Mechanism and kinetics for methanol synthesis from CO₂/H₂ over Cu and Cu/oxide surfaces: Recent investigations by first-principles-based simulation

Abstract

The efficient fixation and utilization of CO₂ has been consistently pursued by chemists for decades. Although Cu-based catalysts, e.g., Cu/ZnO/Al₂O₃, have been widely used in industry for methanol synthesis from CO₂ hydrogenation (CO₂ + 3H₂→H₃COH+ H₂O), many issues on the mechanism and the kinetics remain largely uncertain. For example, the surface site for CO₂ activation and the synergetic effect between Cu and oxide have been hotly debated in literature. In the past few years, theoretical modeling on pure Cu surfaces and Cu/oxide interfaces has been utilized to provide insight into these important questions. Here we will review the recent theoretical advances on simulating this complex heterogeneous catalytic process with first principles density functional theory (DFT) calculations and kinetics modeling. The theoretical results on the mechanism and the kinetics are compared and summarized.

Keywords

density functional theory calculation / CO₂ hydrogenation / methanol synthesis / Cu-based catalysts / review

Introduction

The chemical conversion of CO₂ is of great interest nowadays because it not only removes greenhouse gas but also yields useful fuels and chemicals such as methanol. As one of the most promising route, methanol production by the hydrogenation of CO₂ has been extensively studied in experiment and it was found that this reaction can be catalyzed by metal and metal oxide composites such as Cu supported on ZrO₂, ZnO, etc. Cu/ZnO is a major type catalyst widely used in industry [1-3], which is often further modified by other addictives such as Al₂O₃ [4-6], ZrO₂ [4,7], Ga₂O₃, Cr₂O₃ [4], and metals such as Pd [5]. ZrO₂-supported Cu catalyst also attracted some attention for its good thermal stability as well as the high activity and selectivity [8-12]. Experimental studies on CO₂ hydrogenation over Cu model catalyst surfaces can be dated back to 1980s, in which the active surface structure and the reaction intermediates have been probed via various surface science techniques. For example, Campbell and coworkers demonstrated that pure Cu can already act as a catalyst for methanol synthesis and found that CO is the major product at the low conversion limit [13,14]. One of the major challenges for CO₂ hydrogenation is to control the selectivity to avoid the production of unwanted CO, which is believed to be formed through the reverse water gas shift (RWGS, CO₂ + H₂→CO+ H₂O). A possible solution, for example, is to further convert methanol to methanol formate (HCOOCH₃) by shifting the reaction equilibrium toward methanol related product, as demonstrated recently for CO₂/H₂ over Cu/ZnO catalyst in the solvent of methanol [6,7].

Owing to the fact that Cu itself is capable to catalyze methanol synthesis, the catalytic role of the metal oxide becomes largely unclear. Burch et al. has revealed a paradoxical phenomenon that by adding ZnO (to Cu), the specific activity is shown to be improved by one order of magnitude while the activity can be correlated well with the copper surface area rather than ZnO content [15]. It was also found that the selectivity to methanol varies significantly from catalyst to catalyst and is also sensitive to reaction condition, e.g., from<1% [3,13] to>99% [4]. To understand these puzzles, experimental studies on model catalysts using various techniques, especially in situ infrared spectrum [16], have been carried out to shed light into the microscopic mechanism of the reaction. It was shown that HCOO and H₃CO are present as the common intermediates on different catalysts, implying that the addition of oxides, however, may not change markedly the catalytic mechanism. On the other hand, theoretical simulations, in particular those based on first principles calculations, have demonstrated their ability in providing deep insight into the microscopic mechanism and the kinetics of heterogeneous catalytic process including those at the solid-gas interface and solid-liquid interface [17-22]. Recent years have witnessed the great progress in theoretical simulations to resolve the atomic level picture on the reaction mechanism and kinetics of methanol synthesis from CO₂ hydrogenation [3,23-32].

In this review, we aim to provide an overview of the current understandings on the mechanism and the kinetics of methanol synthesis from CO₂/H₂ based on recent DFT-based theoretical simulations. The theoretical models and computational tools involved will be introduced. The theoretical results obtained on different model catalysts will then be compared and summarized. We intend not only to highlight the consensus achieved, but also to discuss the remaining puzzles on some of the most important issues, including the active site for the activation of CO₂, and the synergetic effect between Cu and oxides.

Theoretical methods for modeling heterogeneous reactions

Of the many electronic structure techniques that have been developed, DFT based on Kohn-Sham equation is the most popular and robust theoretical approach currently available for solving the electronic structures of solid surfaces [33-36]. Pseudopotentials, such as Ultrasoft Vanderbilt pseudopotential [37] and Troullier–Martins norm-conserving pseudopotential [38] were widely utilized to describe valence electrons in solving Kohn-Sham equation. For solid surfaces, the exchange-correlation functionals utilized were generally at the level of local density approximation with the generalized gradient correction (GGA), such as PW91 [39] and Perdew–Burke–Ernzerf (PBE) [40]. A variety of basis sets can be chosen in DFT calculations, and in solid-state surface calculations the typical basis set is plane wave. Recent studies have suggested that carefully constructed numerical atomic orbitals can also be utilized to describe the surface properties of metals at the level of accuracy comparable to the plane-wave calculations [41]. Because of the huge saving in computational power in computing the electronic structure of large systems, the computational methods based on numerical atomic orbitals is particularly useful for simulating large metal/oxide composite systems as encountered in CO₂ hydrogenation.

The mainstream model to model surfaces is the periodic slab model. Conventional techniques utilized in the bulk calculations of solid, such as the Brillouin-zone k-point sampling techniques (e.g., Monkhorst–Pack scheme [42]), can be seamlessly transformed to study the surface properties with the periodic slab approach. For example, it was reported that Cu(111) [3,27] and m-ZrO₂(\overline{\mathrm{2}}12) [23,28] surfaces are modeled by three-layer slabs with a vacuum thickness more than 10 Å, where a part of [3,23,28] or all [27] the substrate atoms are fixed in their bulk truncated positions. A large unit cell is often essential in the modeling of reactions on surfaces in order to minimize the interaction between adsorbates across repeated unit cells. This is also required in simulating the interface between different materials where a good lattice match is needed.

An efficient way to locate the transition state (TS) of reactions is highly desirable in computational simulation since the reactivity depends critically on the barrier height (E_a), the energy difference between the TS and the initial state (IS). Due to the lack of analytic Hessian in calculations with non-Gaussian basis sets, the traditional TS-searching methods by eigenvalue following is not a practical approach in periodic slab calculations. Instead, the first-derivative-based methods are widely utilized for locating the TS of surface reactions, which are in general highly demanding in computational power. The typical methods utilized include nudged-elastic-band (NEB) method [43], the bond constrained methods [44], dimer-based method [45,46], and synchronous transit methods [47]. Rcently, Liu and coworkers developed the Constrained–Broyden–Minimization method [44] and Constrained–Broyden-dimer methods [46], which are found to be efficient for finding the TS of complex reactions on surface.

First-principles based kinetics simulations, such as microkinetic method and kinetic Monte Carlo (kMC) method, have been utilized in recent years to predict the reaction rates and the selectivity of complex surface reactions. These kinetics simulations can deal with complex systems on time scales of the order of seconds or even longer, which cannot be achieved by typical molecular dynamics simulations. In reality, it is the free energy of reactions that determines the reaction kinetics, and thus the calculated DFT total energy must be amended by the necessary zero-point correction and entropy terms to derive the free energy, which can either be calculated from first principles or obtained from standard thermodynamics data in experiment. The use of a mean-field microkinetic model requires that the adsorbates on the surface be well-mixed and diffusion limitations be negligible. The situations should therefore be taken with caution when the reaction occurs only on a minority of surface sites or the diffusion barrier is high for some key species. On the other hand, kMC simulations are capable of accounting for the correct local structure and coverage [48], and are yet much more costly if performed properly. Recently, kMC simulation based on the Bortz–Kalos–Lebowitz algorithm [49,50] was utilized by Liu and coworkers for the study of the CO₂ hydrogenation at the Cu/ZrO₂ interface [28]. It should be emphasized that the lateral adsorbate-adsorbate interactions may strongly influence the stability of intermediates, and thus the coverage-dependence of the kinetics parameters should be considered properly in kinetics simulation. The intrinsic inaccuracy of the current DFT functionals in dealing with the adsorption/desorption energetics should also be paid attention in kinetics simulations [23], although the energy difference between states on surfaces are believed to be more accurate.

Surface models and general consensus

To reveal the mechanism and kinetics of methanol synthesis from CO₂ hydrogenations, various theoretical models have been utilized, ranging from small Cu clusters to extended Cu surfaces and to Cu/ZrO₂ interfaces, as illustrated in Fig. 1. Before we elaborate on more detailed result of each surface model, it is worth outlining first the general consensus achieved from these theoretical investigations.

It was generally accepted that in CO₂ hydrogenation the adsorbed atomic H are present, which comes from the dissociation of H₂. This is because H₂ can readily dissociate to H atoms on Cu surfaces at reaction temperatures (e.g., 500 K). The dissociative adsorption energy is around 0.3 eV per H₂ molecule [3,27] from various DFT calculations. On the other hand, whether CO₂ can adsorb strongly on the catalyst surface is a major question. It is found that on the neutral Cu models including Cu nanoparticles, CO₂ can only weakly physisorb with very low binding energy (<0.1 eV [3,27]). The hydrogenation of CO₂ on pure Cu surfaces is therefore most likely via the Eley-Rideal (ER) mechanism, where the gaseous CO₂ is directly hydrogenated by the adsorbed H atom. In the presence of oxide supports, as CO₂ can adsorb at the Cu/oxide interface (e.g., 0.69 eV on ZrO₂ [23]), it becomes possible that CO₂ hydrogenation occur via the Langmuir–Hinshelwood (LH) mechanism. The alternative pathway via CO₂ decomposition can also not be ruled out, especially in Cu/oxide composite systems.

In Table 1, we summarize the reported key surface species of CO₂ hydrogenation, together with the rate determining steps and their barriers, E_a. From Table 1, it can be seen that three reaction intermediates are shared in different studies, namely HCOO, H₂CO and H₃CO, in consistent with the experimental findings. For the transformation of HCOO to H₂CO, two likely pathways have been reported, either via HCOOH and H₂COOH, or via H₂COO. Although the reported rate determining step differs sharply between different studies, it is in general that CO₂ hydrogenation is highly activated with the calculated reaction barriers being larger than 1 eV.

Mechanism from different models

Cationic Cu

In the 1990s, Kakumoto et al. [24,25] carried out ab initio calculations to study the stability of the reaction intermediates in methanol synthesis by CO₂ over small Cu, CuO and CuZnO clusters [24]. While the models are too simplified compared to real catalyst surfaces, some useful insights have been provided. They found that CO₂ and the following intermediates can adsorb on Cu⁺ sites, while H₂ adsorbs on both metallic copper and ZnO to form H atoms and H⁺/H^-, respectively. The important intermediates for the methanol synthesis were bridging-formate, formaldehyde, and methoxy. The reaction route can be summarized as: CO₂→CO₂*→HCOO^-*→H₂CO*→H₃CO^-*→H₃COH (hereafter * denotes an adsorbed state). From thermodynamics, the most endothermic step is the formaldehyde hydrogenation to methoxy ( + 0.46eV), and this route appears to proceed facilely. It was noticed that the stability of some adsorbates and the reaction barrier of key reactions can be changed by changing the electronic state of the model clusters [25]. For example, the stability of formate and formaldehyde intermediates increases as the total charge increases, while the stability of methoxy species was almost constant for the total charge from 0 to+ 0.4. The total atomic charge distributions on Cu/CuO and Cu/ZnO clusters showed that the cationic copper locate only on the boundary of Cu and the oxide. It is expected that the increase of the interfacial Cu by finely dispersing Cu on oxides can help to increase the number of active sites.

Cu single crystal surfaces

In 1999, Nakatsuji and coworkers [26] studied the mechanism of the hydrogenation of CO₂ to methanol on a Cu(100) surface using the dipped adcluster model (DAM) combined with ab initio Hartree–Fock (HF) and second-order Møller–Plesset (MP2) calculations. A Cu₈(6,2) cluster, which contains six copper atoms in the first layer and two copper atoms in the second layer was used to model the Cu(100) surface (Fig. 1a). The dipped adcluster model [45] was used to include the effects of the bulk metal, such as electron transfer between the admolecule and the surface, and the image force. Based on this Cu cluster model, they evaluated the possible reaction pathway with the LH mechanism starting from the coadsorption of H₂ and CO₂ state. In the DAM of Cu(100), the chemisorbed CO₂ exists as a bent anionic \mathrm{C}{\mathrm{O}}_{\mathrm{2}}^{-} species, with both O atoms bonding with surface copper. About one electron is transferred from the metal surface to the π* orbital of CO₂, making the CO₂ very reactive. (This is however not confirmed later by DFT-slab calculations). Five successive hydrogenations occur later in the hydrogenation of adsorbed CO₂ to methanol involving formate, dioxomethylene, formaldehyde and methoxy: CO₂*→HCOO*→H₂COO*→H₂CO*→H₃CO*→H₃COH. Among the hydrogenation steps, the hydrogenation of adsorbed formate to adsorbed dioxomethylene is the highest barrier elementary step (HCOO* + H*→H₂COO*; E_a = 1.00 eV; ΔE = 0.74eV). Another high barrier reaction calculated is the hydrogenation of adsorbed dioxomethylene to yield formaldehyde (H₂COO* + H*→H₂CO* + OH*; E_a = 0.74eV). The authors suggest that the high energy barrier associated with the unstable dioxomethylene intermediate may account for the lower activity of a clean copper surface compared to the copper-based catalyst in practical methanol synthesis.

Grabow and Mavrikakis [27] recently reported the mechanism of methanol synthesis through CO₂ and CO hydrogenation on Cu(111) using DFT and microkinetics modeling. Some unusual reaction intermediates such as formic acid (HCOOH) and hydroxyl-methoxy (CH₃O₂) were taken into account in the reaction pathway, which allows for the formation of formic acid (HCOOH), formaldehyde (CH₂O), and methyl formate (HCOOCH₃) as by-products. The mean-field microkinetics model was utilized to predict the reaction kinetics, where all kinetic parameters were initially taken from DFT calculations and subsequently fitted to the experimental rate data. It is noticed that the experimental kinetic data are collected in a spinning basket reactor at pressures 15-50 atm and temperatures 483-547 K over a Cu/ZnO/Al₂O₃ catalyst with various H₂/CO₂/CO feed compositions, which is not exactly Cu(111) surface. Theoretically, a CSTR model was employed to simulate the spinning basket reactor; the number of Cu sites used is 300 μmol(sites)/g(catalyst). Because the DFT data are from (111) surface while the experiment was performed on metal/oxide composite catalyst, the authors therefore adjusted the DFT-based kinetic parameters such as the enthalpy of species and the proximity factors of the TS in order to reproduce the experimental kinetic data. By comparing the DFT data with the fitted data, they found that generally the species are better stabilized on the catalyst surface than on Cu(111) from DFT. The deviation is rather large, being 0.3-0.6 eV for most intermediates such as OH*, COOH*, HCO*, HCOO*, HCOOH*, CH₃O₂*, H₂CO*, H₃CO*, and H₃COH*, except for H* and CO* where the deviation is ~0.1 eV.

Both CO and CO₂ hydrogenation pathways are found to be active for methanol synthesis under typical methanol synthesis conditions. The CO₂ pathway for methanol synthesis starts from the formation of formate (HCOO*) from CO₂* and H* on Cu(111) surface, and the following intermediates include HCOOH*, CH₃O₂*, CH₂O*, and CH₃O*. Besides the CO₂ pathway, the hydrogenation of CO can also yield methanol through the pathway: CO* → HCO* → CH₂O* → H₃CO* → H₃COH*. The potential energy surface of the major CO₂ hydrogenation pathway is given in Fig. 2. This reaction mechanism differs from the one on Cu(100) by Nakatsuji and coworkers [26] in three aspects. First, the obvious electron transfer that assists the chemsorption of CO₂ [26] is not observed in the DFT-slab calculation. CO₂ only physisorb on Cu(111) surface with a weak binding energy of 0.08 eV. Second, the hydrogenation of HCOO* leads to HCOOH* instead of dioxymethylene (H₂CO₂*). Third, the hydroxyl-methoxy (H₂COOH, CH₃O₂) intermediate is present in two elementary reactions, namely, HCOOH* + H*→CH₃O₂* and CH₃O₂*→CH₂O* + OH*. The CH₃O₂ is not studied in the work by Nakatsuji and coworkers [26].

The reaction kinetics such as the rate-limiting steps and the contributions of each pathway is analyzed based on the microkinetic simulation. The hydrogenation of H₃CO* is slow, the rate of which qualitatively describes the behavior of overall methanol synthesis rates for a large range of conditions and CO₂-rich feed compositions. However, under some conditions, especially for CO-rich feeds, the formation of CH₃O* (CH₂O* + H*→CH₃O*) can be rate limiting, resulting in a volcano shaped curve for methanol production as a function of the CO₂/(CO+ CO₂) feed ratio. The relative contributions of CO and CO₂ hydrogenation pathways are determined by their corresponding slow steps, namely, HCO* + H* → CH₂O* + * in the CO route and HCOOH* + {\mathrm{H}}^{*}\to \mathrm{C}{\mathrm{H}}_3{\mathrm{O}}_2^{*}+* in the CO₂ route, and also by the feed composition and reaction conditions. Under typical industrial conditions (t = 499.3 K, P = 29.9 atm, molar fractions of CO, CO₂ and H₂ in the feeding gas are 0.053, 0.047, 0.90, respectively), it is found that 2/3 of methanol is produced from CO₂ hydrogenation.

By comparing the binding energies calculated with DFT on the Cu(111) surface and the parameters obtained from fitting, Grabow and Mavrikakis suggested that the merely clean Cu(111) surface cannot provide an accurate representation of the active site on a commercial Cu/ZnO/Al₂O₃ catalyst. Instead, more open and partially oxidized Cu facet, such as Cu(110), Cu(100), Cu(211), might be a more suitable representation of the active site for methanol synthesis. Although it seems likely that only the Cu particles are responsible for the catalytic activity of the real catalyst, the synergistic effects involving the oxide support such as ZnO cannot be excluded. The fitted kinetics model also depends heavily on the experimental data, which are taken mainly from the CO₂-rich conditions. The discrepancies also exist between the theoretical predictions and the data from CO-rich feed compositions in experiment. It might be caused by an adsorbate (CO)-induced surface reconstruction during a shift from oxidizing (CO₂-rich) to reducing (CO-rich) conditions, which, in turn, could affect the dominant reaction mechanism. For more quantitative conclusions about the structure sensitivity of the reaction and the oxidation state of the surface, further studies are needed.

Cu nanoparticles

Yang et al. [3] utilized a Cu₂₉ nanoparticle as the model catalyst to investigate mechanol synthesis from CO₂ hydrogenation. The Cu₂₉ nanoparticle has a pyramidal shape exposing only (100) and (111) facets with a diameter of 1.2 nm (Fig. 1b). Experimentally, they also showed that the clean O-terminated ZnO(000\overline{\mathrm{1}}) surface was inactive as a catalyst [11], and therefore it is important to examine the activity of pure Cu nanoparticles, in which the low-coordinated sites and the shape flexibility of the nanoparticle may play key roles in the catalytic process.

It was found that H₂ can dissociate on Cu₂₉ spontaneously as H–H bond breaks during the geometry optimization. After H₂ dissociation, the atomic hydrogen binds to the top sites of Cu₂₉ with a binding energy of -0.34 eV. In the presence of adsorbed H atoms, CO₂ molecule cannot bind to the nanoparticle. This is consistent with the experimental observation that the increase of adsorbed H coverage hinders only the dissociative adsorption of H₂ on Cu, but not the adsorption of CO₂ [52].

The CO₂ activation is initiated via an Eley-Rideal (ER) mechanism, in which the gaseous CO₂ directly reacts with an adsorbed H atom to form HCOO. The formate pathway occurs via intermediates H₂COO, H₂CO and H₃CO, and the rate-limiting step is the H-assisted H₂COO dissociation, i.e., H₂COO+ H→H₂CO+ OH (E_a = 1.41 eV, ΔE = 0.45 eV). The results on Cu₂₉ nanoparticle are similar to those provided by Mavrikakis et al. [20] on Cu(111) surface except that (i) H₂COO is thermodynamically preferred to HCOOH as an intermediate by 0.82 eV on the nanoparticle, and (ii) H₂COOH may not be present on the nanoparticle since the reaction barrier to form H₂COOH is 1.69 eV, higher than that of H-assisted dissociation of H₂COO (1.41 eV).

Yang et al. [3] also considered the competing RWGS reaction on the Cu₂₉ nanoparticle. In this mechanism, CO₂ is first hydrogenated to HOCO, which splits later to CO and OH. The OH is finally converted to water. The barrier of the rate limiting step of the RWGS (H* + OH*→H₂O*) is 0.27 eV lower than that of methanol synthesis (1.14 eV vs. 1.41 eV). This suggests that at t = 573 K RWGS reaction is 2-3 orders of magnitude faster, and that the dominant product on Cu catalyst is CO rather than methanol. The calculations also indicate that CO produced by the fast RWGS reaction does not undergo subsequent hydrogenation to methanol, but instead, simply accumulates as a product. The possible methanol production from CO hydrogenation may be via the intermediates HCO, H₂CO, H₃CO, but this pathway is hindered by the hydrogenation of CO to HCO. HCO is not stable and tends to dissociate back to CO and H atoms on Cu with a barrier of only 0.03 eV. Therefore, Yang et al. [3] suggested that methanol selectivity over Cu-based catalysts can be improved by adding dopants/promoters that are able to stabilize HCO species or facilitate the hydrogenation of HCOO and H₂COO.

They also performed calculations of the same reaction pathways on Cu(111) surface [3], and found the Cu nanoparticle model shows a higher activity (see Fig. 3). Compared with that on Cu(111), the barrier for the rate-limiting step of methanol synthesis reaction on Cu₂₉ is 0.19 eV lower (1.41 eV compared with 1.60 eV), which corresponds to a reaction rate of ~50 times faster (t = 573 K). They suggested that the higher activity of the nanoparticle is associated with its large stabilization effect on the key intermediates, e.g., formate and dioxomethylene. The theoretical results illustrate that the particle size of Cu catalyst could be important in methanol synthesis.

Cu/ZrO₂ interface

In the series works of Liu and coworkers [23,28,29], the catalytic kinetics of CO₂ fixation to methanol over a binary catalyst Cu/ZrO₂ have been investigated by first principles kinetic Monte Carlo simulation. Instead of the clean Cu models described above, they established a Cu/ZrO₂ interface model and explored the reaction network of CO₂ hydrogenation. In establishing the binary catalyst model, they first screened the possible active surface facetsof ZrO₂ support. It was reported that metal cationic monomers and metal particles anchor preferentially at oxide defected sites. The binding energies of a Cu adatom at ZrO₂(\overline{\mathrm{1}}11) and ZrO₂(\overline{\mathrm{2}}12) are 0.99 and 1.08 eV, less than one third of the calculated cohesive energy of bulk Cu, 3.55 eV. The growth of metallic Cu particles over the oxide surfaces is thus preferred thermodynamically. Furthermore, the DFT calculations for CO₂ adsorption on bare ZrO₂ also showed that CO₂ adsorbs more strongly at the stepped ZrO₂(\overline{\mathrm{2}}12) than at the flat ZrO₂(\overline{\mathrm{1}}11) (1.58 eV compared with 0.83 eV), which implies that the defected ZrO₂ is chemically more active in activating CO₂. To represent the atomistic structure of the Cu/ZrO₂ interface, a two-layer close-packed Cu strip was added onto ZrO₂(\overline{\mathrm{2}}12) stepped surface because the stepped oxide can provide a better geometry for both the adsorption of the Cu strip and the reactant CO₂. This interface structure exposes only the most stable facets of the two components, i.e., the (\overline{\mathrm{1}}11) terrace of ZrO₂ and (111)-like Cu surface. Two kinds of Zr, the five-coordinated (Zr5c) and six-coordinated (Zr6c) lattice Zr are exposed, which are separated by oxygen atoms at the interface (Fig. 1d).

The adsorption of reactants on the interface was reported. CO₂ adsorbs preferentially on the ZrO₂ sites of the Cu/ZrO₂ interface [21]. The calculated Ead of CO₂ is 0.69 eV. The best configuration of CO₂ adsorption features a tri-dented anchoring geometry: its two O and C atoms bond with the Zr_6c, Zr_5c and O_2c, respectively. The strong chemisorption of CO₂ is not observed in pure Cu models, which is an important advantage of the Cu/oxide interface as the fixation and activation of CO₂ can now follow the LH mechanism. H atoms adsorb strongly only on the Cu sites. The H atom adsorption energy at the Cu strip is 2.64 eV with respect to the gas phase H atom, which is much stronger than the H adatoms at the ZrO₂ side (1.76 eV) by forming OH with interface O_2c. Kinetically, H₂ dissociation on Cu only requires a reaction barrier of 0.38 eV on the Cu strip, which is consistent with the general consensus that H₂ adsorbs dissociatively on Cu [53]. Obvioiusly, CO₂ hydrogenation can occur at the interface starting from CO₂ at the ZrO₂ and H on the Cu. Two routes leading to methanol, namely, the formate route and the RWGS route (featuring the splitting of CO₂ in the production of CO and atomic O at the interface) as suggested by experiments [54-56], were both identified.

In the RWGS route, the adsorbed CO₂ dissociates directly at the interface, forming the adsorbed CO and atomic O. By stepwise hydrogenation to formyl (HCO) and formaldehyde (H₂CO), CO is converted to methylate (H₃CO). H₃CO can be further hydrogenated to methanol (H₃COH). Meanwhile, the surface atomic O can react with H to form hydroxyl and water. In parallel with the RWGS pathway, the adsorbed CO₂ can alternatively react with H to yield the surface formate (HCOO). Upon further hydrogenation, HCOO is converted to H₂COO and H₂COOH. The interfacial H₂COOH can dissociate to formaldehyde (H₂CO) and hydroxyl (OH), which can be further hydrogenated to methanol and water, respectively. The HCOO and H₂COO are the most stable intermediates, while H₂COOH and H₂CO are the least stable surface species.

Based on the kinetic Monte Carlo simulation on this model, Liu and coworkers [28] found that the interface sites are largely occupied by oxidative species, such as O atoms, OH and H₃CO groups and, therefore, the interface Cu atoms are cationic. This means that the appearance of cationic Cu is just a natural consequence of the catalytic process. The in situ formation of cationic Cu seen from kMC simulation may help rationalize the experimental observations, where the ionic copper was identified during methanol synthesis from CO₂/H₂. To describe more realistically the interfacial Cu, in the later work [23], Hong and Liu adopted an oxygen-rich interface as the new model to mimic the catalyst under experimental conditions, where the exposed Zr_5c sites are already blocked by the oxygen species. With a switch to the Zr_6c model, the adsorption energies of oxygen-containing species like O and H₃CO are indeed much reduced, which effectively prevents the poisoning of the model catalyst in the kinetics simulation.

In the oxygen-rich interface model, both microkinetics and the kMC [23] simulation were carried out. The temperature, total pressure and molar ratio of CO₂ to H₂ were set at 500 K, 17 atm and 1∶3 respectively. The equilibrium simulation shows that CO₂ conversions to H₃COH and CO are 12.5% and 6.2%, respectively. From the dynamic simulation based on the continuous stirred tank reactor model, the following results are obtained. (i) 90% of the converted CO₂ molecules are hydrogenated to formate, while only a few of them are decomposed to CO. Both methanol and CO are produced dominantly via the formate pathway (see Fig. 4) through the direct hydrogenation of CO₂. The dissociation of CO₂ is the slowest step. H₂CO is a key intermediate species in the reaction pathway, the hydrogenation of which dictates the high temperature of CO₂ hydrogenation. (ii) Formate, H₂COO and CO are formed instantaneously as the reaction starts and their concentrations hardly change during the reaction. On the contrary, H₃CO is low in concentration at the beginning, but its coverage continuously increases at the first ten minutes of the reaction. (iii) The conversion of CO₂ is 1.2% with the selectivity to H₃COH being 68%. The TOF is calculated to be 2.4 × 10^-3 s^-1. (iv) The calculated apparent activation energies for CO and methanol formation are 1.79 and 0.79 eV, respectively, explaining why the selectivity to methanol decreases dramatically as the temperature increases. (v) The secondary reactions due to the readsorption of products lower the overall TOF but increase the selectivity to methanol by 16%.

Hong and Liu also compared the kMC simulation results with the microkinetics results [17]. It was found that the overall rate from microkinetics simulation is by and large overestimated compared to that from kMC simulation. As a mean-field approach, the rate expression in microkinetics assumes that the reactants have been mixed completely before they react with each other. Although this assumption may work well for the reaction in the gas phase or in solution, it is often not valid for heterogeneous catalytic reactions, where the active sites (e.g., the interface of composite) are in minority and the diffusion of reactants is kinetically hindered (e.g., a corrugated potential energy surface or local high surface coverage).

Concluding remarks

In this work, we review the recent theoretical advances of methanol synthesis from CO₂/H₂ over Cu-based catalysts. We summarize the typical theoretical approaches utilized in simulating the complex heterogeneous catalytic reactions. The CO₂ hydrogenation mechanism and kinetics from different model surfaces have been compared and analyzed. We show that this topic is not free from controversies despite many years’ studies,in experiment and theory. This is not least because of the intrinsic complexity of the system, i.e., involving multi-reaction-channels and composite materials. The catalytic role of the charged Cu sites, nanosized Cu particle and the Cu/oxide interface have been highlighted in recent theoretical studies. The direct CO₂ hydrogenation is found to be a major reaction channel for methanol synthesis, while the CO hydrogenation pathway has only a minor contribution.

Due to the limitation of the computational power for direct simulating the complex metal/oxide composite systems, it remains difficult for computational simulation to address fully the “synergic effect” at the atomistic scale and reproduce the experimental kinetics at the real time scale. Although many early works agree that Cu is the only active catalytic component, recent results on Cu/oxide models suggest an alternative picture where the Cu/oxide interface play an important role in activating CO₂. Obviously, more work is needed in the near future to characterize the atomistic structure of the Cu/oxide interface and identify the reaction mechanism toward the rational design of better catalysts for methanol synthesis from CO₂ hydrogenation.

References

[1] Strunk, J.; Kahler, K.; Xia, X.; Muhler, M., Surf. Sci.2009, 603, 1776-1783

[2] Ostrovskii, V. E., Catal. Today2002, 77, 141-160

[3] Yang, Y.; Evans, J.; Rodriguez, J. A.; White, M. G.; Liu, P., Phys. Chem. Chem. Phys.2010, 12, 9909-9917

[4] Saito, M.; Takeuchi, M.; Fujitani, T.; Toyir, J.; Luo, S.; Wu, J.; Mabuse, H.; Ushikoshi, K.; Mori, K.; Watanabe, T., Appl. Organomet. Chem.2000, 14, 763-772

[5] Yu, K. M. K.; Yeung, C. M. Y.; Tsang, S. C., J. Am. Chem. Soc.2007, 129, 6360-6361

[6] Yu, K. M. K.; Tsang, S. C., Catal. Lett.2011, 141, 259-265

[7] Lim, H. W.; Park, M. J.; Kang, S. H.; Chae, H. J.; Bae, J. W.; Jun, K. W., Ind. Eng. Chem. Res.2009, 48, 10448-10455

[8] J. Weigel, A. Koeppel, A. Baiker, A. Wokaun Langmuir 12 (1996) 5319.

[9] Baker, J. E.; Burch, R.; Golunski, S. E., Appl. Catal.1989, 53, 279-297

[10] Koeppel, R. A.; Baiker, A.; Schild, C.; Wokaun, A., Stud. Surf. Sci. Catal.1991, 63, 59-68

[11] Shustorovich, E.; Bell, A. T., Surf. Sci.1991, 253, 386-394

[12] Sun, Y.; Sermon, P. A., Catal. Lett.1994, 29, 361-369

[13] Yoshihara, J.; Parker, S.; Schafer, A.; Campbell, C. T., Catal. Lett.1995, 31, 313-324

[14] Yoshihara, J.; Campbell, C. T., J. Catal.1996, 161, 776-782

[15] Burch, R.; Golunski, S. E.; Spencer, M. S., Catal. Lett.1990, 5, 55-60

[16] Wang, X.; Zhang, H.; Li, W.; Korean, J., Chem. Eng.2010, 27, 1093

[17] Liu, Z. P.; Wang, C. M.; Fan, K. N., Angew. Chem. Int. Ed.2006, 45, 6865-6868

[18] Wang, C. M.; Fan, K. N.; Liu, Z. P., J. Am. Chem. Soc.2007, 129, 2642-2647

[19] Chen, J.; Liu, Z. P., J. Am. Chem. Soc.2008, 130, 7929-7937

[20] Li, Y. F.; Liu, Z. P.; Liu, L.; Gao, W., J. Am. Chem. Soc.2010, 132, 13008-13015

[21] Shang, C.; Liu, Z. P., J. Am. Chem. Soc.2011, 133, 9938-9947

[22] Fang, Y. H.; Liu, Z. P., J. Am. Chem. Soc.2010, 132, 18214-18222

[23] Hong, Q. J.; Liu, Z. P., Surf. Sci.2010, 604, 1869-1876

[24] Kakumoto, T., Energy Convers. Manage.1995, 36, 661-664

[25] Kakumoto, T.; Watanabe, T., Catal. Today1997, 36, 39-44

[26] Hu, Z. M.; Takahashi, K.; Nakatsuji, H., Surf. Sci.1999, 442, 90-106

[27] GrabowL. C. , MavrikakisM. , ACS Catal. 2011, 1, 365.

[28] Tang, Q. L.; Hong, Q. J.; Liu, Z. P., J. Catal.2009, 263, 114-122

[29] Tang, Q. L.; Liu, Z. P., J. Phys. Chem. C2010, 114, 8423-8430

[30] French, S. A.; Sokol, A. A.; Bromley, S. T.; Catlow, C. R. A.; Rogers, S. C.; King, F.; Sherwood, P., Angew. Chem. Int. Ed.2001, 40, 4437

[31] French, S. A.; Sokol, A. A.; Bromley, S. T.; Catlow, C. R. A.; Sherwood, P., Top. Catal.2003, 24, 161-172

[32] Catlow, C. R. A.; French, S. A.; Sokol, A. A.; Thomas, J. M., Philos. Transact. A Math. Phys. Eng. Sci.2005, 363, 913-936, discussion 1035-1040

[33] Payne, M. C.; Teter, M. P.; Allan, D. C.; Arias, T. A.; Joannopoulos, J. D., Rev. Mod. Phys.1992, 64, 1045-1097

[34] Kresse, G.; Hafner, J., J. Phys. Condens. Matter1994, 6, 8245-8257

[35] Soler, J. M.; Artacho, E.; Gale, J. D.; Garcia, A.; Junquera, J.; Ordejon, P.; Sanchez-Portal, D., J. Phys. Condens. Matter2002, 14, 2745-2779

[36] Junquera, J.; Paz, O.; Sanchez-Portal, D.; Artacho, E., Phys. Rev. B2001, 64, 235111

[37] Vanderbilt, D., Phys. Rev. B1990, 41, 7892-7895

[38] Troullier, N.; Martins, J. L., Phys. Rev. B1991, 43, 1993-2006

[39] Perdew, J. P.; Chevary, J. A.; Vosko, S. H.; Jackson, K. A.; Pederson, M. R.; Singh, D. J.; Fiolhais, C., Phys. Rev. B1992, 46, 6671-6687

[40] Perdew, J. P.; Burke, K.; Ernzerhof, M., Phys. Rev. Lett.1996, 77, 3865-3868

[41] García-Gil, S.; García, A.; Lorente, N.; Ordejón, P., Phys. Rev. B2009, 79, 075441

[42] Monkhorst, H. J.; Pack, J. D., Phys. Rev. B1976, 13, 5188-5192

[43] Henkelman, G.; Uberuaga, B. P.; Jonsson, H., J. Chem. Phys.2000, 113, 9901

[44] Wang, H. F.; Liu, Z. P., J. Am. Chem. Soc.2008, 130, 10996-11004

[45] Henkelman, G.; Jonsson, H., J. Chem. Phys.1999, 111, 7010

[46] Shang, C.; Liu, Z. P., J. Chem. Theory Comput.2010, 6, 1136-1144

[47] Halgren, T. A.; Lipscomb, W. N., Chem. Phys. Lett.1977, 49, 225-232

[48] Temel, B.; Meskine, H.; Reuter, K.; Scheffler, M.; Metiu, H. J., Chem. Phys.2007, 126, 204711

[49] Chatterjee, A.; Vlachos, D. G., J. Comput. Aided Mater. Des.2007, 14, 253-308

[50] Levi, A. C.; Kotrla, M., J. Phys. Condens. Matter1997., 9, 299-344

[51] Nakatsuji, H., J. Chem. Phys.1987, 87, 4995

[52] Nakano, H.; Nakamura, I.; Fujitani, T.; Nakamura, J., J. Phys. Chem. B2001, 105, 1355-1365

[53] Wambach, J.; Baiker, A.; Wokaun, A., Phys. Chem. Chem. Phys.1999, 1, 5071-5080

[54] Schilke, T. C.; Fisher, I. A.; Bell, A. T., J. Catal.1999, 184, 144-156

[55] Schild, C.; Wokaun, A.; Baiker, A., J. Mol. Catal.1990, 63, 243-254

[56] Jennings, J. R.; Lambert, R. M.; Nix, R. M.; Owend, G.; Parker, D. G., Appl. Catal.1989, 50, 157-170

RIGHTS & PERMISSIONS

Higher Education Press and Springer-Verlag Berlin Heidelberg