Tuning the oxidation of carbon monoxide using nanoassembled model catalysts

Article abstract of DOI:10.1016/S0301-0104(00)00268-8

While the oxidation of CO has long been considered to be structure insensitive, nano-assembled model catalysts consisting of size-selected metal clusters (Au(n), Pt(n), Pd(n), and Rh(n), n = 1-20), supported on thin MgO(100) films, reveal distinct size-effects. When adding a single Pt atom to Pt₁₄, platinum clusters increase their reactivity by a factor of three. Rh₂₀ shows the highest reactivity of all the investigated clusters, oxidizing about 13 CO molecules per cluster at 350 K. While F-centers on MgO films transform Au₈ from an inert to an active catalyst, the reactivity of Pd₈ is not suppressed when deposited on defect-poor films. These different catalytic properties are rationalized within simple frontier orbital models, whereas for Au₈ the reaction mechanism for the low temperature oxidation of CO is elucidated within first-principle calculations. (C) 2000 Elsevier Science B.V.

Full text of DOI:10.1016/S0301-0104(00)00268-8

Chemical Physics 262 (2000) 189±200
www.elsevier.nl/locate/chemphys
Tuning the oxidation of carbon monoxide using
nanoassembled model catalysts
U. Heiz *, A. Sanchez, S. Abbet, W.-D. Schneider
Institut de Physique de la Matieꢀre Condenseꢁe, Universiteꢁ de Lausanne, CH-1015 Lausanne-Dorigny, Switzerland
Received 29 February 2000
Abstract
While the oxidation of CO has long been considered to be structure insensitive, nano-assembled model catalysts
consisting of size-selected metal clusters (Aun, Ptn, Pdn, and Rhn, n ˆ 1±20), supported on thin MgO(1 0 0) ®lms, reveal
distinct size-e€ects. When adding a single Pt atom to Pt14, platinum clusters increase their reactivity by a factor of three.
Rh20 shows the highest reactivity of all the investigated clusters, oxidizing about 13 CO molecules per cluster at 350 K.
While F-centers on MgO ®lms transform Au8 from an inert to an active catalyst, the reactivity of Pd8 is not suppressed
when deposited on defect-poor ®lms. These di€erent catalytic properties are rationalized within simple frontier orbital
models, whereas for Au8 the reaction mechanism for the low temperature oxidation of CO is elucidated within ®rst-
principle calculations. Ó 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
particles supported on well-characterized oxide
surfaces. In most case studies performed so far, the
active metal particles are composed of hundreds or
thousands of atoms [11,13]. In this size range, the
number of electronic states is large and the elec-
tronic structure of the particles is well described by
electronic bands. This is the reason why intrinsic
bulk properties of a given element do not change
as function of the precise particle size. As a con-
sequence, the observed structure sensitivities and
insensitivities are surface e€ects, and they are de-
pendent on the electronic surface properties and
surface morphologies like densities of kinks, steps
as well as the types of crystalline facets the parti-
cleÕs surface is composed of [14]. For example, the
Ag(1 1 1) surface supports an occupied surface
state as opposed to Ag(1 0 0); such di€erences are
one possible reason responsible for structure sen-
sitivities in catalytic reactions [15]. Likewise, for
Ni(1 1 1) the work function, or in the language of
Boudard de®nes a structure sensitive reaction as
a catalytic process depending on size and mor-
phology of the catalytically active compound [1].
For example, the cyclotrimerization of acetylene is
catalyzed with highest eciency on Pd(1 1 1) facets
[2±9] and reveals distinct di€erences on small Pdn
ꢀn ˆ 1±30† clusters [10]. A structure insensitive
reaction, e.g., the oxidation of carbon monoxide
has long been considered as such [11], depends
neither on size nor shape of the involved particles
[1]. To study di€erent catalytic processes in detail,
the concept of model catalysts was introduced by
Poppa [12]. Such model catalysts consist of metal
* Corresponding author. Tel.: +41-21-692-36-67/53; fax: +41-
21-692-36-35.
E-mail address: heiz@dpmail.ep¯.ch (U. Heiz).
0301-0104/00/$ - see front matter Ó 2000 Elsevier Science B.V. All rights reserved.
PII: S0301-0104(00)00268-8

190
U. Heiz et al. / Chemical Physics 262 (2000) 189±200
molecular orbital theory, the position of the
highest occupied molecular orbital (HOMO) is
5.35 eV, whereas for Ni(1 1 0), it is only 5.04 eV
[16], illustrating the di€erent electronic surface
properties and the concomitant changing chemical
properties of di€erent crystal planes. Finally, the
lattice constants of various crystal planes di€er.
This structural parameter determines whether a
molecular overlayer involved in a catalytic reac-
tion is adsorbed in a commensurate or incom-
mensurate con®guration and whether there is
activation of the molecules due to deformation.
Furthermore, considering the large active metal
particle as a sphere with radius R, the surface-to-
bulk ratio is changing monotonically with 1=R and
thus, the catalytic properties vary accordingly.
Superimposed to this monotonic change of the
chemical reactivity of large particles is their size-
dependent behavior solely characterized by the
particlesÕ morphology, e.g. the density of steps or
kinks [17±19].
For small clusters, however, these simple con-
cepts fail. In nano-scale clusters, the valence elec-
trons are highly con®ned and quantum e€ects
become dominant. Therefore, each cluster with a
speci®c number of atoms has its own intrinsic
chemical and physical properties [20±27] and new
concepts have to be developed to classify size-
dependent catalytic properties of nano-scale clus-
ters. A possible starting point for this classi®cation
is using the large body of data obtained from gas
phase cluster studies. In the early eighties, clusters
of the monovalent atom sodium were ®rst pro-
duced by the group of Schumacher [28] driven by
his knowledge of the important role of small silver
particles in the photographic process and by the
group of Knight [20]. Striking discontinuities in
the mass spectra at magic numbers of 8, 20, 34, 40,
. . . (Na atoms per cluster) were discovered. The
Berkeley group [20] achieved a fundamental un-
derstanding of these abundance spectra within the
Jellium model [20,29±31], well known to nuclear
physicists. In this way, the electronic shell struc-
ture of metallic clusters was established. In anal-
ogy to the classic periodic table of the elements
used to predict chemical and physical properties, a
periodic table for clusters can be constructed
simply by ®lling the electronic shells representing
the di€erent periods [32]. In this way, we ®nd
monovalent clusters, such as Na3, Na9, and Na19
analogous to the alkaline metals or the open-shell
clusters Na7 and Na17 similar to halides. These
clusters might reveal changing reactivities like the
corresponding elements. This simple comparison
suggests, therefore, that for small clusters, where
the electronic structure is dominant in changing
the clusterÕs properties, a new and exciting chem-
istry is waiting for to be discovered. Such a size-
dependent variation of the chemical properties was
observed in several gas phase studies, where reac-
tivities of small metal clusters were investigated.
ꢁ
Berces et al. measured absolute rate coecients for
the reaction of Nbn clusters ꢀn ˆ 2±20† with H2
and N2 at di€erent temperatures [33], con®rming
the low reactivities of Nb8, Nb10, and Nb16 mea-
sured in earlier experiments [34,35]. They recog-
nized an anticorrelation of the reactivity with an
e€ective ionization potential of the cluster that
includes the polarization of the charge on the clus-
ter cation. This correlation between the electronic
structure and the reactivities of niobium clusters
was also observed experimentally by Kietzmann
et al. [36]. These authors estimated the HOMO±
LUMO gap from photoelectron spectra and ob-
served closed shells and a large HOMO±LUMO
gap for the unreactive cluster sizes. In addition,
ꢁ
Berces et al. clearly observed biexponential kinetic
plots, indicative of the presence of di€erent iso-
mers with distinct reactivities. They suggested that
in addition to electronic, steric e€ects also play a
decisive role in determining cluster reactivities [33].
Shi et al. and Hagen et al. even investigated a
whole catalytic cycle and studied the catalytic ox-
idation of CO on free platinum and gold clusters
in a molecular beam experiment [37,38].
In this work, we went a step towards real cat-
alysts and investigated the oxidation of CO on
size-selected supported clusters of Aun, Ptn, Pdn,
and Rhn ꢀ1 6 n 6 20† and discovered a strong size-
dependent behavior. The aim of this work is to
highlight some important examples in the oxida-
tion of CO on cluster catalysts. In most cases, due
to the lack of accurate simulations at the present
time, we rationalize our results with simple mod-
els. First, the observed size-dependency for the
CO-oxidation within an element is compared with

U. Heiz et al. / Chemical Physics 262 (2000) 189±200
191
the evolution of the electronic structure and the
morphology as function of cluster size. The dif-
ferent reactivities of a single cluster size (Au13,
Pt13, Pd13, and Rh13) are then compared and ra-
tionalized with the di€erent electronic structure of
these hypothetically Ôgeometrically closed shellÕ
clusters [39]. Finally, we review results of ®rst-
principle calculations [40,41], which revealed the
complete reaction mechanisms of the oxidation of
CO on deposited Au8.
by a quadrupole mass ®lter with a mass limit of
4000 amu. A typical mass spectrum of Rhn clus-
ters, where the expansion conditions were opti-
mized for the production of small clusters [42], is
shown in Fig. 2. The monodispersed cluster beam
is deposited with low kinetic energy (Ekin < 0:2 eV
per atom [42]) onto thin MgO(1 0 0) ®lms at 90 K
10
under ultrahigh vacuum (UHV) conditions (10
Torr). The total energy of the deposition process
consists of the kinetic energy, the involved chem-
‡
‡
‡
ical binding energy between the Aun , Ptn , Pdn ,
‡
and Rhn clusters and the MgO surface (0.4±1.4 eV
per interacting atom [40,43,44]), the neutralization
energy, as well as a negligible Coulomb interaction
of the incoming cluster ion and its induced polar-
ization charge and image charge on the oxide ®lm
surface and the metal surface, respectively. Con-
sequently, as the kinetic energies of the impinging
clusters correspond to soft-landing conditions
(Ekin 6 1 eV per atom) [45±47], and as the total
energy gained upon deposition is at least a factor
of two smaller than the binding energies of the
investigated clusters ranging from 2 to 5 eV
2. Experimental
The used model catalysts, consisting of size-
selected metal clusters on thin MgO(1 0 0) ®lms are
prepared by recently developed cluster-deposition
facilities [42]. Brie¯y, the clusters are formed by a
high frequency laser evaporation source (Fig. 1).
‡
‡
The produced singly charged clusters (Aun , Ptn ,
‡
‡
Pdn , Rhn ) are transported by ion optics through
di€erentially pumped vacuum chambers, de¯ected
by 90° by a quadrupole de¯ector and mass selected
Fig. 1. Experimental setup composed of high frequency laser evaporation cluster ion source driven by a 100 Hz Nd:Yag laser (Co-
herent In®nity 40=100), ion optics with a quadrupole de¯ector and quadrupole mass spectrometer and the analysis chamber. The latter
is equipped with a mass spectrometer for thermal desorption spectroscopy (TDS), a Fourier transform infrared (FTIR) spectrometer, a
spherical electron energy analyzer used for Auger electron spectroscopy (AES), a gas-handling system equipped with a calibrated
molecular beam doser, and evaporation sources for thin oxide ®lm preparations.

192
U. Heiz et al. / Chemical Physics 262 (2000) 189±200
arity of the deposited clusters are formed. In the
experiments presented in this work, between 0.14%
and 0.40% of a monolayer of size-selected clusters
(1 ML ˆ 2:2  1015 clusters/cm2) are deposited at
90 K, in order to land them isolated on the surface
and to prevent agglomeration of the clusters on the
support [55].
The MgO(1 0 0) ®lms are prepared in situ for
each experiment by epitaxially growing the oxide
®lms on a Mo(1 0 0) surface by evaporating mag-
nesium at 300 K in a 16O2 background [56]. These
®lms show similar properties as the bulk analog as
evidenced with LEED, EELS, and XPS (Fig. 3(a)±
(c)), however, a reproducible density of defects is
present on the surface [40]. These defects are ti-
trated with small molecules like CO or NO (Fig.
3(d)/(e)). In this work, we distinguish between
defect-rich and defect-poor MgO ®lms. The defect-
poor ®lms are characterized by desorption of CO
at 120 K (Fig. 3(d)), which results from adsorption
of CO on domain boundaries most probably.
Defect-rich ®lms are characterized by additional
desorption of about 1% monolayer (ML) of NO
at 650 K (Fig. 3(e)), which is believed to originate
Fig. 2. Mass spectrum of Rhn clusters. The cluster source (He
pressure, laser delay) and the transmission of the ion optics are
optimized for obtaining maximal signal for Rh5.
[48±53], fragmentation of the cluster is excluded.
Soft-landing of the clusters is con®rmed in an in-
dependent experiment, where small nickel clusters
(Ni, Ni2, and Ni3) are deposited under the same
conditions and saturated with CO [54]. It could be
shown that only nickel carbonyls with the nucle-
Fig. 3. Characterization of the MgO(1 0 0) thin ®lms: defect-poor ®lms (see text) show (a) LEED, (b) EELS, and (c) XPS spectra
typical for MgO(1 0 0) bulk systems. However, CO desorbs at an untypically high temperature of 120 K (d). Defect-rich ®lms are
characterized by the desorption of NO at around 600 K (e).

U. Heiz et al. / Chemical Physics 262 (2000) 189±200
193
from NO adsorbed on F-centers [40,57]. Both
oxide ®lms have been proven to be inert towards
the oxidation of CO, e.g., no CO2 is produced with
the experimental conditions used for the cluster
experiments. For the study of the catalytic activity
of the clusters, the model catalysts are ®rst exposed
by a molecular beam doser to an average of 20
18O2 molecules per deposited atom at a sample
temperature of 90 K, and subsequently to the same
number of 13CO or 12CO molecules. This is crucial,
as it is well known from single crystal studies that
the initial amount of oxygen and CO on the sur-
face and the ratio of these molecules during oxi-
dation may in¯uence the reactivity [58±61]. In a
temperature programmed reaction (TPR) experi-
ment, the products of the chemical reaction are
monitored by mass spectrometry. The reactivities
obtained in this one-heating-cycle process are de-
®ned as the produced number of 13C 18O 16O or
12C 18O 16O molecules per cluster. This catalytic
action is determined by integrating the TPR signal
of the CO2 molecules and normalizing to the
number of deposited clusters. For this procedure,
the mass spectrometer has been calibrated using
the known amount of desorbing CO from a
Mo(1 0 0) single crystal [62] and taking account of
the di€erent detection sensitivities for CO and
CO2. The number of deposited clusters is obtained
by integrating the measured ion current on the
substrate over the deposition time.
Fig. 4. CO combustion on Aun, Ptn, Pdn, and Rhn (n ˆ
8; 13; 20) obtained by a one-heating-cycle experiment. The
model catalysts were ®rst exposed to an average of 20 molecules
of 18O2 per deposited metal atom and subsequently by the same
number of carbon monoxide. The CO2 signal was scaled to the
coverage of the deposited metal clusters.
3. Results
ture (500 K). In contrast to these electronically
closed shell clusters, Au13, reveals the lowest re-
activity. Such a strong size-dependency is also
observed for Ptn, where the number of formed CO2
per cluster increases by a factor of 9 when going
from Pt8 (0:6 Æ 0:1 CO2) to Pt20 (5:6 Æ 0:1 CO2)
(Fig. 4). In addition, Pt20 reveals two high tem-
perature mechanisms at around 350 and 500 K,
where for the octamer CO2 is produced only at 300
K. It is interesting to note that only the larger
clusters (Ptn, n > 14) form CO2 at a temperature of
140 K [27]. For the other two metals, palladium
and rhodium, even the octamers show consider-
able reactivity for the oxidation of CO (3:0 Æ 0:1
and 2:8 Æ 0:1 molecules of CO2 per Pd8 and Rh8,
Fig. 4 shows the TPR spectra for the CO-
combustion on supported Aun, Ptn, Pdn, and Rhn
ꢀn ˆ 8; 13; 20† clusters. Surprisingly, each cluster
size shows a di€erent reactivity and reaction tem-
perature, although the oxidation of CO has long
been considered as structure insensitive. As for the
bulk analogs, gold clusters reveal the lowest reac-
tivity, expressed in the number of 13C 18O 16O
formed per cluster. Interestingly, for gold clusters
Au8 is the most reactive cluster (1:1 Æ 0:1 CO2) at
the lowest temperatures (140 and 240 K) when
deposited on defect-rich ®lms, whereas for Au20
(1:9 Æ 0:1 CO2), in addition to the oxidation of CO
at 140 K, CO2 is mainly formed at high tempera-

194
U. Heiz et al. / Chemical Physics 262 (2000) 189±200
respectively). Of all the investigated clusters, Rh20
is by far the most reactive one (12:7 Æ 0:1 CO2).
The interaction of CO with the octamers of the
di€erent elements was investigated by FTIR. Fig. 5
reveals the di€erent vibrational frequencies of
13CO adsorbed on Au8 (2055 cm 1), Pt8 (2020
cm 1), Pd8 (2037, 2014, 1893 cm 1) and Rh8 (2020
cm 1) measured at 90 K. All spectra show domi-
nant frequencies typical for CO adsorbed in an on-
top con®guration, only the spectrum of Pd8 reveals
frequencies typical for bridge bonded CO (1893
cm 1). In the case of Au8, the observed 13CO fre-
quency is the closest to the one of gas phase 13CO
(2096 cm 1), and therefore, CO is the weakest
bound to this cluster. Note the relatively broad
adsorption bands for Pd8 and Rh8, indicative for
several COs bond to the cluster surface. It is in-
formative to compare the reactivities of clusters
with the same nuclearity. Here, we concentrate on
Au13, Pt13, Pd13, and Rh13, in order to reveal the
in¯uence of the di€erent electronic structures on
their chemical properties. Au13 shows by far the
lowest reactivity (0:6 Æ 0:1 CO2), followed by Pt13
(1:9 Æ 0:1 CO2). The highest reactivities in this
series are observed for Pd13 (4:2 Æ 0:1 CO2) as well
as Rh13 (4:4 Æ 0:1 CO2).
For Pt20, Fig. 6 shows desorption of catalyti-
cally produced CO2 and CO, which is not oxidized
Fig. 6. Temperature programmed reaction: CO2 production
and CO desorption for Pt20 as a function of temperature, (Á Á Á)
data. Isotopically labeled 12C 16O and 18O2 are used to unam-
biguously attribute the catalytic activity to the supported clus-
ters. The number of desorbing molecules can be estimated by
integrating the TPR spectra (see text). Inset: FTIR spectra of
adsorbed CO during the catalytic oxidation (e.g. coadsorbed
with O2) are shown. The samples are annealed to the indicated
temperatures and all spectra are then recorded at 90 K.
Fig. 5. Shown are infrared absorption frequencies of 13CO
adsorbed on (a) Au8, (b) Pt8, (c) Pd8, and (d) Rh8. All spectra
are taken at 90 K. Note the high frequency of CO adsorbed on
the gold octamer. Note that the frequency for 12CO can be
calculated by using mꢀ13CO†=mꢀ12CO† ˆ ꢀlꢀ12CO†=lꢀ13CO††
,
1=2
with l being the reduced mass.

U. Heiz et al. / Chemical Physics 262 (2000) 189±200
195
by the cluster. Note that the TPD for CO is only
depicted from 200 to 600 K. The low temperature
range is characterized by the well-known CO de-
sorption typical for these thin MgO-®lms [63]. In
comparison, the evolution of the IR absorption of
CO during the reaction reveals an increase in the
integrated absorption intensity between 100 and
160 K. From 160 to 500 K, a distinct decrease of
the integrated absorption intensity is observed,
which is in concert with the oxidation of CO from
160 to 350 K, whereas between 350 and 550 K the
desorption of CO also contributes to the decrease
of the IR signal. Note that there is no evidence of
adsorbed CO2 as inferred from the missing IR
absorption in the CO2 frequency range.
Finally, the in¯uence of the defect density on
the MgO(1 0 0) thin ®lms was investigated for the
oxidation of CO on Au8 and Pd8. Au8 is only re-
active when deposited on defect-rich ®lms (Fig. 7),
whereas for Pd8 the reactivity expressed in the
number of oxidized CO does not reveal such a
drastic change when varying the defect density on
the MgO ®lms (Fig. 7).
Fig. 7. CO combustion on Au8 and Pd8 deposited on defect-
rich and defect-poor MgO ®lms. Note the decrease in reactivity
of Au8 at 240 K when deposited on defect-poor ®lms. For Pd8
only the reactivity at higher temperature is suppressed on de-
fect-poor MgO ®lms.
4. Discussion
4.1. CO oxidation and its size-dependent evolution
with cluster size
Our results showed that such small clusters are
already catalytically active and revealed element
speci®c and size-dependent reactivities. As only
CO2 molecules with one isotopically labeled 18O
atom are detected, the adsorbed oxygen is disso-
ciated prior or during the oxidation of CO and no
For the oxidation on the cluster surface to take
place, several conditions have to be ful®lled. In an
Eley±Rideal type of mechanism one of the reac-
tant, in this case the O2 molecule, has to be ad-
sorbed on the catalyst. The activation of the
molecule has to be such that the activation energy
for the formation of the product is low enough
that collisions with the incoming CO lead to the
formation of CO2. In a Langmuir±Hinshelwood
type of reaction, both reactants are adsorbed on
the surface and the activation energy for the CO2
formation has to be lower than the desorption
energies of the products. In addition, none of the
reactants ought to block the reactive sites on the
cluster. The one cycle heating experiments pre-
sented here are only sensitive to the Langmuir±
Hinshelwood type of reaction.
13
disproportionation reaction ꢀ C16O ‡ 13C16O !
13C16O2† occurs. In most cases, the reaction is not
stoichiometric as shown by CO desorbing during
the reaction (shown for Pt20 in Fig. 6). As ex-
pected, the reaction heat of the oxidation process
is responsible for the immediate desorption of
CO2 from the cluster. This is inferred experi-
mentally from the absence of IR absorptions at
frequencies typical of adsorbed CO2, and more
directly from a comparison of the evolution of the
vibrational frequencies of CO and the formation

196
U. Heiz et al. / Chemical Physics 262 (2000) 189±200
of CO2 and desorption of CO during the TPR
experiment. The distinct increase in integrated ab-
sorption intensity between 100 and 160 K can be
explained by migration of physisorbed CO to the
cluster and/or reorientation of the adsorbed CO
molecules and a concomitant enhanced absorption
tinuum of states of the metal surface. This cou-
pling with, e.g., the broad sp-bands of the metal
results in broadened adsorbate levels due to the
reduced lifetime of the electrons in the discrete
adsorbate states. The resulting interaction is char-
acterized as a weak chemisorption. The coupling
of the adsorbate states with the narrower d-bands
of the metal results in bonding and antibonding
states; this interaction is called strong chemisorp-
tion. If we replace the molecule±surface interaction
with the molecule±cluster interaction, the follow-
ing size-dependent scenario emerges. For the ox-
ygen molecule, the electronic states of interest are
the bonding, fully occupied pu and rg and the
antibonding, half-occupied pÃg orbitals. Providing
there exists a resonance (energetically and sym-
metrically) between one of these orbitals and the
cluster's density of state (DOS), adsorption and
dissociation of O2 is possible [67]. As shown by
density functional theory (DFT) calculations, the
d-band position is the most decisive parameter,
when describing molecule±surface bonding [68].
Dissociation or activation of the adsorbed O2 oc-
curs when there is enough backdonation from the
cluster into the antibonding pÃg state or a sucient
donation from the pu or rg into the cluster [69].
Both processes reduce the bond order of oxygen. If
we apply this simple argumentation to the obser-
vations of the reactivity of deposited Ptn clusters,
the energy of the HOMO for di€erent cluster sizes,
associated with the energy of the center of the
d-band, can be tuned from 9:0 eV, the ionization
potential (IP) of the atom, to 5.32 eV, the Fermi
energy of bulk platinum. Now, the increase in the
catalytic activity from Pt8 to Pt20 can be correlated
with the decrease of the IP and the change in the
position of the center of d-band with increasing
cluster size and the concomitant enhanced reso-
nance with the antibonding pÃg state of O2. The
maximum reactivity for Pt15 [27] suggests that the
resulting backdonation is largest for this cluster.
Increasing the cluster size further, lowers the
clusterÕs HOMO even more and may result in a
weaker resonance with the antibonding pÃg state of
O2, as the HOMOÕs energy passes the energy of
this antibonding state. For the very small clusters,
this backdonation is small as the energy mismatch
of clusterÕs HOMO and pÃg is large.
1
cross section. At higher temperature, however,
the decrease is in concert with the oxidation of CO,
where between 350 and 550 K desorption of CO
contributes to the changing signal. Very interest-
ingly, Ptn, Pdn, and Rhn are more prone to poi-
soning by carbon monoxide than Aun clusters. In
fact, only gold clusters revealed an unchanged CO
combustion when ®rst exposing to CO and then to
O2 in these one-heating-cycle experiments. This is
related to the bond strength of CO to the clusters.
The stronger the interaction with CO, the more
likely the reactive sites are blocked as observed
for Ptn, Pdn, and Rhn. That this interaction is in-
deed the weakest for gold is shown by infrared
spectroscopy wherein the highest absorption fre-
quency of CO is observed for Au8. The low ad-
sorption energy of CO on gold is also con®rmed
by TPD experiments, which show that CO de-
sorbs already at about 180 K in comparison to,
e.g., 450 K for Pt8.
In many cases, the dissociation of oxygen con-
stitutes the ®rst step in an oxidation reaction
[58,60,64]. Thus, in order to understand to a ®rst
approximation, the size-dependent reactivity of
each element, the knowledge of the evolution of
the intrinsic cluster properties with size and its
in¯uence on the molecular or dissociative adsorp-
tion of oxygen on the clusterÕs surface are crucial.
To ®rst order, the interaction of oxygen with
surfaces can be rationalized within the Newns±
Anderson model [65,66]. In this model, the inter-
action of an adsorbate species with a small group
of atoms of an extended surface is described within
bonding and anti-bonding levels, reminiscent of
the simple valence theory of diatomic molecules.
The discrete adsorbate states couple with a con-
1
IR in re¯ection mode is only sensitive to dynamical dipole
moments perpendicular to the surface. Therefore using inte-
grated absorption intensities to estimate CO concentrations on
the surface have to be taken with care.

U. Heiz et al. / Chemical Physics 262 (2000) 189±200
197
In this simple model, the interaction with the
substrate is completely neglected. As shown by
theoretical investigations, one of the major roles of
the cluster±surface interaction is the charging of
the clusters [40]. Therefore for cluster catalysts,
where the reactivity is similar for di€erent cluster
charge states, this simpli®cation may be valid to a
®rst approximation for rationalizing trends in the
size-dependent reactivities at least when structural
changes of the clusters are small upon deposition.
Gas phase studies indeed indicate that the charge-
state of the clusters does not often alter the general
trends in the size-dependent chemistry, if the
clusters are composed of transition metals with
partially ®lled d-states [70]. On the other hand, we
show below that this interaction completely de-
termines the reactivity of small gold clusters in the
way that charge ¯ows from the substrate to the
cluster [40]. This is in line with gas phase studies
where it was shown that only negatively charged
clusters are reactive towards O2 adsorption.
and the local morphology, which are often used for
solid surfaces, not seem to be sucient to explain
these di€erences.
The observed chemical properties are, thus,
mainly in¯uenced by the varying electronic struc-
tures of the four molecular clusters. Calculations
have shown that their electronic structure can be
viewed as a superposition of atomic d-states and
an electronic shell structure, characterized by the
number of atomic s electrons in the cluster [51]
(Fig. 8). For the nobel metal cluster Au13 each
atom contributes one delocalized valence electron
and a closed d-shell. In this case, the HOMO is
characterized by a discrete electronic shell state
(Fig. 8). For Pd13 and Pt13 with an atomic 4d10 and
5d96s1 con®guration, respectively, the character of
the electronic states around the Fermi level dras-
tically changes (Fig. 8). For these clusters, the
electronic states at the Fermi level are character-
ized by the superposition of atomic d-states. For
Pd13, an additional unoccupied shell-2pà state is
situated just at the Fermi level [51]. This di€erence
in the electronic structure is re¯ected in the
chemical behavior of Au13, Pd13, and Pt13 in the
following manner. In order to oxidize CO mole-
cules eciently at high temperature, several oxy-
gen molecules have to be activated suciently
upon adsorption. This occurs when there is en-
ough hybridization between the antibonding pgÃ
4.2. CO-combustion on the tridecamers: in¯uence of
the geometric and electronic structure
A comparison of the reactivities of the 13-atom
clusters is elucidating as they are considered to
have geometrically closed shell structures. For Pt13,
2
Pd13 [51], and Au13 [50], the structures with cub-
octahedral symmetry are predicted to be slightly
more stable than the icosahedral structures. In
contrast, for Rh13, an icosahedral ground-state
structure is calculated [71]. The average coordina-
tion number for the icosahedral structure is 6.5,
whereas the cuboctahedral structure reveals a co-
ordination number of 5.5 [50]. 2 This suggests the
low-coordinated clusters (Au13, Pt13, and Pd13) to
be more reactive when assuming that the structures
of the clusters are only slightly perturbed by the
interaction with the MgO(1 0 0) ®lm [40]. Such a
trend, however, is not supported by the presented
results. Therefore, purely geometrical arguments
2
Very recent DFT-GGA calculations [39] indicate, however,
that both neutral and anionic Au13 clusters prefer strongly
deformed low-symmetry structures with an average coordina-
tion number per atom of about 5.1.
Fig. 8. The schematic energy level diagram for Pd13, Pt13, and
Au13 [51].

198
U. Heiz et al. / Chemical Physics 262 (2000) 189±200
state of oxygen and the clusterÕs electronic states at
the Fermi level. Thus, the electronic states of oxy-
gen hybridize more eciently with the high den-
sity of d-states of palladium and platinum clusters.
For Au13, the bonding between the discrete shell
states and the molecular electronic states of oxy-
gen is, however, less likely. In fact, ab initio cal-
culations showed that the HOMO and LUMO of
Au13 are energetically below the antibonding state
of oxygen; and therefore, no charge transfer from
the cluster to the oxygen molecule is possible [72].
proposed. In a ®rst direct Eley±Rideal reaction
mechanism, a CO molecule brought from the gas
phase to the vicinity of the peroxo molecule on the
top-facet of the gold cluster reacts spontaneously
forming a CO2 molecule weakly bound to the
catalyst (ꢁ0.2 eV). Another reaction mechanism,
which contributes to the low-temperature oxida-
tion of CO is of the Langmuir±Hinshelwood type,
where the two reactants are initially coadsorbed on
the top-facet of the Au8 cluster. A weakly ad-
sorbed CO2 molecule is formed after overcoming a
small activation barrier of 0.1 eV. The oxidation of
CO at 240 K is attributed to a reaction mechanism,
where the peroxo OÃ2 molecule bonded to the pe-
riphery of the interfacial layer of the cluster is in-
volved. In the case where the gold octamer is
bound to an FC, an activation energy of ꢁ0.5 eV is
found. Interestingly, this activation barrier signif-
icantly increases (ꢁ0.8 eV) when the cluster is
bound to defect-free magnesia, which explains the
experimental observation that the octamer is not
reactive when deposited on defect-free magnesia. It
is interesting to note that the reactivity of Pd8 is
not suppressed upon deposition on defect-poor
MgO(1 0 0) ®lms, but only decreases at high tem-
perature (500 K). These di€erences may again be
explained by the di€erent electronic structure of
the two octamers. Charge transfer to the Pd8 may
have less e€ect, as the density of states around the
Fermi level is characterized by the superposition of
atomic d-states, in addition Pd clusters are already
reactive as neutral species; therefore, subtle
changes in their electronic structure upon deposi-
tion are less important than they are for Au8.
4.3. Cluster-support interaction: the role of defects
The gold octamer is only reactive when depos-
ited on defect-rich MgO(1 0 0) ®lms. These defects
are characterized experimentally and theoretically
by the adsorption properties of NO and are iden-
ti®ed with oxygen vacancies, F-centers (FC)
[40,72]. The role of these defects is threefold [40]:
®rst, the enhanced binding (5.56 eV) of Au8 clus-
ters to the MgO(FC) surface anchors them and
thus, reduces their propensity for agglomeration
and sintering, second, a charge transfer of 0.5 e to
the gold cluster enhances the adsorption of oxygen
and third the activation energy for forming CO2
on the cluster is reduced. This ®nding is in accor-
dance with early experiments of molecular oxygen
adsorption on free gas-phase clusters showing that
only even numbered anionic clusters react with
gaseous O2 molecules. The reaction mechanisms,
occurring at 140 and at 240 K have been eluci-
dated by extensive ab initio simulations [40]. First,
these calculations describe the adsorbed Au8
cluster as a deformed piece of a hexagonal closed-
packed crystal with stacked ®ve (bottom)- and
three (top)-atom layers. The deposited gold oct-
amer adsorbs both O2 and CO molecules. In the
optimized adsorption con®guration, CO binds on
top of a Au atom of the upper triangular facet of
the deposited Au8. On the other hand, the O2
molecule readily adsorbs at several sites, both on
the upper Au8 facet and at the interface between
the adsorbed gold cluster and the underlying
magnesia surface. In both cases, the adsorbed oxy-
gen molecule is found to be activated to a per-
oxo OÃ2 molecular state. For the oxidation of CO
at 140 K, two possible reaction mechanisms were
5. Conclusions
We have shown that the oxidation of CO is
dependent on the size of small gold, platinum,
palladium and rhodium clusters and on their dis-
tinct interaction with the oxide support. For gold
clusters, it was found that the combustion of CO
at low temperature (ꢁ240 K) on Au8 is activated
after deposition on defect sites of the MgO sub-
strate. Interestingly, for Pd8 the combustion of CO
is not suppressed when adsorbed on defect-poor
®lms. The geometrically closed shell clusters Pt13,

U. Heiz et al. / Chemical Physics 262 (2000) 189±200
199
[11] C.R. Henry, Surf. Sci. Rep. 31 (1998) 231±326.
[12] H. Poppa, Catal. Rev. Sci. Engng. 35 (1993) 359±398.
Pd13, and Rh13 as well as Au13 all show a di€erent
reactivity, suggesting that their electronic structure
mainly govern their reactivity. Au13 shows a low
density of states around the Fermi energy and a
coupling with the antibonding state of oxygen is
prevented. On the other hand for Pt13 and Pd13,
the atomic d-states overlap the Fermi-level, re-
sulting in a high density of states, and therefore
these clusters become more reactive. In general,
the evolution of the reactivities with cluster size
correlate with the position of the atomic d-states
with respect to the antibonding pÃg O2 state. These
examples clearly show that the chemical properties
of small metal cluster can be tuned as function
of size and of the distinct cluster-support interac-
tion.
[13] H.J. Freund, Angew. Chem. Int. Ed. Engl. 36 (1997) 452±
475.
[14] S. Dahl, A. Logadottir, R.C. Egeberg, J.H. Larsen,
I. Chorkendor€, E. Toernqvist, J.K. Nùrskov, Phys. Rev.
Lett. 83 (1999) 1814±1817.
[15] E. Bertel, P. Roos, J. Lehmann, Phys. Rev. B 52 (1995)
R14 384±R14 387.

[16] C. Kittel, in: R. Oldenbourg (Ed.), Einfuhrung in die

Festkorperphysik, vol. 7. Verlag, Muenchen, 1988.
[17] L. Piccolo, C. Becker, C.R. Henry, J. E. Phys. D. 9 (1999)
415±419.
[18] M. Frank, S. Andersson, J. Libuda, S. Stempel, A. Sandell,


B. Brena, A. Giertz, P.A. Bruhwiler, M. Baumer, N.
Martensson, H.-J. Freund, Chem. Phys. Lett. 279 (1997)
92±99.
[19] D.W. Goodman, P.M. Holmblad, D.R. Rainer, J. Phys.
Chem. B 101 (1997) 8883±8886.
[20] W.D. Knight, K. Clemenger, W. de Heer, W.A. Saunders,
M.Y. Chou, M.L. Cohen, Phys. Rev. Lett. 52 (1984) 2141.
[21] R.L. Whetten, D.M. Cox, D.J. Trevor, A. Kaldor, Phys.
Rev. Lett. 54 (1985) 1494.
Acknowledgements
^
[22] I.M.L. Billas, A. Chatelain, W.A. deHeer, Science 265
(1994) 1682.
[23] U. Heiz, A. Vayloyan, E. Schumacher, C. Yeretzian, M.

We thank H. Hakkinen and U. Landman for
stimulating discussions, critical reading of the
manuscript and for communicating unpublished
results. This work was supported by the Swiss
National Science Foundation. U.H. thanks the
Alexander v. Humboldt Stiftung for a fellowship.

Stener, P. Gisdakis, N. Rosch, J. Chem. Phys. 105 (1996)
5574±5585.
[24] A.P. Alivisatos, Science 271 (1996) 933±937.

[25] T. Leisner, C. Rosche, S. Wolf, F. Granzer, L. Woste, Surf.
Rev. Lett. 3 (1996) 1105±1108.
[26] U. Heiz, F. Vanolli, A. Sanchez, W.-D. Schneider, J. Am.
Chem. Soc. 120 (1998) 9668±9671.
[27] U. Heiz, A. Sanchez, S. Abbet, W.-D. Schneider, J. Am.
Chem. Soc. 121 (1999) 3214±3217.
References
[28] M.M. Kappes, R.W. Kunz, E. Schumacher, Chem. Phys.
Lett. 91 (1982) 413±418.
[1] M. Boudard, Adv. Catal. 20 (1969) 153.
[2] W.T. Tysoe, G.L. Nyberg, R.M. Lambert, J. Chem. Soc.
Chem. Commun. (1983) 623±625.
[29] W.A. deHeer, Rev. Mod. Phys. 65 (1993) 611±676.
[30] W. Ekardt, Phys. Rev. B 29 (1984) 1558±1564.
[31] K. Clemenger, Phys. Rev. B 32 (1985) 1359±1362.
[32] U. Heiz, W.-D. Schneider, Physical chemistry of supported
clusters, in: K.-H. Meiwes-Broer (Ed.), Metal Clusters at
Surfaces ± Structure, Quantum Properties, Physical Chem-
istry, Springer, Berlin, 2000, pp. 237±273.
[3] W.S. Sesselmann, B. Woratschek, G. Ertl, J. Kuppers,
H. Haberland, Surf. Sci. 130 (1983) 245.
[4] P.M. Holmblad, D.R. Rainer, D.W. Goodman, J. Phys.
Chem. B 101 (1997) 8883±8886.
[5] T.M. Gentle, E.L. Muetterties, J. Phys. Chem. 87 (1983)
2469±2472.
[33] A. Berces, P.A. Hackett, L. Lian, S.A. Mitchell, D.M.
Rayner, J. Chem. Phys. 108 (1998) 5476±5490.
[34] M.D. Morse, M.E. Geusic, J.R. Heath, R.E. Smalley,
J. Chem. Phys. 83 (1985) 2293.
[6] T.G. Rucker, M.A. Logan, T.M. Gentle, E.L. Muetterties,
G.A. Somorjai, J. Phys. Chem. 90 (1986) 2703±2708.
[7] H. Ho€mann, F. Zaera, R.M. Ormerod, R.M. Lambert,
J.M. Yao, D.K. Saldin, L.P. Wang, D.W. Bennett, W.T.
Tysoe, Surf. Sci. 268 (1992) 1±10.
[35] M.R. Zakin, R.O. Brickman, D.M. Cox, A. Kaldor,
J. Chem. Phys. 88 (1988) 3555.

[8] R.M. Ormerod, R.M. Lambert, J. Phys. Chem. 96 (1992)
8111±8116.
[36] H. Kietzmann, J. Morenzin, P.S. Bechthold, G. Gantefor,
W. Eberhardt, J. Chem. Phys. 109 (1998) 2275±2278.
[37] Y. Shi, K.M. Ervin, J. Chem. Phys. 108 (1998) 1757±1760.
[38] J. Hagen, L. Souciou, T. Bernhard, T. Leisner, U. Heiz,
[9] G. Pacchioni, R.M. Lambert, Surf. Sci. 304 (1994) 208±
222.
[10] S. Abbet, A. Sanchez, U. Heiz, W.-D. Schneider, A.M.

L. Woste, submitted for publication.

Ferrari, G. Pacchioni, N. Rosch, Surf. Sci. 454±456 (2000)
984±989.

[39] H. Hakkinen, U. Landman, Phys. Rev. B 62 (2000)
R2287±R2290.

200
U. Heiz et al. / Chemical Physics 262 (2000) 189±200
[40] A. Sanchez, S. Abbet, U. Heiz, W.-D. Schneider, H.
[56] M.C. Wu, J.S. Corneille, C.A. Estrada, J.-W. He, D.W.
Goodman, Chem. Phys. Lett. 182 (5) (1991) 472.
[57] U. Heiz, W.-D. Schneider, J. Phys. D Appl. Phys. 33
(2000) R85±R102.

Hakkinen, R.N. Barnett, U. Landman, J. Phys. Chem. A
103 (1999) 9573±9578.

[41] H. Hakkinen, R.N. Barnett, U. Landman, Phys. Rev. Lett.
82 (1999) 3264.
[58] J. Xu, J.T. Yates, J. Chem. Phys. 99 (1993) 725±732.

[42] U. Heiz, F. Vanolli, L. Trento, W.-D. Schneider, Rev. Sci.
Instrum. 68 (1997) 1986.
[59] K.-H. Allers, H. Pfnur, P. Feulner, D. Menzel, J. Chem.
Phys. 100 (1994) 3985±3998.

[43] I. Yudanov, G. Pacchioni, K. Neyman, N. Rosch, J. Phys.
[60] A. Szabo, M.A. Henderson, J.T. Yates, J. Chem. Phys. 96
(1992) 6191±6201.
Chem. 101 (1997) 2786±2792.
[44] I.V. Yudanov, S. Vent, K. Neyman, G. Pacchioni, N.
[61] A. Rar, T. Matsushima, Surf. Sci. 318 (1994) 89±96.
[62] F. Zaera, E. Kollin, J.L. Gland, Chem. Phys. Lett. 121 (4/
5) (1985) 464.

Rosch, Chem. Phys. Lett. 275 (1997) 245±252.
[45] K. Bromann, C. Felix, H. Brune, W. Harbich, R. Monot,
J. Buttet, K. Kern, Science 274 (1996) 956±958.
[46] H.-P. Cheng, U. Landman, J. Phys. Chem. 98 (1994)
3527.
[63] J.-W. He, A.E. Cesar, J.S. Corneille, M.-C. Wu, D.W.
Goodman, Surf. Sci. 261 (1992) 167±170.
[64] J.A. Rodriguez, D.W. Goodman, Surf. Sci. Rep. 14 (1991)
1.
[47] W. Harbich, in: K.-H. Meiwes-Broer (Ed.), Collision of
Clusters with Surfaces: Deposition, Surface Modi®cation
and Scattering, Springer, Heidelberg, 1999.
[48] G. Valerio, H. Toulhoat, J. Phys. Chem. 100 (1996) 10828±
10830.
[65] D.M. Newns, Phys. Rev. 178 (1969) 1123.
[66] P.W. Anderson, Phys. Rev. 124 (1961) 41.
[67] A.W.E. Chen, R. Ho€mann, W. Ho, Langmuir 8 (1992)
1111±1119.

[49] S. Krueger, S. Vent, N. Rosch, Ber. Bunsenges. Phys.
[68] B. Hammer, O.H. Nielsen, J.K. Nùrskov, Catal. Lett. 46
(1997) 31±35.
Chem. 101 (1997) 1640.
[50] O.D. Haberlen, S.C. Chung, M. Stener, N. Rosch,
J. Chem. Phys. 106 (1997) 5189±5201.

[69] R. Ho€man, Solids and Surfaces: A Chemist's View of
Bonding in Extended Structures, VCH Verlagsgesellschaft,
GmbH: Weinheim, 1988.
[51] N. Watari, S. Ohnishi, Phys. Rev. B 58 (1998) 1665±1677.
[52] S.H. Yang, D.A. Drabold, J.B. Adams, P. Ordejon,
K. Glassford, J. Phys.: Cond. Matter 9 (1997) L39±L45.
[53] A. Sachdev, R.I. Masrl, J.B. Adams, Z. Phys. D. Atoms
Mol. Clusters 26 (1993) 310±312.
[70] M.R. Zakin, R.O. Brickman, D.M. Cox, A. Kaldor,
J. Chem. Phys. 88 (1988) 5943.
[71] B.V. Reddy, S.K. Nayak, S.N. Khanna, B.K. Rao, P. Jena,
Phys. Rev. B 59 (1999) 5214±5222.

[54] U. Heiz, Appl. Phys. A 67 (1998) 621±626.
[55] M.-H. Scha€ner, F. Patthey, W.-D. Schneider, Surf. Sci.
417 (1998) 159.
[72] H. Hakkinen, R.N. Barnett, U. Landman, A. Sanchez, S.
Abbet, U. Heiz, W.-D. Schneider, submitted for publica-
tion.