Oxidation of CO and H2 by O2 and N2O on Au/TiO2 catalysts in microreactors

Article abstract of DOI:10.1016/j.jcat.2008.09.003

We performed steady-state activity measurements in microreactors to obtain the reaction rates for CO and H₂ oxidation. These reactions were studied on three different gold particle sizes (d ≈ 3.6, 5.7, 16.2 nm) using either O₂ or N₂O as oxidizing agents. From our TEM analysis, our CO oxidation rates follow the d^-3 relationship proposed in Hvolbaek et al. [B. Hvolbaek, T.V.W. Janssens, B.S. Clausen, H. Falsig, C.H. Christensen, J.K. Norskov, Nano Today 2 (2007) 14-18]. Density functional theory (DFT) calculations on a Au{532} surface and a Au₁₂ cluster, which model corner sites, reproduced the apparent activation barriers of about 37 kJ mol^-1 for CO oxidation on the smallest nanoparticles by both O₂ and N₂O. For all of the reactions studied, we found the overall activation barrier depended only on the size of the TiO₂ supported gold nanoparticle.

Full text of DOI:10.1016/j.jcat.2008.09.003

Journal of Catalysis 260 (2008) 86–92
Contents lists available at ScienceDirect
Journal of Catalysis
www.elsevier.com/locate/jcat
Oxidation of CO and H2 by O2 and N2O on Au/TiO2 catalysts in microreactors
G. Walther a, D.J. Mowbray a, T. Jiang a, G. Jones a, S. Jensen b, U.J. Quaade c, S. Horch a,∗
a
Center for Atomic-scale Materials Design (CAMD), Department of Physics, Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark
Department of Micro and Nanotechnology (DTU Nanotech), Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark
Center for Individual Nanoparticle Functionality (CINF), Department of Physics, Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark
b
c
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 31 July 2008
Revised 3 September 2008
Accepted 4 September 2008
Available online 1 October 2008
We performed steady-state activity measurements in microreactors to obtain the reaction rates for
CO and H2 oxidation. These reactions were studied on three different gold particle sizes (d ≈
3.6,5.7,16.2 nm) using either O2 or N2O as oxidizing agents. From our TEM analysis, our CO oxidation
−3
rates follow the d
relationship proposed in Hvolbæk et al. [B. Hvolbæk, T.V.W. Janssens, B.S. Clausen,
H. Falsig, C.H. Christensen, J.K. Nørskov, Nano Today 2 (2007) 14–18]. Density functional theory (DFT)
calculations on a Au{532} surface and a Au12 cluster, which model corner sites, reproduced the apparent
Keywords:
Gold
−1
activation barriers of about 37 kJ mol
for CO oxidation on the smallest nanoparticles by both O2 and
N2O. For all of the reactions studied, we found the overall activation barrier depended only on the size
of the TiO2 supported gold nanoparticle.
Nanoparticles
Catalysis
Microreactor
Carbon monoxide
Hydrogen
Oxygen
© 2008 Elsevier Inc. All rights reserved.
Nitrous oxide
Particle size
Sintering
TEM
1. Introduction
CO2 and H2 on gold catalysts [27]. Since there is a keen interest
in converting CH4 into more valuable monocarbon chemicals, it is
necessary to first identify both the reaction pathway and active
sites for CO and H2 oxidation by different oxidizing agents.
The reason for this surprising activity is still under debate. Af-
ter the pioneering work of Haruta [28], the activity of gold has
been linked to several different effects. These include quantum-
size effects [9], support-induced strain [15], charge transfer from
the small gold particles to adsorbed O2 [29] and the role of low-
coordinated gold atoms [6,15,17–19].
It has recently been demonstrated that the CO oxidation activity
of gold nanoparticles is inversely proportional to the cube of the
average particle size (∼d−3), independent of the support material
used [30]. This suggests that it is the corner sites which are the
most active for CO oxidation by O2.
The common oxidizing agent for CO oxidation, used in all the
above studies was O2. Very few studies deal with other oxidation
agents, such as N2O, and they focus on the decomposition and re-
duction of N2O [31]. Since N2O is known to readily dissociate into
adsorbed atomic O and gas phase N2 in the presence of CO, the
oxidation of CO by N2O involves atomic O. On the other hand, CO
oxidation using O2 has been shown to also occur via an alterna-
tive pathway, which does not include O2 dissociation [16,32]. By
using these two different oxidizing agents, we may experimentally
In the past, gold was not thought to be catalytically active, and
was even called “the noblest of all the metals” [1]. However, the
discovery by Haruta et al. [2] that nanoparticulate gold is cat-
alytically active for CO oxidation, has provoked an academic “gold
rush.” This is evidenced by the exponential increase in gold catal-
ysis publications since the early 1990s [3–6].
Many different experimental studies have been performed
to understand gold’s activity for CO oxidation. These investi-
gations considered changes of the particle’s morphology [7,8],
a structure–activity relationship [9–11], the activity of unsupported
nanoporous bulk gold [12,13], and the influence of the support on
the activity [14]. CO oxidation has also been thoroughly studied
theoretically using density functional theory (DFT) [1,6,15–19]. Be-
sides CO oxidation, many other reactions have been studied on
gold catalysts. These include the oxidation of hydrocarbons [5,20–
24] and oxygen-containing hydrocarbons [3–5,21,25] as well as the
catalytic hydrogenation of organic compounds [4,5,26].
In particular, CH4 oxidation includes in the tail of its oxida-
tion pathway CO oxidation and H2 oxidation, since it only forms
Corresponding author.
E-mail address: horch@fysik.dtu.dk (S. Horch).
*
0021-9517/$ – see front matter © 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcat.2008.09.003

G. Walther et al. / Journal of Catalysis 260 (2008) 86–92
87
to those of the commercially available samples. The preparation
◦
was performed below 50 C to evaporate the solvent. After the cat-
◦
alyst had been dried, it was calcined at 200 C.
2.2. Catalytic reactions
We studied two different reactions on the three different gold
catalysts listed above, using the two different oxidizing agents O2
(3.5N) and N2O (2.5N). The overall reactions and heats of formation
ꢀH0 [36] are then
−1
CO + N2O → CO2 + N2, ꢀH0 ≈ −200.9 kJ mol
,
,
,
.
(R1)
(R2)
(R3)
(R4)
−1
CO + 1 O2 → CO2,
ꢀH0 ≈ −283.0 kJ mol
2
−1
H2 + N2O → H2O + N2, ꢀH0 ≈ −159.8 kJ mol
−1
H2 + 1 O2 → H2O,
ꢀH0 ≈ −241.8 kJ mol
Fig. 1. Unloaded microreactor, based on silicon, with two inlets to mix gases on the
chip, a (8.0 × 1.5 × 0.2) mm3 reactor chamber and a single outlet.
2
Due to the high sensitivity of the GC to H2, the reactants in
(R4) and (R3) were diluted with 50% argon with the total flow
compare Au nanoparticle activity for two different CO oxidation
pathways.
−1
kept constant at 1.00 ml min
for all reactions.
To minimize the contribution of self-heating of the catalyst dur-
ing an exothermic reaction, all experiments were performed with
a maximum of 10% conversion, except for CO oxidation where 15%
In the present contribution, we compare the reaction rates for
CO and H2 oxidation on three different particle sizes using either
O2 or N2O as the oxidizing agent. To corroborate our experimen-
tal findings, we also modeled the CO oxidation turnover frequency
using DFT calculations on the Au{532} surface and Au12 cluster
in a microkinetic model. Our findings suggest that such a model
quantitatively describes the kinetics of CO oxidation using small
Au nanoparticles (d ꢀ 5 nm).
◦
◦
was achieved at 60 C on Catalyst A and more than 40% at 80 C
on Catalyst B. The greatest amount of heat liberated during the re-
actions was 26.0 mW.
To ensure reproducibility, we followed the following scheme for
all measurements. First, all catalysts were activated. Catalysts B
and C, containing the larger particles, were pretreated over 20 h at
◦
80 C to ensure steady-state activity measurements. On the other
2. Methods
hand, due to its high activity and the possibility of sintering, Cata-
◦
lyst A was pretreated for 20 h at 50 C. These pretreatments were
2.1. Experimental equipment
done under stoichiometrically supplied CO and O2, for the reac-
1
tion CO + O2 → CO2, and a total gas flow of 1.00 ml min−1. The
2
All experiments were performed in microreactors, based on
deep reactive ion-etched (DRIE) silicon wafers. Details of the fab-
rication process are provided in Ref. [33]. Fig. 1 illustrates the
280 μm deep capillary system that allows mixing of undiluted
gases on the chip, without any danger of explosion. The dimen-
sions of the reactor chamber are (8.0 × 1.5 × 0.2) mm3. The whole
device measures 20 mm × 15 mm × 0.35 mm and is sealed with
a Viton sheet. The inlet and outlet holes are sealed with Viton
O-rings to a heatable interface block that connects the external
tubing to the microreactor. The temperature was measured with
a K-type thermocouple and controlled using a PID-controller (Eu-
rotherm). The gas flow was controlled by mass flow controllers
CO conversion decreased by 33%, 17%, and 5% before stabilizing
for the powder Catalysts A, B, and C, respectively. Temperature
programmed activity measurements were then performed in the
following sequence of reactions for each catalyst: (R2), (R1), (R3),
(R4), (R2), (R1).
Since the maximum temperature reached under these reactions
◦
◦
(120 C on Catalyst B and 160 C on Catalyst C) was higher than
◦
the temperature of the pretreatment (80 C), the last two steps (re-
measurement of (R2) and (R1)) were performed to ensure that the
catalysts had not changed during the preceding reactions. Further-
more, to exclude any influence of the applied gases (by e.g. catalyst
reduction), the reaction sequence was also conducted in reversed
order—starting with (R2)—on a fresh and deactivated catalyst from
the same batch. Hereby, the rate differed for each of the reactions
(R1)–(R4) studied, but the proportions between (R2) and (R1), and
between (R4) and (R3) were similar.
−1
(Bronkhorst), operating in the range from 0.02 to 1.00 ml min
−1
with a precision of 0.02 ml min
at 1 bar.
The reaction products were analyzed using an Agilent gas chro-
matograph (3000A microGC). The GC has combined columns of
10 m molecular sieve and 3 m PLOT U with 1.0 μl back-flushing,
which allows simultaneous analysis of H2, O2, N2, CO, CO2, H2O
and N2O by a thermal conductivity detector (TCD). To avoid con-
densation of water, formed by the catalytic reaction of H2, the
2.3. Particle analysis
Transmission electron microscopy (TEM) was used to study
changes in size of the gold particles. For each catalyst, specimens
were taken as the catalysts were supplied, after the pretreatment
and after measurements of all four catalytic reactions. The spec-
imens were then prepared on a carbon TEM grid by dropping a
suspension of catalyst in ethanol on the grid. These were analyzed
using a JEOL 3000F field emission electron microscope, which was
operated at 300 kV with a LaB6 filament as an electron source.
◦
tubing between the microreactor and the GC was kept at 100 C,
whereas the capillary of the GC itself was kept at 90 C.
◦
The catalysts used were 2.0 nm Au particles supported on TiO2
(AUROlite™, supported by AuTEK [34]), 3.3 nm Au particles sup-
ported on TiO2 powder (reference catalyst, supplied by the World
Gold Council [35]), and 5.0 nm Au particles on TiO2, which we
shall denote as Catalysts A, B, and C, respectively. Catalysts A and
B were both supported on TiO2 P25, as stated by the suppliers.
However, Catalyst C was prepared by deposition–precipitation
of unconjugated gold colloids (Ted Pella, Inc.) on anatase TiO2 pow-
der (Millennium Inorganic Chemicals) with an average BET area of
150–300, according to the supplier. For ease of comparison, the
gold concentration of the in-house catalysts was carefully adjusted
2.4. Theoretical methodology
All theoretical results have been obtained using the DFT code
dacapo [37]. The Kohn–Sham one-electron valence states were ex-
panded in a plane wave basis set with a 340 eV (25 Ry) kinetic

88
G. Walther et al. / Journal of Catalysis 260 (2008) 86–92
Table 1
a Langmuir–Hinshelwood mechanism for the reaction kinetics, we
may write the elementary steps in the form
−1
Activation barriers Ea and adsorption energies Eads in kJ mol
dation by N2O and O2.
for CO and H2 oxi-
Au12 cluster
∗
X(g) + ∗ ꢁ X , ∀X ∈ R,
(R5a)
(R5b)
−1
E (kJ mol
)
Au{532} surface
rds
Eads[O2]
Eads[CO]
Eads[N2O]
−27.0
−74.3
−7.7b
–
−60.8a
−91.7a
−7.7b
27.0a
–
R → P(g) + ∗.
∗
The turnover frequency for the rate determining step (rds) is
then
Ea[CO + O2 → CO2 + O]
Ea[CO + O → CO2]
27.0
ꢁ
a
Ref. [30].
Ref. [43].
+
∗
frds ≈ krds
ΘX
,
(3)
b
X∈R
∗
∗
where ΘX is the fractional coverage of species X on site “ ”, and
krds is the rate constant for the forward rate, which we assume
energy cutoff, and a density cutoff of 680 eV (50 Ry). The core
electrons were described by Vanderbilt type ultrasoft pseudopo-
tentials [38]. The exchange-correlation potential was described us-
ing the RPBE generalized gradient approximation self-consistently
[39]. For the Au{532} surface, a 4 × 4 × 1 Monkhorst–Pack k-point
sampling was applied in the irreducible Brillouin zone. The sur-
face was modeled by a 1 × 1-{532} unit cell containing 24 Au
atoms with periodic boundary conditions, which corresponds to
three close-packed layers. The top most layer and the adsorbents
were allowed to fully relax. The O2 adsorption energy was calcu-
lated relative to the experimentally obtained formation energy of
H2O from O2 and H2 [40]. This avoids difficulties associated with a
DFT treatment of the O2 triplet state in the gas phase [41].
Thermodynamic analysis was carried out using the total en-
ergies obtained from the DFT calculations. It is then possible to
obtain free energies by augmenting the DFT total energies with
the thermodynamics of a classical ideal gas [42]. For a gas-phase
species X at temperature T and pressure p, the Gibbs free energy
GX(p, T ) is given by
+
dominates the total reaction rate.
Assuming the difference in zero point energies of the prod-
ucts and the reactants ꢀEZPE is much smaller than the activation
barrier Ea[rds], and the entropy of the adsorbed species is much
∗
smaller than that of the species in gas phase, SX ꢀ SX(g), we may
express the forward rate constant for the rds as
ꢀ
ꢂ
ꢃ
kBT
h
Ea[rds] − T
SX(g)
+
X∈P
krds
≈
exp −
.
(4)
RT
Since the adsorption steps are assumed to occur in equilibrium,
the net rate
+
rX = kX p Θ − k− Θ = 0,
(5)
∗
∗
X
∗
∗
X
∗
X
so that
+
kX
∗
∗
∗
ΘX
=
pXΘ = KX pXΘ ,
∗ ∗
−
kX
∗
(6)
GX(p, T ) = EX + EZPE + ꢀH(T ) − T S(T ) + RT ln(p/p0),
(1)
∗
where pX is the gas phase pressure of species X, while KX is the
ratio of the forward to backward rate constants. This may be ex-
pressed explicitly in terms of the adsorption energy Eads[X] and
gas phase entropy SX(g) of species X as
where EZPE is the zero point energy, ꢀH(T ) denotes the enthalpy
change due to raising the temperature from 0 K to T , S(T ) is the
entropy at T , R is the universal gas constant, and p0 denotes the
standard pressure (taken to be 1 bar).
ꢂ
ꢃ
Eads[X] + T SX(g)
∗
∗
KX ≈ exp −
,
(7)
The potential energy of an adsorbed species X , EX, is given by
RT
∗
∗
E
− E , where E is the energy of the clean surface and E
∗/X
∗/X
∗
∗
∗
where we again assume SX ꢀ SX and ꢀEZPE is much smaller than
is the energy of the adsorbate and the surface system. In order to
calculate the free energy of this species, we neglect the pressure
term, so that the enthalpy change is replaced by the change in
internal energy. This leads to the following expression for the free
energy GX(p, T ):
the adsorption energy of species X.
3. Results
3.1. Preliminary investigations
∗
GX(p, T ) = EX + EZPE + ꢀU(T ) − T S(T ).
(2)
We shall first focus on the thermodynamically unstable N2O,
when it is applied to the catalyst and to the bare support material
only. At temperatures below 160 C no dissociation of N2O could
be measured on either Catalyst B, Catalyst C or bare TiO2 (Degussa
P25). However, in the presence of a reducing agent, e.g. H2 or CO,
N2O could be reduced over the catalysts, but not over the bare
support material [31].
Additionally, CO and O2 were applied to bare TiO2 in a stoi-
chiometric ratio of 2:1 to check whether (R2) proceeds on the bare
support. This was not the case.
The forward rate constant for a reaction i may then be ex-
pressed in terms of the Gibb’s free energies as ki = vi exp[−ꢀG/
◦
ꢀ
ꢀ
RT ], where ꢀG =
GX
−
GX is the difference of the
X∈P
X∈R
total Gibb’s free energy of the products P and reactants R for re-
action i, and R is the universal gas constant. The pre-exponential
factor in units of s
stant.
−1
is vi ≡ kBT /h, where kB is Boltzmann’s con-
∗
The DFT energies for the adsorbed species EX as well as the
activation barriers are given in Table 1. For the N2O adsorption
energy, we have used the experimental value given in Ref. [43].
3.2. TEM investigation on changes in particle size
2.5. Microkinetic model
Prior to discussing individual reactions, we will first analyze
TEM images of the catalysts (cf. Fig. 2) to evaluate the influence
of pretreatment and catalytic reactions on the gold particles. Nu-
merical values for the change in particle size of each catalyst are
summarized in Table 2.
Using the notation of Ref. [44], we may express any heteroge-
neous catalysis reaction in the form
∗
R(g) → P(g),
(R5)
∗
where R(g) denotes the gas phase reactants, “ ” the active sites on
The TEM images in Fig. 2a show typical images of Catalyst A
from its state as supplied (i), its state after the pretreatment (ii)
the catalyst, and P(g) the products in the gas phase. If we assume

G. Walther et al. / Journal of Catalysis 260 (2008) 86–92
89
Fig. 2. Bright field TEM images of TiO2 supported gold particles of Catalyst A (a), Catalyst B (b) and Catalyst C (c), as they were supplied/prepared (i), after the pretreatment (ii)
and after catalytic reactions (iii). A summary of the change in particle size may be found in Table 2. The scale bars correspond to 10 nm in each image.
Table 2
Table 3
−1
for CO and H2 oxidation by O2 and N2O,
Average TiO2 supported gold particle size d in nm, as determined by TEM, before
and after performing oxidation reactions.
Apparent activation energies in kJ mol
with an uncertainty of 2 kJ mol
−1
.
−1
Ea (kJ mol
)
Catalyst A
Catalyst B
Catalyst C
dAu (nm)
Gold
concentration
Total gold
loading
As supplied†
or prepared‡
Before
reaction
After
reaction
CO + 1 O2
36 (36.4)
37 (37.5)
38
38
40
38
40
60
60
61
60
2
Catalyst A 1.02 wt%
Catalyst B
Catalyst C
62.22 μg
63.96 μg
54.00 μg
2.3 0.6†
3.4 0.8†
13.3 6.5‡
2.6 0.6
3.8 1.1
16.2 6.6
3.6 0.8
5.7 3.4
16.2 6.6
CO + N2O
1.56 wt%
1.60 wt%
H2
H2 + N2O
+
1 O2
2
39
Note. Theoretical values obtained using the microkinetic model in the low tempera-
ture regime are also provided in parentheses for Catalyst A.
and its state after catalytic reactions (iii). The particles may poten-
tially sinter during storage. However, this was not the case, since
the average of the particle size in (a) (2.3 0.6 nm), based on TEM
images of 97 particles, is consistent with that stated by the sup-
plier (2.0 nm). The particles appear predominately in a truncated
cuboctahedron shape. Following the pretreatment, the average par-
ticle size was determined to be 2.6 0.6 nm. The particles them-
selves were still well dispersed on the support, as can be seen in
Fig. 2a(ii). Following the catalytic reactions, the gold particles were
sintered to 3.6 0.8 nm.
The size of the gold particles from Catalyst B stated by the sup-
plier (3.3 nm), was also confirmed, as may be seen in Fig. 2b(i).
Fig. 2b(ii) shows the particles observed after pretreatment, which
were only slightly larger. However, after reactions the gold parti-
cles’ size increased significantly, cf. Fig. 2b(iii).
3.3. Steady-state activity measurements
Fig. 3a shows Arrhenius plots for (R1)–(R4) on Catalyst A. Sur-
prisingly, the activation energies obtained were all approximately
37 kJ mol−1, as may be seen from the parallel linear fits. Table 3
summarizes these and the following activation energies. However,
for CO oxidation the reaction rate differs by a factor of 2.7, when
comparing the different oxidation agents with each other. For H2
oxidation, the catalyst behaves similarly. The linear fits run par-
allel and the reaction rate differs by 3.2. On the Au particles of
Catalyst B, which were sintered to 5.7 nm, the trend depicted on
Catalyst A continues, but is even more distinct, as shown in Fig. 3b.
There is approximately one order of magnitude between the rates
of CO oxidation and for H2 oxidation, using N2O and O2, respec-
tively.
Fig. 2c shows Catalyst C, which was formed in-house. Already
during preparation, the gold particles sintered from 5.0 nm, as
supplied in solution, to 13.3 6.5 nm and formed truncated cuboc-
tahedrons. The pretreatment led to further sintering, with the av-
erage particle size becoming 16.2 6.6 nm. This catalyst did not
show any further changes in size after catalytic reactions.
For CO oxidation by O2 on Catalyst B, the reaction rate was
◦
found to reach a maximum at about 80 C, cf. Fig. 3b. One rea-
son for this may be that the catalyst was operated under mass-
transport controlled conditions, e.g. when too little active material
has been used. On the other hand, this may also be due to CO
or O2 desorption. Despite this, the parallel running fits of the

90
G. Walther et al. / Journal of Catalysis 260 (2008) 86–92
−1
for CO
Fig. 4. Arrhenius plot of the rds turnover frequency frds in kB T /h ≈ 1013
s
oxidation by N2O on Au{532} (blue dashed line) and by O2 on a Au12 cluster (black
dash-dotted line) as obtained from the microkinetic model using the DFT transi-
tion state structures depicted above [30]. The apparent activation barriers (red solid
lines) in the low temperature regime (T < 350 K) are shown for ease of comparison
with Fig. 3, and listed in Table 1.
shape is the top half of a regular cuboctahedron, as is the case for
our catalysts cf. Fig. 2.
On the other hand, the reactivity of the Au–TiO2 interface intro-
−2
duces a d
correction term to the reaction rate, which becomes
important for larger gold nanoparticles similar to Catalyst C. How-
ever, due to the computational complexity of properly treating the
Au–TiO2 interface for large nanoparticles, we shall restrict consid-
eration to reactions which occur on the corner sites of the Au
particles.
We begin by first considering CO oxidation using N2O. It has
been generally accepted that the reaction kinetics for CO oxidation
over a metal surface follow the Langmuir–Hinshelwood mecha-
nism. According to this, the elementary steps of adsorption and
desorption are in equilibrium, and (R1) may be written in terms of
the following elementary steps:
Fig. 3. Arrhenius plots of the activation energies for the four reactions on TiO2 sup-
ported gold particles of Catalyst A (a), Catalyst B (b) and Catalyst C (c): CO + 1 O2
2
("), CO + N2O (Q), H2
+
1 O2 (!) and H2 + N2O (P). An overview of the activation
energies is given in Table 23.
Arrhenius plots again indicate similar activation energies around
39 kJ mol−1, as shown in Table 3. These two catalysts where active
even at room temperature, although O2 was a significantly better
oxidizing agent. For (R1), N2 released from N2O could be balanced
with the CO2 formed, indicating that the CO2 does not originate
from other sources.
∗
CO(g) + ∗ ꢁ CO ,
(R1a)
(R1b)
(R1c)
Fig. 3c shows Arrhenius plots for Catalyst C and demonstrates
that on this catalyst N2O is the better oxidizing agent for CO,
N2O(g) + . ꢁ N2O.,
rds
CO + N2O. → CO2(g) + N2(g) + ∗ + ..
∗
whereas no difference in rate could be found for H2 oxidation. A
−1
very similar apparent activation energy of about 60 kJ mol
was
Here we have assumed that CO and N2O adsorb on different cor-
again found for all the four studied reactions. On these large par-
∗
ner sites, denoted by “ ” and “.” respectively. This is clearly the
◦
ticles an onset of conversion could not be observed below 80 C.
case for CO oxidation, as may be seen from the transition states
depicted in the insets of Fig. 4.
3.4. Theoretical results
Since we found N2O does not dissociate spontaneously on gold,
as has also been reported by Gluhoi et al. [31], we assume this
reaction requires adsorbed CO as a reducing agent. However, the
It has recently been shown that for small Au nanoparticles
∗
(d ꢀ 5 nm) at 273 K, experimentally obtained CO oxidation rates
−3
overall barrier for (R1c) should still be the same as for CO oxida-
tion with atomic O. On the other hand, the desorption of both N2
and CO2 should occur spontaneously.
To model the corner sites of a gold nanoparticle for reaction
(R1c), we have chosen a Au{532} surface, which consists of both
B5 and kink sites. This allows atomic O to adsorb on the preferred
B5 sites while CO adsorbs on the kink sites. This is depicted in
the Au{532} transition state for (R1c), shown in the lower inset
of Fig. 4. Since the Au12 cluster model of a corner site [30] does
not include B5 sites for atomic O adsorption, it yields an activation
follow a d
relationship, independent of the support material
used [6,11,19]. This strongly suggests that corner sites are the ac-
tive sites on small gold nanoparticles, in agreement with recent
DFT studies [11,18]. By extrapolating the linear fit of our CO oxida-
tion rate measurements for Catalyst A to 273 K, cf. Fig. 3a, we find
−1
an estimated reaction rate of approximately 66 μmol g−1 s
d ≈ 3.6 nm. This agrees well with the d
for
−3
relationship of Nørskov
et al. [6], where a fit to data from over 60 different sources on
five different supports is presented. This is not altogether surpris-
ing, since this fit was generated by assuming the gold nanoparticle

G. Walther et al. / Journal of Catalysis 260 (2008) 86–92
91
barrier for (R1c) twice that found for the Au{532} surface. For this
4. Discussion
reason, the Au12 model should not be used to model (R1c).
Employing the microkinetic model described above, the turn-
over frequency for the rate determining step (rds) frds in s−1, may
then be approximated by
Table 3 shows the overall trend in apparent activation energies
for the reactions measured on Catalysts A, B and C. First, there is a
clear size dependence trend from large gold particles with less re-
activity to the significantly more reactive smaller ones for all the
reactions studied. Second, there is very little difference in the ap-
parent overall activation barrier for CO or H2 oxidation, irrespective
of the oxidizing agent used. This is seen for both experimental
and theoretical apparent activation barriers, which are in quanti-
tative agreement for Catalyst A (d ꢀ 5 nm), in the low temperature
regime (T < 350 K), as shown in Table 3. However, there is little
correlation between the experimental reaction rate and theoretical
turnover frequency for the larger Catalyst C. This may be expected,
as for larger Au nanoparticles (d ꢂ 10 nm), the bulk gold properties
may begin to dominate, so that a cluster-based theoretical model
is no longer applicable.
Clearly it would be desirable to get a direct comparison be-
tween measured rates and calculated turnover frequencies. How-
ever, such a comparison requires atomic resolution characterization
of the particles in question. Given the large dispersity in particle
size for our catalysts we are unable to provide such experimental
detail. However, it should be born in mind that the trends for the
theoretical turnover frequency should match those of the experi-
mental rate, as shown in Table 3.
We shall first focus on CO oxidation using N2O according to
(R1). From the preliminary investigations it may be concluded that
in the absence of a reducing agent, the dissociation of N2O is in-
hibited. This may be accounted for by the very weak Au–N2O bond
which leads to rapid desorption of N2O [31]. However, this reac-
tion changes the overall entropy of the system only slightly, since
one diatomic and one triatomic species both adsorb and desorb
from the surface in (R1). This means that although there is only
a small N2O coverage, for high CO coverage experiments the reac-
tion rate should be significant. On the other hand, as seen in Fig. 4,
at higher temperatures (T > 350 K) CO begins to desorb from the
surface and the turnover frequency begins to decrease at higher
temperatures (T > 360 K). We were unable to verify this experi-
mentally since we found gold particle sintering began to occur in
this temperature range, as shall be discussed later.
∗
kBT
h
KCO
K
. pCO p
N2O
N2
O
frds
≈
∗
(1 + KCO
p
CO)(1 + KN . pN
)
2O
2O
ꢂ
ꢃ
Ea[rds] − T (SCO (g) + SN (g)
)
2
2
× exp −
,
(8)
RT
∗
N2O.
where KCO and K
are given in (7). The required activation
and adsorption energies for the Au{532} surface are provided in
Table 1 and the temperature dependent gas phase entropies are
interpolated using data from Ref. [36].
Fig. 4 shows the temperature dependence of the turnover fre-
quency frds, for the rds of CO oxidation by N2O on Au{532}. We
found frds follows an Arrhenius-like behavior in the low temper-
ature regime (T < 350 K). However, at higher temperatures the
turnover frequency decreases with increasing temperature, with an
apex at T ≈ 360 K.
When using O2 instead of N2O to oxidize CO, we may again as-
sume a Langmuir–Hinshelwood mechanism, so that the elementary
steps for (R2) may be written as
∗
CO(g) + ∗ ꢁ CO ,
(R2a)
(R2b)
O2(g) + . ꢁ O.2 ,
rds
∗
∗
CO + O.2 → CO2 + O.,
(R2c)
(R2d)
CO + O. ⇒ CO2(g) + ∗ + .,
∗
where “ ” and “.” denote active sites for CO and O2 respectively,
∗
on the Au particle.
Since O2 is known not to dissociate spontaneously on gold
nanoparticles [45], we assume CO is oxidized directly by O2 (R2c),
as discussed in Refs. [30,46]. We further assume (R2c) is the rate
determining step, so that (R2d) occurs relatively quickly. This is
justified in the case of high CO coverages, since CO2 will quickly
desorb from the surface.
To model the gold nanoparticle corner sites for reaction (R2c),
we have used the Au12 cluster model described in Ref. [30]. As
seen in Table 1, such a model yields much higher adsorption ener-
gies, particularly for O2. This is necessary to correctly model (R2c)
on gold nanoparticles. As shown in the upper inset of Fig. 4, the
transition state for (R2d) on the Au12 cluster model has the O2
molecule strongly adsorbed on the corner site, while CO is ad-
sorbed on the edge.
For the case of CO oxidation by O2 according to (R2), there is a
significant loss of entropy in the overall reaction, as three diatomic
species adsorb, but two triatomic species desorb. Thus a high O2
adsorption energy is required for the reaction rate to be significant.
As this is the case for small gold nanoparticles and the Au12 cluster
model for a corner site, we employ this model for (R2). Even so,
we find at higher temperatures (T > 370 K) the reaction rate de-
creases significantly as both O2 and CO begin to desorb from the
gold nanoparticle. This may be seen in the reaction rate plots for
CO oxidation by O2 on Catalyst B, shown in Fig. 3. Here, the reac-
tion rate begins to flatten at about 370 K, in agreement with the
theoretical turnover frequency shown in Fig. 4.
Besides the activity, stability is the next most important prop-
erty of an industrial catalyst. With a view on the gold loading of
the catalysts used, noted in Table 2, it seems clear that small par-
ticles with a high loading are more likely to sinter than with lower
loadings. One reason for this is that due to the low melting point
of gold [47,48] or quasi-melting [49], gold particles already be-
come mobile at only slightly elevated temperatures. The larger the
particles at a constant loading are, the fewer particles are on the
support, and the larger is the distance to their nearest neighbor.
Fig. 2 illustrates the sintering for different gold particle sizes
during the pretreatment, when comparing image (i) with image
(ii), as well as during the activity measurements. Since the activity
measurements were performed in a sequence (as described above),
Again employing our microkinetic model, frds for (R2) may be
approximated by
∗
.
KO KCO pO pCO
kBT
h
2
2
frds
≈
∗
(1 + KO. pO )(1 + KCO pCO
)
2
2
ꢂ
ꢃ
Ea[rds] − T S
CO2(g)
× exp −
,
(9)
RT
∗
.
where KCO and KO are given in (7). The necessary activation
2
and adsorption energies for the Au12 cluster model are provided
in Table 1 and the temperature dependent gas phase entropies are
interpolated using data from Ref. [36].
The temperature dependence of frds for CO oxidation by O2 on
Au12 is shown in Fig. 4. We again found frds has an Arrhenius
behavior in the low temperature regime (T < 350 K). As with N2O,
we found at higher temperatures the turnover frequency decreases
with increasing temperature, with an apex at T ≈ 370 K.

92
G. Walther et al. / Journal of Catalysis 260 (2008) 86–92
◦
the change in particle size may be related to 40 C more in heat
applied than for the catalytic reactions driven on Catalyst B.
Catalyst A was not heated more than the pretreatment required.
This means that the sintering on this catalyst is related only to
how the reactions proceeded. Across the general observations pub-
lished regarding CO oxidation on TiO2 supported gold using O2 as
an oxidizing agent, sintering of nanoparticulate gold has not been
reported. Gold nanoparticles have also been found to be a stable
catalyst in the presence of H2 and H2O [50]. This suggests that the
sintering observed on Catalyst A is related to the supply of N2O.
In contrast, the particles of Catalyst C did not significantly sinter,
[4] G.C. Bond, D.T. Thompson, Catal. Rev. Sci. Eng. 41 (1999) 319–388.
[5] G.C. Bond, C. Louis, D.T. Thompson, Catalysis by Gold, first ed., Imperial College
Press, London, 2006.
[6] B. Hvolbæk, T.V.W. Janssens, B.S. Clausen, H. Falsig, C.H. Christensen, J.K.
Nørskov, Nano Today 2 (2007) 14–18.
[7] D.E. Starr, S.K. Shaikhutdinov, H.-J. Freund, Top. Catal. 36 (2005) 33–41.
[8] H.-J. Freund, Catal. Today 117 (2006) 6–17.
[9] M. Valden, X. Lai, D.W. Goodman, Science 281 (1998) 1647–1649.
[10] M.S. Chen, D.W. Goodman, Science 306 (2004) 252–255.
[11] T.V.W. Janssens, A. Carlsson, A. Puig-Molina, B.S. Clausen, J. Catal. 240 (2006)
108–113.
[12] C. Xu, J. Su, X. Xu, P. Liu, H. Zhao, F. Tian, Y. Ding, J. Am. Chem. Soc. 129 (2007)
42–43.
◦
[13] C. Xu, X. Xu, J. Su, Y. Ding, J. Catal. 252 (2007) 243–248.
[14] R.J.H. Grisel, B.E. Nieuwenhuys, Catal. Today 64 (2001) 69–81.
[15] M. Mavrikakis, P. Stoltze, J.K. Nørskov, Catal. Lett. 64 (2000) 101–106.
[16] Z.-P. Liu, P. Hu, A. Alavi, J. Am. Chem. Soc. 124 (2002) 14770–14779.
[17] N. Lopez, J.K. Nørskov, J. Am. Chem. Soc. 124 (2002) 11262–11263.
[18] N. Lopez, T.V.W. Janssens, B.S. Clausen, Y. Xu, M. Mavrikakis, T. Bligaard, J.K.
Nørskov, J. Catal. 223 (2004) 232–235.
[19] T.V.W. Janssens, B.S. Clausen, B. Hvolbæk, H. Falsig, C.H. Christensen, T. Bligaard,
J.K. Nørskov, Top. Catal. 44 (2007) 15–26.
[20] B.E. Solsona, T. Garcia, C. Jones, S.H. Taylor, A.F. Carley, G.J. Hutchings, Appl.
Catal. A Gen. 312 (2006) 67–76.
even though the maximum temperature was 80 C higher than the
temperature of the pretreatment.
5. Conclusions
We have investigated two different reactions on three TiO2 sup-
ported gold catalysts with two different oxidizing agents. Experi-
mental results for CO oxidation on Catalysts A and B agree with
both our theoretical model and the literature, following the d
trend advocated by Nørskov et al. [6]. However, for Catalyst C we
found more bulk-like behavior.
−3
[21] D.T. Thompson, Top. Catal. 38 (2006) 231–240.
[22] R.D. Waters, J.J. Weimer, J.E. Smith, Catal. Lett. 30 (1994) 181–188.
[23] S. Ivanova, C. Petit, V. Pitchon, Appl. Catal. A Gen. 267 (2004) 191–201.
[24] S. Ivanova, C. Petit, V. Pitchon, Gold Bull. 39 (2006) 3–8.
[25] C.H. Christensen, B. Jorgensen, J. Rass-Hansen, K. Egeblad, R. Madsen, S.K. Klit-
gaard, S.M. Hansen, M.R. Hansen, H.C. Andersen, A. Riisager, Angew. Chem. 45
(2006) 4648–4651.
[26] M.-C. Daniel, D. Astruc, Chem. Rev. 104 (2004) 293–346.
[27] G. Walther, G. Jones, S. Jensen, U.J. Quaade, S. Horch, Catal. Today, in press.
[28] M. Haruta, Catal. Today 36 (1997) 153–166.
[29] J.A. van Bokhoven, C. Louis, J.T. Miller, M. Tromp, O.V. Safonova, P. Glatzel,
Angew. Chem. Ger. Ed. 118 (2006) 4767–4770.
[30] H. Falsig, B. Hvolbæk, I.S. Kristensen, T. Jiang, T. Bligaard, C.H. Christensen, J.K.
Nørskov, Angew. Chem. 47 (2008) 4835–4839.
[31] A.C. Gluhoi, M.A.P. Dekkers, B.E. Nieuwenhuys, J. Catal. 219 (2003) 197–205.
[32] L.M. Molina, B. Hammer, Phys. Rev. B 69 (2004) 155424.
[33] U.J. Quaade, S. Jensen, O. Hansen, J. Appl. Phys. 97 (2005) 44906.
[34] http://www.mintek.co.za.
[35] http://www.gold.org.
[36] D.R. Lide, Handbook of Chemistry and Physics, 87th ed., CRC Press, Boca Raton,
2006–2007.
Based on our theoretical model, we find oxidizing CO by N2O
involves a CO–O transition state, with atomic O adsorbed on the
gold B5 sites and CO on the corners. On the other hand, CO ox-
idation by molecular O2 occurs via a different reaction pathway,
which instead involves a meta-stable intermediate CO–O2. In this
case, O2 is strongly adsorbed on gold corner sites while CO is ad-
sorbed on the nearby edge sites (cf. Fig. 2).
However, although the two oxidizing agents used proceeded via
different reaction pathways on different active sites, the apparent
overall activation barriers obtained from both theory and experi-
ment were found to be the same. From experiment, we found that
H2 and CO oxidation proceed similarly, with common activation
barriers and rate enhancements when comparing oxidizing agents.
Additionally, from our TEM analysis we conclude that N2O ox-
idation may promote sintering of Au nanoparticles on TiO2, since
significant sintering has not been found for CO and H2 oxidation
on gold nanoparticles using O2 [27,50].
[37] http://wiki.fysik.dtu.dk/dacapo.
[38] D. Vanderbilt, Phys. Rev. B 41 (1990) 7892–7895.
[39] B. Hammer, L.B. Hansen, J.K. Nørskov, Phys. Rev. B 59 (1999) 7413–7421.
[40] J.K. Nørskov, J. Rossmeisl, A. Logadottir, L. Lindqvist, J.R. Kitchin, T. Bligaard, H.
Jonsson, J. Phys. Chem. B 108 (2004) 17886–17892.
Acknowledgments
[41] S. Kurth, J.P. Perdew, P. Blaha, Int. J. Quant. Chem. 75 (1999) 889–909.
[42] J.H. Noggle, Physical Chemistry, third ed., Harper Collins, New York, 1996.
[43] B.E. Nieuwenhuys, A.C. Gluhoi, E.D.L. Rienks, C.J. Weststrate, C.P. Vinod, Catal.
Today 100 (2005) 49–54.
[44] M. Boudart, G. Djéga-Mariadassou, Kinetics of Heterogeneous Catalytic Reac-
tions, Princeton University Press, Princeton, NJ, 1984.
[45] J.K. Nørskov, T. Bligaard, A. Logadottir, S. Bahn, L.B. Hansen, M. Bollinger, H.
Bengaard, B. Hammer, Z. Sljivancanin, M. Mavrikakis, Y. Xu, S. Dahl, C.J.H. Ja-
cobsen, J. Catal. 209 (2002) 275–278.
[46] L.M. Molina, B. Hammer, Appl. Catal. A Gen. 291 (2005) 21–31.
[47] P. Buffat, J.P. Borel, Phys. Rev. A 13 (1976) 2287–2298.
[48] H. Sykes, E. Charles, F.J. Williams, M.S. Tikhov, R.M. Lambert, J. Phys. Chem.
B 106 (2002) 5390–5394.
The authors gratefully acknowledge support in the form of cat-
alysts provided by Project AuTEK. We also thank F.B. Grumsen
for assistance with taking TEM images, and T. Bligaard for useful
discussions. G. Walther and D.J. Mowbray also acknowledge finan-
cial support from NABIIT. The Center for Atomic-scale Materials
Design is funded by the Lundbeck Foundation. The authors also
acknowledge support from the Danish Center for Scientific Com-
puting through grant HDW-1103-06.
References
[49] P.M. Ajayan, D.L. Marks, Phys. Rev. Lett. 60 (1988) 585–587.
[50] P. Landon, J. Ferguson, B.E. Solsona, T. Garcia, S. Al-Sayari, A.F. Carley, A.A. Herz-
ing, C.J. Kiely, M. Makkee, J.A. Moulijn, A. Overweg, S.E. Golunski, G.J. Hutch-
ings, J. Mater. Chem. 16 (2006) 199–208.
[1] B. Hammer, J.K. Nørskov, Nature 376 (1995) 238–240.
[2] M. Haruta, T. Kobayashi, H. Sano, N. Yamada, Chem. Lett. (1987) 405–408.
[3] A.S.K. Hashmi, G.J. Hutchings, Angew. Chem. 45 (2006) 7896.