Control and Manipulation of Gold Nanocatalysis: Effects of Metal Oxide Support Thickness and Composition

Article abstract of DOI:10.1021/ja804893b

Control and tunability of the catalytic oxidation of CO by gold clusters deposited on MgO surfaces grown on molybdenum, Mo(100), to various thicknesses are explored through temperature-programmed reaction measurements on mass-selected 20-atom gold cluster

Full text of DOI:10.1021/ja804893b

Published on Web 12/19/2008
Control and Manipulation of Gold Nanocatalysis: Effects of
Metal Oxide Support Thickness and Composition
†
†
†
Chris Harding, Vahideh Habibpour, Sebastian Kunz,
†
,†
‡
,‡
Adrian Nam-Su Farnbacher, Ueli Heiz,* Bokwon Yoon, and Uzi Landman*
Lehrstuhl f u¨ r Physikalische Chemie, Technische UniVersit a¨ t M u¨ nchen, Lichtenbergstrasse 4,
8
5748 Garching, Germany, and School of Physics, Georgia Institute of Technology,
Atlanta, Georgia 30332-0430
Received June 26, 2008; E-mail: ulrich.heiz@mytum.de; uzi.landman@physics.gatech.edu
Abstract: Control and tunability of the catalytic oxidation of CO by gold clusters deposited on MgO surfaces
grown on molybdenum, Mo(100), to various thicknesses are explored through temperature-programmed
reaction measurements on mass-selected 20-atom gold clusters and via first-principles density functional
theory calculations. Au20 was chosen because in the gas phase it is characterized as an extraordinarily
stable tetrahedral-pyramidal structure. Dependencies of the catalytic activities and microscopic reaction
mechanisms on the thickness and stoichiometry of the MgO films and on the dimensionalities and structures
of the adsorbed gold clusters are demonstrated and elucidated. Langmuir-Hinshelwood mechanisms and
reaction barriers corresponding to observed low- and high-temperature CO oxidation reactions are calculated
and analyzed. These reactions involve adsorbed O
like state through partial occupation of the antibonding orbitals. In some cases, we find activated, dissociative
adsorption of O molecules, adsorbing at the cluster peripheral interface with the MgO surface. The reactant
2
molecules that are activated to a superoxo- or peroxo-
2
CO molecules either adsorb on the MgO surface in the cluster proximity or bind directly to the gold cluster.
Along with the oxidation reactions on stoichiometric ultrathin MgO films, we also study reactions catalyzed
by Au20 nanoclusters adsorbed on relatively thick defect-poor MgO films supported on Mo and on defect-
rich thick MgO surfaces containing oxygen vacancy defects.
Introduction
between, the catalytic properties and the size or dimensionality
of the catalytic material. For example, it has been shown that
6
One of the principal goals of modern research in chemical
catalysis is the development of methods for further control and
manipulation of the activity, selectivity, and specificity of
catalytic systems. This can be achieved in various ways,
including (i) selection of the catalyst material and composition,
for a number of reactions the catalytic activity of single-crystal
surfaces can be modified by using metal nanoparticles with an
average size down to a few nanometers. In this size range, it
was found that nanocrystallites of different size may exhibit
different structural motifs characterized by different proportions
of exposed crystallographic facets (with corresponding different
chemical reactivities), thus resulting in size-dependent variations
1
,2
(
ii) manipulations of the atomic structure, morphology, and
shape of the catalyst, (iii) choice of the support (composition,
structure, and thickness ), and (iv) the use of externally applied
electric or electromagnetic fields. Particle size plays an
3,4
6
-8
5
of the catalytic activity of the nanocrystallites.
Moreover, new catalytic properties emerge that could not have
been anticipated through extrapolation of the behavior known
for larger sizes when the reduction of the size of metal catalyst
particles reaches the ultimate nanocluster regime, that is, clusters
comprised of about 10 to 20 atoms (e1 nm in “diameter”). In
this size range, the nanocluster may be viewed as a quasi-“zero-
dimensional” (0D) quantum dot whose physical and chemical
important role in determining the chemical reactivity of material
aggregates. A main endeavor in current catalytic research aims
at characterizing, understanding, controlling, and utilizing the
effects of the particle size on catalytic properties. With this in
mind, each of the aforementioned ways of controlling and
modifying catalytic activity may be explored also in conjunction
with variations of the size of the catalytic particles under
investigation. Indeed, numerous studies have demonstrated, for
a broad range of size scales, dependencies of, and correlations
1,2,9-14
properties are dominantly affected by quantum size effects
originating from electrons confined in a cluster of finite size
in all three dimensions), including coupling to the supporting
(
†
surface. As aforementioned, in this regime, which is the one
where gold clusters exhibit unique reactivities (see refs 10, 13,
and 14 and the discussion below), extrapolations from larger
Technische Universit a¨ t M u¨ nchen.
Georgia Institute of Technology.
‡
(
(
(
1) Heiz, U.; Landman, U. Nanocatalysis; Nanoscience and Technology;
Springer Verlag: Berlin, 2007.
2) Landman, U.; Yoon, B.; Zhamg, C.; Heiz, U.; Arenz, M. Top. Catal.
007, 44, 145.
3) Ricci, D.; Bongiorno, A.; Pacchioni, G.; Landman, U. Phys. ReV. Lett.
006, 97, 036106.
4) Zhang, C.; Yoon, B.; Landman, U. J. Am. Chem. Soc. 2007, 129, 2228.
5) Yoon, B.; Landman, U. Phys. ReV. Lett. 2008, 100, 056102.
2
(6) Henry, C. R. Surf. Sci. Rep. 1998, 31, 231.
(7) Henry, C. R. In Nanocatalysis; Heiz, U., Landman, U., Eds.;
Nanoscience and Technology; Springer Verlag: Berlin, 2007.
(8) Libuda, J.; Meusel, I.; Hoffmann, J.; Hartmann, J.; Piccolo, L.; Henry,
C. R.; Freund, H. J. J. Chem. Phys. 2001, 114, 4669.
2
(
(
5
38 9 J. AM. CHEM. SOC. 2009, 131, 538–548
10.1021/ja804893b CCC: $40.75  2009 American Chemical Society

Control and Manipulation of Gold Nanocatalysis
A R T I C L E S
sizes using arguments relying on surface-to-volume ratios and
scaling relations based on the enumeration of special sites (e.g.,
near the Fermi level; see the discussion in refs 13 and 14 and
illustrative comparisons, given in ref 18 between the electronic
1
5-17
corner atoms) as a function of the size
are not operative;
density of states of surface-supported Au
as well as of a doped Au Sr cluster, with and without an
adsorbed O molecule), and (ii) the cluster proximity effect,
3 4 8
, Au , and Au clusters,
indeed, on this size scale, almost all of the atoms of the metal
cluster are essentially surface atoms, and almost all can be
classified as undercoordinated (compared to bulk). Instead, in
this important regime, which is the focus of our research in
this paper, one must resort to quantum mechanical calculations
that highlight the correlations between the electronic structure
of the coupled cluster-substrate catalytic system and the
geometrical arrangement of the atoms (including their distortions
in the course of reactions, which we termed as “dynamical
3
2
where the distances between the adsorbed reactant molecules
are restricted by the nanometer (or subnanometer) size of the
cluster, thus lowering (concomitantly with the optimal position-
ing of energy levels noted above and charge-induced activation)
the enthalpic and entropic reaction activation barriers.
It is pertinent to remark here that most recent aberration-
21
corrected transmission electron microscopy investigations used
for identification of the active catalytic gold species among the
many present on real catalysts have found that the high catalytic
activity for CO oxidation is unambiguously correlated with the
presence of bilayer clusters (supported on a metal oxide) that
are less than 1 nm in diameter and contain on the order of only
10 Au atoms. Interestingly, these studies, which are fully
consistent with the early theoretical and experimental investiga-
1
8
fluctionality” ) on an atom-by-atom level, that is, when every
atom counts and where “small is different” in an essential, i.e.,
nonscalable, manner. This is indeed the approach that we have
1
,2,4,5,13,14,18
taken in earlier studies
this study.
and the one that we use in
While the utilization of reactive metals, e.g., transition metals,
as well as some coinage metals (like silver, palladium, and
platinum) in catalytic processes is rather common, the catalytic
properties of gold are less well-known, and they remain largely
unexploited; indeed, until about 2 decades ago, gold was
considered to be strictly inert. While this is the case for bulk
gold, when prepared as aggregates (clusters) of nanometer
dimensions, gold exhibits interesting, potent, and promising
catalytic activity with unique specificity and selectivity
1
3,14,18
tions described above,
found that the catalytically active
subnanometer gold bilayer clusters represented only 1.05 (
0.72% of the total Au loading, with the remaining 98.82 ( 0.8
atomic % of Au being larger particles. The authors further
emphasized their findings by stating that “The observation that
x
the active species in our Au/FeO catalysts consist of subna-
nometer clusters differs from numerous earlier investigations
that identified 2- to 5-nm particles as the critical nanostructures”.
The authors also remarked about experimental difficulties in
detecting these subnanometer clusters, stating that “...it is
probable that these minority Au species would not be easily
detected with traditional ‘bulk’ techniques such as extended
x-ray absorption fine structure or Mossbauer spectroscopy, or
even by surface analysis techniques such as x-ray photoelectron
spectroscopy (XPS), because their contribution to the total signal
would be minimal compared with that of the larger nanopar-
ticles”.
1
,2,13,18-20
characteristics.
In particular, joint experiments and
theoretical investigations of the catalytic oxidation of CO on
size-selected gold clusters supported on relatively thick, defect-
rich (F center, FC), MgO surfaces have shown low-temperature
catalytic combustion to occur at temperatures as low as 140 K
for Au
two-layer) Au
exhibit catalytic activity.
n
clusters with 8 e n e 20 Au atoms; that is, a nonplanar
(
8
nanocluster emerged as the smallest one to
1
3
1
3
We reemphasize here that the emergence of such subna-
nometer-scale bilayer gold clusters (starting with the gold
octamer), supported on thick metal oxide (MgO) surfaces
containing oxygen vacancies (FCs), as the ones exhibiting
catalytic activity, has its origin in two main factors: (i) the
quantum size effect, where the confinement of the electrons
13,14,18
In the earlier theoretical investigations,
charging of the
adsorbed metal cluster through partial electron transfer from
the oxygen vacancy FC defects (or other Lewis basic defect
sites) was found to play a key role in anchoring of the gold
cluster to the metal oxide (MgO) surface (thus enhancing the
stability of the adsorbed gold nanoclusters against coalescence
and sintering) and in the activation and promotion of the reactant
molecules. The change of the charge state of the cluster (through
charge transfer from the underlying substrate, causing an upward
shift of the cluster energy levels and enhanced population near
the Fermi level) and consequent occupation of the antibonding
(
defined by the adsorbed cluster structure and dimensionality
and including the charging effect from the surface defects) serves
to determine the positioning of the chemically active energy
levels (in particular, the location of sd orbitals of gold lying
(
9) Abbet, S.; Sanchez, A.; Heiz, U.; Schneider, W. D. J. Catal. 2001,
1
98, 122.
(
10) Heiz, U.; Bernhardt, T. M.; Landman, U. In Nanocatalysis; Heiz, U.;
Landman, U., Eds.; Nanoscience and Technology; Springer Verlag:
Berlin, 2007.
2π* orbital of O
on the cluster, drops below the Fermi level, E
gold nanocluster and mixes with the sd states near E
activation of the O-O bond and have been identified as
2
(which, upon adsorption of the O
, of the supported
) lead to
2
molecule
F
(
(
(
(
(
(
11) Heiz, U.; Sanchez, A.; Abbet, S.; Schneider, W. D. J. Am. Chem.
F
Soc. 1999, 121, 3214.
12) Judai, K.; Abbet, S.; W o¨ rz, A. S.; Heiz, U.; Henry, C. R. J. Am. Chem.
Soc. 2004, 126, 2732.
2,10,13,14,18
underlying the catalytic activity.
The unusual activa-
13) Sanchez, A.; Abbet, S.; Heiz, U.; Schneider, W. D.; Hakkinen, H.;
Barnett, R. N.; Landman, U. J. Phys. Chem. A 1999, 103, 9573.
14) Yoon, B.; Landman, U.; W o¨ rz, A.; Antonietti, J.-M.; Abbet, S.; Judai,
K.; Heiz, U. Science 2005, 307, 403.
tion of oxygen upon adsorption is portrayed in the elongation
of d(O-O) to about 1.35 Å, corresponding to a superoxo-like
state, or to d(O-O) > 1.4 Å, corresponding to a peroxo-like
state. Interestingly, in most oxidation (combustion) reactions
catalyzed by gold nanoclusters, the O molecule is activated
2
through its interaction with the gold nanocluster but does not
dissociate, whereas for reactions catalyzed on extended metal
15) Hvolbaek, B.; Janssens, T. V. W.; Clausen, B. S.; Falsig, H.;
Christensen, C. H.; Norskov, J. K. Nanotoday 2007, 2, 14.
16) Janssens, T. V. W.; Clausen, B. S.; Hvolbaek, B.; Falsig, H.;
Christensen, C. H.; Bligaard, C. H.; Norskov, J. K. Top. Catal. 2007,
4
4, 15.
(
(
17) Falsig, H.; Hvolbaek, B.; Kristensen, I. S.; Jiang, T.; Bligaard, C. H.;
Norskov, J. K. Angew. Chem., Int. Ed. 2008, 47, 4835.
18) H a¨ kkinen, H.; Abbet, S.; Sanchez, A.; Heiz, U.; Landman, U. Angew.
Chem., Int. Ed. 2003, 42, 1297.
surfaces, dissociation of the adsorbed O
common route to oxidation reactions.
2
molecules is the
(
(
19) Haruta, M. Catal. Today 1997, 36, 153.
(21) Herzing, A. A.; Kiely, C. J.; Carley, A. F.; Landon, P.; Hutchings,
20) Valden, M.; Lai, X.; Goodman, D. W. Science 1998, 281, 1647.
G. J. Science 2008, 321, 1331.
J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009 539

A R T I C L E S
Harding et al.
Another key factor that influences the catalytic activity of
gold nanoclusters is their ability to distort in the course of
interaction with the reactants. This property has been termed
from UPS experiments, where the kinetic energy and intensity (flux)
of electrons ejected from the target material by incident radiation
are measured; in the UPS signal, however, contributions of bulk
electronic states overlap with those coming from the surface region.
In addition to information obtained from UPS spectra about the
electronic density of states, the widths of the UPS spectra yield
good estimates of the work functions of the materials being studied.
“
dynamical fluctionality”, and it is unique to clusters in the
1
8
nanometer size range. Such structural fluctional distortions
enable adsorption and activation of the reactants, as well as
lower the activation barriers of reactions between the adsorbed
reactants.
For all of the defect-poor films investigated in this work, no
contributions in the MIES spectra at binding energies between 1
and 3 eV were observed. This is indicative of the absence of FCs
3,4
Recently, it has been predicted theoretically and confirmed
2
2
experimentally that a dimensionality crossover will occur for
gold clusters deposited on thin metal oxide films grown on
appropriate metal surfaces [e.g., MgO grown on Mo(100) or
Ag(100)] as a function of the metal oxide film thickness. Thinner
MgO films with less than seven to eight layers were predicted
23
on the thin MgO films. The number of FC defects can be varied
through variation of the Mg evaporation rate, with shorter evapora-
tion times (i.e., faster evaporation) being associated with films with
more defects. Such defect-rich substrates were created by us only
for films with larger thicknesses (∼10 ML); as aforementioned,
the presence of these defects was identified by a MIES signal peaked
(
and subsequently found experimentally) to favor two-dimen-
2
3
sional (2D) wetting gold island structures, whereas on thicker
films the dimensionalities of the free clusters are maintained.
Stabilization of the 2D structures has been shown to be caused
by the accumulation of excess electronic charge, originating
from the underlying metal, at the cluster interface with the MgO
at ∼2 eV (see the Supporting Information).
Clusters of Au20 were created through the use of a laser-ablation
2
4
source using the second harmonic of a 120 Hz Nd:YAG laser. A
home-built piezo valve produced a cluster beam from the laser-
induced metal plasma using a He carrier gas. The clusters are
charge-separated with a quadrupole deflector and subsequently
mass-selected using a quadrupole mass spectrometer (ABB-Extrel;
mass limit 4000 amu). The clusters were then soft-landed (with an
impingement translational energy of 1 eV/cluster) onto the MgO
4
film. This has been predicted to result in the catalytic activity
of gold nanoclusters adsorbed on thin MgO films, even when
the films have no Lewis basic defect sites (e.g., FCs), provided
that the films are thin enough.
15
support up to a coverage of 0.3% ML (1 ML ) 2.25 × 10 clusters/
The experimental and theoretical investigations presented
herein aim at demonstrating the effect of the metal oxide film
thickness on the catalytic activity and reaction mechanisms of
adsorbed gold nanoclusters (in particular, Au20). We indeed
show that sufficiently thin MgO films can serve as supports in
gold nanocatalysis, even when they do not contain Lewis basic
defect sites, and identify low- and high-temperature reaction
channels. We also contrast these results with catalytic CO
oxidation on thick, defect-rich MgO films. Conceptually, we
show the active site on the cluster to be characterized by
enhanced electron density, which activates (or, less often,
2
cm ) for temperature-programmed reaction (TPR) studies and 0.5%
ML for Fourier transform infrared (FTIR) measurements.
Following preparation of the Au20 nanocatalysts, TPR and FTIR
13
studies were performed. For the TPR measurements, CO and O
2
were dosed onto the surface (at ∼100 K) to an exact coverage using
a calibrated molecular beam doser. Exposures of one Langmuir (1
13
2
L) O and 1 L CO were administered sequentially. Isotope-labeled
CO was used to improve the signal-to-noise ratio in the measure-
ments. A temperature-programmed ramp of the catalyst was
performed between ∼100 and 800 K using a feedback-controlled
resistive annealing system of home-built design. Products from the
reaction were measured using a quadrupole mass spectrometer. TPR
experiments are particularly well suited for the discussion of
possible reaction mechanisms because reaction temperatures can
be correlated with activation (transition-state, TS) energies obtained
from first-principles simulations of the reaction mechanism. These
combined experimental and theoretical studies give most valuable
information about the reaction mechanism, atomic and molecular
arrangements, vibrational frequencies, and electronic structure. Such
one-cycle experiments do not give information about catalytic
turnover frequencies (TOFs). Determination of the TOF for the
catalytic reaction under investigation requires added experiments;
indeed, in certain cases, we have applied pulsed molecular beam
2
dissociates) the adsorbed O molecule and promotes the bonding
of CO. The location and extent of the enhanced charge density
depend on the characteristics of the underlying metal oxide.
Methods
a. Experimental Methods. The model catalytic systems used
in these investigations consist of Au20 clusters adsorbed on metal
oxide [MgO(100)] films of various thicknesses and stoichiometries
(
that is, with and without oxygen vacancies) supported on Mo(100).
The Mo(100) single crystal was cleaned and a film of MgO was
grown on it by the evaporation of Mg metal in a background of O
2
2
5,26
-
7
scattering measurements for this purpose;
however, in this
at a pressure of 5 × 10 mbar for a specified time interval. The
growth time was calibrated by a break-point analysis and indepen-
dently by estimation of the Mg flux required to achieve a desired
thickness. Rather accurate control of the film thickness can be
achieved through variation of the growth time. The cleanliness and
composition of the films and underlying crystal were assessed
through the use of Auger electron spectroscopy (AES), and further
characterization of the electronic structure of the films was achieved
through metastable impact electron spectroscopy (MIES) and
ultraviolet photoelectron spectroscopy (UPS) (see the Supporting
Information).
paper, we do not employ such experiments, focusing on the reaction
mechanisms and related atomic and electronic structure factors.
Finally, we note that a similar dosing procedure was also carried
out for FTIR studies. Following the coadsorption of O
FTIR measurements were made at a grazing angle of incidence
and 13CO,
2
1
3
and the stretching frequencies of CO were determined.
b. Theoretical Methods. The first-principles calculations are
2
7,28
based on a density functional theory (DFT) approach.
We
(
(
23) Kalmakov, A.; Stultz, J.; Goodman, D. W. J. Chem. Phys. 2000, 113,
MIES is a most surface-sensitive technique where the deexci-
tation of the metastable He* atoms impinging on the surface takes
place 3-10 Å above the surface. For oxide materials, the MIES
spectra give direct information about the electronic density of states
at the surface. The obtained information is similar to that found
7
564.
24) Heiz, U.; Vanolli, F.; Trento, L.; Schneider, W. D. ReV. Sci. Instrum.
1997, 68, 1986.
(25) R o¨ ttgen, M. A.; Abbet, S.; Judai, K.; Antonietti, J. M.; W o¨ rz, A. S.;
Arenz, M.; Henry, C. R.; Heiz, U. J. Am. Chem. Soc. 2007, 129, 9635.
26) Harding, C. J.; Kunz, S.; Habibpour, V.; Teslenko, V.; Arenz, M.;
(
Heiz, U. J. Catal. 2008, 255, 234.
(27) Kresse, G.; Furtmuller, J. Phys. ReV. B 1996, 54, 11169.
(28) Kresse, G.; Hafner, J. Phys. ReV. B 1993, 47, R558.
(
22) Sterrer, M.; Risse, T.; Heyde, M.; Rust, H. P.; Freund, H. J. Phys.
ReV. Lett. 2007, 98, 032235.
5
40 J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009

Control and Manipulation of Gold Nanocatalysis
employed the generalized gradient approximation and ultrasoft
A R T I C L E S
2
9
3
0
pseudopotentials (scalar relativistic ones for gold) with a plane-
wave basis (kinetic energy cutoff of 300 eV). We remark here that
such calculations have been shown to give very accurate bond
lengths (up to 1% too long) and reaction barriers that are accurate
to within 25-30% (usually too low) (see p 87 of ref 10). In
structural relaxations corresponding to minimization of the total
energy, convergence is achieved when the forces on the atoms are
less than 0.001 eV/Å. In the following, we describe certain pertinent
details of the calculations.
(i) In the calculations for a one-layer-thick MgO film on Mo(100),
a (6 × 6) MgO layer containing 36 atoms of Mg and an equal
number of O atoms was used, and the underling Mo(100) substrate
was modeled by four layers (with a lattice constant of 3.151 Å),
each containing 36 Mo atoms. The two bottom layers of the
Mo(100) substrate are held static (in the bulk lattice geometry),
and in structural optimizations, all of the other atoms of the system
(
surface atoms, as well as adsorbed cluster and reactant atoms) are
allowed to undergo unconstrained relaxation; note that the MgO
lattice was stretched in the surface plane by 5% to accommodate
the mismatch between the Mo and MgO lattices.
Figure 1. TPR spectra illustrating 13CO
formation produced using a
(
ii) In calculations involving an eight-layer-thick MgO film on
2
nanocatalyst based on Au20 clusters at a coverage of 0.3% ML deposited
onto MgO films of various thickness (1-10 ML). Markers represent the
experimental data points; the solid line is a multipeak exponential Gaussian
fit drawn to guide the eye.
Mo(100), a 5 × 5 periodic cell containing 25 atoms per layer was
used. In calculations involving a thick MgO crystal (without a metal
support), we used six (6 × 6) MgO layers with one or two oxygen
vacancies (FCs) in the top layer; the bottom MgO layer of the slab
was kept static with the lattice constant taken as the bulk equilibrium
value, 2.998 Å (all other layers, as well as the adsorbed gold
nanocluster and the reactants, were allowed to relax dynamically).
Finally, we note that the large size of the computational supercell
used by us [up to (6 × 6) larger than the planar unit cell of the
MgO(100) and Mo(100) substrate, i.e., equivalent to a 6 × 6 k-point
sampling when a single planar unit cell is used for the substrate]
justifies the k ) 0 (Γ-point sampling) calculations that we employed
here. Comparisons between the k ) 0 calculations and test
computations that included explicit k-point sampling (2 × 2 × 1
and 4 × 4 × 1, i.e., increasing the effective number of k points up
to 24 × 24 × 1) provided clear evidence that the k ) 0 results are
indeed well converged, with differences in the total energies well
below 1% and differences (in comparison with the k ) 0 results)
of at most 1% between the binding energies. We conclude that
added k-point sampling has no discernible effect on the results of
our computations.
verified by TPR measurements on Au20 deposited on partially
oxidized Mo surfaces, where no 13CO
signal (the product of
2
the oxidation reaction) has been recorded. On the other hand,
when MgO films grown on the Mo(100) surface are used as
supports for the gold clusters, the oxidation reaction is found
to occur for various MgO coverages, corresponding to different
film thicknesses (Figure 1b-d). In these experiments, the MgO
film is grown so that the coverage is 3 ML or less. However,
as indicated above, it is not assumed that at a film thickness of
1
-3 ML the film is continuous; in fact, MgO films on Mo are
known to grow via island formation. It is therefore likely that,
although a film is calculated to have 1-3 ML coverage, the
film actually consists of islands that are one to three layers thick,
interspaced by partially oxidized Mo. Because, as aforemen-
tioned, gold clusters deposited on partially oxidized Mo are not
reactive, we conclude that only Au20 clusters on MgO islands
contribute to the observed catalytic activity.
Results and Discussion
On relatively thin MgO films (<3 ML), the observed TPR
a. Experimental Findings. The isotopically labeled product
signals of the TPR experiments are shown in Figure 1, where
variation of the reactivity with the MgO film thickness is
illustrated for films containing no FC defects (labeled as defect-
poor films). In a first experiment, Au20 clusters were deposited
on a Mo support, which was exposed to oxygen at the same
conditions as those in a typical MgO film preparation step (5
shows two regimes of CO
2
formation: a minor peak at ∼180 K
with the major product formation occurring at ∼300 K. The
reactivity associated with the nanocatalyst varies as the film
thickness is increased. From TPR studies made on thick MgO
films (∼10 ML), which are found from break-point analysis to
be continuous, we observe that the reactivity of the gold catalyst
on such films is notably different compared to that of the thinner
MgO film case. In particular, the TPR records for the thicker
films exhibit a single peak at ∼250 K, which is markedly lower
than the higher temperature peak for the aforementioned thin-
film-based nanocatalysts.
To explore the effects of film composition on the reactivity,
measurements were performed on defect-rich and defect-poor
thick films (Figure 2). Unlike the low-temperature reactivity of
the defect-poor thick films (Figure 2a), which exhibits a single
reaction channel, TPR studies on defect-rich films (Figure 2c)
show two reaction regimes. The lower temperature reactivity
is observed at ∼200 K, and the higher temperature reactive
regime is located in the vicinity of 400 K.
-7
×
2
10 mbar of O , T ) 300 K, 20 min). The electronic structure
of this surface was characterized by MIES and UPS and
indicates the presence of a partially oxidized Mo surface (see
the Supporting Information). This experiment is rather important
because thin MgO films of thicknesses of up to about 3 ML
are not necessarily continuous, and consequently Au20 clusters
deposited between MgO islands could be assumed (erroneously
2
as we discuss below) to contribute to the formation of CO in
addition to the reaction catalyzed by the gold clusters anchored
on the MgO film. That the interisland (oxidized Mo) surface
regions do not contribute to the CO oxidation reaction has been
(
29) Perdew, J. P.; Chevary, J. A.; Vosko, S. H.; Jackson, K. A.; Pederson,
M. R.; Singh, D. J.; Fiolhais, C. Phys. ReV. B 1992, 46, 6671.
30) Vanderbilt, D. Phys. ReV. B 1990, 41, 7892.
Further investigations of the effects of defects on the Au20
catalytic activity employed FTIR spectroscopy. Low-temperature
(
J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009 541

A R T I C L E S
Harding et al.
in the underlying material substrate is the source of the
stabilization of the planar geometry of the deposited gold cluster.
However, because in the gas phase the deposited Au20 clusters
have a tetrahedral three-dimensional (3D) structure, it is likely
that not all of the clusters on the thin MgO(100)/Mo film have
attained the optimal 2D structure after deposition. Consequently,
for the thin MgO films, we consider reactions catalyzed either
by the 2D Au20(P) adsorbed cluster (Figure 3a) or by the 3D
Au20(T) isomer (Figure 3b), which are likely to coexist on the
surface. Similarly, for the case of a thick MgO(100) crystal
containing oxygen vacancies (for TPR experiments, see Figure
2
c), we consider two possible isomers of the Au20 cluster. One
of these isomers is the tetrahedral Au20(T) adsorbed on top of
an oxygen vacancy located near the middle of the base facet of
the tetrahedron in Figure 3c. The other isomers that we consider
are (i) a bilayer cluster, Au20(bilayer; FC), adsorbed on an
oxygen vacancy located near the middle of the bottom facet
(
see Figure 3d) and (ii) a bilayer cluster, Au20(bilayer; 2FC),
adsorbed on top of two neighboring FCs located near the middle
of the bottom facet (see Figure 3e). In all of these configurations,
the oxygen vacancies located under the adsorbed gold nano-
cluster are stable against annealing (for example, by one of the
O atoms of the reactant molecule).
Figure 2. (a) TPR measurements made on Au20-based nanocatalysts using
thick, defect-poor MgO films (∼10 ML). Markers show the experimental
data points, and the line is a multipeak exponential Gaussian fit to the data.
(
c) TPR measurements made on Au20-based nanocatalysts using thick,
The optimal structures of the 2D and 3D isomers together
with isosurfaces of the excess electron charge distribution, ∆q,
which is seen to be accumulated mainly in the interface of the
cluster with the MgO(1 L)/Mo(100) surface, are shown in parts
a and b of Figure 3, respectively. The binding energy of Au20(P)
defect-rich MgO films. Markers show the experimental data points, and
the line is a multipeak exponential Gaussian fit to the data. Parts b and d
are FTIR studies corresponding to parts a and c, respectively. All
measurements are made with a 0.5% ML coverage of Au20
.
stretching frequencies of 13CO coadsorbed with 18
in parts b and d of Figure 2. At low temperature (100 K) and
prior to CO combustion, Au20 on both defect-poor and -rich
O
2
are shown
B
(Figure 3a) to the surface is E
while Au20(T) (Figure 3b) is anchored less strongly, with E )
) 12.50 eV with ∆q ) 1.62 e,
B
5.73 eV and ∆q ) 1.06 e; the relaxed structure of Au20(T)
adsorbed on an eight-layer-thick MgO film supported on
Mo(100) was found to be similar to that shown in Figure 3b,
13
MgO films readily adsorbs CO. Three bands at 2048, 2080,
and 2130 cm- are observed for the model catalysts with defect-
1
13
poor support materials. Note that CO on MgO reveals a typical
band at 2127 cm- but with less intensity than is observed in
the spectrum shown in Figure 2b. On defect-rich support
materials, Au20 adsorbs CO with vibrational frequencies at
2
B
with E ) 3.00 eV and ∆q ) 0.73 e.
1
In Figure 3c-e, we show ground and higher energy isomeric
structures of Au20 clusters adsorbed on the (100) surface of a
thick MgO crystal (with no metal support) with oxygen
vacancies, as described above. The binding energies of the
clusters Au20(T) in part c and Au20(bilayer) in parts d and e are
4.36 eV (for part c), 4.92 eV (for part d), and 7.93 eV (for the
2FC case displayed in part e). In the gas phase, the tetrahedral
cluster is more stable than the bilayer one by 1.57 eV.
1
3
-1
063, 2095 (observed as a small shoulder), and 2144 cm . After
initiation of the reaction at temperatures larger than 160 K, CO
desorbs almost totally or reacts to completion on Au20 supported
on defect-poor films, whereas for Au20 supported on defect-
rich films, CO remains adsorbed at these temperatures. This
observation is in concert with the low-temperature catalysis of
the CO oxidation reaction, observed for Au20 clusters supported
on defect-poor films. Interestingly, the shift of the main band
observed for CO adsorbed on Au20 on defect-poor and -rich
Superimposed on the atomic structures in Figure 3c-e, we
display the excess electron densities, which show charge
accumulation in the interfacial region between the cluster and
the surface; this excess charge originates mainly from the FC
defects in the MgO surface. In addition, we show for the Au20(T)
cluster adsorbed on the single FC the electronic charge distribu-
tion corresponding to the highest occupied Kohn-Sham (HOKS)
orbital, which is seen to exhibit maxima at the top (apex) atom
of the tetrahedron, as well as at the interfacial region in the
vicinity of the FC (see the right frame in Figure 3c); adsorption
films is only 15 cm- in comparison to 53 cm in the case of
Au adsorbed on identical films.
8
1
-1
1
4
b. Cluster Structures, Electron Charge Distributions, and
Binding Energies. To elucidate the microscopic mechanisms that
underlie the above observations pertaining to the catalytic
activity of surface-supported Au20 clusters, we have performed
first-principles DFT electronic structure calculations, coupled
with atomic structural optimizations and simulations of the
reaction pathways, for Au20 adsorbed on MgO surfaces. While
for the thicker MgO films the tetrahedral [i.e., Au20(T)] structure
of the cluster is by far the most stable one, on thin MgO films
supported on Mo(100), a “wetting” planar (quasi-2D) island
configuration [Au20(P)] is energetically favorable because of an
increased number of contacts with the substrate, leading to a
larger accumulation of interfacial charge (originating from the
2
of an O molecule is likely to occur in those regions, where the
density of the electrons in the HOKS orbital is higher (see
below).
c. Reaction Mechanisms of CO Combustion on Supported
Au20 Clusters. In this section, we explore several reaction
mechanisms corresponding to the various model catalyst systems
investigated in the TPR experiments (Figures 1 and 2). The
reaction profiles (pathways) that we present were obtained via
first-principles quantum calculations (see the Theoretical Meth-
ods section). In these calculations, a reaction coordinate was
judiciously chosen (that is, the distance between two reacting
3
underlying Mo substrate); as noted previously, the attraction
between the interfacial excess electronic charge and its image
5
42 J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009

Control and Manipulation of Gold Nanocatalysis
A R T I C L E S
atoms, for example, the distance between the C atom of an
adsorbed CO molecule and the nearest O atoms of an adsorbed
O molecule), and the total energy of the system was optimized
2
for given values of the reaction coordinate, through uncon-
strained relaxation of all of the other degrees of freedom of the
system (reactant molecules, gold cluster atoms, MgO surface
atoms, and those of the underlying Mo substrate). The reaction
profiles were obtained via repeating such calculations for various
values of the chosen reaction coordinate. These calculations
yield results that are the same as, or close to, those obtained by
other methods [e.g., the nudged elastic band and variants thereof
(
see the discussion on pp 89 and 90 in ref 10)]. Finally, we
note that while in the following we give calculated reaction
barriers with an accuracy of two decimal figures (e.g., 0.34 eV),
rounding off to one decimal figure (i.e., 0.3 eV) is appropriate.
i. Low-Temperature Mechanisms of Au20 Adsorbed on
Thin MgO Films. We model the thin MgO films used in the
measurements (Figure 2b-d, with MgO coverage up to 1 ML)
by an extended monolayer thin MgO film adsorbed on Mo(100).
We have found in previous studies that the reactivity of metal-
supported MgO films remains similar for film thicknesses of
4
up to three layers. Furthermore, we choose to focus on
systematic trends correlated with the thickness of the MgO
support and not to consider adsorption of gold clusters onto
MgO islands (e.g., adsorption at island step interfaces with the
underlying molybdenum oxide) because gold clusters adsorbed
on molybdenum oxide (between the MgO islands) are catalyti-
cally nonactive; the properties of gold clusters adsorbed on the
top facet of such MgO islands are similar to those adsorbed on
the extended MgO films. In addition, we focus only on MgO
films adsorbed on Mo(100); i.e., we do not model MgO layers
adsorbed on molybdenum oxide (which may also be generated
during the growth process); our demonstrated ability to suc-
cessfully describe the experimentally observed trends (see
below) supports a posteriori our model assumptions.
As depicted in Figure 1b-d, Au20 deposited on thin MgO
2
films forms CO at temperatures below 200 K in a one-heating-
cycle experiment. To model possible reaction mechanisms, we
consider the reactivity of thin film (one-layer MgO) Au20/MgO(1
L)/Mo systems. Molecular oxygen adsorbs at the interface
between the Au20(P) cluster and the MgO surface with a binding
Figure 3. Minimum-energy structures of Au20 clusters adsorbed on
surfaces. (a) Top (left) and side (right) views of a planar gold cluster
adsorbed on a one-layer-thick MgO layer, supported on Mo(100),
Au20(P)/MgO(1 L)/Mo. (b) Same as in that in part a for a 3D tetrahedral
gold cluster. The binding energies of the clusters (with reference to the
separated gas-phase clusters and surface subsystems) are 12.50 eV for
the planar cluster (a) and 5.73 eV for the 3D one (b). Superimposed on
the atomic structures are density isosurfaces corresponding to the excess
electronic charge distributions (blue corresponds to charge accumulation
and pink signifies charge depletion with reference to the isolated cluster
and surface components); these excess electron distributions are obtained
from the difference between the charge distribution of the combined
system [i.e., gold cluster adsorbed on the MgO surface, with or without
a Mo(100) support] and the charge distributions of the isolated
components (in the optimal geometry determined for the combined
system), that is, (i) the Au20 gold cluster and (ii) the MgO surface [with
or without a Mo(100) support]. Note the charge accumulation at the
interfacial periphery regions. (c) Left: Excess electronic charge distribu-
tion for a tetrahedral Au20 cluster adsorbed on an FC (on the surface of
a thick MgO film), located under the middle of the bottom facet of the
cluster. Right: Electron density isosurface of the HOKS orbital, showing
charge accumulation about the top apex atom, as well as near the oxygen
vacancy; the isosurface encompasses 61% of the electron density in the doubly
occupied orbital (i.e., 1.23 e). (d) Bilayer isomer of Au20 with a surface FC located
near the middle of the bottom facet of the adsorbed cluster. (e) Bilayer cluster with
a dimer of neighboring FCs located near the middle of the bottom fact of the
adsorbed cluster. The binding energies of the clusters to the MgO surface [with
reference to the energy of the isolated systems, MgO surface, and the corresponding
gas-phase cluster, i.e., Au20(T) in part c and Au20(bilayer) in parts d and e] are 4.36
eV (for part c), 4.92 eV (for part d), and 7.93 eV (for the 2FC case displayed in
part e). In the gas phase, the tetrahedral cluster is more stable than the bilayer one
by 1.57 eV. Superimposed in parts c and d are equidensity surfaces of the excess
electron charge density, with blue and pink corresponding to charge excess and
charge depletion, respectively. The excess electronic charges are (c) ∆q[Au20(T;
FC)] ) 1.80 e, (d) ∆q[Au20(bilayer; FC)] ) 2.10 e for the bilayer isomer in part
d, and (e) ∆q[Au20(bilayer; 2FC)] ) 2.85 e. Color designation: Au atoms in yellow,
O atoms in red, Mg atoms in green, and Mo atoms in black.
B 2
energy, E [O /Au20(P)], of 3.06 eV; the adsorbed molecule is
activated to a peroxo state [d(O-O) ) 1.515 Å] via occupation
of the antibonding 2π* orbital (see leftmost configuration in
Figure 4a). Clearly, oxygen adsorbs at the location of highest
charge accumulation. Through TS simulations (consisting, as
described above, of ground-state electronic structure calculations
and structural relaxations for adiabatically slow variation of the
reaction coordinate, i.e., the O-O bond in this case), we
determined a dissociation barrier of 0.34 eV with dTS(O-O) )
1.95 Å (Figure 4a). Relaxation from the TS configuration results
in the (rightmost) configuration displayed in Figure 4a, with
d(O-O) ) 3.29 Å. The reaction between this dissociated
2
configuration of the adsorbed O molecule and a CO molecule
adsorbed on top of a surface Mg atom [with a binding energy
of 0.47 eV and d(CO) ) 1.14 Å] at a distance d[C-O(1)] )
3.10 Å between the C atom and the closest O atom of the
adsorbed O
O(1)] is shown in Figure 4c. In this reaction, formation of an
adsorbed CO intermediate (TS configuration in Figure 4c, with
2
molecule [denoted here and in the following as
2
d[C-O(1)] ) 1. 80 Å, d(C-O) ) 1.17 Å, and d[O-O(1)] )
3.285 Å) involves an essentially negligible barrier (0.07 eV),
and the desorption energy of the CO product (from the leftmost
2
J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009 543

A R T I C L E S
Harding et al.
Figure 4. (a and b) Reaction paths for the dissociation of O
2
on 2D and 3D (tetrahedral) Au20 adsorbed on a one-layer thin MgO film supported on Mo(100).
In both cases, the TS activation energy is relatively low: ∆ETS ) 0.34 eV for the 2D gold structure (a) and ∆ETS ) 0.30 eV for the 3D adsorbed cluster (b).
In the inset to frame b, we show the relaxed structure of the adsorbed CO molecule produced by the reaction of the dissociated O molecule on Au20(T)/
MgO(1 L)/Mo with CO; no energy barrier is involved in this reaction, which starts from the dissociated O molecule and a CO adsorbed on the MgO surface
on top of a Mg atom, with an adsorption energy of 0.55 eV), at a distance of 2.48 Å between the C atom and the closest dissociated O atom. The desorption
2
2
2
(
2
energy of the product CO is 0.31 eV. This reaction contributes to the low-temperature peak shown in Figure 1b–d. In parts c and d, we display two other
low-temperature reactions pathways. The reaction in part c corresponds to an adsorbed planar, Au20(P) cluster, and it starts from the adsorbed dissociated
molecule shown in part a (with a dissociation activation energy of 0.34 eV) and a CO molecule adsorbed on top of a Mg atom [E (CO) ) 0.47 eV]. The
generation of adsorbed CO entails a TS barrier of ETS ) 0.07 eV and a desorption energy of the product Edes(CO ) ) 0.27 eV. The reaction in part d
corresponds to an adsorbed 3D Au20(T) cluster, and it involves a peripherally adsorbed activated O molecule and a CO molecule adsorbed on the MgO
surface (E ) 0.41 eV). The reaction in part d entails a TS barrier ∆ETS ) 0.30 eV, and the desorption energy of the product is Edes(CO ) ) 0.16 eV. Color
designation (online): Mg atoms in green, O atoms in red, Au atoms in yellow, C atoms in light gray, and Mo atoms in dark gray.
O
2
B
2
2
2
B
2
configuration in Figure 4c) is Edes ) 0.27 eV. The bottleneck
for this reaction consists of overcoming the relatively low
dissociation barrier (Figure 1a), and thus this reaction channel
contributes to the observed low-T reactivity on the thin MgO
film (Figure 1b-d).
Other reaction channels that may contribute to the low-
temperature reactivity are shown in parts b and d of Figure 4,
with both involving 3D Au20(T)/MgO(1 L)/Mo model catalysts.
The reaction in Figure 4b again involves dissociative adsorption
) 1.58 Å and dTS(O-O) ) 1.54 Å), with subsequent relaxation
yielding an adsorbed CO intermediate (d[C-O(1)] ) 1.30 Å,
d(C-O) ) 1.23 Å) and an essentially dissociated O molecule
(d[O-O(1)] ) 3.03 Å); the desorption energy to form gaseous
CO is 0.16 eV. We note that here, as well as in the other low-
2
2
2
temperature reaction channels discussed above, diffusion of the
reactant CO on the MgO film to the proximity of the adsorbed
3
1,32
cluster is assumed.
ii. High-Temperature Mechanisms on Au20 Adsorbed on
Thin MgO Films. The main contribution of the CO signal in
of O
.30 eV and dTS(O-O) ) 1.85 Å]. Relaxation from the TS
configuration results in the (rightmost) structure shown in Figure
b, with d(O-O) ) 3.33 Å. The reaction of the O(1) atom of
the dissociated O molecule with a CO molecule adsorbed with
2
at the interfacial cluster periphery [with a TS barrier of
2
0
the TPR experiments (Figure 1b-d) is observed at around 300
K. While, as described above, for the 2D Au20(P)/MgO(1 L)/
Mo catalyst reaction of adsorbed CO with predissociated
4
peripherally adsorbed O
when the surface-adsorbed CO reacts with an activated, but
undissociated peripherally adsorbed O , a relatively high TS
reaction barrier of 0.66 eV is found (Figure 5a, with
TS[C-O(1)] ) 1.54 Å and dTS(C-O) ) 1.20 Å). Subsequent
desorption of the adsorbed CO intermediate involves a desorp-
2
leads to a low-temperature reaction,
2
a binding energy of 0.55 eV on the MgO surface is essentially
barrierless. Relaxation of the TS complex yields an adsorbed
2
CO intermediate (see inset in Figure 4b, with d[C-O(1)] )
2
1
.52 Å, d(C-O) ) 1.20 Å, and d[O-O(1)] ) 3.41 Å), whose
d
desorption energy is Edes ) 0.31 eV. The other low-temperature
channel, shown in Figure 4d, does not involve dissociation of
2
tion energy Edes ) 0.29 eV. We note that for this case we
considered the initial surface adsorption of two CO molecules
the adsorbed O
periphery adsorbed O
2
molecule. Instead, it starts from an interfacial-
molecule that is activated to a peroxo
2
(
(
31) Harding, C. J.; Kunz, S.; Habibpour, V.; Heiz, U. Chem. Phys. Lett.
state [d(O-O) ) 1.504 Å] and a CO molecule adsorbed on the
MgO surface with a binding energy of 0.41 eV. Formation of
2
008, 461, 235.
32) Harding, C. J.; Kunz, S.; Habibpour, V.; Heiz, U. Phys. Chem. Chem.
Phys. 2008, 38, 5875.
CO entails a TS barrier of 0.3 eV (occurring at dTS[C-O(1)]
2
5
44 J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009

Control and Manipulation of Gold Nanocatalysis
A R T I C L E S
Figure 5. (a-c) High-temperature pathways of CO with O
2
catalyzed by for Au20 adsorbed on MgO(1 L)/Mo(100). (a) Formation of CO
(CO) ) 0.44 eV, near an activated O molecule adsorbed at the interfacial periphery of a
) ) 3.06 eV and d(O-O) ) 1.52 Å. The reaction entails a TS energy barrier of 0.66 eV, and
product requires Edes ) 0.29 eV. The reaction of the second adsorbed CO with the remaining O atom proceeds with no activation
barrier; however, the desorption of the resulting adsorbed CO molecule requires an energy of 0.60 eV. (b) Pathway of the oxidation reaction catalyzed by
a tetrahedral 3D Au20(T) cluster, with an activated O molecule adsorbed at the peripheral interface of the cluster with the underlying 1 L MgO film [E (O
4.19 eV] and a CO molecule adsorbed on the gold cluster [E (CO) ) 0.6 eV]. The TS barrier for the reaction is 0.58 eV, and desorption of the product
molecule requires 1.06 eV. (c) Enlarged view of the desorbing CO molecule, showing a configuration with the C atom bonded to the gold cluster and
2
through the
reaction of a CO molecule adsorbed on the MgO surface with E
B
2
planar (2D) Au20(P) cluster on the 1 L MgO film with E
B 2
(O
the desorption of the CO
2
2
2
B
2
)
)
B
CO
2
2
one of the O atoms bonded to two Mg surface sites [d(O-O) ) 2.08 Å], resulting in a relatively high desorption energy. (d) CO combustion reaction
pathway catalyzed by a tetrahedral Au20(T) cluster adsorbed on an eight-layer-thick MgO film supported on Mo(100) (see Figure 2a). In the initial configuration,
an activated O
atom of the cluster [E
online): Mg atoms in green, O atoms in red, Au atoms in yellow, C atoms in light gray, and Mo atoms in dark gray.
2
molecule is adsorbed [E
B
(O
2
) ) 0.65 eV] at the cluster interface with the magnesia film and a CO molecule is bonded to a second-layer Au
B
(CO) ) 0.50 eV]. The formation of a CO
2
product entails a 0.14 eV TS barrier, and the molecule desorbs readily. Color designation
(
proximal to the gold cluster (with a binding energy of 0.44 eV
per molecule; see Figure 5a). Formation of a second adsorbed
upper bound estimate of the desorption energy required for the
formation of a gaseous CO product is Edes ) 1.06 eV; this
rather high calculated desorption energy is attributed to the
strong bonding of one of the O atoms of the adsorbed product
2
CO
2
molecule via reaction of the second CO with the remaining
O atom adsorbed at the gold cluster periphery was found to
occur with essentially no activation barrier, but the formation
2
CO molecule to the magnesia surface, with the formation of
of gaseous CO
2
entails a relatively high desorption energy Edes
bonds to two Mg sites of the monolayer MgO film (see Figure
5c), enhanced by the excess accumulation of electronic charge
originating from the underlying metal [Mo(100)] support.
iii. Reaction Mechanisms of Au20 Adsorbed on Thick,
Defect-Free MgO Films. To explore the mechanisms underlying
the relatively low-temperature reactivity peak at T ≈ 250 K,
observed on defect-poor thicker (8-10 ML) films (Figure 2a),
we considered a defectless eight-layer-thick MgO film, i.e., the
model catalyst Au20(T)/MgO(8 L)/Mo (see Figure 5d). We start
)
0.60 eV; note that one of the O atoms of the adsorbed CO
2
molecule is bonded also to a Mg atom of the magnesia surface,
which accounts for the somewhat elevated desorption energy
of the product molecule. We conclude that the reactions of both
O atoms with surface-adsorbed CO are predicted to contribute
to the higher temperature channel observed experimentally.
Another reaction channel that involves a relatively high
activation barrier was found for the system CO/O
MgO(1 L)/Mo (see Figure 5b), where an O molecule is bound
to the periphery of the cluster (where the excess charge
accumulation is highest) with a binding energy E (O ) of 4.19
eV. The O molecule is peroxo-activated [d(O-O) ) 1.50 Å]
and interacts with an adsorbed CO bonded to the second layer
of the 3D gold cluster with E of 0.6 eV. Note that also at this
location a slight charge accumulation is calculated (see Figure
b). The TS barrier is 0.58 eV, and the corresponding TS
2
/Au20(T)/
2
from an equilibrated configuration of peripherally adsorbed O
[E (O ) ) 0.65 eV in a peroxo-activated state and d(O-O) )
1.41 Å] and a CO molecule bonded [E (CO) ) 0.50 eV] to a
2
B
2
B
2
B
2
Au atom in the second layer of the gold cluster. The formation
of a TS (Figure 5d) occurs with a low barrier of 0.14 eV at
d[C-(O(1)] of 2.0 Å, and the desorption of CO involves a
2
very small exit barrier. Note the drastic decrease in the reaction
and exit barriers when compared to Au20(T)/MgO(1 ML)Mo
(see Figure 5b), that is, from ∆ETS ) 0.58 eV and Edes ) 1.06
eV for the thin MgO film to ∆ETS ) 0.14 eV and Edes ) 0 for
B
3
depicted in Figure 5b is characterized by a dTS[C-O(1)] of 2.01
Å, a dTS(C-O) of 1.18 Å, and a d[O-O(1)] of 1.47 Å. An
J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009 545

A R T I C L E S
Harding et al.
the thicker one. This illustrates the marked influence of the
underlying metal surface on the catalytic activity of the adsorbed
Au20(T) cluster. In particular, the thin MgO layer allows for
enhanced charge accumulation [originating from the underlying
Mo(100) substrate], which serves to enhance the activation of
(A) First we display in Figure 6b the pathway for the reaction
on a bilayer isomer anchored on a single FC defect (the binding
energy of the cluster to the surface is calculated to be 4.92 eV).
Here the reaction starts from a peripherally adsorbed O
molecule [E (O ) ) 0.63 eV], which is activated to a peroxo
state [d(O-O) ) 1.43 Å] and a surface-adsorbed CO molecule
[E (CO) ) 0.29 eV] with the distance between the C atom to
the nearest O atom of the O molecule d[C-O(1)] ) 2.48 Å.
The TS barrier for the formation of an adsorbed CO-O
2
B
2
2
adsorbed species (specifically O ). However, the electronic
charge transferred through the thin metal oxide film enhances
binding of intermediates and reaction products to the surface,
and consequently, in certain cases, one finds for this system
higher activation barriers for exit channels corresponding to
dissociation of (TS and post-TS) intermediate reaction com-
B
2
2
complex (characterized by dTS[C-O(1)] ) 1.60 Å, dTS(C-O)
) 1.20 Å, and dTS[O-O(1)] ) 1.49 Å) is 0.29 eV, and
dissociation of the interoxygen bond between the complex and
2
plexes or desorption of the product molecule (e.g., CO ).
2
the other oxygen of the reactant O molecule entails an activation
iv. Reaction Mechanisms on Au20 Adsorbed on Defective
Thick MgO Films. In all of the cases discussed above, the MgO
films were defectless, and the catalytic activity was enabled by
the excess electronic charge (originating from the underlying
metal [Mo(100)] support) accumulated at the interface of the
gold nanocluster with the magnesia film. We turn now to
analysis of the experiments performed on a defect-rich thick
MgO(100) surface, where any effects due to an underlying metal
support are absent. In the data shown for this system in Figure
energy Ediss(O-O) ) 0.43 eV. Desorption of the product CO
2
molecule from the surface is barrierless.
(
B) A somewhat higher temperature reaction pathway is
shown in Figure 6c for a reaction catalyzed by a bilayer cluster
adsorbed on top of two neighboring oxygen vacancies, both
located in the vicinity of the middle of the bottom facet of the
cluster. The binding energy of this isomer to the surface is
E
B
[Au20(bilayer)] ) 7.93 eV (see Figure 3e). The reaction starts
from a peripherally adsorbed [E (O ) ) 0.84 eV], peroxo-
activated [d(O-O) ) 1.51 Å] O molecule and a CO molecule
adsorbed on the bilayer gold cluster [E (CO) ) 0.81 eV]. The
barrier for formation of a TS CO-O complex is 0.26 eV, and
the dissociation process of the complex (see above) entails an
B
2
2
c, the broad distribution of the reactivity (the smaller peak at
2
∼
210 K and the main broad distribution peaking at ∼400 K)
B
can be correlated with the multiplicity of adsorbed cluster
configurations that may coexist on such surfaces. These compet-
ing configurations differ from each other by the isomeric
structure of the adsorbed nanocluster and the nature of the
binding of the cluster to the surface; in particular, we consider
anchoring of clusters to single or double (nearest-neighbor)
surface oxygen vacancy (FC) defects (see Figure 3c-e). First
we considered a single FC defect located under the middle of
the bottom facet of the adsorbed tetrahedral Au20 cluster
2
2
energy Ediss ) 0.51 eV. Desorption of the product CO molecule
is found to be barrierless.
We digress here to illustrate one of the key principles of
1
8
nanocatalysis, namely, “dynamical structural fluctionality”,
which expresses the ability of the catalytic center (here the
surface-supported gold nanocluster) to change its atomic
structure (atomic arrangement) in the course of the reaction.
This property is illustrated by following, in the course of the
oxidation reaction, the relative positions of the three Au atoms
marked (by numbers 1, 2, and 3) in the inset to Figure 6c. In
the initial state (minimum-energy configuration) of the adsorbed
reactants (the rightmost structure in Figure 6c), the angle formed
by the three marked Au atoms (with the atom marked 2 being
at the apex of the triangle) takes the value θ(123) ) 59°. As
the TS is approached (decreasing the distance d[C-O(1)]
[
Au20(T)/MgO(FC)], whose binding energy to the MgO surface
is 4.36 eV. An O molecule adsorbs to the top apex Au atom
of the Au20(T) cluster with a binding energy E (O ) ) 0.28 eV
2
B
2
and a bond distance d(O-O) ) 1.28 Å (see Figure 6a); we
recall here that for this system the HOKS orbital exhibited
electronic charge accumulation at the apex atom (see the right
frame in Figure 3c), where the O
with adsorbed CO [E (CO) ) 0.54 eV], reaching a relatively
low TS barrier of 0.41 eV at dTS[C-O(1)] ) 1.75 Å, dTS(C-O)
1.18 Å, and dTS[O-O(1)] ) 1.35 Å (see Figure 6a) and
resulting in a post-TS CO-O complex (see the leftmost
configuration in Figure 6a). Dissociation of the complex requires
2
molecule adsorbs. It reacts
B
between the C atom and the closest O atom of the adsorbed O
molecule), the above angle increases to a value of θTS(123) )
2
)
2
7
6
6°. The relaxed configuration (the leftmost structure in Figure
c) is characterized by θ(123) ) 79°. We note here that the
an energy Ediss ) 0.49 eV, and desorption of the product CO
molecule occurs with no exit barrier.
2
fluctional structural distortions of the metal cluster, illustrated
above, occur to various degrees in all of the reactions described
in this study, with the largest structural distortions observed
when both reactants adsorb on the metal cluster. These structural
variations serve to enhance the adsorption of the reactants and
to lower the activation barriers for reactions between the
adsorbed reactants. We note here that the activation of the
Other initial adsorption configurations that we have explored
include peripherally adsorbed O with CO adsorbed either on
2
the MgO surface or on the tetrahedral cluster. In all of these
cases, we obtained activation barriers in the range of 0.6-0.75
eV. For example, starting from a peripheral adsorption of O
2
[
[
E
B
(O
d(O-O) ) 1.49 Å] and a CO molecule adsorbed on the MgO
(CO) ) 0.28 eV, d[C-O(1)] ) 3.02 Å, d(C-O) )
.14 Å) resulted in a TS barrier ∆ETS ) 0.62 eV, with
TS[C-O(1)] ) 1.65 Å, dTS(C-O) ) 1.18 Å, and dTS(O-O) )
.56 Å, which relaxes to a state with an adsorbed CO molecule
2
) ) 0.49 eV] yielding a peroxo-activated molecule
2
reactant O molecule (to form a superoxo- or peroxo-like state
through occupation of the antibonding orbital, as discussed
above), which is manifested by elongation of the O-O bond
of the adsorbed molecule (located often at the cluster peripheral
interface with the underlying MgO surface) by up to ∆d(O-O)
) 0.25 Å, is accompanied by changes in the positions (relative
distances) of the Au atoms in the vicinity of the adsorption site.
surface (E
B
1
d
1
2
(
d[C-O(1)] ) 1.18 Å, d(C-O) ) 1.17 Å, and d(O-O) ) 3.20
Å). Desorption of the product CO
2
molecule is found to involve
(C) As an example of a higher temperature reaction, we show
a negligible barrier.
in Figure 6d the pathway of a reaction catalyzed by the bilayer
gold cluster adsorbed on a single surface FC (as in Figure 6b),
Next we consider several reaction pathways involving a bilayer
isomeric form of the adsorbed Au20 cluster (see Figure 3d,e):
starting from a peripherally adsorbed, peroxo-activated, O
2
5
46 J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009

Control and Manipulation of Gold Nanocatalysis
A R T I C L E S
Figure 6. Reaction pathways for Au20 clusters adsorbed on oxygen vacancies, surface FCs, and thick MgO(100) surfaces, corresponding to the experimental
data displayed in Figure 2c. (a) Reaction profile for a reaction catalyzed by a 3D tetrahedral Au20(T) cluster adsorbed on a single FC located under the
middle of the bottom facet of the cluster, whose binding energy to the MgO surface is 4.36 eV (see Figure 3c). An O
atom of the Au20(T) cluster with a binding energy E (O ) ) 0.28 eV, and it reacts with an adsorbed CO [E (CO) ) 0.54 eV], reaching a relatively low TS
barrier of 0.41 eV. Dissociation of the CO-O complex requires a dissociation energy Ediss ) 0.49 eV, and desorption of the CO product is barrierless.
b-d) Reaction pathways catalyzed by bilayer Au20 cluster isomers. (b) Pathway for the reaction on a bilayer isomer anchored on a single FC defect (see
Figure 3d). The reaction starts from a peripherally adsorbed, peroxo-activated, O molecule [E (O ) ) 0.63 eV)] and a surface-adsorbed CO molecule
(CO) ) 0.29 eV]. The TS barrier for the formation of an adsorbed CO-O complex is 0.29 eV, and dissociation of the O-O bond between the complex
and the other oxygen of the reactant O molecule Ediss(O-O) ) 0.43 eV. Desorption of the product CO molecule from the surface is barrierless. (c) CO
combustion reaction catalyzed by a bilayer cluster adsorbed on top of two neighboring oxygen vacancies (see Figure 3e). The reaction starts from a peripherally
adsorbed [E (O ) ) 0.84 eV], peroxo-activated O molecule and a CO molecule adsorbed [E (CO) ) 0.81 eV] on the bilayer gold cluster. The barrier for
the formation of a TS CO-O complex is 0.26 eV, and the dissociation process of the complex entails an energy Ediss ) 0.51 eV. Desorption of the product
CO molecule is found to be barrierless. In the inset, Au atoms whose positions markedly distort during the reaction (that is, exhibiting dynamic structural
2
molecule binds to the top apex Au
B
2
B
2
2
(
2
B
2
[
E
B
2
2
2
B
2
2
B
2
2
fluctionality) are designated as 1, 2, and 3. The distances between these atoms at various stages of the reaction are as follows: initial (rightmost) configuration,
d(1-2) ) 2.95 Å, d(2-3) ) 3.35 Å, d(1-3)) 3.14 Å, and θ(123) ) 59°; TS configuration, d(1-2) ) 2.89 Å, d(2-3) ) 3.11 Å, d(1-3) ) 3.68 Å, and
θ(123) ) 76°; post-TS configuration (corresponding to d[C-O(1)] ) 2.09 Å), d(1-2) ) 2.94 Å, d(2-3) ) 3.27 Å, d(1-3) ) 3.96 Å, and θ(123) ) 79°.
(
d) Pathway of the reaction catalyzed by the bilayer gold cluster adsorbed on a single surface FC (as in Figure 6b), starting from a peripherally adsorbed,
peroxo-activated, O molecule [E (O ) ) 0.96 eV] and a surface-adsorbed CO molecule [E (CO) ) 0.27 eV]. The reaction involves a TS energy barrier of
.68 eV, and breakup of the TS complex resulting in desorption of the product CO molecule occurs with no energy barrier. Color designation (online): Mg
atoms in green, O atoms in red, Au atoms in yellow, and C atoms in light gray.
2
B
2
B
0
2
molecule [E
surface-adsorbed CO molecule [E
a distance d[C-O(1)] ) 3.14 Å between the C atom and the
nearest O atom of the adsorbed O molecule. This reaction
involves a TS energy barrier of 0.68 eV, and breakup of the TS
B
(O
2
) ) 0.96 eV and d(O-O) ) 1.50 Å] and a
defect density. These dependencies are reflected in variations
of the reaction temperatures observed in temperature-pro-
grammed desorption single-heating-cycle experiments, as well
B
(CO) ) 0.27 eV] located at
2
2
as in the amount of CO produced. The first-principles theoreti-
cal investigations presented here show that the observed changes
in the reactivity may be correlated, in part, with a dimensionality
crossover from 3D tetrahedral Au20 in the case of thick films
(g8 ML) to 2D “wetting” planar structures for film thicknesses
of less than ∼3 ML; we note here that we have shown (see
also the Supporting Information) through MIES and UPS
measurements that these thin MgO films are highly stoichio-
metric; that is, they may contain only a very low number density
of (Lewis-base type) oxygen vacancy (FC) defects.
complex resulting in desorption of the product CO
2
molecule
occurs with no barrier.
The above examples, drawn from an ensemble of configura-
tions corresponding to calculated reaction profiles with activation
barriers in the range of about 0.35-0.75 eV, correlate well with
the observed broad temperature distribution measured for the
CO reaction on Au20 clusters deposited on thick, defect-rich
MgO surfaces (Figure 2c).
Underlying the aforementioned structural and dimensionality
variations is the enhanced charge transfer from the underlying Mo
surface through the metal oxide occurring for the thinner films.
This transferred charge accumulates mainly at the interfacial region
of the adsorbed metal cluster with the metal oxide, and it stabilized
Summary
The oxidation of CO on Au20 sensitively depends on the
thickness of the MgO film grown on a Mo(100) single crystal
as well as on the metal oxide stoichiometry, that is, surface
J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009 547

A R T I C L E S
Harding et al.
planar (wetting) configurations of the cluster (through attractive
image charge interactions, which increase with the contact area of
We believe that the concepts and methodologies developed
in this paper, and the demonstrated ability to control and tune
the catalytic reactivity of gold nanoclusters through variations
of the thickness and/or composition of the support [here
MgO(100) grown on the surface of Mo(100)], are of importance
for the development of effective gold-based nanocatalytic
systems. Moreover, it will be of great interest to explore the
extension of these concepts and control methods to nanocatalytic
systems based on other combinations of metal nanoclusters and
oxide supports.
3,4
the metal cluster with the surface). Furthermore, the excess charge
can enhance the chemical catalytic activity of the adsorbed (partially
charged) gold clusters (both 3D and 2D) via transfer of charge to
4
adsorbed reactant molecules; for example, activation of adsorbed
O through population of the 2π* antibonding orbital and formation
2
of a superoxo- or peroxo-activated molecule, which may react with
CO in a Langmuir-Hinshelwood mechanism entailing a lower
activation barrier. On the other hand, the excess interfacial charge
accumulation for thin metal oxide films may enhance the binding
strength of reaction intermediates and/or adsorbed product mol-
ecules, thus causing larger exit reaction barriers. Such reaction
mechanisms have been illustrated in this study for both 2D and
Acknowledgment. The research of U.L. and B.Y. was supported
by a grant from the U.S. Air Force Office for Scientific Research
and the U.S. Department of Energy. The calculations were
performed at the Georgia Tech Centre for Computational Materials
Science and at the National Energy Research Scientific Computing
Center at Berkeley, CA. Research at Technische Universit a¨ t
M u¨ nchen was supported by the Deutsche Forschungsgemeinschaft,
the Deutsch-Franz o¨ sische Hochschule, and the European Union
within the COST D41 program. S.K. acknowledges support of the
Fond der Chemischen Industrie.
3D Au20 nanoclusters adsorbed on thin MgO(1 L)/Mo(100).
For thick films and 3D Au20 adsorbed clusters (both tetra-
hedral and bilayer isomeric structures), charge accumulation,
with the concomitant charging of the adsorbed clusters, can be
induced by defect sites of Lewis-base character (e.g., oxygen
vacancies, FCs). This charge accumulation is local, and it
depends on the type of electron donor. In both cases, charging
through thin metal oxide films and FC-induced charging, the
excess charge is found to be located mainly on the perimeter
atoms of the 2D or 3D adsorbed Au20 clusters; we also find
that in the case of a 3D tetrahedral Au20(T) cluster anchored to
a MgO surface FC the HOKS orbital exhibits an enhanced
electronic density localized on the top apex atom of the
tetrahedron. The charge accumulation defines the location of
the reactive site on the cluster, and thus reactivity can be tuned
as a function of the properties (thickness and stoichiometry) of
the supporting metal oxide film.
Supporting Information Available: Experimental procedure
for MgO film preparation and characterization of the film
thickness and morphology by AES, He* MIES, and UPS as
well as of work functions as a function of the MgO film
thickness measured by UPS. This material is available free of
charge via the Internet at http://pubs.acs.org.
JA804893B
5
48 J. AM. CHEM. SOC. 9 VOL. 131, NO. 2, 2009