Varying oxygen coverage on Cu55 and its effect on CO oxidation

Adsorption of molecular oxygen on a Cu55 cluster and the resulting oxidation effects have been investigated by spin-polarized density functional theory (DFT). The optimal structure for each Cu55O2N (N = 1–20) complex has been obtained via a sequential addition of O2 and systematic screening of the preferable adsorption sites. Upon structural optimization, several O2 molecules dissociate readily on Cu55 at different oxygen coverages, and further DFT molecular dynamics simulations at 300 K confirm the instability (small dissociation barrier) of the remaining O2 and a spontaneous movement of some oxygen atoms from the surface sites towards the cluster interior. The Cu55 cluster and its oxidized derivatives have been placed on a g-Al2O3(100) surface to study the cluster–support interaction, and furthermore, CO oxidation reactions on both Cu55(O)2N and Cu55(O)2N/g-Al2O3(100) have been studied as a function of oxygen coverage. The CO oxidation reaction barrier is rather insensitive to the oxygen coverage regardless of the support, indicating a small increase in activity with the number of surface oxygen atoms.


Introduction
The application of nanoparticles in heterogeneous catalysis is an active area of research. Noble metals, in particular the platinum group metals (PGMs) such as Pd, Pt, Ru, and Rh, have shown high activities and good stabilities for many reactions. [1][2][3][4] However, they are classified as critical metals owing to their high cost, limited availability, and socio-economic issues related to mining processes and supply (environment, working conditions, etc.). This has motivated the active search for more abundant metal catalysts with similar and/or improved performance characteristics for PGM replacement. Such catalysts are highly desirable for a wide variety of applications, not only in heterogeneous catalysis (e.g. acetylene hydrogenation 5 and ammonia decomposition 6 ) but also in electrochemistry (hydrogen energy and CO 2 reduction). 7 Here, Cu-based catalysts have attracted considerable attention due to their low cost (earth-abundant) and a variety of industrial applications, including hydrogenation, 8 oxidation 9 and synthesis of methanol. 10 Much work has been performed over a number of years to optimize their performance. However, under industrial conditions, the catalyst surfaces are readily covered by the reactant, intermediate and product species causing issues with stability. The existence of the adsorbed species or layers may significantly affect the adsorption behavior of other species and reaction mechanisms or induce catalyst reconstruction.
It is well known that the CO oxidation reaction mechanism on a catalyst is influenced by the reactant gas coverage. Under high O 2 pressure, surface oxides have been proposed as the active phases for late-transition and noble metal elements, [11][12][13] which contradicts the well-accepted opinion that the transition or noble metal per se is responsible for CO oxidation. At low CO coverage, dissociated oxygen is the active species for the production of CO 2 from CO, whereas at higher CO coverage, molecular O 2 can directly react with CO via the formation of an OOCO intermediate. 14 Such studies of CO oxidation on transition metal surfaces and supported particles under varying partial pressures are beneficial for understanding the reaction mechanisms under realistic or industrial conditions and also for providing insight on the changing activity on metal surfaces or particles (catalyst stability).
Concerning the adsorption of oxygen on Cu catalysts, most studies have considered only a single O 2 molecule. [15][16][17] The effects  (Table S2); structures and energetics of the catalytic CO oxidation on Cu 55 (O) 40 and Cu 55 (O) 40 /g-Al 2 O 3 (100) by the Eley-Rideal (ER) mechanism (Fig. S8). See DOI: 10.1039/c9cp00974d of several oxygen molecules or atoms on Cu catalysts have been rarely studied, 15 especially on the saturation coverage. For the CO oxidation reaction, there is still controversy regarding the oxidation state of Cu catalysts as the correlation between reactivity and oxidation states is not well established. For example, Huang et al. proposed that Cu 2 O is the most active among the three commercial catalysts, Cu, Cu 2 O, and CuO powders, under oxygen-lean and -rich conditions, and they attribute this to the high mobility of lattice oxygen. 18 In contrast, Fredriksson et al. 19 and Jernigan et al. 20 found that metallic Cu is significantly more active than Cu oxides. On Cu single-crystal surfaces, it is proposed that high oxygen coverage retards the reaction rate. 21,22 On the theoretical side, we recently investigated CO oxidation catalyzed by a Cu 20 cluster and found that O 2 activation upon adsorption (affected by the cluster charge state) is crucial for the rate of CO oxidation. 23 To the best of our knowledge, previous calculations of CO oxidation on a Cu catalyst usually focused on low O 2 concentrations, and the effects of varying the oxygen coverage are not clear.
Typically, the Cu nanoparticles used experimentally have an average diameter of B30 nm for catalytic CO oxidation, 19 while they are in the range of 15-35 nm for copper oxide catalysts. 24,25 A smaller copper oxide catalyst with a particle size of 3.6 nm was obtained by specific preparation methods, 26 and it displays an enhanced catalytic activity. Moreover, previous studies have revealed that the CO oxidation conversion rate is considerably affected by the crystal size of the catalyst where it generally increases up to an optimal crystal size, after which the conversion rate decreases again. 27 In this article, we present the results of density functional theory (DFT) calculations for the icosahedral Cu 55 cluster and explore the interaction of several oxygen molecules/atoms on the cluster surface, the subsequent cluster modifications, and the resulting effect on CO oxidation. Similar 55-atom clusters have been previously used as model structures for larger nanoclusters. 28 Here, the ideal icosahedral structure is the global minimum of Cu 55 where the number of symmetrically inequivalent adsorption sites is limited. This makes it possible to examine all individual sites to find the most favorable adsorption structure, and hence, the icosahedral Cu 55 cluster was chosen as the model structure for copper nanoparticles.
It is known that the support effects play a crucial role in the catalytic activity. Alumina is widely used as a basic catalytic support material because of its high chemical inertness, strength and hardness. 29 g-Al 2 O 3 possesses an excellent surface area owing to its small particle size, which results in high activity of the surface as a catalyst support. g-Al 2 O 3 crystals are terminated by the (100), (110) and (111) planes. The two main orientations (under practical operation conditions) of g-Al 2 O 3 facets are (100) and (110). The high-resolution images show that the (110) face is not atomically flat but significantly reconstructed. The (100) surface has the lowest surface energy, which is considered to be the most representative surface. 30 For these reasons, we have also chosen to study CO oxidation on Cu 55 /g-Al 2 O 3 (100) at different oxygen coverages. In addition, we note that in CO oxidation over metal-based catalysts, O 2 adsorption and activation is often regarded as the rate-limiting step. 31 The use of traditional PGM catalysts with inert supports such as SiO 2 , Al 2 O 3 and zeolites is inefficient. 32 In this type of catalyst, there is a serious competition between the adsorption and activation of CO and O 2 on the metal sites. The strong adsorption of CO seriously hinders O 2 adsorption and activation.

Computational methods
All electronic structure calculations were based on DFT using the spin-polarized, gradient-corrected exchange-correlation functional of Perdew, Burke, and Ernzerhof (PBE) 33 as implemented in the CP2K program package. 34,35 The single-particle Kohn-Sham wave functions were expanded by a combined Gaussian and plane wave (GPW) basis set. A molecularly-optimized double-zeta valence plus polarization (DZVP) basis set was used for the Gaussian expansion, 36 where the basis set superposition error (BSSE) has been reduced during optimization. An additional auxiliary plane wave basis with an energy cutoff of 600 Ry was used for operations involving electron density. The Cu and O valence states are 3d 10 4s 1 and 2s 2 2p 4 , respectively, and the valence electron-ion interaction is based on the analytical pseudopotentials derived by Goedecker, Teter, and Hutter (GTH). 37 The Cu 55 cluster with icosahedral structure is shown in Fig. S1(a) (ESI †). For the support [ Fig. S1(b) and (c), ESI †], a slab model was used by cleaving the bulk g-Al 2 O 3 model 30,38 in the (100) direction. The Al 2 O 3 (100) support was chosen as a fourlayered (4 Â 3) slab (in total 480 atoms) and a vacuum spacing of 25 Å thickness was used within the simulation box. The top two atomic layers of the support were allowed to relax while the bottom two were fixed throughout the calculations. The Bader algorithm was used for analyzing the spatial charge decomposition among atoms. 39 The climbing image nudged elastic band (CI-NEB) method was applied for mapping the reaction pathways. 40 No symmetry constraints were imposed during structural relaxations. Vibrational analysis was performed to identify the stability of the obtained structures and zero-point energy (ZPE) corrections were included for energetics. DFT molecular dynamics (MD) calculations were performed at 300 K using a Nose thermostat 41 and the Born-Oppenheimer mode with a time step of 1 fs.
The electronic ground state of O 2 is a triplet, and correspondingly, the doublet and quartet states were examined for the Cu 55 cluster (odd number of electrons), Cu 55 O 2 (molecular adsorption), and Cu 55 (O) 2 (dissociative adsorption). The spin multiplicity of Cu 55 is a quartet, but this is only 0.03 eV lower than the doublet in total energy. On the other hand, the spin multiplicities of Cu 55 O 2 and Cu 55 (O) 2 are doublets, i.e. the adsorption complexes prefer the lowest spin state.

Varying oxygen coverage on Cu 55
To study the properties of oxygen coverage on copper, we carried out a set of simulations on the Cu 55 cluster for molecular and/or dissociative O 2 adsorption. First, the lowest-energy arrangements of Cu 55 O 2N complexes were determined by sequentially adding O 2 molecules on the cluster surface (Approach 1). Here N is an integer indicating the total number of O 2 molecules adsorbed. An extensive adsorption site sampling was performed in Approach 1, where all symmetrically inequivalent sites on Cu 55 were examined to find the most favourable adsorption structure for each step. The detailed description of the stepwise adsorption of O 2 molecules for N = 1-20 is shown in the ESI † (Fig. S2-S5). The relaxation process led to a spontaneous O 2 dissociation to atomic oxygen in several cases.
After the optimal structure was found for Cu 55 O 2N , we computed the adsorption energy of the last molecular or dissociative adsorption as 42 : where O 2N denotes 2N oxygen atoms adsorbed on Cu 55 molecularly and/or dissociatively. E a (N) 4 0 indicates that the Cu 55 O 2N complex is more stable than the free O 2 molecule and the Cu 55 O (2NÀ2) complex. In addition, the average adsorption energy of the added O 2 molecules was also computed as: By applying Approach 1, the E a (N) value becomes negative at N = 20 ( Fig. S5 and Table S1, ESI †) indicating that the 20th molecule cannot adsorb exothermally (at 0 K). However, this changes if O 2 is allowed to dissociate upon adsorption. Taking into account the pronounced instability of adsorbed O 2 , we decided to apply another method (Approach 2) to generate oxidized Cu 55 geometries based on the stepwise O 2 adsorptive geometries. The initial structures of Approach 2 were based on the O 2 adsorption site scanning in Approach 1. Here, we obtained the corresponding O 2 dissociative geometries by splitting each O 2 upon adsorption one-by-one ( Most of the dissociated O atoms are located on convex adsorption sites binding with three surface Cu atoms. However, the adsorption sites of several O atoms gradually move from convex, to surface, and to concave sites as the number of O atoms increases. As shown in Fig. 1, one O atom completely falls into an endohedral site in Cu 55 at N = 13, forming bonds with surface and core Cu atoms. Furthermore, the endohedral O atoms increase to two at N = 16 and up to five at N = 20. As mentioned, the structural optimizations are performed at 0 K. To see how the adsorption structure changes under more realistic reaction conditions, we carried out a DFT-MD simulation of 20 ps for Cu 55 (O) 40 at room temperature (300 K). During this short period, the number of endohedral oxygen atoms increases up to eight (Fig. 1), which confirms the low-barrier movement of oxygen from the cluster surface towards the interior. In addition, we tested the insertion of a single oxygen atom into a clean Cu 55 cluster. In this case, O always bounces back to the surface after optimization regardless of whether it is placed initially on a sub-surface or at the central interstitial site. This indicates that surface oxidation has to take place first before oxygen can move to the interior sites. Experimentally, 13 the interaction of oxygen with Pd nanoparticles has been studied by temperature-programmed desorption (TPD), temperatureprogrammed low-energy ion scattering (TP-LEIS), and X-ray photoelectron spectroscopy (XPS). A low oxygen exposure leads to surface oxygen adatoms on Pd nanoparticles. Surface O adatoms on the Pd nanoparticles move to subsurface sites starting at 400 K, and they almost all move to the sub-surface at B750 K, desorbing only at higher temperature. High-pressure oxygen exposure readily converts small Pd nanoparticles to PdOlike species. 13 Using eqn (1) and (2), the adsorption energies E a (N) and hE a (N)i for Approach 2 have been calculated for the geometries in Fig. 1 and Fig. S6 (ESI †). The corresponding energies and properties are listed in Table S2 (ESI †). To see the energy changing trends clearly, E a (N) and hE a (N)i are plotted in Fig. 2  for both molecular and atomic adsorption. The dissociative O 2 adsorption is energetically much more favorable than molecular adsorption: E a (N) is above 2.5 eV for dissociative adsorption until N = 16, which is much higher than 1.4 eV for the highest value of O 2 molecular adsorption. For N = 17-20, E a (N) decreases down to 1.6 eV for (O) 2 adsorption, which is still significantly higher than O 2 adsorption (o0.7 eV). The decrease of E a (N) for N Z 17 is consistent with the energy changing trend in Fig. S5 (Approach 1, ESI †), and it is due to a large oxygen concentration on the cluster surface. Our MD simulations for N = 20 reveal that oxygen diffusion to endohedral sites causes an increase in the E a (N) value to 2.58 eV, i.e. back to the original level (Fig. 2). However, one cannot observe any saturation at N = 20 in Approach 2 as E a (N) does not change the sign. For the corresponding geometry of Cu 55 (O) 40 , each triangular facet of Cu 55 has two oxygen atoms. Evidently, it is possible to add even more oxygen to the cluster but we did not pursue further towards the fully saturated Cu 55 (O) 2N due to the computational demands.
For the average adsorption energy hE a (N)i, the fully dissociated adsorption (O) 2N is also higher than the adsorption with a molecule, i.e. O 2 (O) 2(NÀ1) , and they approach each other with N. At the same time, the net charge transfer increases with the number of dissociated oxygen atoms (Table S2, ESI †). The average hCu-Cui bond length in Cu 55 (O) 2N is larger than that in Cu 55 O 2 (O) 2(NÀ1) for each number of N (Table S2,

CO oxidation at different oxygen coverages
Our previous study 23 has demonstrated that CO oxidation proceeds efficiently on Cu 20 when O 2 is pre-adsorbed in a dissociative manner. The results in Section 3.1 also show that O 2 dissociates easily on the Cu 55 cluster at different oxygen coverages. In addition, experimental studies [20][21][22] have suggested that CO oxidation takes place between adsorbed CO molecules and O atoms, following the Langmuir-Hinshelwood (L-H) mechanism. To check the effect of oxygen coverage, we have studied CO oxidation in the presence of fully dissociated O 2 atoms.
3.2.1. Reaction mechanism on Cu 55 (O) 2N . The initial state (IS) of the reaction was obtained by adding CO to Cu 55 (O) 2N . The CO adsorption sites are explored within the triangular facet of Cu 55 which is the one with the last dissociated 2 Â O atoms. The Cu top site is always the most stable for CO adsorption. In contrast to O 2 , the CQO bond is hardly influenced upon adsorption (isolated 1.14 Å, max. 1.16 Å on the cluster). Again N = 1, 10 and 20 are chosen to represent the oxygen coverage from low to high. For N = 20, the annealed geometry from the DFT-MD simulation at 300 K is used which is 0.63 eV more stable than the relaxed starting structure at 0 K.
The catalytic reaction pathways for CO oxidation on Cu 55 (O) 2N are shown in Fig. 3. Upon the reaction, CO and the nearest O move closer to each other, followed by a crossing of the energy barrier (TS) to form an intermediate (IM) complex. Subsequent to the IM complex, there is a barrier-free release of the formed CO 2 molecule. The activation energy barriers decrease as the oxygen coverage increases, while at the same time, the adsorption energies of CO in the corresponding IS also decrease (1.09 eV, 0.83 eV, and 0.78 eV, respectively). Moreover, the binding energies of O in the neighboring site of CO exhibit a similar changing trend to CO adsorption (Table S2, ESI †). Correspondingly, at high oxygen coverage, the adsorbed CO and neighboring O are more weakly bound to Cu 55 giving rise to a more efficient reaction rate. According to CO oxidation experiments on Ru(0001), 11 it has been identified that the highest rates of CO 2 production occurred for high oxygen partial pressures and concomitantly high oxygen coverages.
In addition, the d-band centers (C d ) of the selected Cu 55 (O) 2N clusters in Fig. 3 were calculated from the electronic densities of states, DOS (Fig. S7, ESI †). The C d values are À2.44, À2.66, and À2.78 eV, respectively, displaying a downward shift with oxygen coverage due to the increased charge transfer from Cu 55 . From the above results, we can see that CO oxidation benefits from increasing oxygen coverage, while the d-band center and the CO and O binding energies are reduced.  (Fig. 4). The support binding energies (E b ) of Cu 55 (O) 2N on g-Al 2 O 3 (100) were computed as: where E(X) is the total energy of the corresponding system X.
To visualize the interaction between Cu 55 (O) 2N clusters and g-Al 2 O 3 (100), Fig. 4 also shows the charge density differences (CDDs) of the systems. CDD is calculated as: where r(X) is the electron density of system X. The binding energies, geometric parameters, and the net charge transfer from the substrate to the cluster are listed in Table 1.
As expected, the Cu 55 (O) 2N clusters become more distorted on the Al 2 O 3 substrate (Fig. 4). The number of oxygen atoms at the Cu 55 (O) 2N /Al 2 O 3 interface increases from 1 to 3, then to 6 as N increases. The interfacial oxygen atoms are closer to the Al 2 O 3 surface than Cu atoms. The Cu-O bond lengths across the interface are always greater than 2.0 Å, while the O-Al bonds increase from 1.83 Å to 1.91 Å (Table 1), i.e. the distance between the cluster and substrate increases with N. The corresponding binding energies decrease from 2.87 eV to 2.17 eV, and CDDs in Fig. 4 show how the interaction at the cluster-support interface changes as there is more weight on the cluster itself for Cu 55 (O) 40 . According to the Bader charge analysis (Table 1) To understand how the electronic structure is modified due to oxygen coverage, we have performed a detailed analysis of    [20][21][22] indicate that the reaction order of oxygen is negative over metallic Cu, which is due to the fact that adsorbed atomic O interacts strongly with the metal surface and blocks the active sites so that CO cannot adsorb. However, there is also opposite evidence 11 which shows that the reaction with CO 2 at high gas pressures occurs both via scattering of gas-phase CO molecules and by CO molecules weakly adsorbed at vacancies in the oxygen adlayer, i.e. the Eley-Rideal (E-R) mechanism contributes to the reaction. Here, we tested the E-R mechanism for Cu 55 (O) 40 and Cu 55 (O) 40 /g-Al 2 O 3 (100) (Fig. S8, ESI †), and the reaction barriers are clearly higher than those for the L-H mechanism. In our calculations, as the oxygen coverage increases up to N = 20, the adsorbed CO molecule and the neighboring O atom come close to one another while binding more weakly on Cu 55 , and this is reflected in the modest reaction barrier height.

Conclusion
We have systematically studied the adsorption of several (N = 1-20) oxygen molecules on the icosahedral Cu 55 cluster using the spinpolarized DFT calculations. Herein, we must emphasize the representative nature of the Cu 55 (O) 2N clusters obtained, as it is very difficult to obtain the true ground states for such numbers of atoms via DFT calculations. Based on these model structures, we focus on the behavior of oxygen on Cu 55 , cluster-support interaction, and the effects of oxygen coverage on CO oxidation.
As the oxygen coverage increases, several adsorbed O atoms gradually move from convex sites, to surface, and to concave sites. The first O atom that spontaneously relaxes to an endohedral site of Cu 55 is observed at N = 13. Furthermore, the endohedral O atoms increase with greater oxygen exposure. The O-O bond cleavage processes on Cu 55 with varying numbers of oxygen atoms are analyzed, and it is observed that O 2 gets activated upon adsorption and it dissociates either spontaneously or with a very small barrier. A short DFT-MD simulation (20 ps) at 300 K confirms the small dissociation barrier and the active movement of oxygen atoms from the surface to the interior sites at larger oxygen coverages. Furthermore, CO oxidation reactions have been studied on Cu 55 (O) 2N and Cu 55 (O) 2N /g-Al 2 O 3 (100) (N = 1, 10, and 20) systems with fully dissociated oxygen molecules. For Cu 55 (O) 2N /g-Al 2 O 3 (100), the interaction between the cluster and the substrate decreases as the distance between them increases with the number of oxygen atoms, N, and the charge transfer across the Cu 55 (O) 2N /Al 2 O 3 interface is weak. The CO oxidation reaction order is positive over oxygen coverage on both Cu 55 (O) 2N and Cu 55 (O) 2N /g-Al 2 O 3 systems with similar reaction barrier heights (0.50-0.65 eV). This is due to the inherently small adsorption energies of the reactants and their modest variation as a function of the varying oxygen coverage and the presence of the Al 2 O 3 support.
Finally, it should be noted that the range of adsorbed oxygen is still limited in this work. The Cu 55 cluster reaches a maximum content of 40 oxygen atoms (i.e. beyond Cu 2 O but not yet at CuO), and, as evidenced by the results (small number of endohedral O atoms; modest reaction barriers), it appears that the cluster can host even more oxygen without losing its catalytic activity. In particular, it is likely that the number of endohedral oxygen atoms will increase further with increasing oxygen content.

Conflicts of interest
There are no conflicts of interest to declare.