Applied Catalysis A: General 291 (2005) 13–20

www.elsevier.com/locate/apcata

CO oxidation on gold nanoparticles: Theoretical studies

Ioannis N. Remediakis a, Nuria Lopez b, Jens K. Nørskov a,*
a
Center for Atomic-scale Materials Physics, Department of Physics, Technical University of Denmark, 2800 Lyngby, Denmark
b
Department Quimica Fisica, Universitat de Barcelona, C/Marti i Franques 1, 08028 Barcelona, Spain
Received 11 November 2004; received in revised form 3 January 2005; accepted 3 January 2005
Available online 27 June 2005

Abstract

We present a summary of our theoretical results regarding CO oxidation on both oxide-supported and isolated gold nanoparticles. Using
Density Functional Theory we have studied the adsorption of molecules and the oxidation reaction of CO on gold clusters. Low-coordinated
sites on the gold nanoparticles can adsorb small inorganic molecules such as O2 and CO, and the presence of these sites is the key factor for the
catalytic properties of supported gold nanoclusters. Other contributions, induced by the presence of the support, can provide parallel channels
for the reaction and modulate the final efficiency of Au-based catalysts. Finally, our theoretical simulations allow us to discuss the selectivity
of supported Au nanoparticles.
# 2005 Elsevier B.V. All rights reserved.

Keywords: Density functional calculations; Gold; Oxidation; Titanium oxide; Nanoparticle

1. Introduction at specific ensemble sites at the interface between the metal

and the support. The proposed mechanisms can be classified
Gold is known to be the noblest of all the metals [1], in terms of the degree of dependence on the support as
although, since the work performed in the group of Haruta follows:
et al. [2] it is well-established that the nobility of gold
disappears for nanometric-size particles. In fact, those
Gold-only mechanisms: This category includes mechan-
particles are active for CO oxidation at room temperatures; isms where low-coordinated Au atoms in gold nanopar-
partial oxidation processes, i.e. propylene oxidation; NO ticles act as the active sites, and those that benefit from the
reduction by CO; the water gas shift (WGS) reaction and effect of odd–even transitions and quantum size effects
selective elimination of CO from H2 containing gases. [3] [4]. Undercoordinated gold sites can act as centers for the
For CO oxidation, the turn-over frequency (TOF) adsorption of molecules [10–15], therefore the roughness
depends mainly on the size of the gold nanoparticle, the of the surface has been proposed as an indicator of its
most reactive ones being in the 2–4 nm range. [4] The TOF activity [16]. Due to the finite size of the supported
also depends on the preparation method and the support. [3] nanoparticles, undercoordinated centers are more abun-
Nowadays, active Au nanoparticles are prepared as dant in a nanoparticle compared to a single-crystal [10].
dispersions on reducible (TiO2, NiO or Fe2O3 [2]) or On the other hand, odd–even transitions have been
irreducible (Al2O3 and MgAl2O4 [5–7]) oxides, or, more reported to have an important role in the reactivity of
recently, on activated carbon fibers [8] or even without any small gas-phase gold clusters and gold subnanometer
support, in the form of gold nanotubes [9]. A multitude of structures [17,18].
factors have been proposed in order to explain the enhanced
Support induced mechanisms: In these cases, the support
catalytic activity of gold nanoparticles, and the active site for could either charge the gold nanoparticle [17] or induce
CO oxidation has been suggested to be either on the metal or strain in the Au structures that can affect the reactivity of
the supported metal [10,19]. EXAFS experiments
* Corresponding author. performed on real catalysts [20] give no indication of

0926-860X/$ – see front matter # 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.apcata.2005.01.052
14 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20

significant changes in the gold–gold distances. With electrons were described by ultra soft pseudopotentials [30].
respect to charging, it has been found that it can be Exchange and correlation effects were taken into account via
relevant in the case of isolated metal atoms or the Generalized Gradient Approximation (GGA) and the
subnanometer structures [17,21]. For larger metallic revised Perdew–Burke–Ernzerhof (RPBE) functional [31].
structures, however, charge is screened quite efficiently The wave function was expanded in a plane-wave basis with
and its effect decays exponentially with the distance from a kinetic energy cutoff of 25 Ry. The valence electron
the charged center [22]. density is obtained by self-consistent iterative diagonaliza-

Support driven and support-assisted mechanisms: This tion of the Hamiltonian [32], with Pulay mixing of the output
group of mechanisms include situations where the support and input densities [33]. Occupation of the one-electron
provides activated oxygen [23], or stabilizes reactants and states was calculated using a temperature of kBT = 0.2 eV;
intermediates [15,24]. The role of the support as O2 all energies have been extrapolated to T = 0 K. The ionic
provider has been ruled out as the main contribution in degrees of freedom were relaxed using a conjugate-gradient
gold-based catalysts, as the reaction can also take place on minimization, until the root of the mean squared force
reducible carriers, like MgAl2O4 [7]. However, reducible component was smaller than 0.05 and 0.2 eV/Å for the rigid
supports can trap incoming O2 molecules and in that way and the flexible clusters, respectively. For the large systems
enhance the performance of some catalysts by improving containing the flexible clusters a further relaxation of
mass transport of the reactants to the reaction site [25]. In selected structures results in insignificant changes in the
addition, the support can stabilize some species along the calculated energies. When the transition states have been
reaction path either by polarization effects [26] or by located through constrained minimization, the forces on the
direct bonding, e.g. of O2 to one Au atom and one atom atoms are smaller than 0.3 eV/Å leading to energies
from the support [15,24,27]. In the latter case, if the converged up to 0.05 eV. The rigid unsupported clusters
reaction ensemble is placed at the interface, the were built with a fcc structure keeping the (1 1 1) and (1 0 0)
mechanism is usually referred to as metal/oxide boundary facets observed in the supported particles. They were
mechanism. calculated in a 15 Å cubic box with a single k-point. The
TiO2(1 1 0) substrate was modeled by a two-layered slab and
However, the above classification is rather schematic, a (4  2) surface unit cell. Subsequent slabs are separated by
since in a given system more than one of the above about 15 Å of vacuum; even when CO is adsorbed on the
mechanisms may be taking place. For instance, the substrate gold cluster, the minimum distance between slabs is more
redox properties affect the dispersion and the shape of the than 6.5 Å. Although the energies of vacancy formation and
supported nanoparticles [20,28] but they also control the O2 adsorption are strongly dependent on the number of TiO2
charge transfer at the interface and the ability of the support layers [34], the huge size of the calculation prevented us to
to trap the reactants. This is the reason why it is a very use thicker supercells. However, the final structure
challenging task to distinguish experimentally between the employed contains the structural and chemical complexity
aforementioned mechanisms, and find the dominant one in of dispersed supported nanoparticles and allows the
each case. On the other hand, first principles Density identification of the relevant terms in the reaction process.
Functional Theory (DFT) calculations can provide a clearer The computational effort required for the detailed theore-
insight into the CO oxidation on supported gold nanopar- tical study of even a simple chemical process as CO
ticles. The procedure we follow is to increase the complexity oxidation is enormous.
of the model employed by obtaining the contributions from
the finite size effect of the gold particle and the effect of the
support in a subsequent step. This procedure allows the 3. CO oxidation
identification of the different aspects involved in the catalyst
by gold and provides the quantitative contribution of the To unravel all possible contributions coming from the
different factors linking them to the structure and composi- presence of the support in a gold catalyst, we compare the
tion of the system [11,27]. The system chosen is Au/TiO2 reactivity of a gold nanoparticle in different environments.
since it is the most deeply investigated Au-based catalyst, Our starting structure is a rigid model, Au10 particle, that
and several aspects of its activity and its structure, up to the mimics the shape of a flat nanoparticle and shows the (1 1 1)
atomistic level, have been investigated [4,28]. CO oxidation and (1 0 0) characteristic faces of larger particles. In a
is taken as the model reaction for the same reasons. second step, the same Au10 particle has been supported on an
oxygen deficient rutile TiO2(1 1 0), and all gold atoms and
the first rutile layer have been relaxed without constraints.
2. Technical details To identify the electronic effect of the support we have
considered the same structure that the gold cluster would
The calculations were carried out using Density have if supported, but without the support. This procedure
Functional Theory. The DACAPO package was employed deconvolutes the electronic and the geometric constraints
[29]. The ionic cores and their interaction with valence induced by the support. In the two latter cases, contributions
 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20 15

from the metal cluster fluxional character are, at least first one is the direct CO–O2 coupling that shows an energy
partially, included. barrier of 0.40 eV. The final O atom that remains on the
particle can be eliminated by a second CO molecule at a
3.1. Isolated rigid Au10 nanoparticle much lower cost; the energy barrier is only 0.16 eV. The
indirect path takes place through the dissociation of the O2
The calculations for a rigid Au cluster illustrate the role molecule. For the rigid Au10 particle, O2 can easily
that low-coordinated atoms play in the enhanced reactivity dissociate to form two adsorbed O atoms, at the corner
of gold. The starting geometry of the Au10 nanoparticle site. The energy barrier for this process is dissociation
shows a basal plane of (1 1 1) character and lateral faces pathway is very sensitive to the geometry. For instance,
with (1 1 1) and (1 0 0) patterns. The Au coordination dissociation of O2 from the m2 state to two adsorbed O
number is 5 for the gold atoms in the top layer and 4 for the atoms, one of which is adsorbed on the basal Au plane while
periphery metal atoms on the basal layer. During relaxation the other is adsorbed on the top plane, shows a much larger
of the atomic positions, the symmetry of the cluster as well barrier, close to 1 eV. Again, the adsorbed isolated O atoms
as the z-coordinates of the basal Au atoms are kept fixed. can be easily removed by the incoming CO molecules as in
After relaxation, we observe a small reduction in the Au–Au the second step for the direct mechanism.
distance of the order of 0.08 Å with respect to bulk Au–Au
distance, in agreement with the EXAFS results on real 3.2. Supported gold nanoparticle
catalysts [20].
For the unsupported rigid particle, CO and O2 adsorption In a second step, we deposited the Au10 cluster on the
and CO oxidation were analyzed; the reaction path is shown defective rutile TiO2(1 1 0) [27]. At room temperature and
in Fig. 1 [11]. The adsorption of CO on the particle is rather pressure, gold nanoparticles do not bind to the perfect rutile
strong: CO adsorption energy is 0.6 eV, much lower than TiO2(1 1 0) [36]. However, under the same conditions,
that for a step-edge site on Au(2 1 1), which is approxi- reducible oxides like rutile contain a certain amount of
mately 0.25 eV [20]. This structure can further adsorb an oxygen vacancies that act as anchoring points [28]. For a
oxygen molecule in a m2 configuration, where the two O normal sample preparation used in STM experiments, about
atoms are bonded to a Au center each, with an adsorption 7–10% of the surface O atoms are missing [37]. Low
energy of 0.3 eV. The adsorption of O2 on the gold temperature STM experiments have shown that the most
nanoparticle can only proceed at the very low-coordinated frequently observed gold nanostructures, containing up to 30
sites on highly defective Au surfaces. This correlates with gold atoms, present a ratio of about three Au atoms per
the upward shift (about 0.75 eV) of the d-band center from oxygen vacancy [28]. To reproduce the effect in our model,
the Au(1 1 1) surface to the Au atoms in the periphery of the the Au10 cluster is set up to bond through three missing O
Au10 cluster. It has been demonstrated that for a given metal atoms from the bridging oxygen rows. The relaxed geometry
atom in different environments the closer the d-band center is shown in Fig. 2. After relaxation of the supported Au10
is to the Fermi level the more prone towards adsorption the cluster the Au–Au distances are similar to the original, but
site is [1]. the average coordination number of the Au atoms has
After adsorption, the reaction can proceed via two increased and thus the surface energy of the system has been
different pathways similar to those reported for Pt [35]. The

Fig. 1. Reaction path for CO oxidation on the isolated rigid Au10 cluster. Fig. 2. Relaxed geometry for an Au10 cluster supported on reduced
Au, C and O atoms are represented by yellow, gray and red spheres, TiO2(1 1 0). Au, Ti and O atoms are represented by yellow, gray and red
respectively. The blue line represents the indirect path while the black line spheres, respectively. Structure is periodic in both lateral dimensions. (For
corresponds to the direct path. (For interpretation of the references to color interpretation of the references to color in this figure legend, the reader is
in this figure legend, the reader is referred to the web version of this article.) referred to the web version of this article.)
16 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20

reduced. Only one Au reduces its coordination number when configurations. This structure has an energy of 1.15 eV
compared to the rigid model (atom number 1 in Fig. 2). relative to gas-phase CO and O2, similar to the 0.9 eV
For this supported cluster, we only consider adsorption of stand-alone Au10 particle and constitutes the starting point
CO, O2, O and CO2 and the direct CO + O2 oxidation path. for a gold-only mechanism.
To reduce the size of the calculations only some The energy barrier for direct oxidation, CO + O2 !
characteristic geometries, resembling of those found for CO2 + O, on the rutile-supported Au cluster through the
the rigid cluster, were explored, and only Au and adsorbate gold-only mechanism was estimated by considering CO and
atoms were relaxed. The adsorption energies for all O2 in a configuration similar to that found for the isolated
considered configurations are shown in Table 1. CO binds rigid Au10 particle. The C–O bond length was fixed at the
with the same adsorption energy atop the highest Au atom transition state value, 2.8 Å, while all other degrees of
and between the two highest ones. The adsorption energy is freedom of the adsorbate and the gold atoms were relaxed.
0.95 eV, higher than the energy that reported for the rigid The so calculated energy barrier is 0.36 eV with respect to
cluster 0.60 eV, due to the reduced coordination of the Au adsorbed CO and O2. For the metal/oxide boundary
atom in the supported cluster, 4 versus 5 in the previous case. mechanism the CO + O2 coupling path has been located
Recently, Meier and Goodman [38] have reported singleton by modifying the C–O distance until we the maximum
CO binding energy for CO on TiO2 supported Au bilayer energy is found, while the O2 is placed at the interface. The
clusters of 0.8 and 0.5 eV for unsupported ones. Within the calculation was performed in both ways, once increasing the
error bar of the experiments, these values are essentially the C–O bond length, going from CO2 to CO and the second
same as the computed ones. time decreasing the distance, thus going from CO + O to
On the other hand, O2 has a preference for binding in the CO2. In such a way the actual energy barrier should be close
contact area between the Au cluster and the Ti(5c) on the to the value we obtain for the highest energy point which is
surface. One of the O atoms of the O2 molecule binds to a similar in both directions. At the transition state, CO is
periphery Au atom, while the second bonds to a Ti atom of bonded to the top gold atom in Fig. 2. The O–O distance is
the support. The binding energy to this site is 1.59 eV 1.4 Å and the C–O distance is 1.8 Å. The second O atom that
relative to gas-phase CO and O2. This value is only does not participate directly to the oxidation, sits on top of a
marginally affected, by 0.05 eV or less, by changes in the ‘‘bare’’ Ti atom, with an O–Ti distance of 2.3 Å. This path
redox characteristics of the neighboring sites, i.e. by has an energy barrier (from adsorbed CO and O2) of 0.40 eV
removing a bridging O from TiO2 or adding a spectator similar to the reported in for the gold-only mechanism. For
O2 molecule far away from the adsorbates. This configura- both transition state structures the forces on the ions were
tion is the initial point for a support-assisted or metal/oxide small, indicating that the system is indeed in a saddle
boundary mechanism, where the substrate stabilizes the point.
reactants. Other low-coordinated gold atoms can provide a The two different final states, CO2 + O, corresponding to
site for O2 binding without any direct interaction to the the metal/oxide boundary mechanism and to the gold-only
oxide. The reduced size of the cluster employed makes the mechanism were also identified. In the first case, the Oads is
location of such a center difficult but the bridge site between sitting at the metal-oxide boundary, and is placed at
Au atoms 5 and 8, with the constraint on the z position of the 3.88 eV, while in the second case Oads it is placed at the
O atom closer to the support, can be representative of such particle, with an energy of 4.06 eV. In any case, the final
state CO2 + O has a lower energy than for the isolated rigid
Table 1 Au10, which was 3.20 eV. In fact, for gas-phase Au clusters
Adsorption energies for configurations of the considered species on rutile- the adsorption of O atoms largely distorts the cluster
supported Au10 structure [14]. This phenomenon is not observed in the rigid
Adsorbate (adsorption sites) Energy (eV) model due to the geometry constraints imposed. O-induced
CO (atop, 1) 0.95 distortions might contribute to the rapid deactivation
CO (bridge, 1,2) 0.95 observed for the activity of supported gold clusters.
CO (atop, 1) + O2 (bridge, 4,5) 1.59 Moreover, since adsorbate induced reforming is strongly
CO (atop, 1) + O2 (bridge, 5,6, constrained) 1.15 related to the average coordination number of all the atoms
CO (atop, 1) + O2 (bridge, 5,6) 1.32
in the metallic nanoparticle, reformation will preferentially
CO (atop, 1) + O2 (bridge, 3,5) 0.57
CO (hollow, 1,2,3) + O2 (bridge, 1,4) 0.10 affect very small particles. This could explain the low
TS: CO (atop, 1) + O2 (atop, 5) 0.79 activity of clusters below 1 nm [4].
TS: CO (atop, 5) + O2 (atop, 5 and Ti) 1.19
CO2 (atop, 1) + O (bridge, 1,5) 4.06 3.3. Unsupported flexible Au10 cluster
CO (bridge, 2,3) + O (bridge, 1,5) 0.63
CO2 (atop, 1) 3.66
In order to estimate the oxide-induced contributions to
The numbers next to each molecule refer to the labels of Au atoms in Fig. 2.
Energies are relative to gas-phase CO and O2. The configurations whose the gold-only mechanism, the energies of the initial, final
energies are used in Fig. 3 are typed in bold. The adsorption energy for CO2 and transition states for an unsupported flexible Au10 cluster
is 0.48 eV relative to the gas phase. were calculated. The presence of the oxide was simulated by
 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20 17

Fig. 3. Compared geometries (left) and energy profiles (right) for the CO oxidation on a Au10 cluster. Upper panel, blue line: Au10 supported on TiO2(1 1 0), CO
oxidation takes place on the Au/TiO2 interface. Lower panel, black line: Au10 supported on TiO2(1 1 0), CO oxidation takes place solely on Au particle. Red
line: Unsupported cluster with the bottom three atoms kept fixed at the positions they would have if the oxide was present. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

fixing the bottom three atoms (numbered 8–10 in Fig. 2) at barrier (0.75 eV) due to the large stabilization of the
the positions they had in the initial state of the supported reactants. In the present case, the noble nature of gold
cluster. As a consequence, geometric contributions are only prevents a strong adsorption of O atoms and molecules even
partially taken into account while the electronic effects when very low-coordinated atoms exists. This mild bonding
coming from the support are completely removed. All other character is at the basis of gold-based catalysts in oxidation
Au atoms and adsorbates are fully relaxed, except for the processes at low temperatures. Moreover, a broad range of
transition state, where the C–O bond was kept fixed. The binding energies, about 1 eV, going from slightly repulsive
comparison between the energetics of the unsupported and to weakly attractive can be attainable by controlling the
the supported cluster is shown in Fig. 3. The adsorption coordination number of Au.
energy for CO on the unsupported gold cluster is 0.90 eV, Gold-based catalysts may also benefit from the presence
similar to that of the supported cluster (0.95 eV). This also of the support. First, the support favors and stabilizes metal
holds for the co-adsorbed CO and O2 state: 1.12 eV for the dispersion and a given shape. As shown by Mavrikakis et al.
unsupported cluster (1.15 eV for the supported). The [10], a high dispersion increases the relative number of low-
transition state energy, however, is 0.15 eV higher for the coordinated sites and therefore the number of reaction sites
unsupported cluster, yielding an energy barrier of 0.48 eV. for gold-only mechanisms. The electronic contributions due
The final state CO2 and O, has an energy closer to that to the to the presence of the rutile support to the gold-only
rigid cluster, 3.20 eV, due to the geometric constraints mechanism can be investigated from the density difference
imposed. between the supported particle and the free particle, shown
in Fig. 4. In this figure, we plot the difference between the
3.4. Discussion charge density of the Au10/reduced TiO2(1 1 0) system
minus the sum of the charge densities of Au10 (at the same
For the above gold-based systems, all the CO oxidation geometry) and reduced TiO2(1 1 0). Charge is accumulated
paths reported show similar activation energies 0.36– between Au and Ti atoms, and the electrons trapped in the O-
0.40 eV, for the rate limiting step. This is similar to the deficient substrate are employed to furnish the bonds at the
value measured by Haruta et al. [2], close to the range 0.15– interface. The Au atoms of the second layer, however, are
0.25 eV given for STM characterized catalysts [39] and not affected by the presence of the support. In addition, the
within the broad range of 0.16–0.60 eV reported by other added charge resides at the Au-substrate interface. This is
authors [40,41]. As a consequence, all the aforementioned mirrored in the positions of the d-band centers for Au atoms
Au nanoparticles appear to be better catalysts for CO that are changed only marginally between the free and the
oxidation than any single-crystal transition metal catalyst. supported cluster. These observations do not lend any
For example, on Pt(1 1 1), O2 adsorbs in a precursor state support to the possibility of a support driven mechanism
and dissociates through a small barrier 0.46 eV [35]. The based on metal charging. This explains why only small
adsorbed O atoms are, however, strongly bond to Pt(1 1 1) differences are found for supported and unsupported Au10 in
and thus the next step, CO–O coupling, shows a large energy the previous sections and indicates that both dispersion
18 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20

through the external circuit and protons that react with O2

at the cathode. The anode catalyst is severely poisoned by
CO present in the H2 gas feed. CO concentrations as small
as 100 ppm can completely block the active sites on Pt or
modified Pt surfaces typically employed as anode
electrodes. To avoid this problem, it has been proposed
to selectively oxidize CO from the H2 feed by means of
Au-based catalysts [2,42]. The so-called preferential
oxidation of CO, PROX, reaction has been addressed
theoretically by Mavrikakis et al. They studied the
adsorption of several intermediates in the oxidation
reactions for Cu, Pt and Au on the (1 1 1) surfaces, where
water formation shows larger energy barriers than CO
oxidation [43]. Moreover, in kinetic experiments it has
been proposed that the H2 activation/dissociation can be
the rate determining step [42]. For Au(1 1 1) surfaces the
reaction is endothermic by 0.5 eV and has a barrier of
1.2 eV (within the PW91 approach) [1], but for very small
gas-phase, Aun (n = 2 and 3) clusters, dissociation barriers
Fig. 4. Contour plot of the density difference along the (0 0 1) direction of 0.55 and 0.12 eV, with respect to gas-phase reactants,
between the Au10/reduced TiO2 minus the sum of charge densities of Au10
and the reduced TiO2(1 1 0). Background gray color denotes zero differ-
have been reported [44]. Negative barriers arise from the
ence. Lighter (darker) color denotes higher (lower) local electron density. presence of a molecularly adsorbed H2 species previous to
dissociation.
To oxidize either hydrogen or CO, the molecules should
(size) and shape are the main contributions from the support be adsorbed before they react with O2. CO adsorption on
in gold-only mechanisms. Au10 is a non-activated process while a barrier is found for
Moreover, the presence of the support can provide the dissociative adsorption of H2 on several metal surfaces
additional channels for the reaction to occur, as happens for including Au(1 1 1). We have investigated several paths for
the Au/TiO2 system. In that case, the metal/support interface the adsorption of H2 to the rigid Au10 cluster. The final states
presents special complex ensembles that contain both under- considered are such that H atoms are adsorbed in between
coordinated Au and support-atoms act as reaction sites in two Au atoms either on the top layer or at the basal plane.
metal/oxide boundary mechanisms. These complex ensem- For a site on the rigid Au10 cluster with coordination number
bles depend on the support structure and properties and of 5 (top layer), the dissociated state is still endothermic for
therefore should be analyzed for each substrate. H2, +0.47 eV, and hindered by a sizeable barrier close to
In addition, the presence of a gold-only reaction pathway 1 eV, while CO adsorption at the same site is exothermic by
that is common for all supports explains why very different 0.6 eV. In the case of a competitive activated (H2) and non-
supports have been successfully employed; since there is activated (CO) adsorption process the ratio of initial
always an open channel based only on the chemical activity adsorption rates on the catalyst surface can be written as
of low-coordinated Au atoms. The simultaneous presence of follows [45]:
a second mechanism taking place at the metal/oxide rCO pCO
boundary, that enhances the activity of the system and is  (1)
rH2 pH2 expðDE=RTÞ
substrate dependent, explains the different activity observed
for different gold-based catalyst. In both reaction paths a where rCO and rH2 are the initial adsorption rates, p the
high concentration of low-coordinated gold atoms is pressures of CO and H2 and DE are the barrier for H2
required to bind the reactants. As a consequence, the gold adsorption. With the previous values for the adsorption of
particle size, a measure of the number of low-coordinated CO and H2 barrier we find that at room temperature and with
Au atoms, is found to be the main parameter governing the a H2 feed of 1 atm containing 100 ppm of CO, the initial
reactivity of gold-based catalysts. [13] adsorption rate is still several orders of magnitude larger for
CO than for H2. Therefore, the selective CO elimination
3.5. CO versus H2 oxidation, selectivity from H2 feeds can be understood, at least partially, from the
fact that a larger fraction of sites would be covered by CO
One of the potential uses for Au-based catalysts is in the preventing H2 oxidation.
cleaning of the H2 gas feed to fuel cells. In a Proton Finally, we want to point out that the barriers for H2
Exchange Membrane (PEM) fuel cell, hydrogen produced dissociation were found to be dependent on the gold
by hydrocarbon and alcohol reforming is supplied to the structure. For example, Morikawa [47] has found a much
anode where it oxidizes, thus generating electrons that flow lower barriers for H2 dissociation on gold chains. A large
 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20 19

contribution from reforming in very small Au clusters is also References

expected from the results of Varganov et al. [44]. This large
dependence on the coordination number in the very small [1] B. Hammer, J.K. Nørskov, Nature 376 (1995) 238.
cluster regime might explain the selectivity observed for [2] M. Haruta, S. Tsubota, T. Kobayashi, H. Kageyama, M.J. Genet, B.
Delmon, J. Catal. 144 (1993) 175.
other reactions. For instance, propylene epoxidation in the [3] M. Haruta, Catal. Today 36 (1997) 153.
presence of O2 and H2 yields the oxidation product (for 2 nm [4] M. Valden, X. Lai, D.W. Goodman, Science 281 (1998) 1647.
Au nanoparticles) or the hydrogenated one (for particles [5] M. Okamura, S. Nakamura, S. Tsubota, T. Azuma, M. Haruta, Catal.
smaller than 1 nm) [48]. Lett. 51 (1998) 53.
[6] S.-J. Lee, A. Gravriilidis, J. Catal. 206 (2002) 305.
[7] J.-D. Grunwaldt, H. Teunissen, Eur. Patent EP1209121, 2002.
[8] D.A. Bulushev, I. Yuranov, E.I. Suvorova, P.A. Buffat, L. Kiwi-
4. Conclusions Minsker, J. Catal. 224 (2004) 8.
[9] M.A. Sanchez-Castillo, C. Counto, W.B. Kim, J.A. Dumesic, Angew.
Theoretical methods are capable of analyzing the Chem. Int. Ed. 43 (2004) 1140.
possible contributions to the enhanced activity of gold. [10] M. Mavrikakis, P. Stoltze, J.K. Nørskov, Catal. Lett. 64 (2000) 101.
We have investigated several of these contributions by [11] N. Lopez, J.K. Nørskov, J. Am. Chem. Soc. 124 (2002) 11262.
[12] C. Lemire, R. Meyer, S. Shaikhutdinov, H.-J. Freund, Angew. Chem.
analyzing CO oxidation on a small (0.7 nm) gold cluster, Int. Ed. 43 (2004) 118.
standalone and supported on rutile TiO2. According to the [13] N. Lopez, T.V.W. Janssens, B. Clausen, Y. Xu, M. Mavrikakis, T. Bli-
calculations, and in agreement with the experimental results, gaard, J.K. Nørskov, J. Catal. 223 (2004) 232.
the presence of low-coordinated Au sites seems to be crucial [14] G. Mills, M.S. Gordon, H. Metiu, Chem. Phys. Lett. 359 (2002) 493.
for CO oxidation. Moreover, for supported gold catalysts [15] L.M. Molina, M.D. Rasmussen, B. Hammer, J. Chem. Phys. 120
(2004) 7673.
two types of mechanisms have been identified. Gold-only [16] G. Mills, M.S. Gordon, H. Metiu, J. Chem. Phys. 118 (2003) 4198.
mechanisms for which the reaction takes place only on the [17] A. Sanchez, S. Abbet, U. Heiz, W.-D. Schneider, H. Häkkinen, R.N.
nanoparticle, in that case, the role of the support is that of Barnett, U. Landman, J. Phys. Chem. A 103 (1999) 9573.
changing the dispersion and shape of the cluster and the [18] B. Yoon, H. Hakkinen, U. Landman, J. Phys. Chem. A 107 (2003)
electronic contributions are found to be minor. Those 4066.
[19] S. Giorgio, C. Henry, B. Pauwels, G.V. Tendeloo, Mat. Sci. Eng. A 297
mechanisms are therefore present in all the chemical (2001) 197.
preparations of Au-based catalyst irrespective of the support [20] N. Lopez, J.K. Nørskov, T.V.W. Janssens, A. Carlsson, A. Puig-
nature. A second branch of support-assisted metal/oxide Molina, B.S. Clausen, J.-D. Grunwaldt, J. Catal. 225 (2004) 86.
boundary mechanisms may exist for which the reaction site [21] Q. Fu, H. Saltsburg, M. Flytzani-Stephanopoulos, Science 301 (2003)
is at the metal/oxide interface and the ensemble needed for 935.
[22] L. Giordano, J. Goniakowski, G. Pacchioni, Phys. Rev. B 64 (2001)
O2 adsorption requires both low-coordinated Au atoms in 075417.
the gold particle and other centers on the oxide. In any case [23] M. Schubert, S. Hackenberg, A. van Veen, M. Muhler, V. Plzak, R.
for the Au/TiO2(1 1 0) system the calculated barriers of Behm, J. Catal. 113 (2001) 197.
about 0.4 eV are in excellent agreement with the experi- [24] Z.P. Liu, X. Gong, J. Kohanoff, C. Sanchez, P. Hu, Phys. Rev. Lett. 91
mental results. For the preferential oxidation of CO in H2 (2003) 266102.
[25] E. Walhström, E.K. Vestergaard, R. Schaub, A. Rønnau, M. Vester-
rich gases it is found that the barrier for H2 dissociation is gaard, E. Læsgaard, I. Stengsgaard, F. Besenbacher, Science 303
still too large, preventing a significant H2 adsorption even in (2004) 511.
gases with a high H2 partial pressure. All these aspects [26] L.M. Molina, B. Hammer, Phys. Rev. Lett. 90 (2003) 206102.
indicate that the particular chemistry of gold nanoparticles is [27] I.N. Remediakis, N. Lopez, J.K. Nørskov, Angew. Chem. Int. Ed. 44
still based on the noble character of gold, since the (2005) 1824.
[28] E. Wahlström, N. Lopez, R. Schaub, P. Thostrup, A. Rønnau, C.
intermediates are not strongly bound to the surface and can Africh, E. Lægsgaard, J.K. Nørskov, F. Besenbacher, Phys. Rev. Lett.
easily react to form products. Finally, the challenging task in 90 (2003) 026101.
the development of new gold-based catalysts is the [29] The DACAPO plane wave/pseudopotential DFT code is available as
generation and stabilization of low-coordinated centers at Open Source Software at http://www.fysik.dtu.dk/CAMPOS/.
the Au particles [46]. [30] D. Vanderbilt, Phys. Rev. B 41 (1990) 7892.
[31] B. Hammer, L.B. Hansen, J.K. Nørskov, Phys. Rev. B 59 (1998)
7413.
[32] M.C. Payne, M.P. Teter, D.C. Allan, T.A. Arias, J.D. Joannopou-los,
Acknowledgements Rev. Mod. Phys. 64 (1992) 1045.
[33] G. Kresse, J. Furthmüller, Comput. Mater. Sci. 6 (1996) 15.
We acknowledge support from the Danish Center for [34] M.D. Rasmussen, L.M. Molina, B. Hammer, J. Chem. Phys. 120
Scientific Computing through grant no. HDW-1101-05. (2004) 988.
[35] A. Eichler, J. Hafner, Phys. Rev. B 59 (1999) 5960.
Discussions with B.S. Clausen, L.M. Molina, B. Hammer, F.
[36] N. Lopez, J.K. Nørskov, Surf. Sci. 515 (2002) 175.
Besenbacher, M. Mavrikakis, H. Metiu and D.W. Goodman [37] R. Schaub, P. Thostrup, N. Lopez, E. Lægsgaard, I. Stensgaard, J.K.
are gratefully acknowledged. N.L. thanks MEC for financing Nørskov, F. Besenbacher, Phys. Rev. Lett. 87 (2001) 266104.
her work through the RyC program and the Generalitat de [38] D.C. Meier, D.W. Goodman, J. Am. Chem. Soc. 126 (2004) 1892.
Catalunya for a mobility grant BE2004. [39] M. Valden, S. Pak, X. Lai, D.W. Goodman, Catal. Lett. 56 (1998) 7.
20 I.N. Remediakis et al. / Applied Catalysis A: General 291 (2005) 13–20

[40] G.R. Bamwenda, S. Tsubota, T. Nakamura, M. Haruta, Catal. Lett. 44 [44] S.A. Varganov, R.M. Olson, M.S. Gordon, G. Mills, H. Metiu, J.
(1997) 83. Chem. Phys. 120 (2004) 5169.
[41] T.V. Choudary, C. Sivadinarayana, C.C. Chusuei, A.K. Darye, J.P. [45] I. Chorkendorff, J.W. Niemantsverdriet, Concepts of Modern Catalysis
Fackler, D.W. Goodman, J. Catal. 207 (2002) 247. and Kinetics, Wiley–VCH, 2003.
[42] B. Schumacher, Y. Denkwitz, V. Plzak, M. Kinne, R.J. Behm, J. Catal. [46] M.S. Chen, D.W. Goodman, Science 306 (2004) 252.
224 (2004) 449. [47] Y. Morikawa, private communication.
[43] S. Kandoi, A.A. Gokhale, L.C. Grabow, J.A. Dumesic, M. Mavrikakis, [48] T. Hayashi, K. Tanaka, M. Haruta, J. Catal. 178 (1998) 566.
Catal. Lett. 93 (2004) 93.