Mechanism of selective alcohol oxidation to aldehydes on gold catalysts: Influence of surface roughness on reactivity

Article abstract of DOI:10.1016/j.jcat.2010.11.013

The complete reaction pathway for the selective alcohol oxidation to aldehyde has been obtained from periodic density functional theory (DFT) calculations on a series of catalyst models having Au atoms with different coordination number. A clear trend exists in the reactivity of the surfaces: it increases with decreasing the Au coordination number, suggesting that a distribution of Au sites with different activity might exist on the surface of real catalysts. This hypothesis emerging from the theoretical study has been experimentally confirmed by measuring the kinetics of benzyl alcohol oxidation on a series of Au/MgO catalysts having different particle diameter and therefore different concentration and distribution of low coordinated Au sites.

Full text of DOI:10.1016/j.jcat.2010.11.013

Journal of Catalysis 278 (2011) 50–58
Contents lists available at ScienceDirect
Journal of Catalysis
journal homepage: www.elsevier.com/locate/jcat
Mechanism of selective alcohol oxidation to aldehydes on gold catalysts: Influence
of surface roughness on reactivity
b
Mercedes Boronat a, Avelino Corma a, , Francesc Illas , Juan Radilla b,c, Tania Ródenas a, María J. Sabater a
⇑
a Instituto de Tecnología Química, Universidad Politécnica de Valencia – Consejo Superior de Investigaciones Científicas, Av. los Naranjos, s/n, 46022 Valencia, Spain
b Departament de Química Física & Institut de Química Teòrica i Computacional (IQTCUB), Universitat de Barcelona, C/Martí i Franquès 1, 08028 Barcelona, Spain
c Departamento de Ciencias Básicas, Área de Química, Universidad Autónoma Metropolitana-Azcapotzalco, Av. San Pablo 180, CP. 02200 México, D.F., Mexico
a r t i c l e i n f o
a b s t r a c t
Article history:
The complete reaction pathway for the selective alcohol oxidation to aldehyde has been obtained
from periodic density functional theory (DFT) calculations on a series of catalyst models having Au atoms
with different coordination number. A clear trend exists in the reactivity of the surfaces: it increases with
decreasing the Au coordination number, suggesting that a distribution of Au sites with different activity
might exist on the surface of real catalysts. This hypothesis emerging from the theoretical study has been
experimentally confirmed by measuring the kinetics of benzyl alcohol oxidation on a series of Au/MgO
catalysts having different particle diameter and therefore different concentration and distribution of
low coordinated Au sites.
Received 6 October 2010
Revised 18 November 2010
Accepted 19 November 2010
Available online 6 January 2011
Keywords:
Catalysis
Au
Ó 2010 Elsevier Inc. All rights reserved.
DFT
Alcohol oxidation
Selectivity
1. Introduction
has been recently reported that a bifunctional Pd/MgO catalyst
performs the one-pot selective N-monoalkylation of amines with
The selective oxidation of alcohols to carbonylic compounds
over heterogeneous recyclable catalysts is an important process
in green organic chemistry [1–3]. Most efforts in recent years have
focused on the aerobic oxidation that uses molecular oxygen as
oxidant generating water as the only byproduct, and good results
have been obtained with metal-based catalysts containing Ru
[4,5], Ag [6,7], Cu [8,9], Au [10–14] or Pd [15–17]. Different studies
[1–6,11,14] agree that the mechanism for aerobic alcohol oxidation
starts with the formation of a metal-alkoxy species that, in a sec-
ond step, suffers a b-hydride elimination giving rise to the carbony-
lic product and a metal-hydride intermediate. Although it has been
shown that in some cases the alkoxy intermediate is not formed on
the metal particle but on the oxide support [6,10,14], there are
many other examples demonstrating that naked metal particles
in colloidal solutions [18] or supported on carbon [7,19,20], silica
[12,14,16,17] or metal organic frameworks (MOF) [21,22] are
highly active for alcohol oxidation, even in the absence of a base
[21]. Moreover, instead of using molecular oxygen to finally obtain
nontoxic but valueless H2O as a byproduct, it would be more inter-
esting to use the carbonyl compound and metal-hydride species
obtained from oxidative alcohol dehydrogenation in consecutive
reactions yielding products of high added value. In this sense, it
alcohols. This occurs through a series of consecutive steps in a cas-
cade mode involving alcohol dehydrogenation, condensation of the
carbonyl compound with the amine yielding an imine, and hydro-
genation of the imine with the hydrogen atoms from the metal hy-
dride obtained in the first step [23]. Thus, it is of interest to
understand the mechanism of the metal-catalyzed dehydrogena-
tion of alcohols in the absence of O2 or of a base.
In the case of gold-catalyzed alcohol oxidation, it has been
found that catalytic activity and selectivity depend on metal parti-
cle size, and that the best performance is exhibited by the smaller
particles [14,22,24]. This observation could be explained by taking
into account that, as the particle size decreases, the number of low
coordinated atoms that are considered to be the active sites in
many gold-catalyzed reactions increases. To confirm this hypothe-
sis, we have attacked the problem from two different and comple-
mentary perspectives. On one hand, the complete mechanism of
ethanol oxidation to ethanal has been theoretically investigated
by means of periodic density functional theory (DFT) based
calculations on a series of catalyst models having Au atoms with
different coordination number. On the other hand, different gold
catalysts with particle size between 2 and 10 nm have been syn-
thesized, and the kinetics of benzyl alcohol dehydrogenation have
been measured. By combining the theoretical and experimental
results, it is shown that the presence of low coordinated Au atoms
enhances alcohol adsorption on the catalyst and decreases the
⇑
Corresponding author. Fax: +34 96 387 79 444.
E-mail address: acorma@itq.upv.es (A. Corma).
0021-9517/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcat.2010.11.013

M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
51
activation barrier for dehydrogenation, thus explaining the influ-
ence of particle size on the oxidation activity of gold catalysts.
the study because its behavior in relation to ethoxy adsorption is
not completely equivalent to that of the Au(5 1 1) surface. Finally,
an Au38 nanoparticle having a typical cuboctahedral shape and
ꢂ1 nm diameter was placed in a 20 ꢁ 20 ꢁ 20 Å cubic box, large
enough to allow a separation larger than 10 Å between periodically
repeated particles. In the geometry optimization of the extended
surfaces the uppermost layers were fully relaxed, while the atomic
positions of the two innermost layers were kept fixed as in the bulk
to provide an adequate environment to the surface atoms. In the
case of the Au38 nanoparticle, the positions of all atoms were al-
ways fully relaxed.
Ethanol adsorption on the different Au models described above
was considered and the complete reaction pathway for its oxida-
tion to acetaldehyde was obtained. This includes geometry optimi-
zation of the adsorbed ethanol reactant, the ethoxy reaction
intermediate plus a hydrogen atom, the adsorbed acetaldehyde fi-
nal reaction product plus two adsorbed hydrogen atoms, and the
transition states connecting all these minima. In all calculations,
the atomic positions of C, O, and H atoms, as well as those of the
Au atoms in the uppermost layers of the surfaces as described ear-
lier and all Au atoms in the nanoparticle were fully relaxed.
Total energies were obtained from periodic density functional
calculations using the Perdew–Wang (PW91) implementation of
the Generalized Gradient Approach (GGA) for the exchange–corre-
lation potential [26–28]. The valence electron density was ex-
panded in a plane wave basis set with a kinetic energy cutoff of
415 eV. The effect of the core electrons in the valence density
was taken into account by means of the Projected Augmented
Wave (PAW) formalism [29] as implemented by Kresse and Joubert
[30] in the VASP code [31,32]. Integration in the reciprocal space
was carried out using a 5 ꢁ 5 ꢁ 1 mesh of special Monkhorst–Pack
k-points [33] for the gold surfaces, while calculations with the gold
2. Experimental section
2.1. Catalyst preparation
Au–MgO catalysts were prepared by impregnation of 1 g of a
MgO sample with a surface area of 670 m2/g from Nanoscale Mate-
rials with 10 ml of HAuCl4ꢀ3H2O solutions containing different
amounts of gold (0.4%, 0.8%, 1.5%, 3.0%, 5.0%, and 10% weight)
under stirring for 2 h. The solvent was evaporated at reduced
pressure, and the solid was dried at 333 K during 6 h. Finally, the
sample was calcined at 573 K under air flow for 3 h and then under
N2 flow for 3 h.
2.2. Kinetic experiments
The kinetic experiments were carried out by working at differ-
ent initial concentrations of benzyl alcohol (0.4, 0.8, 1.0, 1.2, and
1.6 mmol/ml) and with a catalyst concentration corresponding in
all cases to 0.0075 mmol of Au. The reactants and the catalyst were
introduced into an autoclave containing 20 lL n-dodecane and
1 ml trifluorotoluene. Once closed, the reactor was purged with
N2 and the reaction mixture was heated to the desired temperature
(413, 423, 453, 463, 473, and 483 K) and stirred at 500 r.p.m.
during the whole experiment. The reaction was monitored by GC.
2.3. Model catalysts and computational details
Perfect and stepped single crystal Au surfaces as well as an iso-
lated Au nanoparticle were represented by the usual periodic slab
model approach, using super cells large enough to avoid interac-
tion either between the periodically repeated gold surfaces (or par-
ticles) or between the periodically repeated adsorbates. The
extended surfaces were modeled by (2 ꢁ 2) super cell slabs con-
taining a material thickness of ꢂ8 Å and a vacuum region of
ꢂ12 Å, as shown in Fig. 1. The slab model for the Au(1 1 1) surface
contains 16 atoms arranged in four layers. The stepped Au(5 1 1)
surface has (1 0 0) terraces and (1 1 1) steps, and the unit cell con-
tains 44 atoms distributed in eleven metallic layers. Another
stepped surface hereafter termed Au-rod and containing 36 atoms
in the unit cell was built from the 4 ꢁ 2 unit cell of the (1 0 0) slab
as described in previous work [25]. The Au-rod model has two
types of (1 0 0) terraces (upper and lower) separated by a mono-
atomic step with (1 1 1) orientation, and it has been included in
nanoparticle were carried out at the
Ck-point of the Brillouin zone.
The atomic positions were optimized by means of a conjugate-gra-
dient algorithm until atomic forces were smaller than 0.01 eV/Å.
Transition states were located using the DIMER algorithm [34,35]
and stationary points thus found were characterized by pertinent
frequency analysis calculations. Vibrational frequencies were cal-
culated by diagonalizing the block Hessian matrix corresponding
to displacements of the C, O, and H atoms of the organic fragment,
and the zero point vibrational correction was added to the calcu-
lated energies. Charge distributions were estimated using the
quantum theory of atoms in molecules (AIM) of Bader using the
algorithm developed by Henkelman et al. [36,37]. Calculations
have been carried out without taking spin polarization into ac-
count except when needed, such as in the case of the isolated O2
molecule.
Fig. 1. Side view of the unit cell of the different gold surfaces and Au38 nanoparticle employed in this work.

52
M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
Table 1
3. Results and discussion
Optimized values (in Å) of the most relevant distances of the structures involved in
the mechanism of ethanol oxidation on gold. The atom labeling is given in Fig. 2.
3.1. Theoretical study of the reaction mechanism
Au(1 1 1)
Au(5 1 1)
Au-rod
Au38
The mechanism of selective oxidation of ethanol to ethanal con-
sists of two elementary steps: (1) dehydrogenation of the hydroxyl
group of adsorbed ethanol to give an ethoxy species and a hydro-
gen atom adsorbed on the metal surface and (2) transfer of a sec-
ond hydrogen atom from the C atom bonded to O (Hb) in the
ethoxy intermediate to the Au surface, yielding ethanal and a sec-
ond hydrogen atom adsorbed on the catalyst surface. The struc-
tures involved in the reaction mechanism on Au(5 1 1) surface,
together with a definition of the most relevant geometric parame-
ters to be analyzed, are depicted in Fig. 2. Table 1 summarizes the
optimized values of these parameters on all catalyst models con-
sidered. The calculated adsorption, reaction and activation ener-
gies are given in Table 2, and the energy profiles obtained on the
different catalyst models are schematized in Fig. 3.
Etanol
rAu–O
rOH
rAu–H
rCO
3.231
0.978
2.820
1.439
2.696
0.978
2.819
1.449
2.690
0.980
2.835
1.449
2.468
0.979
3.054
1.455
TS1
rAu–O
rOH
rAu–H
rCO
2.167
1.927
1.823
1.418
2.275, 2.323
1.703
1.671
2.335
1.587
1.908
1.440
2.087, 3.348
2.045
1.619
1.441
1.425
Ethoxy + H
rAu–O
rAu–H
rAu–Hb
rCO
2.277, 2.361
1.875, 1.944
2.943
2.201, 2.194
1.780
3.312
2.199
1.777
3.220
1.441
2.198, 2.202
1.816
3.365
1.425
1.437
1.434
TS2
Ethanol adsorbs on all surfaces and on the Au38 particle with the
O on top of an Au atom, although adsorption on highly coordinated
Au atoms present at regular surfaces is fairly weak. Therefore, only
sites located at the step edges and having coordination number 7
were considered for ethanol adsorption on the Au(5 1 1) surface
and on the Au-rod model. On the Au38 nanoparticle, ethanol
adsorption was found to occur at corner sites having coordination
number 6. The bond lengths and angles in ethanol are practically
unmodified upon adsorption, but the Au–O distance becomes
shorter as the coordination number of the Au atom with which
ethanol is interacting decreases. The calculated adsorption ener-
gies (Eads in Table 2 and Fig. 3) follow the opposite trend and in-
crease in a continuous way from Au(1 1 1) to Au38. However, the
Eads value obtained on the Au-rod model is considerably larger than
on the stepped Au(5 1 1) surface, indicating that coordination
number is not the only factor determining energies. When compar-
ing the step sites in the Au(5 1 1) and Au-rod models, one can see
that the coordination and local environment around the adsorption
site is similar along the step, but quite different across the step,
rAu–O
rC–Hb
rAu–Hb
rCO
2.411
1.614
1.712
1.274
2.271
1.616
1.689
1.279
2.273
1.618
1.692
1.280
2.202
1.698
1.685
1.279
Aldehyde + 2H
rAu–O
rCO
4.735
1.223
2.757
1.224
3.010
1.224
2.375
1.235
Table 2
Calculated adsorption, activation and reaction energies (in kcal molꢃ1) involved in the
reaction mechanism schematized in Fig. 3. The negative values indicate that the
process is exothermic.
Au(1 1 1)
Au(5 1 1)
Au-rod
Au38
Eads
Eact1
ꢃ2.9
32.7
27.2
5.2
ꢃ4.9
27.9
14.0
10.2
ꢃ1.4
ꢃ9.6
24.1
11.4
13.7
ꢃ1.3
ꢃ12.1
19.2
2.1
9.7
1.7
D
E1
Eact2
E2
D
ꢃ16.0
Fig. 2. Structures involved in the reaction mechanism on Au(5 1 1): (a) adsorbed ethanol, (b) transition state for the first (TS1) dehydrogenation, (c) co-adsorbed ethoxy and H
and (d) transition state for the second (TS2) dehydrogenation.

M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
53
Fig. 3. Calculated energy profile for the dehydrogenation of ethanol on different gold catalyst models.
Fig. 4. Structures involved in the reaction mechanism on the Au-rod model (left) and the Au38 nanoparticle (right): (a) transition state for the first dehydrogenation TS1, (b)
co-adsorbed ethoxy and H and (c) transition state for the second dehydrogenation TS2.
because the (1 0 0) terraces in the Au(5 1 1) surface are larger than
in the Au-rod model. The coordination of the Au atoms next to
those in the step in the Au(5 1 1) surface is 8, while in the Au-rod
model, the Au atoms next to those in the step where ethanol is ad-
sorbed belong to another step and are therefore coordinated to 7
Au atoms. As a result, the Au-rod model is more flexible than the
Au(5 1 1) surface and allows a better relaxation upon molecular
adsorption, which is reflected in larger adsorption energies and
lower activation barriers (see Table 2).
In the transition state for the first elementary step of the etha-
nol oxidation mechanism, TS1, the OH bond is broken and the
hydrogen atom of the hydroxyl group and the resulting ethoxy
fragment bind to different sites on the Au surfaces and nanoparti-
cles. The optimized OH (1.5–2.0 Å) and Au–H (1.6–1.9 Å) bond
lengths summarized in Table 2 reflect the hydrogen transfer
process. Moreover, the Au–O distances become shorter, while the
CO bond remains almost unaltered. After dissociation of the OH
group, the ethoxy fragment adsorbs with the C–O bond almost

54
M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
the OH bond on the Au(5 1 1) surface, the H atom occupies a bridge
35
30
25
20
15
(a)
position between two Au atoms of the Au(1 0 0) terrace having
coordination number 8. However, the same process on the Au-
rod model leaves the H atom adsorbed on the step edge, in a more
stable situation. The dissociation of the OH group of ethanol on the
Au38 nanoparticle, depicted in Fig. 4, deserves a more detailed
description. Despite all our efforts to find a transition state TS1
equivalent to those obtained on Au(5 1 1) and Au-rod models, in
which OH dissociation occurs through the square formed by the
Au atoms of the Au(1 0 0) terrace, the breaking of the hydroxyl
group on Au38 nanoparticle occurs at the edge of the Au(1 0 0)
upper facet of the particle, and only involves in a direct way two
Au atoms. Thus, the OH distance in TS1 on Au38 is the longest,
the H atom is on top of one Au atom in a corner position, and
the ethoxy fragment is on top of a second corner Au atom. The dis-
tance between these two Au atoms, which were initially bonded
with a Au–Au bond length of 2.824 Å, increases to 3.828 Å in the
transition state and reaches a value of 3.479 Å in the ethoxy + H
intermediate. This large geometrical distortion of the particle is
energetically compensated by the interactions with the organic
fragments, leading to the lowest activation and reaction energies
in the series of model catalysts studied. This is an important point
since it shows that the fluxional structure of the Au nanoparticle is
also a key factor in reducing the activation energy toward O–H dis-
sociation. It is also important to remark that, after dissociation of
the hydroxyl group, the H atom occupies a bridge position between
two Au atoms beside adsorbed ethoxy, but can easily move to the
particle surface or to other more stable bridge positions farther
away from ethoxy. The energy difference between the least and
the most stable structures obtained for ethoxy + H intermediate,
depicted in Fig. 4, is ꢂ3 kcal molꢃ1, and the data given in Table 1
correspond to the most stable system, which has also been consid-
ered as starting point for the second deprotonation step.
-15
-10
-5
0
E
ads (kcal/mol)
45
35
25
15
5
(b)
-5
-15
5
6
7
8
9
10
Au coordination number
The optimized structure of TS2 is almost similar on all surfaces
and particles considered. The ethoxy fragment is displaced from its
bridge position and moves closer to one of the Au atoms to which it
was bonded, leaving the other Au atom free to interact with the Hb
that is being abstracted. The hydrogen atom being transferred to
the Au surface is halfway between the C and the Au atoms, with
optimized C–Hb and Au–Hb distances between 1.6 and 1.7 Å. At
the same time, the CO bond length shortens to ꢂ1.28 Å, indicating
that it is being converted into a carbonyl group. Finally, ethanal is
formed and desorbs, leaving two hydrogen atoms occupying 3-fold
hollow positions on Au(1 1 1) surface and bridge positions on the
stepped surfaces and the nanoparticle. The activation barriers in-
volved in the abstraction of the Hb are smaller than those corre-
sponding to the dehydrogenation of the hydroxyl group, and do
not seem to follow any clear trend with the coordination of the
Au atoms or surface roughness.
Fig. 5. Plots of: (a) activation energy of the rate-determining step versus energy of
adsorption energy of ethanol on the different gold catalyst models and (b)
adsorption (squares), activation (circles) and reaction (triangles) energies versus
Au coordination number.
perpendicular to the surface and the oxygen atom on a 3-fold hol-
low position on Au(1 1 1) surface, and with the oxygen atom inter-
acting with two atoms of the step edge in a bridge configuration on
the Au(5 1 1), Au-rod and Au38 models (see Figs. 2 and 4). In all
cases, the calculated Au–O distances are ꢂ2.2 Å.
The calculated activation energies reported in Table 2 clearly
show that this dehydrogenation is the rate-determining step of
the reaction. Eact1 values decrease from 32.7 kcal molꢃ1 for the per-
fect Au(1 1 1) surface to 19.2 kcal molꢃ1 on the Au38 nanoparticle
and are not equivalent in the stepped Au(5 1 1) and Au-rod sur-
faces. A similar but still more pronounced trend is found for the
In order to rationalize the results obtained, we have searched
for relationships between adsorption, reaction and activation ener-
gies for the rate-determining step and the coordination of the Au
atoms involved in the process. As discussed earlier, Au atoms in
reaction energies
DE1, which reflect the higher stabilization of
the ethoxy + H intermediate as the coordination of Au atoms on
the active site decrease. As depicted in Fig. 2, after dissociation of
Table 3
Characterization data (gold loading in% weight and measured by ICP-OES analysis, particle diameter in nm and area in (m2/g) and calculated turnover frequencies (TOF in hꢃ1) for
the different catalysts considered in this work together with calculated adsorption enthalpies of benzyl alcohol (DH) and activation energies (Eact) for benzyl alcohol oxidation.
a
Au (%)
Au (ICP-OES)
Diameter (nm)
Area (m2/g)
TOF (hꢃ1
)
D
H (kcal molꢃ1
)
Eact (kcal molꢃ1
)
0.4
0.8
1.5
3.0
5.0
0.3784
0.8216
1.4371
2.9360
5.1321
10.1885
1.9
3.3
4.1
5.9
8.1
9.7
212.2
175.9
53.3
52.6
38.6
37.9
113
324
74
31
30
7.3 1.3
6.5 0.9
5.2 0.3
4.7 0.6
4.3 0.5
3.8 0.3
7.9 1.3
8.6 1.1
9.6 0.6
9.8 0.8
10.0 1.0
10.4 0.6
10.0
26
a
Calculated at T = 453 K and [alcohol] = 1 mmol/L.

M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
55
Fig. 6. TEM micrographs and particle size distribution of Au/MgO catalysts with different gold loading: (a) 0.4%wt, (b) 0.8%wt, (c) 1.5%wt, (d) 3.0%wt, (e) 5.0%wt and (f)
10.0%wt.

56
M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
where k1 is the kinetic constant corresponding to the first dehydro-
genation of the alcohol to give an alkoxy intermediate and a hydro-
gen atom adsorbed on the Au surface and k2 is the kinetic constant
corresponding to the second dehydrogenation step yielding the
aldehyde and a new hydrogen atom adsorbed on the gold surface.
However, in practice, these two steps cannot be measured indepen-
dently and a global kinetic constant kc was employed in the kinetic
rate equation:
the step edge of the Au(5 1 1) and Au-rod models have coordination
number 7, but Au atoms next to those in the step and also directly
participating in the mechanism have coordination number 8 on
Au(5 1 1), and therefore an average coordination number of 7.5
has been applied to Au(5 1 1) surface. The relevant relationships
thus found are depicted in Fig. 5. On one hand, we observe that
ethanol adsorption energies and activation barriers are linearly
correlated, evidencing that a Brønsted–Evans–Polanyi (BEP) rela-
tionship [38,39] holds for this elementary step (Fig. 5a). Therefore,
the term reactivity can be in this case associated either to a stron-
ger interaction with adsorbates or to a lower activation barrier in-
volved in the reaction. On the other hand, and as deduced from the
discussion above, there is a clear trend in the reactivity of the mod-
el surfaces; it increases with decreasing the Au coordination num-
ber (Fig. 5b). This means that not all active sites are equivalent and
that real gold catalysts might have heterogeneous surfaces with Au
atoms in different coordination state showing different reactivity
towards alcohol dehydrogenation. In order to confirm this hypoth-
esis, a series of Au catalysts with different particle diameter were
prepared and characterized, and their catalytic activity for the oxi-
dation of benzyl alcohol to benzaldehyde was measured. These
experiments will be described in detail in the forthcoming section.
Kadskc½catꢄ½alcoholꢄ0
r0
¼
ð4Þ
1 þ Kads½alcoholꢄ0
which after linearization is converted into:
1
r0
1
1
¼
þ
ð5Þ
K
adskc½catꢄ½alcoholꢄ0 kc½catꢄ
Thus, by plotting the inverse of the initial reaction rate versus
the inverse of the initial concentration of benzyl alcohol, a straight
line should be obtained, and from the slope and intercept at the
origin of the fitted equation, the values of kc and Kads can be de-
duced. The fit of the experimental data to Eq. (5) for the different
catalysts considered in this work at different temperatures is good,
as depicted in Fig. S1. The calculated kinetic and adsorption con-
stants are summarized in Table S1 in the Supporting information.
Activation energies Eact were calculated from the measured rate
constants at different temperatures according to the Arrhenius
equation:
3.2. Kinetic study of the reaction mechanism
The kinetics of the dehydrogenation of benzyl alcohol to benz-
aldehyde on a series of Au/MgO catalysts containing different gold
loading and particle size were studied using the initial reaction rate
method. The characterization data summarized in Table 3 and the
TEM images shown in Fig. 6 indicate that the amount of gold in the
catalyst determines the particle size, so that the higher the gold
loading the larger the metal particles are.
The kinetic rate equation was deduced by considering that ben-
zyl alcohol adsorbs on the gold nanoparticles with a equilibrium
constant Kads defined as:
act=RT
kc ¼ AeꢃE
ð6Þ
ð7Þ
ꢀ ꢁ
Eact
R
1
T
lnðkcÞ ¼ lnðAÞ ꢃ
where A is the pre-exponential factor, R is the molar gas constant
and T is the reaction temperature. Adsorption enthalpies H of ben-
D
zyl alcohol on the different catalysts were calculated using the van’t
Hoff equation:
ꢀ ꢁ
D
H
1
T
DS
½alcoholꢄads
lnðKadsÞ ¼ ꢃ
þ
R
ð8Þ
Kads
¼
ð1Þ
R
½alcoholꢄ½catꢄ
where R is the molar gas constant, T is the reaction temperature,
and S is the entropy change of the reaction. Fig. 7 shows the var-
and that the mechanism for adsorbed alcohol dehydrogenation
involves two consecutive steps according to:
D
iation of the kinetic kc and adsorption Kads constants with tempera-
ture. The fitting of the experimental data to Eqs. (7) and (8) is good
in all cases, and the calculated activation energies and adsorption
enthalpies are summarized in Table 3. Activation energies linearly
increase with increasing particle size, while benzyl alcohol adsorp-
k1
ðalcoholÞads !ðalkoxyÞads
ð2Þ
ð3Þ
k2
ðalkoxyÞads !ðaldehydeÞads
Fig. 7. Variation of the kinetic constants kc and adsorption constants Kads for benzyl alcohol oxidation with temperature on different Au/MgO catalysts.

M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
57
gold catalysts might have heterogeneous surfaces, with Au atoms
in different coordination state showing different reactivity towards
alcohol dehydrogenation. That is, alcohol dehydrogenation on gold
catalysts might be a structure-sensitive reaction. This proposal has
been experimentally confirmed by measuring the kinetics of ben-
zyl alcohol oxidation on a series of Au/MgO catalysts having differ-
ent particle diameter and therefore different concentration of low
coordinated Au sites. Comparison of the trends observed in DFT
calculated alcohol adsorption energies and activation barriers with
those found in the values obtained from the kinetic experiments,
together with the analysis of the variation of the TOF with the
number of accessible Au sites in different samples, allows us to
conclude that the oxidative dehydrogenation of alcohols to car-
bonyl compounds is structure sensitive and that the activity of
the accessible Au sites is heterogeneous and varies with the coor-
dination number.
hemispherical particles
500
450
400
350
300
250
200
150
100
50
0
0
0.5
1
1.5
2
2.5
-19
º
n surface atoms (x10
)
Fig. 8. Turnover frequency (TOF) in hꢃ1 for the oxidation of benzyl alcohol at 453 K
versus number of surface atoms on the series of Au/MgO catalysts described in
Table 3.
Acknowledgments
We thank Consolider-Ingenio 2010 (project MULTICAT),
Spanish MICINN (Projects FIS2008-02238 and MAT2006-14274-
C02-01), Generalitat Valenciana (Project PROMETEO/2008/130),
Generalitat de Catalunya (grants 2009SGR1041 and XRQTC) and
COST Action D41 ‘‘Inorganic oxides: surfaces and interfaces’’ for
financial support and Red Española de Supercomputación (RES) for
computational resources and technical assistance. J.R. expresses
his gratitude to the Mexican CONACyT for a postdoctoral fellow-
ship. T. R. thanks Consejo Superior de Investigaciones Científicas
for I3-P fellowships. FI acknowledges support received through
the ‘‘2009 ICREA Academia’’ prize for excellence in research.
tion enthalpies follow the opposite trend. This is in complete agree-
ment with the results obtained from the theoretical study described
in Section 3.1 and suggests that the direct relationship between par-
ticle size and catalytic activity cannot only be attributed to a larger
concentration of undercoordinated Au sites in the smallest parti-
cles, but also to the fact that not all undercoordinated Au sites are
equally active. To further confirm this hypothesis, turnover frequen-
cies (TOF) were measured at different temperatures for the series of
catalysts considered, and the results obtained at 453 K are plotted
versus the number of surface metal atoms in Fig. 8. The TOF is de-
fined as the number of molecules reacting per surface active site per
unit time and is calculated by dividing the measured initial reaction
rate r0 by the number of surface metal atoms in each particle. It
indicates the number of sites, among those accessible, that are ac-
tive for the reaction under study. The number of surface Au atoms
was estimated assuming a hemispherical shape for all particles
and a surface density of 13.9 atoms nmꢃ2. The plot of TOF against
the number of accessible sites provides information about the reac-
tivity of the accessible sites. If all active sites were equivalent, a hor-
izontal straight line should be obtained. However, Fig. 8 shows that
the TOF for benzyl alcohol oxidation is high for particles with a
small number of accessible sites, and decreases considerably as
the number of accessible sites increases. This behavior of the TOF
indicates that the reaction is structure sensitive, that is, that the
activity of all the accessible sites is not equivalent, as suggested
by the theoretical study presented in the previous section. There-
fore, by comparing the experimental kinetic values and the results
from calculations, it is found that the design of an active and selec-
tive Au catalyst should involve the synthesis of Au particles with
size and shape that maximize the number of atoms with the lowest
coordination number.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.jcat.2010.11.013.
References
[1] R.A. Sheldon, Catalytic Oxidation, in: R.A. Sheldon, R.A. van Santen (Eds.),
World Scientific Publishing, Singapore, 1995.
[2] R.A. Sheldon, J.K. Kochi, Metal-catalyzed Oxidations of Organic Compounds,
Academic Press, New York, 1981.
[3] T. Mallat, A. Baiker, Chem. Rev. 104 (2004) 3027.
[4] K. Yamaguchi, N. Mizuno, Angew. Chem. Int. Ed. 41 (2002) 4538.
[5] K. Yamaguchi, N. Mizuno, Chem. Eur. J. 9 (2003) 4353.
[6] Z.X. Zhang, P. Chen, J. Liu, Y.H. Zhang, W. Shen, H.L. Xu, Y. Tang, Chem.
Commun. (2008) 3290.
[7] K. Shimizu, K. Sugino, K. Sawabe, A. Satsuma, Chem. Eur. J. 15 (2009) 2341.
[8] F. Zaccheria, N. Ravasio, R. Psaro, A. Fusi, Chem. Eur. J. 12 (2006) 6426.
[9] L. Huang, Y. Zhu, C. Huo, H. Zheng, G. Feng, C. Zhang, Y. Li, J. Mol. Catal. A:
Chem. 288 (2008) 109.
[10] S. Carrettin, P. McMorn, P. Johnston, K. Griffin, G.J. Hutchings, Chem. Commun.
(2002) 696.
[11] A. Abad, P. Concepción, A. Corma, H. García, Angew. Chem. Int. Ed. 44 (2005)
4066.
[12] A. Abad, A. Corma, H. García, Chem. Eur. J. 14 (2008) 212.
[13] C. Della Pina, E. Falletta, M. Rossi, J. Catal. 260 (2008) 384.
[14] Y. Guan, E.J.M. Hensen, Appl. Catal. A: Gen. 361 (2009) 49.
[15] K. Mori, T. Hara, T. Mizugaki, K. Ebitani, K. Kaneda, J. Am. Chem. Soc. 126
(2004) 10657.
[16] J.D. Grunwaldt, M. Caravati, A. Baiker, J. Phys. Chem. B 110 (2006) 25586.
[17] D. Ferri, C. Mondelli, F. Krumeich, A. Baiker, J. Phys. Chem. B 110 (2006) 22982.
[18] M. Comotti, C. Della Pina, R. Matarrese, M. Rossi, Angew. Chem. Int. Ed. 43
(2004) 5812.
[19] A. Villa, D. Wang, N. Dimitratos, D.S. Su, V. Trevisan, L. Prati, Catal. Today 150
(2010) 8.
[20] S. Kanaoka, N. Yagi, Y. Fukuyama, S. Aoshima, H. Tsunoyama, T. Tsukuda, H.
Sakurai, J. Am. Chem. Soc. 129 (2007) 12060.
[21] T. Ishida, M. Nagaoka, T. Akita, M. Haruta, Chem. Eur. J. 14 (2008) 8456.
[22] H.L. Liu, Y.L. Liu, Y.W. Li, Z.Y. Tang, H.F. Jiang, J. Phys. Chem. C 114 (2010)
13362.
[23] A. Corma, T. Ródenas, M.J. Sabater, Chem. Eur. J. 16 (2010) 254.
[24] A. Quintanilla, V.C.L. Butselaar-Orthlieb, C. Kwakernaak, W.G. Sloof, M.T.
Kreutzer, F. Kapteijn, J. Catal. 271 (2010) 104.
4. Conclusions
The complete molecular mechanism for ethanol oxidation to
ethanal has been elucidated from periodic DFT-based calculations
carried out on different catalyst models including perfect and
stepped single crystal Au surfaces as well as an isolated Au38 nano-
particle, which is a prototype for this kind of systems. A clear trend
exists in the reactivity of the model surfaces, measured either by
the adsorption energy of the adsorbates or by the activation barri-
ers involved in the reaction. These two quantities are indeed line-
arly correlated in a BEP relationship. Hence, the activation energy
decreases (and the adsorption energy increases) with decreasing
the Au coordination number. From the model calculations, it can
be concluded that not all active sites are equivalent, and that real

58
M. Boronat et al. / Journal of Catalysis 278 (2011) 50–58
[25] J. Radilla, M. Boronat, A. Corma, F. Illas, Phys. Chem. Chem. Phys. 12 (2010)
6492.
[26] 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.
[27] J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson, M.R. Pederson, D.J. Singh, C.
Fiolhais, Phys. Rev. B 48 (1993) 4978.
[28] J.P. Perdew, Y. Wang, Phys. Rev. B 45 (1992) 13244.
[29] P.E. Blöchl, Phys. Rev. B 50 (1994) 17953.
[31] G. Kresse, J. Furthmuller, Comput. Mater. Sci. 6 (1996) 15.
[32] G. Kresse, J. Furthmueller, Phys. Rev. B 54 (1996) 11169.
[33] H.J. Monkhorst, J.D. Pack, Phys. Rev. B 13 (1976) 5188.
[34] A. Heyden, A.T. Bell, F.J. Keil, J. Chem. Phys. 123 (2005) 224101.
[35] G. Henkelman, H. Jónsson, J. Chem. Phys. 111 (1999) 7010.
[36] E. Sanville, S.D. Kenny, R. Smith, G. Henkelman, J. Comp. Chem. 28 (2007) 899.
[37] G. Henkelman, A. Arnaldsson, H. Jónsson, Comput. Mater. Sci. 36 (2006) 254.
[38] J.N. Brønsted, Chem. Rev. 5 (1928) 231.
[30] G. Kresse, D. Joubert, Phys. Rev. B 59 (1999) 1758.
[39] M.G. Evans, M. Polanyi, Trans. Faraday Soc. 34 (1938) 11.