, the electronic structure of the activated CO₂^⋅− is greatly changed, and the lowest unoccupied molecular orbital of CO₂^⋅− can be even lower than the conduction band (CB) minimum of TiO₂, which greatly facilitates CO₂ reduction. For the activated CO₂^⋅− seen in configuration C, we can notice that CO₂ adsorbs with a bent configuration and at an OCO angle of 136.9°. The electron spin density largely accumulates around the oxygen of the adsorbed CO₂ [see Fig. 1(b)], which indicates the photogenerated electron transfers from the TiO₂ surface to CO₂ forming CO₂^⋅−. The adsorption and activation of CO₂ can be influenced by several factors including surface defects and the solvation effect, which will be discussed in Secs. II B 1 and II B 2.

Surface defects commonly occur in the preparation of photocatalytic materials and occasionally act as the active reaction sites since they change the geometrical and electronic structures of photocatalysts. For example, the oxygen vacancy (V_O) in a TiO₂ system has been found to participate in adsorption and activation of CO₂ molecules during photoreduction. A brief schematic diagram of CO₂ adsorption at the V_O site on the TiO₂ (110) surface is shown in Fig. 2(a).¹⁸ One of the oxygen atoms of CO₂ is located at the V_O site and the CO₂ molecule is tilted away from the surface by 57° along [110] direction. This particularly linear adsorption configuration was found to have an adsorption energy of 0.44 eV. Using scanning tunneling microscopy (STM), Lee et al. investigated the adsorption of CO₂ on the reduced rutile (110) surface at 55 K. As illustrated in Fig. 2(b), they found that CO₂ molecules (bright spots) occupy all the V_O sites rather than OH_b (dull spots). The inset STM images illustrate two CO₂ molecules before and after thermal diffusion from their V_O sites (dotted ellipses), which depict the CO₂ molecules adsorption upon the V_O sites. Their STM results are consistent with other work that showed CO₂ can adsorb at the V_O site on the reduced TiO₂ (110) surface with good stability and is also supported further with the results of thermal desorption studies.^54,55

Huygh et al. comprehensively studied the influence of oxygen vacancies on CO₂ adsorption and activation upon an anatase (001) surface using density functional theory (DFT) calculations.⁵⁶ Four different oxygen vacancies were identified and named as V_O1, V_O2, V_O3, and V_O4 as shown in Fig. 3(a). The corresponding formation energies were 4.42, 4.95, 4.76, and 5.71 eV, respectively. However, V_O4 was discovered to be highly unstable and was not considered. Although direct formation of the oxygen vacancy is very difficult, it is much easier to form oxygen vacancies from the oxidization of H₂O to O₂ in the presence of holes. The activation barrier is as low as 0.07 eV.⁵⁷ Therefore, the oxygen vacancies are possible to be generated. With respect to CO₂ adsorption, both bent and linear adsorption configurations were identified. The linear physisorbed configurations were found to easily convert to the bent chemically adsorbed configurations with small activation barriers, which are consistent with the work conducted by He et al.¹⁰ Considering CO₂ adsorption on the different oxygen vacancies, six stable bent adsorption configurations of CO₂ were found, as shown in Fig. 3(b). The relevant results for each configuration are summarized in Table I. In terms of V_O1, three configurations of CO₂ adsorption were identified. V_O1(I) shows a monodentate carbonate configuration for CO₂ adsorption with C binding to the formed Lewis base center (Ti³⁺). V_O1(II) has a bent adsorption configuration in which one oxygen of CO₂ occupies the oxygen vacancy. In contrast, V_O1(III) shows CO₂ located near the surface oxygen vacancy. The adsorption structure of V_O1(I) is the most stable with an adsorption energy of −2.11 eV resulting from the largest charge transfer from the surface to the CO₂ molecule. The adsorption of CO₂ upon V_O2 was also explored. One adsorption configuration [V_O2(I)] was found with CO₂ bonding with two Ti³⁺ ions located in the same [100] row. The adsorption energy for this configuration was calculated to be −1.10 eV. Finally, two monodentate carbonate configurations for CO₂ adsorption on V_O3 were identified, and in both configurations, the C atom bonds with the Ti site near the oxygen vacancy, however with different orientations of CO₂. The adsorption energies are −1.04 and 1.01 eV, respectively. Therefore, the most stable adsorption configuration of CO₂ on the defective surface is CO₂ adsorbing on the V_O1 site with a bent monodentate adsorption configuration. It is also more stable than CO₂ adsorbing on the perfect anatase surface (0.68 eV). In conclusion, the oxygen vacancy can be seen to highly promote the adsorption of CO₂ molecules.

The adsorption of CO₂ in the presence of oxygen vacancies was also investigated on a Cu₂O (111) surface using DFT by Wu and co-workers.⁵⁸ The adsorption of CO₂ can be promoted when the CO₂ molecule occupies the oxygen vacancy and forms the reactive CO₂^⋅− radical. When CO₂ adsorbs on the Cu and O sites near the formed oxygen vacancy, the adsorption energies are even smaller than those corresponding with adsorption on the perfect surface. Therefore, the oxygen vacancy only delivers enhanced activity for CO₂ adsorption and if CO₂ adsorbs anywhere else, only limited activity will be observed.

In general, the oxygen vacancies play an important role in the photoreduction of CO₂ on the surface of photocatalysts. The vacancies greatly improve the adsorption and activation of CO₂ when CO₂ molecules adsorb upon them, which is beneficial in the conversion of CO₂ to useful products.

In numerous experiments reported within the literature, the photocatalytic reduction of CO₂ is proceeded with water because of its hydrogen resource. Using H₂O as a reducing agent over (Au, Cu)/TiO₂ materials, Neatu et al. reported a high production rate of 2.2 ± 0.3 mmol g⁻¹ h⁻¹ for CO₂ photoreduction to CH₄.^5,59 Wang et al. also found that the photocatalytic reduction of CO₂ can be significantly enhanced in the presence of H₂O using a Pt@Cu₂O co-catalyst loaded on TiO₂.⁶⁰ Accordingly, the effect of water should be elucidated. The interactions of CO₂ and water over titania nanoparticles were thoroughly investigated using electron paramagnetic resonance.⁴³ The detection of redox products including H atoms, OH$⋅$ radicals, and CH₃$⋅$ radicals indicates water participating in CO₂ photoreduction.

Aside from acting as a H donator, the solvation effect of H₂O is also an important issue and has been extensively studied.^11,61,62 Yin et al. simulated the adsorption of CO₂ on rutile TiO₂(110) situated in a water environment using the periodic continuum solvation model (PCSM).⁴⁶ In total, four typical configurations of CO₂ adsorption were examined on the rutile (110) surface as shown in Fig. 4(a). C-1 is a tilted monodentate configuration with one oxygen of CO₂ adsorbing at a Ti_5c site; C-2 is a horizontal bidentate configuration adsorbing along the surface Ti_5c row; C-3 shows CO₂ adsorbing between two adjacent bridge oxygen atoms; similarly, C-4 is the adsorption of CO₂ on the top of one bridge oxygen atom. From Table II, C-2 is the most stable adsorption configuration in the gas phase with an adsorption energy of −0.34 eV. When the solvation effect is considered, the adsorption energy of CO₂ can be increased by about 0.3 eV relative to its adsorption in the gas phase. Interestingly, they also found that the most stable configuration of CO₂ adsorption changes from C-2 to C-1 with solvation effect, which is consistent with the results from low temperature scanning tunneling microscopy (STM).^63,64

To further explore the role of water, they also investigated the co-adsorption of CO₂ and H₂O on the rutile (110) surface. From Fig. 4(b), six configurations were considered and denoted as C_w-i, i = 1–6. As shown in Table II, the co-adsorption effect was also found to greatly affect the adsorption of CO₂ on rutile (110). Relative to single CO₂ adsorption, the co-adsorption energy of CO₂ and H₂O increases from −0.06 to −0.12 eV. It was suggested that the increased co-adsorption energy may result from hydrogen bonding between CO₂ and H₂O. When the solvation effect is also considered, the adsorption energies change by about 0.3 eV and the most stable adsorption configuration changes from C-2 to C_w-2. Furthermore, the interactions of the surface with adsorbate are also changed with parameters shown in Table II. Compared to single CO₂ adsorption in the gas phase, the distances of the O–Ti bond in C_w-1 and C_w-3 are significantly decreased by about 0.11 Å. When solvation is included, the corresponding distances are also decreased by about 0.1 Å. The O–C–O angle is also affected by the solvation and co-adsorption effect. The co-adsorption effect enlarges the O–C–O angle by 0.5° in the gas phase, while the solvation effect changes the O–C–O angle by 0.1°–0.7°. A similar trend was discovered by Sorescu et al. from co-adsorption with H₂O on rutile (110) at 1/8 monolayer (ML) and 1 ML coverages.¹¹ They found that co-adsorbed H₂O molecules generally lead to a small increase in the stability of CO₂ on the surface through the formation of hydrogen bonds. The adsorption of CO₂ is greatly influenced by H₂O and the solvation effect can even change the most stable adsorption configuration of CO₂. In conclusion, the solvation effect and co-adsorption of H₂O significantly affect the adsorption of CO₂.

Following the adsorption of CO₂, it is worth stressing that the subsequent mechanism of photocatalytic CO₂ reduction, including electron and proton transfers, is also vital. The chosen pathway affects the efficiency of these types of reactions and even determines the final products. As shown in Eqs. (2)–(6), the conversion of CO₂ to useful chemicals consists of steps involving C=O bond breaking and C–H bond formation. The conversion reaction can contain as many as eight electron and four proton transfers for CH₄ formation. Depending on the different adsorption states of the CO₂ molecules and the number of electron and proton transfers, the reaction can lead to the formation of many different products including HCOOH, HCHO, CH₃OH, and CH₄. Therefore, the photocatalytic mechanism of CO₂ is considerably complicated.

To date, many researchers have been dedicated to explicate the reaction mechanism of CO₂ reduction.^{4,6,60,65–67} Based on different CO₂ adsorption models, He et al. investigated the photoreduction of CO₂ to HCOOH on a perfect anatase (101) surface. They discovered two competing reaction mechanisms.¹⁰ The first mechanism involves a single-electron (1e⁻) reduction process (CO₂ + e⁻ → CO₂^⋅−; CO₂^⋅− + H⁺ + e⁻ → HCOO⁻; HCOO⁻ + H⁺ → HCOOH), in which the first step is the chemical adsorption of CO₂ to CO₂^⋅− by one electron followed by the conversion of CO₂^⋅− to HCOO⁻ by a proton-coupled electron mechanism [see Fig. 5(a)]. The other mechanism is a concerted two-electron process from a linear physisorption configuration of CO₂ (CO₂^⋅− + H⁺ + 2e⁻ → HCOO⁻; HCOO⁻ + H⁺ → HCOOH) as shown in Fig. 5(b). For the first reduction mechanism, CO₂ is initially activated to the bent CO₂^⋅− bidentate by one electron from a physisorbed CO₂ configuration. The activation barrier is 0.87 eV. This step is followed by a proton-coupled electron mechanism to form HCOO⁻ with a barrier of 0.46 eV. Then HCOO⁻ receives a proton from the breaking of one of the Ti–O bonds to form HCOOH. The barrier for this step is 0.46 eV. Thus, the rate determining step is the activation of CO₂ to CO₂^⋅−. The second mechanism is a concerted proton-coupled electron process of CO₂ photoreduction, in which CO₂ is activated by two electrons and one proton simultaneously. The structure of the transition state (TS) is illustrated in Fig. 5(b), in which CO₂ is bent (∠OCO = 140°) with carbon bonded toward an adsorbed proton by a hydrogen bond of 2.1 Å. The energy barrier along this pathway (0.82 eV) is slightly lower than that of pathway I (0.87 eV) and is therefore more favorable. Interestingly, this mechanism leads to a linear HCOO⁻ adsorbed configuration which has a low proton hydrogenation barrier of 0.01 eV. Thus, it is thermodynamically favorable to CO₂ to reduce to HCOOH. However, there is a low possibility that two electrons will participate in the reaction simultaneously, and therefore this step may be difficult to occur. Accordingly, the relevant micro-kinetic studies including both the kinetics and electron concentration are worth being investigated in the future.

The mechanism of photoreduction for CO₂ conversion to CH₄ was also investigated theoretically on the anatase TiO₂(101) surface by Ji and Luo.³⁶ Similarly to the photoreduction of CO₂ to HCOOH, two reaction pathways to form CH₄ were identified: pathway I proceeds via an intermediate of HCOOH (CO₂ → HCOOH → H₂CO → CH₃OH → CH₄), while pathway II gives rise to the formation of CH₄ by the intermediate of CO (CO₂ → CO → H₂CO → CH₃OH → CH₄). Both pathways prefer the two-electron mechanism where two photogenerated electrons concertedly participate in one elemental step. For pathway I, the initial step is the photoreduction of the CO₂ molecule to form HCOOH as shown in Eq. (2), and this was also reported in the work of He et al.¹⁰ The rate-determining step is the activation of CO₂, in which the linear CO₂ converts to a bent HCOO with a proton and an electron. The barrier is 1.49 eV and it is higher than that in the work of He et al. The different barriers for this step might be due to the different sizes of supercell used in their studies. In the second step, the hydrogenation of HCOO has virtually no barrier, meaning that the effective barrier of HCOOH formation from CO₂ is 1.49 eV (see Fig. 6). Comparatively, the deoxygenation of CO₂ to CO in pathway II appears to be a little easier with an energy barrier of 1.41 eV. Although the energy barrier for the formation of HCOO from CO is higher, 0.88 eV, the effective barrier for this path is lower than pathway I by 0.08 eV. This indicates that HCOOH is more favorable to form from pathway II than CO from pathway I. The subsequent H₂CO formation also has two pathways. The first is the formation of H₂CO from the deoxygenation of HCOOH and another is from the hydrogenation of CO. In the photoreduction of HCOOH to H₂CO, it is more inclined to have a concerted mechanism to form H₂CO directly with two electrons participating simultaneously (HCOOH + 2e⁻ + 2H⁺ → H₂CO + H₂O). The energy barrier was calculated to be 1.84 eV. For the hydrogenation of CO, a one-electron photoreduction is observed, in which a HCO⋅ intermediate forms first and then it couples with a proton as well as an electron to form H₂CO. However, it should be noted that the HCO intermediate is difficult to generate with an energy barrier as high as 2.08 eV. This step is also thermodynamically unfavorable. For the hydrogenation of CO to H₂CO (CO + 2e⁻ + 2H⁺ → H₂CO), the two-electron path is also more favorable than the one-electron mechanism. In the two-electron reduction process, HCO⁻ will form after one proton transfers concertedly with two electrons to CO and then couples with a proton to form H₂CO. The effective reaction barrier is 1.07 eV and is a little lower than that of the one-electron mechanism by 0.12 eV. Considering the HCOOH path, the hydrogenation of CO is highly favorable. Subsequent reactions to form CH₄ proceed with relative ease as shown in Fig. 6. The energy barriers of H₂CO to CH₃OH and CH₃OH to CH₄ are 0.60 and 0.95 eV, respectively. From their study, the rate-determining step for pathway I is the photoreduction of HCOOH to H₂CO, while that of pathway II is the photoreduction of CO₂ to CO. The lower barrier of pathway II compared with that of pathway I indicates that it is more favorable to CO₂ to be reduced to CH₄ via the CO intermediate. The rate-determining step of CO₂ to CO also suggests that a surface oxygen vacancy could be an active site for photocatalytic CO₂ reduction and reduce the energy required for the deoxygenation of CO₂. This possibility will be discussed later.

In aqueous photocatalytic systems, water can act as donors of both protons [Eqs. (2)–(6)] and electrons [Eqs. (7)–(9)]. Taking the TiO₂ system as an example, H₂O dissociates on the surface of TiO₂ through its oxygen atom adsorbing at a surface Ti site, forming terminal and bridging OH groups.^{11,46,68–70} The dissociated OH and H on the surface can interact with the external molecular H₂O network via hydrogen bonding.⁷¹ Possible interactions with the large water environment can have a considerable impact on the mechanism of CO₂ photoreduction.

Dimitrijevic and co-workers addressed the role of water in the overall mechanism of CO₂ reduction on the TiO₂ surface from both experimental work and theoretical calculations.⁴³ They used a 1:1 molar ratio of CO₂/H₂O in the system, and upon illumination, H atoms, ⋅OCH₃ radicals, and ⋅CH₃ radicals were detected. This suggests competitive transfer of electrons to adsorbed protons and adsorbed CO₂ on the surface of TiO₂. Thus, water acts as an electron acceptor for the H atoms formed from the reaction of photogenerated electrons with protons on the surface. In their study, the initial step of CO₂ photoreduction was found to be the photocatalytic reduction of CO₂ to HCOO⁻ and was also verified by the first principles calculation (see Fig. 7). In the presence of two photogenerated electrons and one adsorbed proton next to a CO₂ molecule, CO₂ can be activated and converted to a formate on the TiO₂ surface (CO₂ + 2e⁻ + H⁺ → HCOO⁻). This involves a concerted two-electron, one-proton mechanism. In the transition state (TS, see Fig. 7), the distance between carbon and the adsorbed proton is 2.1 Å. The configuration of CO₂ is bent (∠OCO = 140°) and bonds strongly to the surface five-fold Ti site, with a Ti–O bond length of 2.05 Å. The effective barrier for the formation of formate was calculated to be 0.82 eV. The relatively high activation barrier of 0.82 eV also explains the relatively low efficiency for CO₂ photoreduction on the pure surface of TiO₂ at room temperature.

Furthermore, in their study, they also identified that water induces a better charge separation in this system. It was found that the number of spins increases from 0.8 × 10¹⁵ spins/cm³ for the dry TiO₂ system to 1.0 × 10¹⁵ spins/cm³ for the aqueous TiO₂ system. From Fig. 8(a), the intensities of signals associated with trapped electrons in the aqueous TiO₂ system are evidently higher than those in the dry TiO₂ system. To give a better understanding, the signals of oxidation products normalized for different spectra are presented [see Fig. 8(b)]. As the EPR spectra were recorded under illumination, the intensities of the signals correspond to the steady-state concentrations of products. This is ruled by the steady recombination of electrons and holes and the formation of products from subsequent reactions. In the presence of H₂O, the charges are stabilized due to the dipolar interaction with water and the band bending at the interface of TiO₂/H₂O. This suppresses the recombination of charges.

The mechanism of CO₂ photoreduction at oxygen vacancies on the TiO₂ surface was investigated by Ji and Luo.⁴⁷ As shown in Fig. 9(a), they found that CO₂ is more favorable to dissociate at V_O (0.73 eV) than on the perfect surface (1.19 eV). This is consistent with the previous work conducted by Sorescu et al.¹¹ The subsequent hydrogenation of CO to CH₃OH and CH₄ has been thoroughly studied. The hydrogenation of CO to CH₃OH follows four steps: CO → HCO → CH₂O → CH₃O → CH₃OH [see Fig. 9(c)]. The rate-determining step was found to be the hydrogenation of CH₂O to CH₃O at the V_O site. The energy barrier was calculated to be 1.26 eV. It is lower than the rate-determining step on the perfect surface by 0.15 eV,³⁶ indicating that the V_O site is more active than the Ti site. In addition, the hydrogenation step of CH₃O to CH₃OH at the V_O site is endothermic by 0.65 eV with a barrier of 0.71 eV. The adsorption energy of the formed CH₃OH at the V_O vacancy is 1.33 eV. It indicates that CH₃OH prefers to dissociate to CH₃O by overcoming a small barrier of 0.06 eV. For CH₄ formation [see Figs. 9(b) and 9(c)], the path is as follows: CO → HCO → CH₂O → CH₃O → CH₃⋅→ CH₄. The rate-determining step is also the hydrogenation of CH₂O to CH₃O at the V_O site. However, the subsequent path from CH₃O to CH₄ is exothermic with an effective barrier of 0.71 eV. Therefore, CH₄ is favorable to form at the V_O site. From their study, the oxygen vacancy on the defective surface is more active than the surface Ti site. This can greatly lower the energy barrier of the deoxygenation processes so that the photoreduction of CO₂ can be achieved more easily. The regeneration of oxygen vacancies was also investigated. Experimentally, Fujishima et al. found that O_V can be generated under UV illumination during the generation of O₂.⁷² Theoretically, Li and Selloni reported that O_V can be generated with a low barrier in the presence of holes.⁵⁷ Therefore, the oxygen atom is possible to be removed to regenerate the O vacancy in the TiO₂ system.

Since the photoreduction of CO₂ was first reported on TiO₂ in 1979,²⁵ numerous studies have been focused on the photocatalytic mechanism, activity, and selectivity of CO₂ reduction. However, the efficiency of CO₂ conversion is also largely controlled by the performance of the photocatalysts.^13,73–80 The ability to enhance the photocatalytic efficiency by screening the proper semiconducting photocatalysts is also an important issue.^24,81–86 Semiconductor photocatalysts play two main roles in photocatalytic reactions. The first role is producing the photogenerated electrons and holes to participate in the reduction and oxidation reactions. Another is providing the active sites for reactions to occur. The use of semiconductors as photocatalysts is attractive because of their appropriate band structure. The band structure consists of a conduction band (CB) and a valence band (VB) with a bandgap inside (see Fig. 10).^7,24,87 The VB lies below the gap and is occupied by electrons at the ground state, while the CB lies above the gap and is unoccupied. Upon illumination, an electron is excited from the VB into the CB, thereby leaving an empty state in the VB (this constitutes a hole). Once the charge carriers are separated, they may migrate to the surface of the photocatalyst and the adsorbed molecules, thereby initiating the redox reactions. For the photocatalytic reduction of CO₂, the photogenerated electrons transfer from the photocatalyst to the adsorbed CO₂ to reduce CO₂. Oppositely, the holes participate in the oxidizing reactions, for example, the oxidation of water. This requires the semiconductor to have a higher CB energy level than the CO₂ reduction potential to drive the photogenerated electrons from the CB to the surface CO₂ species. Therefore, an ideal photocatalyst for CO₂ reduction should satisfy the following fundamental requirements: (1) high light absorption efficiency; (2) excellent ability for spatial charge carrier separation; (3) small electron/hole effective masses; (4) narrow bandgap with suitable CB and VB energy levels; and (5) active sites for CO₂ adsorption on the surface. A single semiconductor does not usually satisfy all these requirements. Moreover, because of the fast recombination of photogenerated electrons and holes plus the lack of appropriate reaction sites, single semiconductor photocatalysts usually do not show high efficiency in photocatalytic reactions. This is verified by a higher activity observed on the metal-loaded TiO₂ photocatalysts compared with pure TiO₂ during the photocatalytic reduction of CO₂.⁵ Although the pure TiO₂ shows low activity, it is worth noting that heterojunction structures were identified in the TiO₂ system recently, linking anatase with rutile.⁸⁸ This kind of structure allows the migration of holes and hinders the electron transfer from anatase to rutile. This would improve the separation of charge carriers and enhance the photocatalytic efficiency. Thus, it also deserves to investigate CO₂ photoreduction on this kind of heterojunction structure in the future.

To improve the efficiency of photocatalytic reactions, composite photocatalysts were widely developed by loading proper co-catalysts such as metal or metal oxides on semiconductor photocatalysts. In composite photocatalysts, the roles of co-catalysts include^87,89 (1) providing trapping sites and promoting the separation of the photogenerated charges to enhance the efficiency; (2) modifying the electronic structures of photocatalysts to promote the adsorption of reactants; and (3) catalyzing the reactions by lowering the activation energy. Numerous kinds of co-catalysts have been incorporated into the photocatalysts to improve the activity of photocatalytic CO₂ reduction,^90–94 including metal and metal oxide co-catalysts.

Loading noble metals such as Pt, Au, Ag, and Cu on photocatalysts have been reported and have shown an enhanced activity of CO₂ photoreduction.^{17,26,59,95–97} Umezawa et al. investigated the effect of Pt clusters on the photoreduction of CO₂ to fuel with water on a rutile TiO₂(110) surface based on the reaction CO₂ + H₂O + 2e⁻ → HCOO⁻ + OH⁻.⁹⁶ In their study, the co-adsorption of CO₂/H₂O and HCOO/OH species on the TiO₂(110) surface in the absence and presence of Pt clusters was fully explored. Relative to the physical adsorption of CO₂ and H₂O, the chemical adsorption energy of HCOO and OH is highly promoted when these species adsorb on the TiO₂ surface with Pt clusters in the vicinity. This is most apparent when they absorb on a Pt cluster [see Figs. 11(a) and 11(b)]. It is consistent with the experimental observation which revealed that Pt nanoparticles loaded on the TiO₂ surface can improve the efficiency of CO₂ photoreduction to CH₄.¹³ Figure 11(c) illustrates three models of HCOO and OH adsorption. Model A shows a structure of HCOO and OH adsorbing on a pure TiO₂ surface. HCOO adopts a bidentate adsorption configuration, whilst OH takes up a monodentate configuration. Model B also presents the same adsorption configuration with a Pt cluster neighboring. Finally, Model C illustrates the adsorption of HCOO and OH on the Pt sites. From analyzing the density of states (DOS’s), they could rationalize the change of adsorption of HCOO and OH in the presence of the Pt cluster. They reported that two electrons were transferred from the HOMO of the Pt cluster to the empty states at the top of the valence band of adsorbates. This two-electron transfer helps facilitate the ionization of HCOO⁻ and OH⁻. The strong chemisorption of HCOO⁻ and OH⁻ on a Pt cluster further stabilizes the system. Therefore, in this system, the introduction of a Pt co-catalyst provides active sites for the adsorption of various reactants and intermediates whilst simultaneously acting as a reservoir of electrons to promote the adsorption of adsorbates. Furthermore, Yang et al. investigated the effect of different sizes of Pt clusters from 2D to 3D morphology toward CO₂ photoreduction on the anatase TiO₂ (101) surface.⁴² Compared to the 2D Pt cluster, they found that the 3D Pt cluster is more beneficial to the formation of the bent CO₂⁻ configuration, with electronic charge accumulation at the carbon atom. The differences can be attributed to the following reasons: (1) the 3D Pt cluster provides more accessible Pt–TiO₂ interfacial active sites which are beneficial to the activation of CO₂ and charge transfer; (2) the 3D cluster possesses higher structural relaxation than the 2D cluster, which highly promotes the binding strength of CO₂.

In the presence of a Pt co-catalyst, the addition of MgO oxide onto the TiO₂ surface was also found to highly enhance the efficiency of photoreduction of CO₂ to CH₄ with H₂O [see Fig. 12(a)].⁴⁵ A positive correlation between the activity of CH₄ formation and the amount of adsorbed CO₂ has been observed in the ternary nanocomposites containing TiO₂, Pt, and a basic metal oxide [see Fig. 12(b)]. The interface between TiO₂, Pt, and MgO plays a crucial role in the photocatalytic reaction. The addition of MgO was found to largely enhance the amount of CO₂ adsorption on the catalyst surface. The adsorbed CO₂ molecules are subsequently reduced to CH₄ on the adjacent Pt nanoparticles by the photogenerated electrons enriched from TiO₂.

One of the main problems associated with photocatalytic CO₂ reduction in aqueous solution is the photoreduction of protons to hydrogens (2H⁺ + 2e⁻ → H₂). This can take place simultaneously and consequently competes with CO₂ reduction.^5,98 However, the reduction of CO₂ is much more difficult than the reduction of protons. The reasons are shown in the following: (1) breaking the O=C=O double bonds is very difficult due to the high antibonding orbital of the CO₂ molecule; (2) hydrogen formation from protons is a two-electron transfer process, whereas photocatalytic CO₂ reduction to methanol and methane is six-electron and eight-electron transfer processes, respectively. In aqueous solution, competition between photoreduction of CO₂ and protons exists. Designing and synthesizing photocatalysts with high selectivity for CO₂ photoreduction and low selectivity for proton reduction are crucial. Neatu et al. co-deposited an Au and Cu alloy on TiO₂ films and found that the photocatalytic efficiency and selectivity for CO₂ photoreduction to methane by H₂O increased dramatically,⁵ especially when the ratio of Au and Cu is 1:2. At this ratio, the formation rate of methane is 2.2 ± 0.3 mmol·g⁻¹·h⁻¹. It is substantially larger than that of H₂ formation which was only 0.29 mmol·g⁻¹·h⁻¹. The selectivity of CH₄ formation can be as high as 97%. This is a major increase when comparing to the formation rates of CH₄ and H₂ on the pure TiO₂ surface, which were only 0.05 and 0.048 mmol g⁻¹ h⁻¹, respectively. The results demonstrate that an (Au, Cu) alloy is an extremely efficient material for CO₂ photoreduction to CH₄ with H₂O by effectively inhibiting the competitive reaction of H₂ formation. The high selectivity of photocatalytic production to CH₄ is due to the presence of Cu bonding to CO on the photocatalyst, while the visible light photo-response would be introduced by not only the TiO₂ photocatalyst but also the surface plasmon band of Au.

Co-catalysts can improve the efficiency and selectivity of CO₂ photoreduction. However, their limited ability to harvest visible-light and the rapid charge recombination still constitutes serious obstacles for these photocatalysts to efficiently generate solar fuels. Therefore, a solution to expand the light absorption of harvesters and enhance the separation and transfer of photogenerated charge carriers is very important. Instead of focusing on co-catalysts, some researchers have turned their attention to prepare photocatalysts consisting of different semiconductors, such as TiO₂/ZnO,⁹⁹ CdS/TiO₂,¹⁰⁰ 13 g-C₃N₄/ZnO,¹⁰¹ and Ag₃PO₄/g-C₃N₄.¹⁰² The formed Z-scheme heterojunctions can considerably diminish the recombination of electron-hole pairs and can also utilize visible light more efficiently. As shown in the Z-scheme photocatalytic system (see Fig. 13), the reaction involves two steps of photoexcitation. First, the electrons in the VB of photocatalyst I (PC I) are excited into its CB under sunlight. Similarly, the electrons in the VB of PC II are excited into its CB. The photogenerated electrons in the CB of PC I are then transferred to the VB of photocatalyst II and recombined with holes (PC I). As a result, the photogenerated electrons of PC II are located in its CB and the holes of PC I are left in its VB. Finally, the photogenerated electrons in the CB of PC II proceed with the reduction of CO₂, while the photogenerated holes in the VB of PC I are used to oxidize H₂O. Thus, this kind of Z-scheme photocatalytic system illustrates a strong reducibility through PC II and a strong oxidizability by PC I. The Z-scheme system can be composed by two photocatalysts with narrow bandgaps and this is beneficial to enhance the absorption efficiency of visible light.

He et al. designed an Ag₃PO₄/g-C₃N₄ composite photocatalyst (see Fig. 14) and proposed that the addition of Ag₃PO₄ increased the photo-absorption performance of g-C₃N₄ by using diffuse reflectance spectroscopy.¹⁰² They also found that this kind of Z-scheme structure highly promotes the separation of electron-hole pairs and enhances the efficiency of photocatalytic CO₂ reduction. Under simulated sunlight irradiation, the optimal Ag₃PO₄/g-C₃N₄ photocatalyst can generate a CO₂ conversion rate of 57.5 μmol h⁻¹ g_cat⁻¹, which is 6.1 and 10.4 times higher than those of g-C₃N₄ and P25, respectively. They also found the formation of Ag nanoparticles in the composite and proposed that they act as a charge transmission bridge in this system. The photogenerated electrons in the CB of Ag₃PO₄ migrate to metallic Ag, where they combine with the holes transferred from the VB of g-C₃N₄. This is because the conduction band minimum of Ag₃PO₄ is more negative than the Fermi level of Ag nanoparticles. This kind of charge transmission effectively improves the separation of electron-hole pairs. This enables the trapping of electrons in the CB of g-C₃N₄ and creation of holes in the VB of Ag₃PO₄. Furthermore, the negative E_CB of g-C₃N₄ also results in strong reducibility and facilitates the reduction of CO₂ to hydrocarbons. Based on the g-C₃N₄/ZnO system, Yu et al. also found enhanced photocatalytic CO₂ reduction when the system follows a direct-scheme mechanism.¹⁰¹ This kind of photocatalytic system exhibited a 2.3-fold increase in the photocatalytic activity for CO₂ reduction when compared to the pure g-C₃N₄ system, and the original selectivity of pure g-C₃N₄ to convert CO₂ directly into CH₃OH was maintained. Therefore, the Z-scheme shows an excellent activity for CO₂ photoreduction and highly improves the efficiency of reactions.

This work has summarized the present status of the photocatalytic reduction of CO₂ to fuels in heterogeneous catalysis from both experimental and theoretical perspectives. Although significant achievements have been made toward the conversion of CO₂, research is still far from achieving artificial photosynthesis. This is due to the complex photocatalytic mechanism and low efficiency and selectivity of products. A more comprehensive understanding of the reaction mechanism and modification of photocatalysts is required. Because of the inert character of the CO₂ molecule, the adsorption of CO₂ on the surface of catalysts is a key step to activate CO₂ and determines the subsequent photoreduction mechanism. The advancements in DFT have enabled the adsorption and activation of the CO₂ molecule to be widely investigated at the atomic level. The activation of CO₂ is from a linear physisorbed configuration to a bent chemical adsorbed CO₂^⋅− configuration by coupling with an electron. Surface defects such as oxygen vacancies formed during the preparation and fabrication of photocatalysts can effectively promote the adsorption and activation of CO₂. Photocatalytic CO₂ reduction generally proceeds with aqueous dispersion. Results prove that the solvation effect has considerable influence on CO₂ photoreduction. It even changes the most stable adsorption configuration of CO₂. Therefore, the influence of surface defects and solvation effect is pivotal to reveal and improve the understanding of CO₂ photoreduction.

The photoreduction of CO₂ with H₂O involves as many as eight electron and six proton transfers resulting in the formation of different products such as CO, HCOOH, HCHO, CH₃OH, and CH₄. This makes the mechanism of CO₂ photoreduction extremely complicated. In the gas phase, a concerted two-electron process for CO₂ photoreduction with adsorbed protons was found on the pure TiO₂ surface and the formation of CH₄ was achieved via a CO intermediate. However, in the actual aqueous environment, protons widely dissolve in the solution. It is still unknown how the protons participate in the hydrogenation reaction. Will they adsorb on the surface first and then react with CO₂, or directly attack CO₂? Therefore, an in-depth understanding elucidating the role of protons in aqueous CO₂ conversion is much desired.

Furthermore, designing of proper photocatalysts is another important issue as it determines the activity and selectivity of CO₂ photoreduction. The early research in this field employed suspensions of simple semiconductor materials such as pure TiO₂, which exhibit low efficiency. Except the nanostructuring techniques leading the increasing surface area and active reaction sites, recent developments have also focused on various composite photocatalysts by introducing metal and metal oxide co-catalysts, as well as Z-scheme systems. The use of composite photocatalysts greatly improves the charge separation since the photogenerated electrons and holes can be trapped efficiently and effectively. These kinds of composite photocatalysts also provide active sites for activation and reaction of the adsorbates. For example, in Au-Cu alloys loaded on TiO₂ photocatalysts, the role of Cu is to promote CO₂ adsorption and activation, while Au inclines to promote the hydrogenation reaction and act as a light harvester due to a surface plasmon band in the visible light region. Additionally, Z-scheme photocatalytic systems have shown strong reducibility and oxidizability, whilst their narrow bandgap is extremely beneficial to enhance the adsorption efficiency of visible light. Although various photocatalysts have shown good performance, the photocatalytic reduction of protons to hydrogens, a competitive reaction in aqueous CO₂ photoreduction, remains a major obstacle for the selectivity of the reaction. Therefore, the design of ideal photocatalysts to enhance the photocatalytic reduction of CO₂ and suppress H₂ formation need to be explored further.

This project was supported by the National Key Basic Research Program of China (Grant No. 2013CB933201), the National Natural Science Foundation of China (Grant Nos. 21333003, 21303052, and 21622305), Shanghai Rising-Star Program (No. 14QA1401100), “Chen Guang” Project (No. 13CG24), Young Elite Scientist Sponsorship Program by CAST, and Fundamental Research Funds for the Central Universities.