1 Introduction

Catalysts based on gold particles with a size of a few nanometers supported on metal oxides have gained enormous attention in the area of Surf. Sci. and catalysis since Haruta et al. reported their remarkably high activity for low temperature CO oxidation [1]. Since this breakthrough, the reaction has been studied extensively, on supported particles [2–5], atomically dispersed species [6, 7], and Surf. Sci. model systems [8].

Although it is generally accepted that the catalytic activity of Au depends to a large extent on the size of the nanoparticles [9], the question which property of the nanoparticles is responsible for the reactivity has not yet been answered conclusively and the issue of gold’s high activity is still under debate [10]. Several explanations have been proposed, including the role of the support [11], quantum size effects, charge transfer to and from the support, support induced-strain, oxygen spill over to and from the support, the Au oxidation state [12], and the role of very low coordinated Au atoms on Au nanoparticles [13–15]. Of course, it is possible that several of the aforementioned effects occur simultaneously.

With respect to the mechanism of the CO oxidation, both atomic and molecular oxygen have been suggested as active species [5, 16, 17]. The energetically difficult step in a mechanism based on O-atoms is the dissociation of O2 on the gold, as the activation energy for this step usually exceeds the small heat of adsorption of the O2. In fact, using DFT calculations on stepped Au surfaces, Liu et al. [18] and Fajin et al. [19] have suggested that both atomic and molecular adsorbed oxygen can oxidize CO, the former having the lower activation barrier. Under experimental conditions, water and water derived species such as hydroxyls may play a role [20–22], and reducible support oxides may provide active oxygen for the reaction [23]. It seems plausible that multiple reaction channels and, therefore, multiple O2 activation mechanisms may exist. O2 molecules may directly dissociate on the support, leaving oxygen atoms to diffuse to the gold particles, or they may diffuse molecularly to the gold and react or dissociate at the particle-support interface, or somewhere on the gold particles. Essential intermediates may be atomic, or molecular oxygen, or some type of oxygen-containing species such as peroxides or carbonates [24, 25].

In this paper we ask the question what type of ensemble of gold atoms would be needed to either dissociate O2 into O-atoms, or enable a direct reaction between O2 and CO. To this end we have investigated the relevant reaction steps on various Au surfaces including low index, stepped and diatomic Au rows created on Au(100) employing periodic self-consistent DFT calculations. As some papers report that unsupported clusters may interact strongly with O2 and even catalyze CO oxidation [26–29], we also include calculations on a gold cluster. We confirm that O2 does not dissociate on a flat surface or even the stepped, relatively reactive (310) surface of gold that we used in previous work [30–32]. However, we find that extended gold surfaces consisting of diatomic rows of Au-atoms on a Au(100) surface are able to dissociate O2 and allow CO oxidation, while a Au38 cluster terminated by a similar double dimer facet as the diatomic row structure shows even more favorable energetics for the CO oxidation from CO and O2.

2 Computational Details

We used the Vienna ab initio simulation package (VASP) [33], which performs an iterative solution of the Kohn–Sham equations in a plane-wave basis set. Plane-waves with a kinetic energy below or equal to 400 eV were included in the calculations. The exchange-correlation energy was calculated within the generalized gradient approximation (GGA) proposed by Perdew and Wang (PW91) [34]. The electron–ion interactions for C, O and Au atoms were described by the projector-augmented wave (PAW) method developed by Blöchl [35]. This is essentially a scheme combining the accuracy of all-electron methods and the computational simplicity of the pseudo-potential approach [36].

The relative positions of the Au metal atoms were initially fixed as those in the bulk, with an optimized lattice parameter of 4.18 Å (the experimental value is 4.08 Å) [37]. The optimized lattice parameter was calculated using the face-centred cubic (fcc) structure unit cell and its reciprocal space was sampled with a (15 × 15 × 15) k-point grid generated automatically using the Monkhorst–Pack method [38]. A first-order Methfessel–Paxton smearing-function with a width ≤0.1 eV was used to account for fractional occupancies [39]. Partial geometry optimizations were performed including the RMM-DIIS algorithm [40]. Geometry optimizations were stopped when all the forces were smaller than 0.05 eV/Å. Vibrational frequencies for transition states (TS) were calculated within the harmonic approximation. The adsorbate-surface coupling was neglected and only the Hessian matrix of the adsorbate was calculated [41]. The climbing-image nudged elastic band (cNEB) method [42] was used in this study to determine minimum-energy paths.

Molecules in gas phase were simulated in a 10 × 12 × 14 Å3 orthorhombic unit cell at the Γ-point. Non-spin polarized calculations were done for closed shell CO and CO2 molecules and spin-polarized calculations were performed for open shell species, O and O2.

2.1 Spin Polarized Calculations were Done for Adsorbates on Gold

A four layer slab model, with the two top most layers relaxed, was chosen for constructing diatomic rows on the Au(100) surface by removing one row of Au atoms along the vector b as shown in Fig. 1. A p(3 × 2) unit cell was used, and its reciprocal space was sampled with (3 × 5 × 1) k-point meshes with a vacuum gap >10 Å. The Au(100) surface was represented with a slab model using five-metal layers of which the top two were relaxed) and for Au(310) we used 11 layers with the top four relaxed, both with a vacuum gap of >10 Å to separate the periodic slabs. For the (100) slab, we used a p(2 × 2) unit cell with the reciprocal space sampled with (5 × 5 × 1) k-point meshes, and for the Au(310) surface (3 × 9 × 1) k-point meshes were used for sampling the reciprocal space of the (310) p(2 × 1) unit cell.

Fig. 1
figure 1

Top view of the two relatively reactive Au surfaces used in this paper with O2 and O adsorbed, a O2 adsorbed on diatomic rows on Au(100), b O2 on Au(310), c O-atoms on diatomic rows, d O-atoms on Au(310). Adsorption energies and characteristic distances are indicated

Full size image

The Au38 nano particle is a three-dimensional metal crystallite cut from metal bulk with low index surfaces (using 111 and 100 planes as basis) to have a cubo octahedral shape. The structure of the particle has been optimized in a fully relaxed state inside a 19 × 20 × 21 Å3 orthorhombic unit cell, and its reciprocal space was sampled with (1 × 1 × 1) k-point mesh generated automatically using the Monkhorst–Pack method. The choice of unit cell allows for vacuum gaps of >10 Å between particles.

3 Results

The ability of molecular oxygen to dissociate, be it in adsorbed O2 or in a complex with CO, is crucial for CO2 formation. We will first discuss O2 adsorption and dissociation and then the reaction with CO, all on extended surfaces. Finally we discuss the reaction on the Au38 cluster.

3.1 CO Oxidation on Au via O2 Dissociation

3.1.1 Adsorption of Molecular Oxygen

Figure 1 shows the two surfaces considered in this paper, namely the Au(100) with a diatomic row of Au-atoms (Fig. 1a, c) and the stepped Au(310) (Fig. 1b, d) with molecularly and dissociatively adsorbed oxygen. The interaction of molecular oxygen with gold surfaces varies markedly depending upon the surface structure, location and mode, and for many geometries the interaction is repulsive. This result is an artefact from the calculations, mainly because DFT fails to describe the attractive long-range van der Waals interactions with respect to the short-ranged Pauli repulsion [43]. The hollow-bridge configuration of O2 on the diatomic row in Fig. 1a represents the most stable adsorption mode of O2 on these surfaces (−0.52 eV); single bridges as well as adsorption in double or single bridged modes are a few tenths of an eV less stable. The distances show that the Au atoms of the row relax in the direction perpendicular to the row, with Au–Au distances significantly larger than the 2.96 Å of bulk gold. The O–O bond is clearly activated in the hollow adsorption geometry, as evidenced by the elongated O–O distance of 1.44 Å, which is almost 0.2 Å larger than the bond in gas phase molecular O2 (calculated at 1.24 Å; the experimental value is 1.207 Å [44]). Additionally, interaction of O2 for this configuration results in a significant decrease of the stretching frequency (ν O–O = 748 cm−1) with respect to the gas phase reference (1520 cm−1 calculated vs. 1556 cm−1 experimental [44]) indicating substantial bond weakening.

For O2 on the Au(310) the single bridge mode in Fig. 1b is the most stable one (−0.19 eV), although adsorption on the long bridge at the step comes close. Both should be considered weak chemisorption modes.

Adsorption of O2 on Au(100) and on Au/Au(100), which has an additional Au atom placed in every fourth hollow site, is even weaker and is on the order of −0.1 eV. Modes of adsorption with the O–O molecular axis normal to the surface are predicted to adsorb endothermically.

In summary, O2 interaction with gold, if it exists at all, is weak except for the diatomic row configuration presented in Fig. 1a, where O2 is chemisorbed with an adsorption energy of about 0.52 eV (about 50 kJ/mol).

We conducted Bader charge analysis to quantify the amount of charge transfer from gold to O2 [45]. Au atoms having direct bonding with O2 on the diatomic rows are substantially more oxidized than those on the stepped (310) surface. For instance, for the most stable configuration of Fig. 1a, O-atoms are reduced by 0.53e and 0.32e and the Au atoms are oxidized by 0.23e. However, for the most favorable configuration on (310), O-atoms gained a charge of 0.20e only, with a loss of charge on the relevant Au atoms only up to 0.13e. Hence we conclude that the strength of the O2 binding to Au strongly depends upon that structure of Au which is capable of donating charge to the anti-bonding 2π* orbitals of O2. The charge transfer weakens the O–O bond and consequently the bond elongates.

Xu and Mavrikakis [25] have studied adsorption and dissociation of O2 on the (111) and (211) surfaces of Au (GGA = PW91). They found no adsorption on (111) but observed weak interaction (−0.15 eV) on the stepped surface with O2 in a top-bridge-top configuration at the step. An adsorption energy of −0.12 eV was reported in an earlier paper by Mavrikakis et al. [46] for O2 Au(211) with the molecular O–O axis parallel to the surface. Again, no adsorption was found on Au(111). Fajin et al. [47] have recently investigated the adsorption of O2 on Au(321) in detail (GGA = PW91). The highest E ads (−0.17 eV) occurred with O2 on the bridge site. A value of < −0.05 eV is reported on Au(221) (GGA = PBE) [18]. A recent experimental study also reports a weak interaction of O2 with Au(111) [48]. Experimental results for the diatomic rows of Fig. 1 are not available.

3.1.2 Adsorption of Atomic Oxygen

Atomic oxygen, once available, adsorbs readily on the gold surfaces of Fig. 1. As we are interested in dissociation, we immediately consider the adsorption of two O-atoms on adjacent sites. Figure 1c shows the most stable configuration with two O-atoms on outer bridge positions of the diatomic row structure. The joint adsorption energy amounts to −6.64 eV per two atoms (for comparison, the strongest adsorption bond for a single O-atom, which resides in the inner bridge along the b-vector amounts to −3.49 eV; the adsorption energy for a single O-atom in the fourfold hollow site is −3.42 eV). This configuration is exothermic by −0.38 eV) with respect to molecular O2 in the gas phase. The adsorption energy is −3.32 eV per oxygen atom with respect to gas phase atomic oxygen. Coadsorption of two O-atoms on the fourfold hollow positions is significantly less favorable (−5.58 or −2.79 eV per O atom).

On Au(310) the most stable geometry of two O-atoms is as shown in Fig. 1d. A single O-atom would be adsorbed with energy of −3.32 eV. However, for two adjacent O-atoms, the joint adsorption energy of −6.18 eV is 0.08 and 0.27 eV endothermic with respect to O2 in the gas phase or adsorbed O2 respectively.

Adsorption energies of −2.47 to −2.71 eV for O on the fcc hollow site of Au(111) have been reported in literature [18, 49, 50]. Energies reported by Liu et al. [18] (GGA = PBE) on the step bridge site of Au(221) and Au(211) are −2.91 and −3.09 eV, respectively. Fajin et. al. [47] (GGA = PW91) reported −3.30 eV for O on the most stable sites of Au(321). All these values are in good agreement with our results.

We conclude that coadsorption of two O-atoms on the diatomic rows as in Fig. 1c is expected to provide the least unfavorable thermodynamic driving force for dissociation, with an endothermicity of about 0.14 eV with respect to adsorbed molecular O2, see Fig. 2.

Fig. 2
figure 2

Potential energy diagram for CO oxidation on the diatomic Au rows on Au(100). Top (below) and side (above) views for O2 adsorption and dissociation, CO and O co-adsorption and CO2 formation are shown. The critical step is O2 dissociation which has an activation barrier almost equal to that of O2 desorption. All further steps towards CO2 however, are thermodynamically very favorable. Note that the process ends with an O-atom on the surface, making the next reaction with adsorbed CO a facile one (0.02 eV activation energy and 2.22 eV exothermic)

Full size image

The high adsorption energies of O-atoms on and near the steps clearly demonstrate how Au atoms of low coordination are more reactive. Again, the strongest adsorption of O-atoms is on the diatomic rows. For comparison we note however that the highest adsorption energy on this corrugated Au surface (−3.5 eV) is still ~0.50 eV lower than that on the least reactive surface of platinum, (4.00–4.30 eV for O/Pt(111)), where the O2 is experimentally seen to dissociate [51].

3.1.3 Dissociation of Molecular Oxygen

For O2 to dissociate it is imperative that the transition state for dissociation has a lower activation energy than that for desorption of the O2. Among the surfaces and geometries of adsorbed O2 and the two co-adsorbed O-atoms considered in this paper we found only one feasible route for dissociation, namely on the diatomic Au rows on Au(100). All other combinations led to transition states with activation energies much higher than the O2 heat of adsorption.

Figures 1a and c appear as likely start and end configurations for O2 dissociation. Both are enthalpically stable with respect to gas phase oxygen, and the reaction is only slightly endothermic (0.14 eV). We calculate a transition state with an activation barrier of 0.5 eV, which is about the same as the activation energy for desorption of O2 (0.52 eV), see Fig. 2.

In the transition state, which is symmetric, the O–O bond length increases from 1.44 to 2.08 Å and each O-atom is at a distance of 2.07 Å from the nearest Au atoms. To ascertain whether the above arrangement is a saddle point or a local minimum, vibrational analysis was carried out. The transition state was characterized by a unique imaginary frequency of 265i cm−1 and the normal mode analysis shows that the O-atoms move in opposite directions. As the activation energy for dissociation is almost similar to that for desorption, both processes will compete. In general, desorption has a higher pre-exponential factor than surface dissociation, caused by the entropy difference between adsorbed and free molecules. Nevertheless, it is well possible that part of the O2 will dissociate into O-atoms on the surface.

3.1.4 CO Adsorption

CO adsorbs weakly on close-packed surfaces of gold, but when the surface contains steps, adsorption energies become appreciable. On Au(310) CO bound linearly through the C-atom to the low-coordinated Au atoms at the step yields the highest adsorption energy, namely −0.73 eV. For details we refer to our previous work [32].

On the diatomic rows CO binds preferentially on bridge sites, with an adsorption energy of −0.75 eV. The bridge configuration is a local minimum with real frequencies, the CO stretch being at 1890 cm−1. The values of E ads are on the order of 0.2 eV higher than on the corresponding sites of Au(100) [32]. Adsorption on the top site (−0.63 eV) has two imaginary modes at 50i, 95i cm−1 and corresponds to a second order saddle point. As the adsorption energies on bridge and top sites are very similar, we conclude that the CO molecule will be able to diffuse easily on the diatomic rows.

3.1.5 CO Oxidation on Diatomic Rows

The diatomic row may catalyze the CO oxidation reaction as illustrated in Fig. 2. Once CO and O-atoms are coadsorbed on the diatomic row, reaction takes place spontaneously with a negligible activation barrier of 0.02 eV. Similarly low activation energies have been observed in other studies [19, 52]. In the most favorable coadsorption state CO and O are both on bridge positions (see Fig. 2). In the transition state the carbon atom of CO and O-atom are 2.41 Å apart. It has one imaginary frequency of 112i cm−1, and is thus a true transition state. Two other combinations were tried, one with CO and O both on bridge locations along the vector b and the other with positioning CO on top and O on adjacent bridge. The former geometry is 0.45 eV less stable indicating substantial repulsion. The latter leads to the formation of CO2 right away. CO2 formation is highly exothermic (−2.22 eV) and thus thermodynamically favorable. As its adsorption energy is very small, it desorbs instantaneously upon formation.

3.2 CO Oxidation via OCOO Complex Formation and Decomposition

We have also investigated the direct reaction between CO and O2. Fig. 3 shows the reaction pathway on the (310) surface. A few initial configurations for carbon monoxide and molecular oxygen coadsorbed on the (310) surface were considered. Knowing that O2 is adsorbed weakly on this gold surface, and that CO preferentially chemisorbs at the outer step [32], the combination of coadsorbates as well as the transition state shown at the start of the reaction in Fig. 3 come out as the most likely. The adsorption energy for CO and O2 equals −0.7 eV. It is important to note that the differential adsorption energy of the O2 in this configuration is −0.19 eV, which effectively corresponds to weak chemisorption. In the transition state, the carbon atom of the CO molecule interacts directly with the O–O bond of the O2 molecule, with a transition state characterized by a small activation barrier of 0.20 eV and a unique imaginary frequency of 113i cm−1.

Fig. 3
figure 3

Energy profile for CO2 formation on Au(310) via a direct reaction of CO with molecular O2. The zero level corresponds to gas phase CO, O2 and the clean slab

Full size image

Hence, the transition state for the direct reaction between CO and O2 has an activation energy of 0.2 eV, i.e. of the same magnitude as the energy with which O2 binds to the Au (310) surface. Our results are rationalized in terms of enthalpy, and we do not consider entropy. O2 molecules on the surface may react with CO, but the barrier to desorption, which is entropically favoured, is similar. Hence desorption will be more likely than reaction.

This OCOO species on the surface is 0.7 eV more stable than the co-adsorbate configuration. It has been invoked in other studies on Au(321) and Au(211) surfaces [19]. The structural parameters of the OCO–O complex are in excellent agreement with these studies. However, on Au(321) and (211) the activation energies for OCOO formation are substantially higher (0.58–0.68 eV) than the 0.20 eV on Au(310) in our work. Interestingly, an even lower barrier of 0.08 eV was calculated for OCOO formation on a Au strip supported on ZrO2 [52].

Once the OCOO intermediate forms, it reacts via a 0.25 eV barrier to CO2 and an adsorbed O-atom. Literature reports barriers of 0.33–0.43 eV for dissociation of the OCOO species [19, 52]. The remaining O-atom reacts with CO almost without barrier (0.04 eV). The process is highly exothermic (−2.85 eV).

While the diatomic rows represent the most reactive Au surface for O2 splitting reported thus far, this structure is almost inactive for OCOO complex formation. Thus the diatomic row structure is active enough for O2 dissociation and oxidation of CO, but not via the direct reaction between O2 and CO. Conversely, on the (310) surface both OCO-O compound formation and subsequent O–O bond cleavage are thermodynamically favorable and proceed by surmounting small energy barriers of 0.20 and 0.25 eV for the two steps. Hence this channel may be operative for catalyzing the CO oxidation on nanoparticles with appropriate steps.

3.3 CO Oxidation on a Au38 Cluster

The Au38 cluster shown in Fig. 4 exposes (111) facets consisting of seven atoms and (100) facets of four atoms. The latter bear similarity to the diatomic row structure on Au(100) in Figs. 1 and 2, and is therefore of particular interest for the purpose of this study.

Fig. 4
figure 4

Energy scheme for CO oxidation via O2 dissociation on a Au38 cluster

Full size image

Figure 4 starts with the adsorption and dissociation of O2 on the (100) facet, in geometries that are equivalent to the ones investigated on the extended diatomic row surface. It is seen that O2 adsorbs with an appreciable strength (−0.78 eV) on the cluster, while the activation energy for dissociation is only 0.43 eV, the dissociation step itself being exothermic. Hence this cluster successfully binds and dissociates O2 in a similar fashion as the diatomic row structure in Figs. 1 and 2. The two adjacent O-atoms after dissociation show some repulsion, hence we continue the calculations by taking twice the energy of a single O-atom per cluster. Coadsorption of CO on the same (100) facet is feasible, with a differential adsorption energy of −0.76 eV. In the transition state the CO moves towards the O-atom, which involves a small barrier of 0.24 eV after which CO2 forms and desorbs.

We compare our results with those of Róldan et al. [29, 53], who reported O2 adsorption energies in the range of −0.19 to −0.91 eV on Aun (n = 25, 38, 55, and 79) clusters. Au38 had the highest E ads for O2 and was found to dissociate O2 with an activation barrier of 0.46 eV [29] in excellent agreement with our values.

Róldan et al. also investigated the effect of exchange–correlation functional on the adsorption and dissociation energy of O2 [53]. Values calculated using PW91 were larger than the PBE by roughly 0.20 eV, however, the effect of changing the functional on the activation barriers is smaller. Overestimation of about 0.30 eV in binding of O2 to gold clusters employing GGA = PW91 has been reported in previous works as well [54, 55].

4 Discussion

In this paper we address CO oxidation on extended surfaces of particular structure, chosen such to maximize their reactivity towards either O2 dissociation, after which CO + O is usually a facile reaction step, or to CO2 formation between CO and molecular oxygen. We identified two configurations of Au atoms for which the energetics of these reactions appear feasible.

Following the steps of normal Langmuir–Hinshelwood mechanisms, the reactants first need to adsorb on the surface. Figure 5 presents an overview of how CO and O2 adsorb on a number of gold surfaces. Carbon monoxide chemisorbs in principle on all gold surfaces, with the exception of the close-packed Au(111) surface, where the adsorption energy is characteristic for weak chemisorption. Figure 5 clearly shows the trend that CO adsorbs more strongly when the coordination of the gold atom decreases. The strongest bond (−0.88 eV) is formed with the four-fold coordination of the additional Au atom on the Au(100) surface, but adsorption of CO on the stepped (310), the diatomic row structure and the Au38 cluster also show considerable bond strengths between −0.73 and −0.79 eV. For O2 the situation is more critical. Irrespective of whether the mechanism of CO oxidation proceeds via dissociated or molecular O2, the molecule needs to bound sufficiently strong that the activation energy of the subsequent step will not exceed the adsorption energy, otherwise the O2 will probably desorb. Of all the surfaces considered in Fig. 5, only the diatomic row and the Au38 cluster bind the O2 sufficiently strong. Interestingly the surface exposing the single Au atoms with the lowest coordination of all (N = 4) does not interact appreciably with O2. The diatomic row and the Au38 cluster have in common that they expose ensembles of four Au atoms of relatively low coordination in a square array to which the O2 binds in the “hollow-four centre bridge” configuration of Fig. 1a with the molecular axis parallel to the surface. This site can also accommodate the O-atoms after dissociation. On the diatomic row, the four Au atoms in the square ensemble each have a coordination number of seven (N = 7). O2 binds appreciably (−0.52 eV) as do the two O-atoms after dissociation, although the dissociation is slightly endothermic (0.14 eV) and the activation energy for dissociation is with 0.5 eV about equal to the O2 heat of adsorption. However, the same configuration on the Au38 cluster offers the lowest coordination of the Au atoms (N = 6 for each of the four Au atoms), and here the dissociation of O2 becomes exothermic, with the concomitantly lower activation energy of 0.43 eV, which is considerably smaller than the O2 adsorption energy.

Fig. 5
figure 5

Overview of adsorptions and reactions relevant for CO oxidation on different gold surfaces

Full size image

Our results are in good agreement with those of Róldan et al. [29, 53], who studied the O2 dissociation on a number of Aun clusters (n = 25, 38, 55, and 79), among which Au38 shows the highest reactivity towards O2. The second best is Au25 which interestingly also exposes the square Au4 arrangement.

Although CO oxidation via O2 dissociation on the Au38 cluster appears as a realistic possibility, the feasibility of this reaction on the extended diatomic rows of Au on Au (100) seems less certain. This surface does enable the adsorption and dissociation of O2 in a way that the activation energy of dissociation is similar as that for desorption of O2. Hence, on the basis of energy one may expect that dissociation can compete with desorption. The uncertainty, however, is in the entropy change for both reactions. Desorption often has preexponential factors exceeding those of dissociation, thus favoring the former [53]. Nevertheless, we feel that our result of O2 dissociation on diatomic Au rows on Au(100) being energetically feasible is significant, as this is the only extended gold surface we have found sofar which comes close to enabling the dissociation of O2.

Another mechanism for the CO oxidation involves the direct reaction between molecular O2 and CO to CO2 and an adsorbed O-atom. The stepped Au(310) surface provides a pathway for this reaction with a barrier about equal to the O2 adsorption energy. Nevertheless, we cannot be sure that this mechanism operates, as the O2 is bound to the surface in a weakly chemisorbed state. Although reaction with CO may be energetically feasible, the molecule can diffuse freely over the surface, and hence instead of surmounting the activation barrier to form the OCOO complex, the molecule has the alternative of moving away at no energy cost.

The results in this paper highlight the importance of sites consisting of low coordinated atoms in a particular geometry, in this case a square of four Au atoms. This ensemble appears capable of binding both molecular and dissociated oxygen with sufficient strength, that dissociation of O2 is energetically favorable. As these sites also bind CO sufficiently strong, reaction with atomic oxygen to form CO2 proceeds readily with a small activation barrier.