Mechanisms of carbon dioxide reduction on strontium titanate perovskites © The Royal Society of Chemistry

Strontium titanate (SrTiO 3 ) is a promising material for the light-driven conversion of carbon dioxide (CO 2 ) into renewable fuels. However, the mechanisms of the relevant reactions are not yet well understood. In this work, we have used density functional theory calculations to explore CO 2 reduction on the (001) surface of the SrTiO 3 photocatalyst. Our results indicate that, in contrast to COOH, the formation of a HCOO or CO 2 (cid:1) intermediate is thermodynamically hindered, which is consistent with the fact that formic acid (HCOOH) is not a major product in the experiments reported in the literature. We show that a pathway to carbon monoxide (CO) is instead possible, and that the formation of COOH is the rate-limiting step. Finally, we suggest that substitutional doping of Sr ions represents a promising approach to lower the energy barrier of the COOH formation.


Introduction
To reduce the still increasing CO 2 emissions responsible for global warming and the concomitant climate change remains one of the biggest challenges of our era. The catalytic technologies currently being explored to utilise the CO 2 molecule by converting it into useful hydrocarbons include heterogeneous [1][2][3][4] and electrochemical [5][6][7] catalysis. Approaches such as photocatalysis, 8,9 mimicking the natural process of photosynthesis, are particularly attractive, as they are achieved at room temperature and require only the harnessing of solar light as a source of energy. However, the achievement of photocatalytic conversions suitable to industrial applications remains a difficult challenge, 10 which motivates the continuous search for new and improved photocatalytic materials.
Since their rst use in heterogeneous catalysis in 1952, 11 perovskite oxides have attracted signicant interest in a diverse range of reactions, 12 especially because they are cheaper and easier to prepare than metal-based catalysts such as platinum or palladium. 13 Compared to other widely investigated oxide photocatalysts, e.g. ZrO 2 and Ga 2 O 3 , perovskite oxides have bandgaps with a better overlap with the solar spectrum. In addition, owing to the unique exibility of their structure, perovskite oxides have bands edges which can be tuned to facilitate both the reduction of CO 2 and the oxidation of water during the photocatalytic process. 9 Because of their striking photochemical stabilities, titanate perovskites are particularly promising for CO 2 photocatalysis. 14 Taking inspiration from leaves, Zhou et al. have developed an articial photosynthetic system based on titanate perovskites which generates carbon monoxide and methane from CO 2 and water. 15 These materials have also been investigated in combination with other semiconductors, co-catalysts and dopants, in the effort to improve their conversion efficiencies. For example, Li et al. obtained CO and H 2 by loading SrTiO 3 with noble metals, 16 while CaTiO 3 samples loaded with Ag exhibited stable photocatalytic activities for CO 2 reduction to CO using water as an electron donor. 17 The photoreduction of CO 2 on SrTiO 3based nanocomposite systems has also been demonstrated, with CH 4 as the major product. 18 The generation of oxygen vacancies enhances the adsorption of CO 2 on SrTiO 3 , leading to higher yields of CH 4 , 19 while NaTaO 3 doped with Ba, Ca, and Sr has shown high activity towards CO 2 reduction to CO in water. 20 Detailed atomistic knowledge of the light-triggered catalytic reaction mechanism is crucial to help design better photocatalysts. 21 For example, it allows us to ascertain the limiting reaction steps, featuring the higher activation energy barriers in a catalytic pathway, which elucidates ways of ne-tuning the perovskite structure to lower these activation barriers. However, in contrast to the most investigated anatase TiO 2 photocatalysts, 22 the elementary reaction steps occurring aer the adsorption of CO 2 on titanate perovskites are yet to be claried.
Here, we provide this missing piece of information on the textbook SrTiO 3 perovskite, one of the most commonly investigated members of the titanate perovskite family, which, with a tunable bandgap of 3.2 eV (ref. 23) (identical to that of anatase) and remarkable charge transport properties, 14 holds great potential as a photocatalyst. 19,[24][25][26][27] Using simulations within the density functional theory (DFT), we rst aim to achieve the description of a realistic reaction pathway for the experimentally observed reduction of CO 2 to CO. Aerwards, we take advantage of the fresh insights into the system to put forward a doping approach aimed at improving the efficiency of the photocatalytic reaction.

Models and methods
The most-stable SrTiO 3 (001) surface can have two different competitive terminations, namely SrO and TiO 2 . 28 In this study, we have focussed on the SrO termination only, since its equivalent has been shown generally to be the favoured plane in a wide diversity of perovskites. 29,30 Our (001) surface model, built from an optimised primitive cell, consists of a 2 Â 2 Â 4 supercell, and contains 80 atoms located within 8 layers. Surface slabs were separated by a vacuum region of 15Å along the normal direction to avoid spurious interactions between periodic images. All structures were drawn with the visualization for electronic and structural analysis (VESTA) program suite. 31 The adsorption energy of CO 2 was evaluated according to the formula where E slab/mol is the energy of the supercell with the adsorbed CO 2 , while E slab and E mol are the energy of the same supercell with, respectively, only the surface slab or a CO 2 molecule. In addition to the DFT energy E, in order to take into account the effect of temperature and pressure in the reaction mechanism, we have considered the Gibbs free energy G. For a condensed phase structure, G is dened as where F vib is the vibrational contribution to the free energy, which we have estimated using the frequencies at the G-point.
The pV term can be neglected, as it is of the order of meV under standard working conditions. 32 The Gibbs free energy of a gasphase molecule corresponds to its chemical potential m, which can be derived from 33 where DH and DS are the enthalpy and entropy variations from the reference chemical potential m ref and k B T log(p/p 0 ) includes the effect of pressure on the chemical potential with respect to the standard one. The substitutional energy for the dopant X (X ¼ Ba or Ca) was calculated according to where E(doped) and E(undoped) are the energies of doped and pure SrTiO 3 surfaces, respectively, while E(Sr) and E(X) are the energies of single Sr and X atoms. All DFT calculations were performed with the VASP 5.3 package 35,36 using the Perdew-Burke-Ernzerhof (PBE) functional. 37 We have employed the projector augmented wave method to model the core-electron interaction, 38 treating explicitly the following electrons: Sr 4s 4p 5s; Ti 3d 4s; O 2s 2p; Ba 5s 5p 6s; Ca 3s 3p 4s. All calculations were performed with a plane wave cutoff of 500 eV, which guarantees that the absolute energies of bulk SrTiO 3 are converged within 2 meV per atom. Calculations were stopped when the forces acting on the ions were less than 10 À2 eVÅ À1 . Aer nding that a 6 Â 6 Â 6 Monkhorst-Pack grid 39 ensures that the absolute energies of bulk SrTiO 3 are converged to better than 1 meV per atom, we scaled the grids of the surface calculations inversely with the dimension of the unit cells to 3 Â 3 Â 1. We have also undertaken test calculations to make sure that the atomic relaxations in the surface slab are converged with respect to its thickness. Transition states were located with the climbing image nudged elastic band algorithm. 40 The number of images used varied from one to four, depending on the elementary reaction step. We have calculated vibrational frequencies under the harmonic approximation and with Hessian matrix elements estimated from nite differences of analytical gradients with displacements of 0.015Å in length. Only the degrees of freedom of the adsorbate species and the atoms in the topmost atomic layer of the surface have been included in the vibrational analysis.

Adsorption of CO 2
A sufficiently large number of different adsorption congurations for CO 2 on the SrO-terminated SrTiO 3 (001) surface have been probed to ensure that realistic minimum energy structures would be obtained. Following optimisation, all congurations relaxed towards the structure depicted in Fig. 1 41 Both the geometry and the energetics, which are summarised in Table 1, are comparable to those found in a previous DFT investigation by Sopiha et al.,42 who also showed that the adsorption geometry of CO 2 slightly changes, while its interaction becomes stronger, when the size of the surface unit cell is increased from a 2 Â 2 to a 3 Â 3 supercell. As the structure in Fig. 1 is the only stable on the 2 Â 2 unit cell which we adopted, we assume that any reaction pathway must include it as a reactant conguration. However, we stress that other starting congurations, potentially leading to different reduction mechanisms, may also be possible at lower CO 2 coverages. It is instructive, at this point, to compare the adsorption energy of this structure with that of a water molecule. Regardless of the initial conguration and in line with the literature, 43 we found that water spontaneously dissociates on the SrOterminated (001) surface of SrTiO 3 . The dissociated water molecule is adsorbed with an energy of À1.26 eV, which is considerably less negative than the À1.82 eV value found for CO 2 . As a result, we expect SrTiO 3 to be preferentially covered by CO 2 rather than by water, which in turn suggests that the CO 2 reduction process should be more favourable than the competing hydrogen evolution reaction. 10 Future detailed studies can now build on these ndings to conrm the selectivity of SrTiO 3 towards the reduction of CO 2 , rather than water.

Thermodynamics of intermediates
The photocatalytic conversion of CO 2 takes place in the presence of a photoexcited electron which reduces the SrTiO 3 surface. When the carbonate-like structure described above is adsorbed on the reduced surface, which we have modelled by adsorbing a H atom on SrTiO 3 as in previous works, 22,44 a hydrogen bond between a carbonyl oxygen and the proton on the surface is formed, which is illustrated in Fig. 2a. However, the geometry changes with respect to the neutral surface, e.g. the elongation of the two C-O bonds, are only marginal, as can be seen from the parameters reported in Table 2.
TiO 2 is arguably the prototype photocatalyst, 45 and many efforts have been devoted to the elucidation of the molecular mechanisms yielding fuels from CO 2 using TiO 2 polymorphs. He et al. concluded that the formation of an activated CO 2 À radical adsorbed in a bridging conguration is the rst step towards hydrogenation and the formation of formic acid on anatase. 22 In search of a possible similar mechanism, we have relaxed such a structure. While not stable on a neutral surface, interestingly, in the presence of a reducing electron, this bent geometry converged to one with a O 1 -C-O 2 angle of 140.4 , see Fig. 2b and Table 2, which highlights the importance of photogenerated electrons on the activation of the molecule. However, its energy resulted to be 1.96 eV higher than that of the carbonate-like structure in Fig. 2a, which makes it an unlikely candidate as a possible intermediate, while indicating that catalytic mechanisms different than those of TiO 2 are at play in SrTiO 3 .    Fig. 2c. However, we found the energy of the resulting conguration to be 2.55 eV higher than that of the original structure, which clearly excludes its participation in any reaction mechanism.
Finally, we relaxed the structure in Fig. 2d, where we simulated the H transfer from the surface site of structure of Fig. 2a to the closest CO 2 oxygen. The formation of a COOH group changes the adsorption geometry to that of a bridging bidentate, having an energy 1.15 eV higher than the initial structure, taken as reference, as shown in Table 2.
On the basis of thermodynamic (energetic) considerations, we can conclude that only the COOH intermediate can participate in the reduction of CO 2 , while pathways involving either CO 2 À and/or HCOO should be discarded. However, we cannot rule out the existence of other, in principle less favoured, pathways where two or more electrons and/or proton are transferred. 22 The high energy penalty for hydrogenation through the carbon atom explains why HCOOH is not detected in CO 2 photochemical reactions on SrTiO 3 . Instead, the COOH species that can be reached energetically points to a pathway towards CO, 10 in agreement with experimental ndings on titanate perovskites. 15,16

Reaction pathway to CO
With the information gained in the previous section, we have derived a realistic pathway for the reduction of CO 2 to CO, which we present in the top panel of Fig. 3, whereas reaction geometries are shown in Fig. 4. The formation of the COOH intermediate proceeds via TS1 and has an activation energy barrier for the hydrogen transfer of 1.22 eV. Only 0.75 eV is required to reach TS2 (where the C-O distance is 1.56Å), break the C-O bond and form the CO and OH species at the surface. Note, however, that the barrier for CO and OH recombination is lower, which may indicate that, from a kinetic point of view, a substantial part of reactions will go backwards, with forward direction favoured by fast CO and OH diffusions. The subsequent release of CO is endothermic by merely 0.12 eV, and the process leaves a hydrogen-bonded H + and OH À couple. In the last step of the reaction mechanism, as a result of a proton transfer in TS3, a H 2 O molecule forms. This water molecule interacts with the surface by means of two hydrogen bonds, and its desorption raises the energy by 0.86 eV. The entire reaction cycle, CO 2 + 2H + + 2e À / CO + H 2 O, is endothermic by 0.73 eV. Obviously, the energy prole provides a preliminary picture of the overall mechanism, allowing to discard the HCOO intermediate. However, to unravel the amount of CO formed and the effect of H 2 O requires further analysis that is beyond the scope of the present work. Note also, that the models used represent the ideal (001) surface, whereas the real material may be more complex. In any case, the present results provide some clear and unbiased hints that explain experimental observations. In the bottom panel of Fig. 3, we plot the Gibbs free energy of the pathway above, taking into account temperature (300 K) and pressure (1 bar) effects. The prole attens compared to that in the top panel of Fig. 3, although the endothermicity of the reaction is not much affected. The attening can be explained by the larger entropy contributions carried by the gaseous species compared to their adsorbed counterparts. For example, the gain due to the adsorption of CO 2 increases drastically, from À2.11 to À1.40 eV, when the Gibbs free energy, rather than the DFT energy, is considered. However, the rate limiting step remains the formation of the COOH group, whose activation energy barrier remains virtually unchanged at 1.23 eV, which is in line with the low photoreduction rates typical of titanate perovskites. 46 Note, however, that once pressure and temperature effects are taken into account, the thermodynamics will tend to drive the reaction backwards since now some transition states are located above the desorption limit. This is an issue deserving further attention through macroscopic simulations including all elementary steps and searching for the optimum operating conditions as well.

Substitutional doping
Having established that the formation of the COOH intermediate (more favoured than its desorption from the surface by around 0.1 eV, as shown in Fig. 3), is the rate-limiting step in the production of CO on the pure SrTiO 3 (001) surface, we have carried out further calculations to explore how the activation energy of this elementary reaction step is affected by surface doping. Similar to the study by Carlotto et al., 47 we have only focussed on the top surface layer. In particular, we have considered the substitution of one of the four Sr ions by a Ba or Ca ion. We have chosen these dopants because their ionic radii in the 2+ charge state, at 149 and 114 pm, respectively, are close to the 132 pm ionic radius of Sr, 48 which implies that they should be accommodated easily within the SrTiO 3 material, both geometrically and in terms of charge. Moreover, BaTiO 3 and CaTiO 3 are the most common titanate perovskites aer SrTiO 3 , which should ensure the successful replacement of Sr by Ba and Ca ions during the synthesis of the photocatalyst. The stability of the doped materials is suggested by the negative substitutional energies of both dopants, i.e. À0.26 and À0.09 eV, respectively, for Ba and Ca. Fig. 5 shows the activation energy barriers of the elementary steps and the difference in Gibbs free energy between the two intermediates. While Ba dopants change the thermodynamics of the reaction step without altering its kinetics, the substitution of Sr by Ca ions has the opposite effect, with the activation barrier being lowered from 1.23 to 1.15 eV. Although the effect is small, it clearly indicates that cation doping is a conceivable way to accelerate the reaction rate for CO production over the SrTiO 3 photocatalyst, as shown here by Ca.

Conclusions
We have presented a theoretical investigation of the adsorption and reaction of CO 2 on the SrO-terminated (001) surface of SrTiO 3 , a perovskite compound of signicant technological interest. We have shown that the possible intermediates towards the formation of HCOOH are very high in energy, which explains why the formation of HCOOH is not normally reported in photocatalytic experiments. Instead, the reduction of CO 2 to CO appears to be feasible, in agreement with experimental observations, thus pointing towards to the use of SrTiO 3 as a photocatalyst for the reduction of CO 2 . The reaction occurs via the formation of a COOH species (the associative mechanism in the reverse water gas shi reaction 3 ), with an activation energy of about 1.2 eV, corresponding to the rate-limiting step of the overall reaction.
However, a more rm statement will require a systematic analysis of adsorption/desorption rates, the rates for diffusion of the various species and their comparison to reaction rates for all the steps in the mechanism. While such simulations are le for subsequent work, the present results provide compelling evidence that the barrier for the likely rate determining step is sensitive to the nature of the surface cations, and therefore can be lowered by means of a doping strategy, for example where smaller Ca ions substitute for Sr ions. Future experimental studies along with macroscopic simulations will be important to validate these predictions.