Adsorption of O2 and oxidation of CO at Au nanoparticles supported ...

JOURNAL OF CHEMICAL PHYSICS

VOLUME 120, NUMBER 16

22 APRIL 2004

Adsorption of O2 and oxidation of CO at Au nanoparticles supported by TiO2 „110… L. M. Molina, M. D. Rasmussen, and B. Hammera) Interdisciplinary Nanoscience Center (iNANO) and Department of Physics and Astronomy University of Aarhus, DK-8000 Aarhus C, Denmark

共Received 4 December 2003; accepted 27 January 2004兲 Density functional theory calculations are performed for the adsorption of O2 , coadsorption of CO, and the CO⫹O2 reaction at the interfacial perimeter of nanoparticles supported by rutile TiO2 (110). Both stoichiometric and reduced TiO2 surfaces are considered, with various relative arrangements of the supported Au particles with respect to the substrate vacancies. Rather stable binding configurations are found for the O2 adsorbed either at the trough Ti atoms or leaning against the Au particles. The presence of a supported Au particle strongly stabilizes the adsorption of O2 . A sizable electronic charge transfer from the Au to the O2 is found together with a concomitant electronic polarization of the support meaning that the substrate is mediating the charge transfer. The O2 attains two different charge states, with either one or two surplus electrons depending on the precise O2 adsorption site at or in front of the Au particle. From the least charged state, the O2 can react with CO adsorbed at the edge sites of the Au particles leading to the formation of CO2 with very low 共⬇0.15 eV兲 energy barriers. © 2004 American Institute of Physics. 关DOI: 10.1063/1.1687337兴

I. INTRODUCTION

Au/Ceria catalysts for water–gas-shift reaction may not be due to metallic nanoparticles, but rather to adsorbed single Au atoms.17 From the theoretical point of view, the explanations of the low-temperature activity of supported Au nanoparticles are also several. By now, a number of density functional theory 共DFT兲 studies have been carried out for O2 adsorption and CO⫹O2 reaction at various Au aggregates. Ha¨kkinen, Landman and co-workers18 共HL兲 were the first to identify from their calculations the charging of the Au particles as an important parameter. Mavrikakis et al. and Lopez and Nørskov19,20 consider the high relative abundance of lowcoordinated Au sites to be the most important property for obtaining high reactivity at small particles. This view is supported by Hu and co-workers21 as well as by the authors of the present paper,22,23 on the basis of DFT calculations demonstrating a CO–O2 reaction intermediate to form much more readily over edged and stepped than over flat Au surfaces. Mavrikakis et al.24 have found an effect of introducing strain in the Au system, and HL25 emphasize the fluxionality, the ability of cluster systems to undergo large deformations upon adsorption of the reactions, to be of importance. In our previous work for the CO⫹O2 reaction at MgO supported Au nanoparticles, we further propose that reaction sites in which both the metal and the oxide support interact with the reactants play an important role.22,23 Such reaction sites are present only at the perimeter of the Au-support interface and their precise conformation is expected to be very sensitive to the particle size. With the large number of suggestions for important parameters determining the exceptional high reactivity of supported Au particles, there is a call for the theoretical investigation of more systems. In the present paper we present a density functional theory study of the O2 and CO adsorption

Recently, oxide supported Au nanoparticles have attracted considerable attention due to their catalytic activity at low temperatures,1,2 for an increasingly larger number of oxidation reactions.3– 6 The precise activity and durability obtained for a given set of supported Au nanoparticles depend on factors like the choice of oxide support and the particle size distribution, which again depends on the synthesis procedure.7 From the experimental point of view, several suggestions have been put forward for the cause of the exceptional activity of Au nanoparticles. Goodman et al.8 highlight size and band structure effects and measure the onset of the reactivity to appear when the Au particles become as thin as two monolayers at which point they undergo a transition from a metallic to an isolating state. Haruta9 emphasizes the shape effect on the basis of TEM investigations showing better performance of hemispherical particles as opposed to more spherical particles. Along these lines, Pietron et al.10 have observed high reactivity for Au–TiO2 aerogels despite of relatively large particle sizes, proposing that the enlargement of the Au/TiO2 contact area is the important factor. Finally, the support effect is generally present in the experimental studies, showing larger activity of Au clusters supported at reducible oxides like, e.g., TiO2 and Fe2 O3 , yielding larger CO2 production rates than other nonreducible oxides like MgO, Al2 O3 , and SiO2 , under similar conditions.11–13 For reducible oxide supports, the oxygen vacancy defects have been speculated either to improve O2 trapping by the support14,15 or to change the charge state of the Au nanoparticles.16 Finally, Fu et al. have recently proposed that the reactivity of a兲

Author to whom correspondence should be addressed. Electronic mail: [email protected]

0021-9606/2004/120(16)/7673/8/$22.00

7673

© 2004 American Institute of Physics

Downloaded 15 Apr 2004 to 130.225.27.190. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp

7674

J. Chem. Phys., Vol. 120, No. 16, 22 April 2004

and reaction at Au nanoparticle models supported by TiO2 (110) surfaces. The choice of TiO2 (110) as a support is motivated by its widespread use in the experiments26 and by the fact that it is reducible, which allows us to investigate the support effect to some extent. Spurred by the recent STM investigation of O2 diffusion on TiO2 (110) 共Ref. 27兲 the O2 – TiO2 (110) interaction has been considered in great detail by now.28,29 We build on this insight and model the presence of Au nanoparticles at the TiO2 (110) surface with and without bridging oxygen vacancies in the support. In our work on the O2 ⫹CO/Au/MgO system22,23 we found that the O2 was unbound and only became bound in the presence of coadsorbed CO on the Au cluster. In the present work on the O2 ⫹CO/Au/TiO2 system we see that the role of having a reducible oxide is to assist the trapping of O2 molecules, which may now bind either directly to the support or to the Au nanoparticles even in the absence of coadsorbed CO. II. METHOD A. Density functional theory

The DFT calculations are performed using supercells,30 ultrasoft pseudopotentials31 and a plane-wave basis set 共cutoff energy, 25 Rydbergs兲 for the Kohn–Sham electronic states.32 Exchange–correlation 共XC兲 effects are described in the generalized gradient approximation 共GGA兲 with the revised Perdew–Burke–Ernzerhof 共RPBE兲 form.33 Judging from the non-self-consistent Perdew–Wang-91 共PW91兲34 energetics also obtained in the calculations, the relative stability of the systems treated remains largely unaffected by the choice of XC functional, only the RPBE functional shows smaller binding than the PW91 functional.33 For every system reported in this work, an explicit search for nonzero spin states is performed, but only for the system of O2 at the stoichiometric TiO2 (110) surface a finite spin moment is found. The theoretical lattice constants for rutile TiO2 共4.65 Å and 2.98 Å兲, are used throughout this work. For the TiO2 (110)-p(3⫻2) surface unit cells we use two k points in the surface Brillouin zone for the k integrals 关four k points in the case of the TiO2 (110)-p(2⫻2) cell used for the study of CO oxidation兴. The Broyden–Fletcher–Goldfarb–Shanno 共BFGS兲 algorithm35 is employed for structural optimizations until the square-root of the square-sum of the forces on all the relaxing atoms is less than 0.01 eV/Å times the number of relaxing atoms. This effectively ensures that the average force on each atom upon convergence is smaller than 0.01 eV/Å. Transition state searches are done using constrained minimization.36 –38 Here, a chosen reaction coordinate is fixed at a sequence of values while all other (3N⫺1 with N dynamic atoms兲 coordinates are allowed to relax. We verify in each case that a continuous path is obtained by plotting the potential energy—not as a function of the constrained coordinate—but as a function of the length in 3N dimensional space between the relaxed structures. B. Modeling Au clusters

Since nanosized Au clusters are presently too large to be described explicitly in the framework of our DFT setup, we

Molina, Rasmussen, and Hammer

choose to model only the Au-nanocluster/TiO2 (110) interface perimeter. Following Ref. 22, the nanoparticle is replaced with a one-dimensional rod, which on one side has Au atoms in exactly the same local bonding situation as the Au atoms at the edge of the nanoparticle. The Au atoms on the other side of the rod only serve to setup the correct boundary conditions. As we are interested in simulating large clusters with an edge length of the order to 2 nm 共⬇7 Au atoms兲, the approximation of a finite particle by a rod becomes meaningful, as it describes the majority of active sites at the perimeter of the cluster 共with the exception of corner sites兲. For computational reasons, we consider only epitaxially grown Au rods. Considering the experimental lattice constants 共4.08 Å for Au, 4.58 Å and 2.95 Å for TiO2 ) we realize that such Au rods must be strained by 2% and 11% ¯ 0 兴 directions, rewhen oriented in the TiO2 - 关 001兴 and -关 11 spectively. In order to minimize the strain, we therefore choose the TiO2 - 关 001兴 orientation of the Au rods, noting, however, that the opposite orientation would be also interesting since the Au rod would then be intersecting the Ti5c troughs, that have been shown to be the main diffusion channels of adsorbed O2 . 27 For extended, strained Au layers at TiO2 (110), the Au atoms have been shown to be in registry with the bridging oxygen atoms.39 In the present work we therefore align the Au rods so that the side which models the edge of a Au nanocluster has Au atoms atop the bridging oxygen atoms. Moving in our calculations the Au rods by half a lattice spacing in the TiO2 - 关 110兴 direction causes a noticeable energy increase of 0.3 eV 共per unit cell along the rod direction兲 which further justifies this choice. The structure of nanosized Au clusters supported by TiO2 (110) depends on the preparation conditions.7 It is however believed that, as no strong metal-support interaction is present,39 the Au共111兲 facet 共being the most stable one兲 often forms the interface to the oxide. This is consistent with a Wulff–Kaichew construction40,41 which further would predict either Au共111兲 or Au共100兲 facets at the edge of the cluster. Experimentally, Madey’s group has found coexistence of Au共111兲 and Au共112兲 faces at the interface42 关the latter one, presumably, driven by the intrinsic corrugation of the rutile TiO2 (110) surface兴 with unstrained Au islands. Henry and co-workers have, on the other hand, found strained, epitaxially grown, small Au particles on anatase TiO2 (101). 43 In the present work, we choose to investigate the two cluster edges depicted in Figs. 1共a兲 and 1共b兲. These edges cover the situations of both large and small contact angles between the Au and the oxide support, which would be the result of large and small Au–oxide adhesion energies, respectively. The larger contact angle or adhesion energy in Fig. 1共a兲 compared to Fig. 1共b兲 could be caused by, e.g., the presence of oxide defects at the metal–oxide interface, or could be the consequence of finite size effects as argued in Refs. 22 and 23. The five-layer fcc–Au rods shown in Figs. 1共a兲 and 1共b兲 have been relaxed in a p(1⫻2) surface unit cell at a three tri-layer TiO2 (110) slab. The bright Au atoms and the first TiO2 (110) tri-layer atoms are fully relaxed, while the dark Au atoms are only relaxed in the vertical direction. For adsorption and reaction studies, Au rods in p(N⫻2), N⫽2,3



Adsorption of O2

7675

FIG. 1. 共a兲 and 共b兲 One-dimensional Au aggregates supported on the rutile TiO2 关 110兴 with edges 共on the left-hand sides兲 modeling edges of Au particles having either large or small substrate-Au contact angles 共i.e., either ‘‘sharp’’ or ‘‘rounded’’ Au particle terminations兲. Bright Au atoms are fully relaxed, while dark Au atoms are only allowed to relax perpendicular to the support surface. 共c兲 View of a 1D Au rod supported on the TiO2 (110) surface, aligned along the 关001兴 direction. The shaded rectangle 共and dark Au atoms兲 highlights the size of the employed unit cell along the 关001兴 direction. The bridging oxygen rows where vacancies are introduced are highlighted as either hatched 共for O vacancies just below the Au edge兲 or plain white circles 共for O vacancies outside of the Au/TiO2 interface兲. 共d兲 Top view of the p(3⫻2) unit cell, defining the T1 and T2 O2 binding sites relative to an O vacancy 共shaded circle兲.

surface unit cells over four (two relaxed⫹two static) trilayer TiO2 (110) slabs are required. The longer cells are needed to accurately model the presence of isolated bridging oxygen vacancies in the TiO2 (110) surface,28 and to reduce adsorbate–adsorbate interactions along the rod edge. The thicker TiO2 (110) slab is mandatory for an accurate description of the relaxation 共and hence energetics兲 of the bridging oxygen vacancy as well as of the O2 adsorption at the TiO2 surface—the use of only three tri-layers in the support, e.g., cause errors of the order of 0.5–1.0 eV in the vacancy formation energy.28 To make such larger super cells feasible, we have to reduce the size of the Au rods. From the relaxed five-layer rod structures we therefore extract the positions of the bright Au atoms and freeze most of the Au atoms relaxing now only a single or two rows of Au atoms along the edge of the rod 共which ones will become apparent below兲. The final computational setup is shown in Fig. 1共c兲 and two different O2 adsorption sites are defined in Fig. 1共d兲. The present calculations 共with 144 substrate Ti⫹O atoms and either 15 or 21 Au atoms depending on the shape of the Au particle兲 require more than one wall clock week for relaxing the ionic coordinates when run in parallel on eight 2.0 GHz PC’s.

FIG. 2. 共Color兲共a兲 Partially relaxed structure of O2 on top of a Ti trough atom at the stoichiometric TiO2 surface, with the O2 adsorption potential energy and O–O bond length indicated below. Lighter small circles represent the adsorbed O2 molecule. 共b兲 and 共c兲 Relaxed structures, binding energies, and O–O bond lengths for O2 bonded at the Ti trough or leaning towards a supported Au rod, respectively. 共d兲 Density of states 共DOS兲 around the Fermi energy 共0 eV兲 for O2 adsorbed at stoichiometric TiO2 关see panel 共a兲兴. As there is a small (0.5 ␮ B ) magnetic moment on O2 , both spin-up and spin-down DOS are shown. 共e兲 DOS around the Fermi energy for O2 adsorbed on top of a Ti trough atom at stoichiometric TiO2 with a supported Au rod 关see panel 共b兲兴. Labels indicate the individual character of each O2 -derived or Au-derived eigenstate. 共f兲 Contour plot for the square modulus of the two occupied states closest to the Fermi energy in 共e兲关labeled as 共**兲兴. The norm has been z integrated and plotted along x and y, for easier visualization. Transparent circles indicate the location of the O2 molecule, the Ti trough atom, and the Au rod atoms. 共g兲 Real part of one of the O2 2 ␲ * eigenvalues 关labeled as 共*兲 in 共d兲兴 plotted in a vertical plane passing through the axis of the adsorbed O2 molecule 关dashed line in 共f兲兴. Transparent circles indicate the location of adsorbed O2 and several substrate atoms.

III. RESULTS A. O2 at the stoichiometric TiO2 „110… surface

We start by investigating the adsorption of molecular O2 at the stoichiometric TiO2 (110) surface without and with

some Au cluster present. Without Au, the stoichiometric surface is incapable of binding an O2 molecule, with relaxation of the system leading to desorption of O2 from the surface. In Fig. 2共a兲 we plot the structure and adsorption potential


7676


energy after some initial relaxation, with an O2 -substrate distance comparable to the bound configurations found in the presence of either Au or substrate vacancies. We find the adsorption strongly endothermic 共by more than 1 eV兲 as has been reported in detail previously.28,29 The unbound O2 molecule retains a molecular bond length of 1.28 Å close to the calculated value for the gas phase O2 , indicating a negligible charging. Inspection of the density of states 共DOS兲 for this system around the Fermi energy 关see Fig. 2共d兲兴 shows the presence of O2 2 ␲ * antibonding molecular orbitals (1 ␲ * g in the homonuclear notation兲 within the band gap of TiO2 . Although there is a sizable interaction with the substrate, that results in a small O2 magnetic moment, the O2 2 ␲ * orbitals are largely unoccupied, meaning that O2 remains very weakly charged. With a Au cluster present the situation changes dramatically: O2 adsorption on top of a Ti trough atom 关Fig. 2共b兲兴 is strongly stabilized rendering a net O2 binding energy of ⫺0.45 eV. The O2 bond length is increased to 1.41 Å indicating a charge state of the O2 molecule close to O⫺2 2 . The role of the Au cluster at the stoichiometric TiO2 surface is therefore that of donating electrons to the molecule. This charge transfer from Au to adsorbing O2 can be identified in the calculated DOS and charge density rearrangements. In Fig. 2共e兲 we plot the DOS for the system shown in Fig. 2共b兲. The presence of the supported Au rod results in several Auderived states located around the Fermi energy, that strongly interact with the O2 2␲* orbitals 共the location of both Au related and O2 2␲* orbitals are indicated in the figure, including those ones located within the TiO2 valence band兲. Figure 2共f兲 shows the z-integrated square modulus 兩 ␺ 兩 2 for the two eigenstates closest to the Fermi energy 关labeled as 共**兲兴. These states have both a mixed Au/O2 character 共with different weights on O2 and the Au rod in each case兲, clearly indicating that electrons are transferred into O2 from the Au rod. As a result of this interaction, the O2 2␲* orbitals now become fully occupied. This electron transfer is apparently mediated by the support through polarization effects: Fig. 2共g兲 shows the real part 共plotted in a vertical plane passing through the O2 molecular axis兲 of one of the O2 2␲* orbitals 关labeled as 共*兲兴. A substantial interaction with in-plane oxygen 2p orbitals is evident. It is also interesting to note that upon O2 adsorption there is a strong relaxation of the TiO2 support, mainly localized at the Ti trough atom, which is lifted up by as much as 0.8 Å. Finally, we also find a stable adsorption configuration with the O2 molecule coordinating with one end to the Ti5c trough site and with the other end to an Au atom 关see Fig. 2共c兲兴. This configuration with O2 ‘‘leaning’’ towards the Au edge is characterized by a 0.35 eV weaker binding and a shorter O–O bond length 共1.34 Å兲 than the one with O2 adsorbed in the Ti trough. The shorter bond length correlates with a smaller degree of charging of the O2 molecule, that would correspond to an O⫺ 2 charge state. B. O2 at the reduced TiO2 surface

Next, we study the adsorption of O2 at a reduced TiO2 (110) surface, modeled by introducing a bridging oxy-


FIG. 3. Equilibrium structures, adsorption potential energies and O–O bond lengths of O2 adsorbed on top of a Ti trough atom (T1 position兲 for reduced TiO2 surfaces. 共a兲 Clean TiO2 reduced surface. 共b兲 A single Au atom adsorbed at the vacancy. 共c兲 and 共d兲 A ‘‘sharp’’ supported Au edge with the vacancy below or outside the Au edge, respectively. 共e兲 A ‘‘rounded’’ Au edge with the vacancy below it. The semitransparent dashed circle highlights the position of the vacancy.

gen vacancy in the p(3⫻2) unit cell 共 61 coverage兲. In Fig. 3 we present the calculated structures and energetics for various situations with O2 bonded at a Ti trough site. Now, the O2 adsorption at this site becomes highly exothermic even in the absence of Au 关Fig. 3共a兲兴. With the bridging oxygen vacancy, the TiO2 (110) surface has two unbalanced electrons which prefer to reside on the O2 molecule that consequently expands to a bond length of 1.42 Å.28,29 This O2⫺ 2 adsorbate becomes less stable by about 0.4 eV if the TiO2 surface further accommodates a Au atom in the vacancy site 关Fig. 3共b兲兴, while it becomes more stable by ⬇0.5 eV if some Au cluster resides at the surface either over the vacancy or at a distance from it 关Figs. 3共c兲–3共e兲兴. The destabilization of the O2 by the single Au atom we attribute to the competition between the highly reactive Au atom and the O2 for the



FIG. 4. Same as Fig. 3, for ‘‘leaning’’ configurations where O2 binds simultaneously a Ti trough atom and a Au one. The special case where O2 binds directly at the vacancy site 共a兲 is also included. In 共a兲–共d兲 the T1 position for the O2 is used while in 共e兲–共f兲 the T2 position is used.

electrons of the vacancy. The Au atom itself adheres by 1.1 eV more strongly to the vacancy site of the reduced surface than to the Ti5c site of the stoichiometric surface, suggesting an accumulation of charge at the Au atom44 that makes the transfer of the vacancy electrons to O2 more difficult. On the contrary, the Au rods adhere with only 0.05 eV difference 共and with an overall adhesion energy virtually thermoneutral兲 at or far from the vacancy since in these systems all Au atoms are in more metallic surroundings; besides, when supported at the vacancy the Au atom right on top of it prefers to bind to the rod rather than move down and attach to the substrate Ti atoms. This reflects that these larger Au systems do not interact strongly with the vacancy electrons.45 Actually, the enhanced binding of O2 and its larger bond distance 共1.46 Å兲 suggest that the Au rod provides additional charge to O2 共through the same charge-transfer process observed for the stoichiometric TiO2 surface, see Fig. 2兲, instead of taking charge from the vacancy.

Adsorption of O2

7677

Inspection of Figs. 3共c兲 and 3共e兲 shows that the O2 interaction with the large and small contact angle Au particles is very similar. The bond length becomes 1.46 Å in both cases, only the adsorption at the small contact angle Au particle edge is less stable by 0.24 eV. The activity of the two edges will, however, become much different since the CO adsorption strongly discriminates between these two particles edges as we shall report below. We next consider for a reduced surface adsorption configurations of O2 analogous to the one shown in Fig. 2共c兲. Figure 4 illustrates these most important configurations of the O2 molecule where it coordinates with one end to the Ti5c trough site and with the other end to either an uncovered vacancy 关Fig. 4共a兲兴, a single Au atom adsorbed at a vacancy 关Fig. 4共b兲兴, or a Au atom within a Au rod 关Figs. 4共c兲 and 4共f兲兴. For a clean reduced TiO2 surface, we find the configuration with O2 leaning into a vacancy site only 0.1 eV short of being the most stable O2 configuration at the reduced surface. The most stable configuration involves an O2 adsorbed symmetrically in the vacancy site, still with a molecular axis perpendicular to the bridging oxygen row.28 A comparison of the results in Figs. 3共a兲 and 4共a兲 shows that the binding energy increases by about 0.3 eV when the O2 molecule is moved from the trough towards the vacancy, with only a small change in the O–O bond length. On the contrary, a 0.5–1.0 eV destabilization of the O2 molecule is found upon moving it from over the Ti5c site 关Figs. 3共b兲–3共d兲兴 to the leaning configurations of Figs. 4共b兲– 4共d兲. Also, when the O2 molecule bonds to Au the O–O bond length is shortened to ⬃1.36 Å indicating that the charge state of the molecule reduces to O⫺ 2 . In Fig. 4共e兲 we study an adsorption configuration in which O2 is leaning towards the model Au cluster at the T2 position, i.e., one lattice site away from the vacancy. This site turns out to be slightly preferred by about 0.1 eV over the T1 position 关Fig. 4共d兲兴 showing that the precise O2 -vacancy registry is of minor importance. Finally, in Fig. 4共f兲 we evaluate the effect of having a different Au particle edge. As for the O2 adsorption in the Ti trough, the two different Au particle edges modeled cause very similar adsorption energetics. For the leaning adsorption configuration, comparison of Figs. 4共e兲 and 4共f兲 shows even a smaller binding energy difference. The identification in our work that the charge state of the O2 is different for O2 at the TiO2 (110) trough from what it is for O2 leaning against and bonding to the Au has profound implications for the understanding of how O2 can be activated at the Au systems. The charge state of a TiO2 surface supporting some Au nanoparticles must depend on the density of bridging oxygen vacancies, and on the amount of adsorbed O2 . Very high O2 coverages are indeed expected at the working conditions of real catalysts. The vacancies are electron charge donors, which however are passified by O2 adsorption. Also these donor states may be depleted if the electron bands are bent upwards at the surface under the influence of an external potential, or if other acceptor states 共impurities, for example兲 are also present in the vicinity of the surface. Thus, depending on the O2 surface coverage and external potential, an adsorbing O2 might not be able to obtain two electrons from the support as in Fig. 3, but only one


7678


FIG. 5. Structures and CO binding energies at 共a兲 a single Au atom adsorbed site, 共c兲 a ‘‘sharp’’ Au edge on top of a vacancy site, and 共e兲 a ‘‘rounded’’ Au edge on top of a vacancy. 共b兲, 共d兲, and 共f兲 CO binding with O2 preadsorbed at the Ti trough.

electron from the Au nanoparticle as in Fig. 4. A straightforward relationship is found between the amount of charge available to be transferred to O2 from a vacancy and the energy required to activate O2 to configurations leaning against the Au systems: 0.8 –1.0 eV when two electrons are available 关Au particle on reduced TiO2 , cf. e.g., Figs. 3共c兲 and 4共c兲兴, 0.45 eV when a Au atom sits in a vacancy and therefore only one electron is available 关cf. Figs. 3共b兲 and 4共b兲兴, and only 0.35 eV when the vacancy is absent 关Au particle on stoichiometric TiO2 , cf. Figs. 2共b兲 and 2共c兲兴. Having realized that the leaning configurations of Fig. 4 are attainable under realistic conditions, it is natural to consider them as the starting configurations of O2 that is reacting with CO, since here O2 is only moderately strongly bound, and the O2 is in the immediate proximity of any CO adsorbed at the Au.


FIG. 6. Structures and reaction energetics for the oxidation of CO at the perimeter of a Au nanoparticle with ‘‘rounded’’ shape. Left column, bridging oxygen vacancy situated just below the nanoparticle edge; right column, vacancy situated outside of the edge.

C. CO oxidation

We will finally consider the adsorption of CO at supported Au atoms and clusters, and its reaction with O2 . In Fig. 5 we show the results for the coadsorption of CO and O2 at a single Au atom adsorbed at a vacancy and either ‘‘sharp’’ or ‘‘rounded’’ Au rods supported on the reduced TiO2 surface. For a single Au atom at an oxygen vacancy 关see Figs. 5共a兲 and 5共b兲兴, we find a very weak CO binding 共⫺0.11 eV兲. This fact has already been noted by Vittadini and Selloni46 for Au atoms in oxygen vacancies at the anatase TiO2 (101) surface, and explained in terms of a negative charging of the Au atom that opposes the electron donation from CO and prohibits CO from binding. With an additional O2 molecule adsorbed next to the Au atom, we find that the binding of CO is enhanced 共⫺0.32 eV兲, although still relatively moderate. This enhancement can be attributed to the charge redistribution occurring upon O2 adsorption, with O2 being able to remove part of the charge transferred to the Au atom from



the vacancy. With the weak binding of the reactants 共0.64 eV for O2 ; 0.32 eV for CO兲 the single Au atoms adsorbed in the vacancies appear promising for low temperature catalysis. Next, we have evaluated CO binding at ‘‘sharp’’ Au cluster boundaries 关Figs. 5共c兲 and 5共d兲兴. Both with or without preadsorbed O2 we find a very weak binding of CO to the Au edge in contact with the substrate. Such anomalous low binding has also been encountered for similar Au edges supported on MgO,22 and is due to an unfavorable relative orientation of the CO molecule around the edge, forced by the presence of the substrate. The binding of CO at the second Au layer is even more unfavorable, due to the high coordination 共9兲 of that atom, that virtually belongs to a Au共111兲 facet 关since the second layer Au atoms are unrelaxed, the present ⫹0.26 eV result is more endothermic than the ⫹0.05 eV found on relaxed Au共111兲 facets22兴. Finally, we find that CO is able to bind rather strongly 共⬇⫺0.5 eV兲 to the low-coordinated atoms of the ‘‘rounded’’ cluster situated at the second Au layer. The binding is only weakly dependent by 0.06 eV on the presence of O2 at the substrate. As rounded Au particles are the only case where strong CO–Au binding is found, we focus the study of the CO ⫹O2 →CO2 ⫹O(ads) reaction on this system. The equilibrium structures and energetics of the reaction at such cluster terminations, with an oxygen vacancy either below the Au particle 共left column兲 or outside of it 共right column兲 are shown in Fig. 6. In these simulations, smaller slab thickness 共three tri-layers兲 and a p(2⫻2) unit cell were employed, in order to reduce the computational cost of the simulations and allow for an extensive search of the transition states of the CO⫹O2 reaction. Favorable error cancellation from the use of a smaller substrate both in the 关001兴 and 关110兴 directions 关i.e., O2 adsorption energies in the smaller system are comparable to the ones found in the larger one, cf. Fig. 4兴 renders the use of this model meaningful in connection with the models used up to now. We find very similar results in the two situations considered with the vacancy either outside or below the Au particle. In both cases, CO is able to move down from the upper Au edge and react with O2 by overcoming a very small energy barrier of around 0.15 eV. Most of the energy barrier comes from the initial distortion downwards of the CO molecule from its most stable location on top of a Au edge atom; after that distortion, CO is found to approach O2 moving in a flat potential energy surface, that is, the total energy of the system barely changes with decreasing CO–O2 distances. Then, when the distance between the carbon atom in CO and the oxygen atom in O2 is below 1.8 Å the energy suddenly drops, a finding that correlates with the formation of CO2 and the dissociation of the O2 molecule. We also find that as CO approaches O2 , the O2 bond length uniformly expands, from the initial value of 1.36 Å 共with CO adsorbed at the upper edge兲 to a value around 1.45 Å in the vicinity of the transition state. Afterwards, this bond length further expands and leads to the complete dissociation of O2 , with the remaining oxygen atom bonded on top of a Ti trough atom. In our previous work on CO oxidation over Au particles supported by stoichiometric MgO 共Ref. 22兲 we stressed the importance of the formation of a meta-stable CO•O2 reac-

Adsorption of O2

7679

tion intermediate, that binds to several low-coordinates Au edge atoms and to the support. In the present case of Au particles supported on TiO2 , the CO•O2 is no longer formed as a metastable reaction intermediate, only as a snapshot along the CO–O2 →CO2 ⫹O reaction pathway. Also, there is another more fundamental difference between the reaction at these two different substrates: for MgO, the reaction follows an Eley–Rideal mechanism, O2 coming from the gas phase and its binding being assisted by preadsorbed CO; on the contrary, on TiO2 a Langmuir–Hinshelwood mechanism is possible, with both CO and O2 having stable adsorption configurations and reacting once they have reached them. Overall, the role of substrate interactions is much more dramatic for TiO2 : in the case of MgO, a charge transfer from the substrate to the adsorbates is found, thereby stabilizing them, whereas for TiO2 the adsorbates 共and intermediate states兲 actually bind to the substrate. IV. CONCLUSIONS

We have presented DFT calculations for the adsorption of O2 and CO and for the CO⫹O2 reaction at Au aggregates modeling the interfacial perimeter of nanoparticles supported by rutile TiO2 (110). For the stoichiometric TiO2 (110) surface, the surface cannot bind O2 without Au nanoparticles. With Au nanoparticles 共or Au atoms兲 the O2 binds strongly to the TiO2 (110) surface. The Au provides electron charge transfer to the O2 and the TiO2 (110) supports the resulting charged O2 through charge polarization. For the reduced TiO2 (110) surface, the Au nanoparticles enhance the O2 binding for binding sites at the Ti trough. Also, a ‘‘leaning’’ O2 configuration appears in which the O2 bridges a Ti trough atom and a Au atom in the nanoparticle. The O2 charges with about two and one electrons in the trough and leaning configurations, respectively. From the least charged state, the O2 can react with CO adsorbed at the edge sites of the Au particles leading to the formation of CO2 with very low 共⬇0.15 eV兲 energy barriers. ACKNOWLEDGMENTS

This work was supported by The Danish Research Councils and Dansk Center for Scientific Computing. We thank A. Selloni for providing a preprint of Ref. 29. 1

M. Haruta, T. Kobayashi, H. Sano, and N. Yamada, Chem. Lett. 2, 405 共1987兲. 2 M. Haruta, Catal. Today 36, 153 共1997兲. 3 G. C. Bond, P. A. Sermon, G. Webb, D. A. Buchanan, and P. B. Wells, J. Chem. Soc., Chem. Commun. 13, 444 共1973兲. 4 D. Andreeva, V. Idakeiv, T. Tabakova, A. Andreev, and R. Giovanoli, Appl. Catal., A 134, 275 共1996兲. 5 T. Hayashi, K. Tanaka, and M. Hatura, J. Catal. 178, 566 共1998兲. 6 A. Ueda and M. Haruta, Gold Bull. 共Geneva兲 32, 3 共1999兲. 7 A. Wolf and F. Schu¨th, Appl. Catal., A 226, 1 共2002兲. 8 M. Valden, X. Lai, and D. W. Goodman, Science 281, 1647 共1998兲. 9 M. Haruta, CATTECH 6, 102 共2002兲. 10 J. J. Pietron, R. M. Stroud, and D. R. Rolison, Nano Lett. 2, 545 共2002兲. 11 R. J. H. Grisel and B. E. Nieuwenhuys, J. Catal. 199, 48 共2001兲. 12 M. M. Schubert, S. Hackenberg, A. C. van Veen, M. Muhler, V. Plzak, and R. J. Behm, J. Catal. 197, 113 共2001兲. 13 A. C. Gluhoi, M. A. P. Dekkers, and B. E. Nieuwenhuys, J. Catal. 219, 197 共2003兲. 14 M. Manzoli, A. Chiorino, and F. Boccuzzi, Surf. Sci. 532, 377 共2003兲.


7680 15


F. Boccuzzi, A. Chiorino, M. Manzoli, P. Lub, T. Akita, S. Ichikawa, and M. Haruta, J. Catal. 202, 256 共2001兲. 16 G. C. Bond and D. T. Thompson, Gold Bull. 共Geneva兲 33, 41 共2001兲. 17 Q. Fu, H. Saltsburg, and M. Flytzani-Stephanopoulos, Science 301, 935 共2003兲. 18 A. Sanchez, S. Abbet, U. Heiz, W. D. Schneider, H. Ha¨kkinen, R. N. Barnett, and U. Landman, J. Phys. Chem. A 103, 9573 共1999兲. 19 M. Mavrikakis, P. Stoltze, and J. K. Nørskov, Catal. Lett. 64, 101 共2000兲. 20 N. Lopez and J. K. Nørskov, J. Am. Chem. Soc. 124, 11262 共2002兲. 21 Z.-P. Liu, P. Hu, and A. Alavi, J. Am. Chem. Soc. 124, 14770 共2002兲. 22 L. M. Molina and B. Hammer, Phys. Rev. Lett. 90, 206102 共2003兲. 23 L. M. Molina and B. Hammer, Phys. Rev. B 共to be published兲. 24 Y. Xu and M. Mavrikakis, J. Phys. Chem. B 107, 9298 共2003兲. 25 H. Ha¨kkinen, S. Abbet, A. Sanchez, U. Heiz, and U. Landman, Angew. Chem., Int. Ed. Engl. 42, 1297 共2003兲. 26 U. Diebold, Surf. Sci. Rep. 48, 53 共2003兲. 27 R. Schaub, E. Wahlstro¨m, A. Rønnau, E. Lægsgaard, I. Stensgaard, and F. Besenbacher, Science 299, 377 共2003兲. 28 M. D. Rasmussen, L. M. Molina, and B. Hammer, J. Chem. Phys. 120, 988 共2004兲. 29 X. Wu, A. Selloni, M. Lazzeri, and S. K. Nayak, Phys. Rev. B 68, 241402 共2003兲. 30 M. C. Payne, M. P. Teter, D. C. Allan, T. A. Arias, and J. D. Joannopoulos, Rev. Mod. Phys. 64, 1045 共1992兲.

Molina, Rasmussen, and Hammer D. Vanderbilt, Phys. Rev. B 41, 7892 共1990兲. W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 共1965兲. 33 B. Hammer, L. B. Hansen, and J. K. Nørskov, Phys. Rev. B 59, 7413 共1999兲. 34 J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. J. Singh, and C. Fiolhais, Phys. Rev. B 46, 6671 共1992兲. 35 D. C. Liu and J. Nocedal, Math. Program. 45, 503 共1989兲. 36 B. Hammer, K. W. Jacobsen, and J. K. Nørskov, Phys. Rev. Lett. 69, 1971 共1992兲. 37 J. J. Mortensen, Y. Morikawa, B. Hammer, and J. K. Nørskov, J. Catal. 169, 85 共1997兲. 38 A. Alavi, P. Hu, T. Deutsch, P. L. Silvestrelli, and J. Hutter, Phys. Rev. Lett. 80, 3650 共1998兲. 39 N. Lopez and J. K. Nørskov, Surf. Sci. 515, 175 共2002兲. 40 G. Wulff, Z. Kristallogr. 34, 449 共1901兲. 41 W. L. Winterbottom, Acta Metall. 15, 303 共1967兲. 42 F. Cosandey and T. E. Madey, Surf. Rev. Lett. 8, 73 共2001兲. 43 S. Giorgio, C. R. Henry, B. Pauwels, and G. Van Tendeloo, Mater. Sci. Eng., A 297, 197 共2001兲. 44 A. Vijay, G. Mills, and H. Metiu, J. Chem. Phys. 118, 6536 共2003兲. 45 E. Wahlstro¨m, N. Lopez, R. Schaub, P. Thostrup, A. Rønnau, C. Africh, E. Lægsgaard, J. K. Nørskov, and F. Besenbacher, Phys. Rev. Lett. 90, 026101 共2003兲. 46 A. Vittadini and A. Selloni, J. Chem. Phys. 117, 353 共2002兲. 31 32
