Adsorption and dissociation of O2 on Cu(111 ): thermochemistry, reaction barrier and the effect of strain

Adsorption and dissociation of O2 on Cu(111 ): thermochemistry, reaction barrier and the effect of strain

Surface Science 494 (2001) 131±144 www.elsevier.com/locate/susc Adsorption and dissociation of O2 on Cu(1 1 1): thermochemistry, reaction barrier an...

505KB Sizes 4 Downloads 42 Views

Surface Science 494 (2001) 131±144

www.elsevier.com/locate/susc

Adsorption and dissociation of O2 on Cu(1 1 1): thermochemistry, reaction barrier and the e€ect of strain Ye Xu, Manos Mavrikakis * Department of Chemical Engineering, University of Wisconsin ± Madison, 1415 Engineering Drive, Room 2010, Madison, WI 53706, USA Received 3 January 2001; accepted for publication 23 July 2001

Abstract The adsorption and dissociation of dioxygen on the Cu(1 1 1) surface have been studied using periodic self-consistent density functional calculations. Two types of di-r-type chemisorbed molecular precursors are identi®ed: a non-magnetic type located over threefold hollows and a paramagnetic type over bridge sites, both with a binding energy of ca. 0.50 eV with respect to a gas-phase O2 molecule. Atomic oxygen is found to prefer threefold hollows, with a binding energy of ca. 4.3 eV with respect to a gas-phase O atom. Possible pathways for the dissociation of O2 are explored, and the lowest activation energy is calculated to be ca. 0.20 eV. Expansive strain parallel to the surface plane is shown to enhance the binding of atomic and molecular oxygen on the surface as well as to decrease the transition state energy of O2 dissociation. Ó 2001 Published by Elsevier Science B.V. Keywords: Density functional calculations; Catalysis; Chemisorption; Oxidation; Surface chemical reaction; Surface stress; Copper; Oxygen

1. Introduction Dioxygen (O2 ) dissociation on transition metal surfaces is of considerable research and practical interest both because many important industrial oxidation processes are heterogeneously catalyzed by transition metals and feature O2 as the oxidizing agent and because O2 dissociation is the ®rst step toward the corrosion of various metals. Different mechanisms for O2 dissociation have been identi®ed on di€erent metal surfaces [1], including direct dissociation upon adsorption and precursormediated dissociation. Furthermore, the precur* Corresponding author. Tel.: +1-608-262-9053; fax: +1-608262-5434. E-mail address: [email protected] (M. Mavrikakis).

sor may be physisorbed or chemisorbed. In the chemisorbed-precursor mechanism, the incident O2 molecule initially chemisorbs intact. Subsequent thermally driven kinetics determines selectivity between desorption and dissociation. Chemisorbed molecular oxygen species have been detected on a number of transition metal surfaces (see Table 1), generally at temperatures below 150 K. Although the exact nature of the molecular precursors cannot be determined with complete con®dence, vibrational spectroscopy has suggested the existence of a peroxo form (O22 ), with a stretching frequency of 610±650 cm 1 , and a superoxo (O2 ) form, with a stretching frequency of 810±870 cm 1 . A third form, with a stretching frequency of 950±1035 cm 1 , is observed on the Pd(1 1 1) surface [6]. The precursors generally

0039-6028/01/$ - see front matter Ó 2001 Published by Elsevier Science B.V. PII: S 0 0 3 9 - 6 0 2 8 ( 0 1 ) 0 1 4 6 4 - 9

132

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

Table 1 Transition metal surfaces on which molecularly adsorbed O2 has been detected experimentally Surface

m…O±O† (cm 1 )

Cu(1 1 1) Pd(1 1 1) Ag(1 1 0) Ir(1 1 1) Pt(1 1 1) Other surfaces

610 635±650 640 740 650±690 Ni(1 1 1) [15,16]

References 810±870 810±850

950±1035

805 870 Cu(1 1 0) [17]

Ir(1 1 0) [18]

[2] [3±7] [8±10] [11] [12±14]

O±O stretching frequencies, measured with HREELS, are shown here to distinguish between di€erent molecular precursor states.

adopt orientations in which their molecular axes are parallel to the surface (the di-r con®guration). Signi®cant charge transfer from the substrate to the O2 p2p orbital weakens the O±O bond, causing it to lengthen by up to 20% of its gas-phase length and reducing its stretching frequency. Such structural changes suggest that these molecular states are likely precursors to O2 dissociation. Because the existence of molecular O2 precursors in¯uences surface reaction kinetics, it is important to ascertain their characteristics and behavior. Molecularly adsorbed O2 has been detected on the Cu(1 1 1) surface experimentally. Using UV photospectroscopy (UPS) Spitzer et al. found a chemisorbed molecular O2 species that disappeared at 160 K [19]. Prabhakaran et al. reported in an electron energy loss spectroscopy (EELS) study two possible species of molecularly adsorbed O2 on polycrystalline Cu at 80 K [20]. They were characterized by O±O stretching frequencies of 610 and 880 cm 1 , and their bond lengths were esti respectively. Both mated to be 1.65 and 1:48 A species disappeared upon annealing to 160 K. Later in an XPS/UPS study, Rajumon et al. found evidence for the existence of at least one molecular O2 species at 100 K [21]. More recently, Iwasawa and co-workers identi®ed two molecular precursors on Cu(1 1 1) with O±O stretching frequencies of 610 and 820±870 cm 1 in HREELS spectra [2]. The molecular precursors dissociated around 170 K, and in any event no evidence was found for their existence at room temperature. By comparison to similar studies done on other metals, these

authors identi®ed the two precursors as a bidentate peroxo species located over bridge sites and an atop monodentate peroxo species. The properties of molecular oxygen on Cu(1 1 1) have been studied theoretically in the past. In their cluster-type density functional theory (DFT) studies of ammonia oxidation on Cu(1 1 1) [22,23] Neurock et al. found that O2 could adsorb both perpendicular and parallel to the surface, the parallel mode being slightly preferred. The structural and electronic properties of the precursors, however, have still not been well understood, and the dissociation of O2 on Cu(1 1 1) has not been studied theoretically. The adsorption of atomic oxygen on Cu(1 1 1), on the other hand, has been extensively studied under a variety of experimental conditions. Extensive surface reconstruction is observed at medium to high oxygen coverage and room temperature and above, although there is no conclusive evidence for ordered overlayer structures until much higher temperature, probably with the onset of Cu2 O formation [24]. We present here periodic self-consistent DFT calculations performed to investigate the characteristics of various molecular and atomic oxygen con®gurations on Cu(1 1 1) and to explore the most likely pathways of O2 dissociation. This is done both on a ®xed bulk-truncated surface and on a relaxed surface. Because industrial heterogeneous catalysts consist of nanometer-sized metal particles supported on metal oxides that often induce strain on the surfaces of the metals [25±27], we also examine the e€ect of strain on the thermochemistry and on the reaction barrier for O2 dissociation on Cu(1 1 1). The results are discussed in the context of relevant experimental ®ndings. 2. Methods The calculations were performed using DACAPO [28]. Adsorption was allowed on only one of the two surfaces of a …2  2† unit cell used to construct the Cu(1 1 1) surface, with the electrostatic potential adjusted accordingly [29]. The metal slab consists of four layers of Cu atoms, and successive slabs are separated by a vacuum equivalent to six layers of Cu. The Kohn±Sham

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

one-electron valence states are expanded in a basis of plane waves with kinetic energies below 25 Ry. The exchange-correlation energy and potential are described by the generalized gradient approximation (GGA-PW91) [30,31], and ionic cores are described by ultrasoft pseudopotentials [32]. The surface Brillouin zone is sampled at 18 special k points. The self-consistent PW91 density is determined by iterative diagonalization of the Kohn±Sham Hamiltonian, Fermi-population of the Kohn±Sham states …kB T ˆ 0:1 eV†, and Pulay mixing of the resulting electronic density [33]. All total energies have been extrapolated to kB T ˆ 0 eV. The calculated equilibrium PW91 lattice con in good agreement stant for bulk Cu is a ˆ 3:66 A,  [34]. with the experimental value of a ˆ 3:62 A The calculated bond energy and bond length for  respectively, in gas-phase O2 is 5.64 eV and 1:24 A reasonable agreement with the experimental values  [36]. The di€erence of 5.23 eV [35] and 1:21 A between our bond energy of the gas-phase O2 molecule (5.64 eV) and the corresponding number given in Ref. [35] (6.2 eV) is probably due to the fact that the latter result was obtained with the O± O bond ®xed at its experimentally determined length. Adsorption of atomic and molecular oxygen at various sites, as well as possible dissociation pathways, are explored ®rst on a ®xed ideal bulktruncated surface, at the equilibrium lattice con The binding energies …Eb ˆ stant of a ˆ 3:66 A. Etotal Esubstrate Egas-phase adsorbate †, geometries, and magnetic moments of the adsorbates are calculated with respect to a clean ®xed surface. The change in the surface work function is also calculated. The e€ect of surface relaxation on these results is investigated by relaxing the top two layers of the ®xed surface, and the various properties are re-calculated with respect to a clean relaxed surface. The ®xed bulk-truncated surface is then compressed or stretched, whereon the various properties are calculated again. As was performed in a previous study [37], strain is applied only in the plane of the surface and not perpendicularly to it. To compute activation energies, the O±O bond length is chosen as the reaction coordinate. At a number of ®xed points along this coordinate the energy of the system is minimized with respect to

133

all other degrees of freedom in order to produce the optimal intermediate con®gurations along the dissociation pathway. The choice of the O±O bond length as the major component of the reaction coordinate has been validated by calculations on similar systems [38], using the nudged elastic band method [39±42]. The O±O stretching frequency is calculated using the harmonic oscillator approximation. Based on the minimized structure on a relaxed surface, a series of static calculations are performed for an O2 precursor, in which the O±O bond is stretched or compressed to up to 2% of its equilibrium length. The resultant data points (binding energy vs. bond length) are then ®tted to a parabola E ˆ 2 k…x b† =2 , where k is related to the vibrational frequency of the harmonic oscillator via m ˆ p k=M =2p, M being the reduced mass of the O2 molecule.

3. Results and discussion 3.1. Atomic oxygen on Cu(1 1 1) The properties of adsorbed atomic oxygen are calculated at a coverage of 1=4 ML in this study. Four high symmetry sites are explored, as indicated in Fig. 1. Table 2 lists the results. Threefold hollows are clearly preferred to top and bridge sites, with the fcc site slightly more favorable than

Fig. 1. A top view of adsorption sites for atomic oxygen on a Cu(1 1 1) surface. Open circles denote Cu atoms and ®lled circles denote O atoms.

134

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

Table 2  Properties of atomic oxygen adsorbed on a Cu(1 1 1) surface with equilibrium lattice constant …a ˆ 3:66 A†   Site Eb (eV/O) z (A) dCu±O (A) l …lB †

De/ (eV)

Fixed substrate

top bridge hcp fcc

2.35 3.75 4.11 4.20

1.76 1.33 1.21 1.19

1.76 1.86 1.92 1.91

0.34 0 0 0

‡2.49 ‡1.59 ‡1.30 ‡1.23

Relaxed substrate

hcp fcc

4.17 4.29

1.15 1.13

1.92 1.91

0 0

‡1.27 ‡1.23

Experimentala

threefold hollowb; c

4.47d

0:2  0:2c

1:83  0:2c

6 0.02e; f

Oxygen coverage is 1=4 ML. Eb is the binding energy, calculated as Eb ˆ Etotal Esubstrate EO…g† , where the substrate is either a clean ®xed ideal bulk-truncated Cu(1 1 1) surface or such a surface with its top two layers relaxed, and where EO…g† is the total energy of an O atom in the gas phase; z is the height measured from the plane of the centers of mass of the top-layer (®xed ideal bulk-truncated) copper nuclei to the center of mass of the O atom; dCu±O is the distance between O and the nearest Cu; l is the residual magnetic moment of O; and De/ is the change of work function with respect to the clean Cu(1 1 1) surface. a Experimental data taken at room temperature. b Ref. [44]. c Ref. [45]. d Ref. [43]. e Ref. [19]. f Ref. [47].

the hcp site by 0.1 eV. The top site is found to be unstable with respect to small perturbation. Adsorption on the bridge site is not possible on a relaxed surface because Cu atoms would move eventually to accommodate the O atom in an fcc site. The di€erences in the binding energies of the O atom introduced by surface relaxation are small ( 6 0.1 eV). For adsorption in an fcc or hcp site, the three Cu atoms bound to the O atom in the  There is also unit cell relax outward by 0.01 A.  of these Cu atoms lateral movement (0.05 A) away from the O atom to accommodate its presence. The fourth Cu atom of the unit cell, which is  (all not bound to O, relaxes inward by 0.15 A relative to the positions of Cu atoms in the ®xed ideal bulk-truncated (1 1 1) surface). The di€usion of an O atom on the Cu(1 1 1) surface most likely starts from one threefold hollow and proceeds over a bridge site to an adjacent threefold hollow. The minimum energy barrier to di€usion is estimated to be 0.45 eV on both the ®xed and the relaxed surface. As the residual magnetic moments of the adsorbed O atoms (Table 2) and the commensurate increases in the surface work function suggest, there is a net charge transfer to the O atoms, which makes these species ionic in character.

In experiments threefold hollows are generally found to be where O atoms adsorb on Cu(1 1 1) [44,45]. At low temperature a mixture of chemisorbed molecular oxygen and atomic oxygen species are found on the surface. The molecular species are not detectable at temperatures above 170 K. The estimated saturation coverage of atomic oxygen ranges from 0.3 ML [46] to 0.5 ML [45,47,48] at room temperature. At room and moderately higher temperature and below an O2 exposure of 105 L, no ordered oxygen overlayer is observed in most LEED experiments [19,44,45,47,48]. The adsorption of O atoms is found to be either just above or in the top Cu layer. In addition, SEXAFS [45] and ion scattering [47] techniques show that Cu atoms re upon adsorption of O lax outward by 0.3±1.0 A atoms. The formation of copper oxide islands on the surface commences at an O2 exposure of 106 L [44]. Evidence is also found at this exposure level that suggests the incorporation of O atoms into the Cu bulk [49]. Even higher exposures …>109 L† result in the growth of bulk cupper oxide, mostly Cu2 O. Raising the substrate temperature above room level at low pressure (10 5 ±10 3 Pa) also results in the incorporation of O atoms into the Cu bulk [50], while doing so at higher pressure

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

135

(102 ±105 Pa) produces a ®lm of Cu2 O several  thousand Angstroms thick [51]. Superstructures of large dimensions (29 and 44 times the area of the …1  1† substrate unit cell) have been observed by Besenbacher et al. at 670 K, using atomic-resolution STM [24,52±54]. The authors are able to explain those structures in terms of local O±Cu±O coordination, which suggests the onset of Cu2 O formation. The high complexity of the reconstructed surface and the observed large unit cell make it dicult to compare our results with these experimental ®ndings. 3.2. Molecular oxygen precursors on Cu(1 1 1) Calculations for molecular oxygen precursors are carried out with one O2 molecule in every …2  2† unit cell. A number of high symmetry sites are explored, including both perpendicular and parallel (i.e., di-r type) con®gurations. The perpendicular con®gurations are named after the sites they occupy (i.e. bridge, fcc), and the parallel ones are named after the sites occupied by the two O atoms and spanned by the O±O bond. For brevity, site names are abbreviated to their ®rst letters (thus, for instance, t±f±b stands for top±fcc± bridge). The parallel con®gurations that we have identi®ed are shown in Fig. 2, and their properties are listed in Table 3. Self-consistent total-energy calculations based on DFT have recently been used to investigate O2 precursors on metal surfaces such as Ag(1 1 0) [55], Ni(1 1 1) [56], Pt(1 1 1) [57], Pd(1 1 1) [58], and Cu(1 1 0) [59]. The results of these studies have shed much light on the nature of the precursors and found good agreement with experimental results (see Table 4). Our results bear similarity to those of the previous studies. The parallel O2 precursors are more strongly bound on Cu(1 1 1) than the perpendicular O2 precursors are. The latter are calculated to be only marginally stable. The di€erence in binding energy between the perpendicular and the parallel precursors is calculated to be 0.5 eV/O2 . Neurock et al. have found a much smaller di€erence in the binding energy of the two precursors (Eb ˆ 0:10 eV/O2 for the perpendicular mode and 0.18 eV/ O2 for the parallel mode), although the binding energies of the perpendicular precursors that we ®nd

Fig. 2. A top view of adsorption sites for O2 precursors on a Cu(1 1 1) surface: (a) con®gurations, and (b) schematic showing the de®nitions of the spherical angles h and / (as used in Tables 3 and 5). Open circles denote Cu atoms and ®lled circles denote O atoms.

agree closely with their result [23]. The perpendicular precursors only interact with the surface weakly and resemble the gas-phase O2 molecule, as can be seen from their residual magnetic moments and short bond lengths. Their gas-phase-like characteristics, combined with small binding energies, suggest that the perpendicular precursors, if stable at all, would represent physisorbed and not chemisorbed states. Very recently Kresse and coworkers have observed in ab initio molecular dynamics simulations of O2 adsorption on Cu(1 1 0) that at a low translational incident energy of 0.05 eV O2 molecules impact the surface with their bond axes perpendicular to it [59]. Once on the surface, the O2 molecules undergo extensive steering so that eventually most of them end up adsorbing parallel to the surface in the more favorable fourfold hollows. It should be pointed out that, although perpendicularly adsorbed O2 molecules are unlikely to dissociate in their native con®guration, they can present their own reactivity. For instance, Neurock et al. have concluded that the perpendicularly adsorbed O2 molecule acts as the precursor for NH3 decomposition on

136

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

Table 3  Properties of molecular O2 precursors on a Cu(1 1 1) surface with equilibrium lattice constant …a ˆ 3:66 A†   Site Eb (eV/O2 ) z (A) h …°† / …°† dO±O (A) m…O±O† (cm 1 ) l …lB † O2 (g) Fixed substrate

Transition state Relaxed substrate

Transition state Experimental

1.24 top bridge hcp fcc t±b±t t±h±b t±f±b b±h±b b±f±b b±h±b pathway

0.02 0.04 0.04 0.05 0.40 0.49 0.49 0.53 0.52 0.26

2.99 2.35 2.22 2.20 1.89 1.68 1.68 1.61 1.61 1.45

top bridge hcp fcc t±b±t t±h±b t±f±b b±h±b b±f±b b±h±b pathway

0.02 0.06 0.07 0.09 0.45 0.52 0.52 0.56 0.55 0.36

2.99 2.33 2.17 2.16 1.88 1.65 1.65 1.55 1.57 1.35

0.1 10.0 10.2 0.3 0.3 0.3

0.0 10.6 10.7 0.6 0.4 0.9

gas-phase precursor 1 precursor 2

1607

De/ (eV)

2

30.9 0.0 0.0 29.7 29.7 29.4

1.26 1.30 1.31 1.32 1.36 1.44 1.44 1.47 1.47 1.79

1.75 1.26 1.15 1.10 0.93 0 0 0 0 0

‡0.94 ‡1.48 ‡1.63 ‡1.67 ‡1.76 ‡2.04 ‡2.04 ‡1.91 ‡1.93 ‡2.17

30.4 0.0 0.0 29.0 29.6 29.4

1.26 1.30 1.32 1.32 1.35 1.44 1.44 1.48 1.47 1.77

1.76 1.27 1.15 1.10 0.99 0 0 0 0 0

‡0.93 ‡1.49 ‡1.65 ‡1.68 ‡1.77 ‡2.07 ‡2.07 ‡1.85 ‡1.91 ‡1.99

1.21a

1085 954 789 729

1555b 820±870c 610c

Coverage of O2 is 1=4 ML. Eb is the binding energy, calculated as Eb ˆ Etotal Esubstrate EO2 …g† , where the substrate is either a clean ®xed ideal bulk-truncated Cu(1 1 1) surface or such a surface with its top two layers relaxed, and where EO2 …g† is the total energy of an O2 molecule in the gas phase; z is the height measured from the plane of the centers of mass of the top-layer (®xed ideal bulk-truncated) copper nuclei to the center of mass of the O2 molecule; h and / are the spherical angles; dO±O is the bond length; m…O±O† is the O±O stretching frequency; l the residual magnetic moment of the adsorbed O2 ; and De/ is the change in the surface work function. DFT calculations have been performed to obtain the properties of O2 in the gas phase, shown in the ®rst entry, as a reference. a Ref. [36]. b Ref. [14]. c Ref. [2].

Cu(1 1 1) by extracting one of its hydrogen atoms [23]. Also, perpendicularly adsorbed O2 molecules have been proposed to form peroxide anions (OOH ) on Pt in an acidic environment in the electrochemical reduction of oxygen. This takes place, for instance, on the cathode of a fuel cell that uses acidic electrolyte [60]. As was found in DFT studies of parallel O2 precursors on Pt(1 1 1) [57] and Ni(1 1 1) [56], we determined that t±b±t and t±h(f)±b precursors too exist on Cu(1 1 1). However, a third con®guration,

b±h(f)±b, is in fact the most energetically favorable on Cu(1 1 1), although t±h(f)±b and b±h(f)±b states are almost energetically degenerate. On the other hand, as mentioned earlier, Iwasawa and coworkers postulated an atop parallel precursor with both O atoms bound to the same Cu atom [2]. Our calculations indicate that while such a con®guration is possible on the ®xed surface, the O2 molecule cannot retain this position on a relaxed surface. Therefore it is excluded from our consideration. Charge transfer from the substrate has

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

137

Table 4 Molecular O2 precursors identi®ed on several other metal surfaces in previous DFT studies Surface

Precursor species

Properties

Precursor species

Properties

Precursor species

Properties

Ag(1 1 0)a

Chemisorbed in fourfold hollow

[0 0 1] 0.50 eV  1.45 A  1.20 A

Chemisorbed in fourfold hollow

‰1 1 0Š 0.42 eV  1.48 A  0.96 A

Physisorbed on long-bridge site

[0 0 1] 0.37 eV  1.32 A  2.07 A

top±fcc±bridge 0.68 eV  1.43 A  1.78 A

Chemisorbed in threefold hollow

top±hcp±bridge 0.58 eV  1.42 A  1.81 A

Chemisorbed at bridge site

0.12lB

b

Pt(1 1 1)

Chemisorbed in threefold hollow

0.0lB 690 cm

Ni(1 1 1)c

Chemisorbed in threefold hollow

0.25lB

0.0lB 710 cm

1

top±fcc±bridge 1.65 eV  1.47 A  1.62 A

Chemisorbed in threefold hollow

0.22lB

d

1.23lB

top±bridge±top 0.72 eV  1.39 A  1.92 A 0.4lB 850 cm

1

top±hcp±bridge 1.67 eV  1.46 A  1.62 A

Chemisorbed at bridge site

0.22lB

1

top±bridge±top 1.41 eV  1.42 A  1.77 A 0.44lB

Cu(1 1 0)

Chemisorbed in fourfold hollow

[1 1 0] 1.52 eV  1.54 A  1.04 A 0.0lB

Chemisorbed on long-bridge site

[0 0 1] 1.08 eV  1.41 A  1.71 A 0.0lB

Chemisorbed on short-bridge site

[1 1 0] 0.93 eV  1.40 A  1.87 A 0.95lB

Pd(1 1 1)e

Chemisorbed in threefold hollow

top±fcc±bridge 1.01 eV  1.39 A  1.75 A

Chemisorbed in threefold hollow

top±hcp±bridge 0.92 eV  1.41 A  1.79 A

Chemisorbed at bridge site

top±bridge±top 0.89 eV  1.36 A  1.91 A

0.0lB 890 cm

1

0.0lB 830 cm

1

0.3lB 960 cm

1

In the case of Ag(1 1 0) and Cu(1 1 0) where more than three precursors are identi®ed, the stable and energetically more favorable three are shown. The properties, in the order listed, are: axis direction/con®guration; binding energy; O±O bond length; adsorption height; residual magnetic moment; and O±O stretching frequency. All results shown obtained on ®xed substrates. a Ref. [55]. b Ref. [57]. c Ref. [56]. d Ref. [59]. e Ref. [58].

quenched the magnetic moment of all but the t±b±t state, which still carries half …1:0lB † of what a gasphase O2 molecule possesses. This pattern is similar to that of the O2 precursors on Pt(1 1 1) and is in contrast to the Ni(1 1 1) case, where all O2 precursors are magnetic because of the magnetic properties of Ni. The O±O bond of the parallel precursors on Cu(1 1 1) is lengthened up to about 19% of the gas-phase O2 bond length. Surface relaxation introduces a small gain of 0.03 eV in

binding energy for most precursors. Cu atoms not bound to the precursors sink between 0.05 and  in place. Those that are directly bound to 0:14 A the precursors do not show appreciable vertical displacement but move away from the precursors  horizontally (all relative to the posiby 0.05 A tions of Cu atoms in the ®xed ideal bulk-truncated surface). O±O stretching frequencies of t±b±t, t±h±b and b±h±b states are calculated (see Table 3). The

138

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

calculated frequencies are higher than experimental ®ndings, but the two are in reasonable agreement given that GGA-PW91 calculations regularly over-predict vibrational frequencies by 5±10% [61]. The stretching frequencies of adsorbed O2 that we calculated span a range of 220 cm 1 , in good agreement with the range of 260 cm 1 spanned by experimental data. It is observed experimentally that although the 610 cm 1 band has a greater intensity than that of the other species (initially at 820 cm 1 ) at low coverage, the 820±870 cm 1 band dominates at medium to high coverage [2]. Change in the ratio of intensities of the two bands reveals a faster growth for the 820±870 cm 1 species as oxygen coverage increases. This suggests that the 820±870 cm 1 species may be lower in coordination to the surface than the 610 cm 1 species. Since the t±b±t precursor is bound to two Cu atoms, whereas the b±h(f)±b precursor is bound to three Cu atoms, given also the much lower O±O stretching frequency of the b±h(f)±b species, we suggest that the b±h(f)±b state produces the 610 cm 1 band in HREELS and that the t±b±t state is responsible for the 810±870 cm 1 band. Based on their bond lengths and magnetic properties, the t±b±t precursor can be viewed as superoxo-like, and the b±h(f)±b precursor can be viewed as peroxo-like. The latter is taken to be the most likely precursor to dissociation. The last column in Table 3 gives the change in Cu(1 1 1) surface work function due to O2 adsorption. All chemically important adsorbed O2 states increase the surface work function, which implies a net charge transfer from substrate Cu atoms to the adsorbate. So far we have not found experimental data to compare with these results. 3.3. O2 dissociation on Cu(1 1 1) The dissociation of O2 is also studied with one O2 molecule in every …2  2† unit cell. Thus the initial state represents a molecular coverage of 1=4 ML, leading to a dissociated ®nal state with an atomic coverage of 1=2 ML. We ®nd that a set of adjacent fcc and hcp sites cannot be simultaneously occupied by two O atoms because they would experience very strong repulsion. Therefore immediately after the O±O bond of a precursor is

broken the two O atoms will have to occupy two adjacent fcc or two adjacent hcp sites. A probable pathway starts with the most thermodynamically favorable precursor, b±h±b, and then the O±O bond is stretched until it is broken, upon which the two separate O atoms fall into two adjacent fcc sites. From there they can di€use away from each other on the surface. The energy evolution along this pathway, both on the ®xed and on the relaxed surface, is given in Fig. 3. It can be seen that the transition state occurs early in the process, with a small energy barrier of 0.26 eV/O2 above the initial b±h±b state on the ®xed surface. On the relaxed surface, the activation energy barrier is lowered to 0.20 eV/O2 , in reasonable agreement with the estimated value of 0.08±0.18 eV/O2 of Habraken et al. [48] based on kinetics measurements. The activation energy of O2 dissociation on Cu(1 1 1) is

Fig. 3. Reaction pathway for O2 dissociation from b±h±b to two O atoms in fcc sites on the static and on the relaxed Cu(1 1 1) surface. The binding energy of O2 is calculated as Eb Etotal Esubstrate EO2 …g† where the substrate is either a clean ®xed ideal bulk-truncated Cu(1 1 1) surface or such a surface with its top two layers relaxed. The binding energy of the ®nal point (in®nite separation) is calculated as Eb ˆ 2  Eb;O…fcc† BEO2 …g† where Eb;O…fcc† is the adsorption energy of an O atom in the fcc site and BEO2 …g† is the bond energy of an O2 molecule in the gas phase. The indicated activation energy and heat of reaction are those calculated for the reaction on the relaxed surface. The zero of the energy axis corresponds to the total energy of an O2 molecule in the gas phase and the clean slab at in®nite separation from each other. Alphabetical labels correspond to the selected points shown in Figure 4. Lines are drawn only as a guide for the eye.

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

very similar in size to that calculated on Ni(1 1 1) surface, both of which are smaller by a factor of 4±5 than the activation energy calculated for O2 dissociation on Pt(1 1 1) and Pd(1 1 1) surfaces [58]. The overall O2 dissociation process on Cu(1 1 1) is very exothermic, releasing 2.4 eV/O2 of heat. A similar pathway that leads from an initial b± f±b state to two adjacent hcp sites has also been probed. It shares the same characteristics as the pathway already discussed, only with a slightly higher activation energy. Given the relative magnitudes of the precursor binding energy ( 0.55 eV), activation energy (0.20 eV) and overall reaction exothermicity (2.4 eV), O2 precursor molecules will dissociate upon heating of the surface, instead of desorbing, to yield chemisorbed O atoms. This agrees with the experimental fact that O2 desorption is not detected in TPD [2]. The relaxed surface displays a high level of ¯exibility in the dissociation process. Surface Cu atoms show extensive movement, particularly after the transition state, indicating their substantial involvement in the dissociation process (see Fig. 4 for snapshots of the system along the pathway of Fig. 3). As shown in the snapshot depicting the ®nal state of the dissociation process, the Cu atoms of the top two layers have been substantially displaced by 0.5 ML of atomic oxygen with respect to their bulkterminated positions. This result is in accord with

139

the massive adsorbate-induced surface reconstruction reported by high temperature STM experiments [24]. 3.4. E€ect of strain Strain is introduced in the substrate slab by increasing or decreasing the equilibrium lattice constant of a bulk Cu crystal by a small percentage in the direction parallel to the surface (Fig. 5). We  (4.4% less than the equilihave used a ˆ 3:50 A  to create a compressed brium value of 3:66 A)  (3.8% more) to create a surface and a ˆ 3:80 A stretched surface. The same set of calculations performed before on the equilibrium surface are repeated for the adsorption of atomic and molecular oxygen, as well as for the dissociation pathway, on both the expanded and compressed surfaces. The results are compiled in Table 5. We ®nd that lattice expansion leads to increased stability for all adsorbates. The binding energy of both atomic oxygen and molecular precursors are modi®ed by 0:1 eV by 4% strain. Table 6 shows an additional way to quantify this trend. Since three data points of E vs. ln…a=aeq †, namely  are available for at a ˆ 3:50, 3:66, and 3:80 A, every energy quantity that we calculated, these three points could be ®t to a line for each energy quantity and the slope of this line calculated to obtain dE=d ln…a=aeq †. It can be seen that for a 1%

Fig. 4. Snapshots along the O2 dissociation pathway from the b±h±b state to two fcc sites on the relaxed surface Cu(1 1 1) surface. The coverage is 1=4 ML of O2 initially and corresponds to 1=2 ML of O at the end of the dissociation process. For clarity, only one O2 molecule is shown in each image. Periodically repeated images of adsorbates have been omitted. The numbers that follow denote the  (a) b±h±b (initial state), (b) 1.77 (transition state), (c) 1.78, and (d) 2.10. Panel I gives a side view, and Panel II O±O bond length in A: gives a top view.

140

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144 Table 6 E€ect of strain on the binding energies and on the activation energy of O2 dissociation dE=d ln…a=aeq †

E is . . . Binding energy of O in fcc Binding energy of t±b±t Binding energy of t±h±b Binding energy of b±h±b Activation energy of O2 dissociation Heat of reaction

3.33 2.07 3.46 3.46 1.01

eV eV eV eV eV

3.20 eV

 a is the lattice constant and aeq its equilibrium value (3.66 A).

Fig. 5. E€ect of strain on O2 dissociation from O2 (b±h±b) to two O atoms in fcc sites on a Cu(1 1 1) surface. Surface is not allowed to relax. Zero of the energy axis corresponds to the total energy of an O2 molecule in the gas phase and the clean slab at in®nite separation from each other.

increase in the lattice constant, the binding energies of various adsorbates increase by 0.03 eV. We also calculated the position of the d-band center of the Cu(1 1 1) surface along the b±h±b dissociation pathway. The results are plotted in

Table 5  and a stretched (a ˆ 3:80 A)  Cu(1 1 1) surface (Cu Properties of atomic O and molecular O2 precursors on a compressed (a ˆ 3:50 A) atoms in ideal bulk-truncated con®guration and not relaxed)   Site Eb (eV) z (A) h …°† / …°† dO±O (A) m…O±O† De/ (eV) l …l † B

(cm 1 )

Compressed surface  a ˆ 3:50 A

Atomic O

top bridge hcp fcc

2.39 3.68 3.99 4.06

1.76 1.39 1.29 1.28

Molecular precursor

t±b±t t±h±b t±f±b b±h±b b±f±b b±h±b pathway

0.34 0.36 0.36 0.39 0.40 0.08

1.95 1.74 1.74 1.67 1.67 1.48

Atomic O

top bridge hcp fcc

2.37 3.84 4.23 4.34

1.76 1.28 1.13 1.12

Molecular precursor

t±b±t t±h±b t±f±b b±h±b b±f±b b±h±b pathway

0.51 0.65 0.65 0.68 0.67 0.45

1.86 1.63 1.63 1.54 1.54 1.39

Transition state Stretched surface  a ˆ 3:80 A

Transition state

0.0 8.6 8.4 1.9 0.3 0.1

0.0 11.0 11.3 1.6 0.7 1.5

29.9 0.0 0.0 26.7 30.0 29.8

32.1 0.0 0.0 27.6 29.3 27.6

1.35 1.43 1.43 1.47 1.46 1.82

1.36 1.45 1.45 1.48 1.48 1.79

926 775 736

0 0 0 0

‡2.62 ‡1.74 ‡1.48 ‡1.41

0.95 0 0 0 0 0

‡1.79 ‡2.13 ‡2.13 ‡2.05 ‡2.04 ‡2.35

0.50 0 0 0

‡2.26 ‡1.36 ‡1.04 ‡0.99

0.90 0 0 0 0 0

‡1.60 ‡1.85 ‡1.86 ‡1.66 ‡1.68 ‡1.91

Coverage for atomic and molecular oxygen is 1=4 ML. Eb is the binding energy, calculated as Eb ˆ Etotal Esubstrate EO2 …g† , where the substrate is either a clean compressed or clean stretched Cu(1 1 1) surface, and where EO2 …g† is the total energy of an O2 molecule in the gas phase; z is the height measured from the plane of the centers of mass of the top-layer (®xed bulk-truncated) copper nuclei to the center of mass of the O atom or O2 molecule; h and / are the spherical angles; dO±O is the bond length; m…O±O† is the O±O stretching frequency; l is the residual magnetic moment of the adsorbed oxygen species; and De/ is the change in the surface work function.

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

Fig. 6. The position of the d-band center of the Cu(1 1 1) surface. The d atomic states are projected onto the Cu atom that is bound to both oxygen atoms (except for the clean-slab case). eF is the Fermi level. Represented are the clean slab as well as the initial, transition and ®nal state of the b±h±b dissociation  equilibrium and pathway on the compressed …a ˆ 3:50 A†,  surface. All three surfaces are held at stretched …a ˆ 3:80 A† ®xed ideal bulk-truncated con®guration.

Fig. 6, which shows a consistent upshift of the dband center, and thus increasingly strong surface± adsorbate interaction, with lattice expansion. This is in good agreement with the ®ndings of previous theoretical studies on the e€ect of strain on the reactivity of several metal surfaces [37,62]. As expected, in moving from the clean surface to the surface with the molecular state to the ®nal dissociated state, the center of the d-band for Cu moves downward away from the Fermi level as the surface becomes progressively less reactive. The details of this progressive change in the electronic characteristics of the dissociation pathway are shown in Fig. 7. Because the O2 molecule experiences di€erent coupling to the surface at di€erent stages of the b± h±b dissociation pathway [63], the surface±adsorbate interaction is not identically a€ected by strain throughout the pathway (Fig. 6), and so the activation energy and heat of reaction are also a€ected by strain. Table 6 indicates that for a 1% increase in the lattice constant the exothermicity of the reaction …DH † is increased by 0.03 eV, while the corresponding gain in reactivity, a lowering of the activation energy …Ea † by 0.01 eV, is somewhat less pronounced. The activation energy appears to be less a€ected by strain, but that is simply because

141

the e€ect of strain has almost exactly the same magnitude and sign on the b±h±b and the transition state. This, however, does not mean that the e€ect of strain on the dissociation kinetics is negligible. Because the transition state becomes signi®cantly more stable in going from a ˆ 3:66  which means that the desorption barrier to 3:80 A, …Edes † for the transition state becomes correspondingly higher, the rate of O2 dissociation can outperform the rate of O2 desorption by 1±2 orders of magnitude (Table 7), depending on temperature. Higher temperature tends to diminish this advantage by reducing the di€erence between Ea and Edes in the exponential. According to a recent study of Au supported on TiO2 by Giorgio et al., gold can support lattice expansions of up to 12% [64]. If highly dispersed Cu particles sustain lattice expansion to a similar extent [25±27], our results would suggest that the kinetics and thermochemistry of O2 dissociation and other oxidation reactions on Cu could be substantially a€ected by lattice strain. 4. Conclusions Periodic self-consistent density functional calculations have been performed to study the adsorption and dissociation of dioxygen on the Cu(1 1 1) surface. Atomic oxygen is found to preferentially adsorb in threefold hollows, slightly favoring the fcc site …Eb ˆ 4:3 eV† at hO ˆ 1=4 ML. Surface Cu atoms show appreciable relaxation upon adsorption of atomic O, in accord with experimental ®ndings. Molecular O2 binds to Cu(1 1 1) in several di-r-type, energetically quasidegenerate states. The binding energy of the most favorable state, b±h(f)±b, is found to be 0.55 eV/ O2 at hO2 ˆ 1=4 ML. The O±O stretching frequency of b±h±b is 729 cm 1 and its magnetic moment is completely quenched. For another molecular precursor state, t±b±t, we have calculated a binding energy of 0.45 eV/O2 at hO2 ˆ 1=4 ML, an O±O stretching frequency of 954 cm 1 , and a remaining magnetic moment of 1:0lB . These results suggest that the b±h(f)±b precursor is peroxo-like (O22 ) and correspond to the experimentally observed precursor that produces the

142

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

Fig. 7. Density of states (DOS) for various steps in the O2 dissociation process on Cu(1 1 1). Darker lines represent local d-DOS projected onto a Cu atom in the top layer that is bound to the O2 precursor or O atom; lighter lines represent local s- and p-DOS projected onto one of the two O atoms. The ®rst column of panels (a1±a4) corresponds to the compressed static surface system, the second column (b1±b4) corresponds to the equilibrium static surface system, and the third column (c1±c4) corresponds to the stretched static surface system. The ®rst row (a1±c1) represents the clean surface and gas-phase O2 (discrete states) and serves as a reference. The second row (a2±c2) corresponds to the initial state (b±h±b), the third row (a3±c3) corresponds to the transition state, and the last row (a4±c4) corresponds to the ®nal state (O atoms in two adjacent fcc sites).

610 cm 1 band in HREELS, whereas the t±b±t precursor is superoxo-like (O2 ) and is responsible for the other experimentally observed band located at 810±870 cm 1 .

Dissociation of the molecular O2 precursor encounters a small activation barrier of 0.26 eV/O2 on the ®xed surface. Surface relaxation has a moderate stabilizing e€ect on the precursors (0.03 eV)

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

143

Table 7 E€ect of strain on the kinetics of O2 dissociation via the b±h±b pathway on the ®xed Cu(1 1 1) surface Lattice constant  (A)

Ea (eV)

Edes (eV)

3.50 3.66 3.80

0.31 0.26 0.23

0.08 0.26 0.45

Ea ˆ Eb;transition

state

Eb;b-h-b is the activation energy, and Edes ˆ

and decreases the dissociation barrier by 0.06 eV. Expansive lattice strain is shown to increase the stability of adsorbed atomic oxygen, molecular precursors and transition state as well as to increase the heat of reaction. Expansive strain also greatly enhances the kinetics of O2 dissociation. Acknowledgements This research was supported in part by NSF cooperative agreement ACI-9619020 through computing resources provided by the National Partnership for Advanced Computational Infrastructure. Part of the calculations were performed on DoE supercomputing facilities. Y.X. gratefully acknowledges partial ®nancial support from an NSF-GRT grant (contract # EHR-9554586). M.M. thanks Shell Oil Company Foundation for a Faculty Career Initiation Award, which provided for partial ®nancial support. References [1] P.D. Nolan, M.C. Wheeler, J.E. Davis, C.B. Mullins, Acc. Chem. Res. 31 (1998) 798. [2] T. Sueyoshi, T. Sasaki, Y. Iwasawa, Surf. Sci. 365 (1996) 310. [3] R. Imbihl, J.E. Demuth, Surf. Sci. 173 (1986) 395. [4] P. Sj ovall, P. Uvdal, Chem. Phys. Lett. 282 (1998) 355. [5] P. Sj ovall, P. Uvdal, J. Vac. Sci. Technol. A 16 (1998) 943. [6] P.D. Nolan, B.R. Lutz, P.L. Tanaka, C.B. Mullins, Surf. Sci. 419 (1998) L107. [7] B.A. Sexton, R.J. Madix, Chem. Phys. Lett. 76 (1980) 294. [8] L. Vattuone, M. Rocca, P. Restelli, M. Pupo, C. Boragno, U. Valbusa, Phys. Rev. B 49 (1994) 5113. [9] L. Vattuone, M. Rocca, U. Valbusa, Surf. Sci. 314 (1994) L904. [10] J. Pawelacrew, R.J. Madix, J. Stohr, Surf. Sci. 339 (1995) 23.

D ˆ Ea

Edes

0.23 0.00 0.22 Eb;transition

state

exp… D=kT† T ˆ 400 K

T ˆ 700 K

1:3  10 3 1.0 5:9  102

2:2  10 1.0 3:8  10

2

is the barrier to desorption from the transition state.

[11] J.E. Davis, P.D. Nolan, S.G. Karseboom, C.B. Mullins, J. Chem. Phys. 107 (1997) 943. [12] C.T. Campbell, G. Ertl, H. Kuipers, J. Segner, Surf. Sci. 107 (1) (1981) 220. [13] P.D. Nolan, B.R. Lutz, P.L. Tanaka, J.E. Davis, C.B. Mullins, J. Chem. Phys. 111 (1999) 3696. [14] J.L. Gland, B.A. Sexton, G.B. Fisher, Surf. Sci. 95 (1980) 587. [15] J.T. Stuckless, C.E. Wartnaby, N. Al-Sarraf, St. J.B. Dixon-Warren, M. Kovar, D.A. King, J. Chem. Phys. 106 (1997) 2012. [16] M. Beutl, K.D. Rendulic, G.R. Castro, Surf. Sci. 385 (1997) 97. [17] A. Hodgson, A.K. Lewin, A. Nesbitt, Surf. Sci. 293 (1993) 211. [18] D. Kelly, R.W. Verhoef, W.H. Weinberg, J. Chem. Phys. 102 (1995) 3440. [19] A. Spitzer, H. L uth, Surf. Sci. 118 (1982) 136. [20] K. Prabhakaran, P. Sen, C.N.R. Rao, Surf. Sci. 177 (1986) L971. [21] M.K. Rajumon, K. Prabhakaran, C.N.R. Rao, Surf. Sci. 233 (1990) L237. [22] W. Biemolt, A.P.J. Jansen, M. Neurock, G.J.C.S. van de Kerkhof, R.A. van Santen, Surf. Sci. 287/288 (1993) 183. [23] M. Neurock, R.A. van Santen, W. Biemolt, A.P.J. Jansen, J. Am. Chem. Soc. 116 (1994) 6860. [24] F. Jensen, F. Besenbacher, I. Stensgaard, Surf. Sci. 269/270 (1992) 400. [25] A. Knop-Gericke, M. Havecker, T. Schedel-Niedrig, R. Schl ogl, Catal. Lett. 66 (2000) 215. [26] I. Bottger, T. Schedel-Niedrig, O. Timpe, R. Gottschall, M. Havecker, T. Ressler, R. Schl ogl, Chem.-Eur. J. 6 (2000) 1870. [27] A. Knop-Gericke, M. Havecker, T. Schedel-Niedrig, R. Schl ogl, Top. Catal. 10 (2000) 187. [28] B. Hammer, L.B. Hansen, J.K. Nùrskov, Phys. Rev. B 59 (1999) 7413. [29] J. Neugebauer, M. Sche‚er, Phys. Rev. B 46 (1992) 16067. [30] J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson, M.R. Pederson, D.J. Singh, C. Fiolhais, Phys. Rev. B 46 (1992) 6671. [31] J.A. White, D.M. Bird, Phys. Rev. B 50 (1994) 4954. [32] D.H. Vanderbilt, Phys. Rev. B 41 (1990) 7892. [33] G. Kresse, J. Furthm uller, Comput. Mater. Sci. 6 (1996) 15.

144

Y. Xu, M. Mavrikakis / Surface Science 494 (2001) 131±144

[34] R.W.G. Wycko€, Crystal Structures, vol. 1, Wiley, New York, 1963. [35] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865. [36] D.R. Lide (Ed.), CRC Handbook of Chemistry and Physics, seventy sixth ed., CRC Press, Boca Raton, FL, 1995. [37] M. Mavrikakis, B. Hammer, J.K. Nùrskov, Phys. Rev. Lett. 81 (1998) 2819. [38] Y. Xu, M. Mavrikakis, unpublished data. [39] A. Ulitsky, R. Elber, J. Chem. Phys. 92 (1990) 1510. [40] G. Mills, H. J onsson, G.K. Schenter, Surf. Sci. 324 (1995) 305. [41] G. Henkelman, H. J onsson, J. Chem. Phys. 113 (2000) 9978. [42] G. Henkelman, B.P. Uberuaga, H. J onsson, J. Chem. Phys. 113 (2000) 9901. [43] E. Shustorovich, A.T. Bell, Surf. Sci. 268 (1992) 397. [44] L.H. Dubois, Surf. Sci. 119 (1982) 399. [45] J. Haase, H.-J. Kuhr, Surf. Sci. 203 (1988) L695. [46] M. Wiegel, A.R. Balkenende, O.L.J. Gijzeman, G.W.R. Leibbrandt, F.H.P.M. Habraken, Surf. Sci. 254 (1991) L428. [47] H. Niehus, Surf. Sci. 130 (1983) 41. [48] F.H.P.M. Habraken, E.P. Kie€er, G.A. Bootsma, Surf. Sci. 83 (1979) 45. [49] R.W. Judd, P. Hollins, J. Pritchard, Surf. Sci. 171 (1986) 643. [50] J. Bloch, D.J. Bottomley, S. Janz, H.M. van Driel, R.S. Timsit, J. Chem. Phys. 98 (1993) 9167.

[51] K.R. Lawless, A.T. Gwathmey, Acta Met. 4 (1956) 153. [52] F. Jensen, F. Besenbacher, E. Lñgsgaard, I. Stensgaard, Surf. Sci. Lett. 259 (1991) L774. [53] F. Besenbacher, C. Klink, E. Lñgsgaard, L.P. Nielsen, L. Ruan, I. Stensgaard, J. Vac. Sci. Technol. A 11 (Series 2) (1993) 1637. [54] F. Besenbacher, P.T. Sprunger, L. Ruan, L. Olesen, I. Stensgaard, E. Lñgsgaard, Top. Catal. 1 (1994) 325. [55] P.A. Gravil, D.M. Bird, Phys. Rev. Lett. 77 (1996) 3933. [56] F. Mittendorfer, A. Eichler, J. Hafner, Surf. Sci. 433±435 (1999) 756. [57] A. Eichler, J. Hafner, Phys. Rev. Lett. 79 (1997) 4481. [58] A. Eichler, F. Mittendorfer, J. Hafner, Phys. Rev. B 62 (2000) 4744. [59] S.Y. Liem, J.H.R. Clarke, G. Kresse, Surf. Sci. 459 (2000) 104. [60] G.J.K. Acres, J.C. Frost, G.A. Hards, R.J. Potter, T.R. Ralph, D. Thompsett, G.T. Burstein, G.J. Hutchings, Catal. Today 38 (1997) 393. [61] B. Hammer, J.K. Nùrskov, Phys. Rev. Lett. 79 (1997) 4441. [62] M. Mavrikakis, P. Stoltze, J.K. Nùrskov, Catal. Lett. 64 (2000) 101. [63] B. Hammer, J.K. Nùrskov, in: R.M. Lambert, G. Pacchioni (Eds.), Chemisorption and Reactivity on Supported Clusters and Thin Films, Kluwer Academic Publishers, Dordrecht, The Netherlands, 1997, pp. 285±351. [64] S. Giorgio, C.R. Henry, B. Pauwels, G. van Tendeloo, Mater. Sci. Eng. A 297 (2001) 197.