Oscillatory CO oxidation on Pd(1 0 0) studied with in situ scanning tunneling microscopy

Article abstract of DOI:10.1016/j.susc.2004.01.025

We have studied the catalytic oxidation of CO on Pd(100) at atmospheric pressures, using a combined flow reactor and scanning tunneling microscope (STM). By switching from a CO-rich flow to an O₂-rich flow, and vice versa, we could reversibly oxidize and reduce the surface, as we have observed with the STM. Both during the initial oxidation and during the catalytic reaction on the oxide, the surface became rough. Coinciding with the formation of the palladium oxide there was a step up in the CO oxidation rate. Whereas the reaction rate on the metallic surface followed traditional Langmuir-Hinshelwood kinetics, the reaction rate on the oxide was proportional to the CO-pressure and did not depend on the O₂-pressure. We suggest that the high reaction rate on the oxide resulted from the low stability of the oxide, making it highly reactive, and that the reaction followed the Mars-Van Krevelen mechanism, in which the oxide is continually 'consumed' and reformed. The switching between the metallic surface and the oxide showed significant hysteresis in the CO-pressure for O₂-rich conditions. This resulted in a CO-pressure window where reaction was stable both on the metal and on the oxide. We have observed that the reaction switched spontaneously between these branches, leading to oscillations in the reaction rate. Our interpretation is in strong disagreement with well-established models for CO oxidation on Pt-group metal surfaces.

Full text of DOI:10.1016/j.susc.2004.01.025

Surface Science 552 (2004) 229–242
www.elsevier.com/locate/susc
Oscillatory CO oxidation on Pd(1 0 0) studied with
in situ scanning tunneling microscopy
B.L.M. Hendriksen, S.C. Bobaru, J.W.M. Frenken *
Kamerlingh Onnes Laboratory, Leiden University, P.O. Box 9504, 2300 RA Leiden, The Netherlands
Received 5 November 2003; accepted for publication 16 January 2004
Abstract
We have studied the catalytic oxidation of CO on Pd(1 0 0) at atmospheric pressures, using a combined flow reactor
and scanning tunneling microscope (STM). By switching from a CO-rich flow to an O -rich flow, and vice versa, we
2
could reversibly oxidize and reduce the surface, as we have observed with the STM. Both during the initial oxidation
and during the catalytic reaction on the oxide, the surface became rough. Coinciding with the formation of the pal-
ladium oxide there was a step up in the CO oxidation rate. Whereas the reaction rate on the metallic surface followed
traditional Langmuir–Hinshelwood kinetics, the reaction rate on the oxide was proportional to the CO-pressure and
2
did not depend on the O -pressure. We suggest that the high reaction rate on the oxide resulted from the low stability of
the oxide, making it highly reactive, and that the reaction followed the Mars–Van Krevelen mechanism, in which the
oxide is continually ÔconsumedÕ and reformed. The switching between the metallic surface and the oxide showed sig-
nificant hysteresis in the CO-pressure for O -rich conditions. This resulted in a CO-pressure window where reaction was
2
stable both on the metal and on the oxide. We have observed that the reaction switched spontaneously between these
branches, leading to oscillations in the reaction rate. Our interpretation is in strong disagreement with well-established
models for CO oxidation on Pt-group metal surfaces.
ꢀ
2004 Elsevier B.V. All rights reserved.
Keywords: Models of surface chemical reactions; Scanning tunneling microscopy; Catalysis; Oxidation; Surface structure, morphology,
roughness, and topography; Palladium; Carbon monoxide; Oxygen
1
. Introduction
tions (i.e. oxygen-rich conditions) [1] has been
attributed to the fact that next to metallic Pd, also
PdO is an active oxidation catalyst. Datye et al.
have studied the effect of the transformation of Pd
Palladium is a widely used catalyst for oxida-
tion of CO and volatile organic compounds and
for the reduction of NO in automotive catalysis.
The high catalytic activity under fuel-lean condi-
2 3
to PdO on Al O -supported Pd catalysts on the
reactivity in methane oxidation [2,3]. Still, it is
unknown whether Pd or PdO is the active phase in
the oxidation of hydrocarbons [4,5]. For CO oxi-
dation on ruthenium based model catalysts it has
been shown that, whereas a metallic Ru surface
is not active for this reaction, ruthenium oxide
has superior catalytic activity [6]. The question
*
Corresponding author. Tel.: +31-71-527-5603; fax: +31-71-
5
27-5404.
E-mail address: frenken@phys.leidenuniv.nl (J.W.M. Fren-
ken).
URL: http://www.physics.leidenuniv.nl/sections/cm/ip.
0
039-6028/$ - see front matter ꢀ 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.susc.2004.01.025

230
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
whether the metal or the oxide is the more cata-
lytically active phase exists also for CO oxidation
on Pd, especially in relation to the kinetic oscilla-
tions observed for this reaction. During these ki-
netic oscillations of the reaction rate the catalyst
switches spontaneously between a high- and a low-
reactivity state. The oscillatory behavior has been
studied for various forms of Pt-group catalysts,
e.g. single crystals, polycrystalline foils, wires, and
supported catalysts, over a broad range of condi-
tions, both in ultrahigh vacuum (UHV) and at
atmospheric pressures. Under UHV conditions the
oscillations on Pt(1 0 0) and Pt(1 1 0) have been
ascribed to structural changes like surface recon-
structions and faceting. Since clean Pd(1 1 0) does
not exhibit a surface reconstruction, it has been
suggested that on this surface oscillatory behavior
at low pressures involves the formation of sub-
surface oxygen [7], and this has been modeled
accordingly [8,9]. On the other hand, an oxygen-
induced reconstruction of the Pd(1 1 0) surface has
been observed [10]. Bondzie et al. have questioned
the subsurface-oxygen models, which attribute the
oscillations to the periodic formation and deple-
tion of subsurface oxygen [11], and they have
shown that PdO formation was possible at con-
ditions for oscillatory reactions [12]. They have
suggested that oscillatory behavior on Pd(1 1 0)
may be due to an oxidation and reduction mech-
anism. It was found that the oxide species were
weakly bound or unstable and their decomposition
appeared to be very fast rather than slow, as is
usually assumed because of the time scales of the
oscillations, which can have periods of several
minutes [8,9]. The reaction kinetics for the for-
mation and reduction of the PdO showed these
processes to be weakly activated, indicating a
spontaneous reaction [13]. Turner et al. have
investigated oscillatory CO oxidation on poly-
crystalline wires of Pt, Pd and Ir at atmospheric
pressures [14,15]. They have suggested a general
mechanism for the oscillations based on the for-
mation of an inactive oxide in the near-surface
region and the reactivation of the catalyst via
reduction of the oxide by CO. The oxidation and
reduction were proposed to make the reaction
system switch between two branches of Langmuir–
Hinshelwood kinetics, where the reaction rate de-
pends directly on the surface coverage of the
reactant molecules or atoms and therefore on the
competition in adsorption of these gases [16] (see
also Section 4.4), namely a mainly CO-covered
surface with relatively low reaction rate and an
mainly oxygen-covered surface with a higher rate.
This has also been proposed for dispersed palla-
dium embedded within a zeolite matrix under non-
isothermal conditions [17].
The oxidation of Pd(1 1 1) has been studied by
Zheng and Altman using in situ STM at relatively
ꢀ
8
low pressures (10 mbar) [18]. Their results indi-
cate that the oxidation proceeds through three
stages, involving four distinct states of the surface,
the ejection of Pd atoms from the surface, and
eventually the formation of a bulkoxide with
no crystalline order. Lundgren et al. have stud-
ied the structure of the two-dimensional oxide on
Pd(1 1 1) to greater detail and found an incom-
mensurate surface oxide, which has no resem-
blance with bulkoxides of PdO and can be
considered as an intermediate phase between an
oxygen overlayer and a bulkoxide [19]. The more
open Pd(1 0 0) surface is far more reactive towards
2
O than Pd(1 1 1). The oxidation of Pd(1 0 0) pro-
ceeds through four stages involving up to five
distinct oxygen structures [20]. The first stage is
dissociative adsorption, which results in a chemi-
sorbed oxygen overlayer, involving three different
structures. In the second stage, Pd atoms are
ejected onto the terraces and either nucleate is-
lands or attach to preexisting steps. These ejected
Pd atoms are thought to be displaced by oxygen
filling subsurface sites. When all subsurface sites
are filled the island growth stops and in the third
stage the surface reconstructs to either a (5 · 5)
structure for temperatures below 400 K, or a
pffiffi pffiffi
( 5 Â 5)R27ꢁ structure for temperatures above
400 K, both structures having been attributed to
PdO surfaces. In the fourth stage, three-dimen-
sional clusters appear on the surface, which has
been ascribed to bulkoxidation, and the surface
roughens. It has been shown by Todorova et al.
pffi ffiffi pffiffi
that the ( 5 Â 5)R27ꢁ structure was actually not
a rumpled PdO(0 0 1) film [21], but a strained
PdO(1 0 1) layer on top of Pd(1 0 0) [22]. In a sep-
arate study Zheng and Altman concluded that at
low pressures the oxygen overlayer structures had

B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
231
pffiffi
the highest reactivity towards CO. The ( 5 Â
pffiffi
active to reduction with CO [23]. Recently, the
Pt(1 1 1) [29]. Post-reaction analysis therefore may
not be representative for the in situ structure and
was not attempted.
5
)R27ꢁ and the bulkPdO were relatively unre-
pffi ffiffi pffiffi
formation of a thin ( 5 Â 5)R27ꢁ surface oxide
and the growth of a rough, poorly ordered bulk
oxide film of PdO has been observed with high-
pressure surface X-ray diffraction (SXRD) [24].
In Ref. [25] we concluded for Pt(1 1 0) that the
switching in reactivity for CO oxidation at atmo-
spheric pressures, when changing from a CO-rich
to an oxygen-rich flow and vice versa, is due to the
oxidation and reduction of the Pt(1 1 0) surface,
the platinum oxide having the higher reactivity.
The area of the part of the Pd(1 0 0) surface on
which the catalytic reaction tookplace, was 0.4
2
cm . The Pd(1 0 0) single crystal surface was
cleaned in UHV by cycles of ion-bombardment
þ
using 600 eV Ar at 300 K, followed by short
annealing at a temperature of 1100 K. This re-
sulted in a clean surface with the (1 · 1) structure,
as was observed using low-energy electron dif-
fraction (LEED). Additional UHV preparation of
the Pd(1 0 0) surface between the high-pressure
experiments was not required since exposing the
surface to a high pressure of CO at temperatures
around 425 K effectively removed all observable
contamination and roughness.
We have used a cut Pt/Ir alloy tip for STM
imaging and we have routinely checked that the tip
did not influence either the surface structure or the
reaction kinetics, e.g. by imaging parts of the
surface, which had not been scanned previously,
and by measuring the reaction kinetics without
STM imaging. By performing experiments on
various sample materials, e.g. Pt and Ru, we have
made sure that the results presented here were
characteristic for Pd(1 0 0).
The STM observations strongly suggest that CO
2
is formed by reaction of CO molecules with oxy-
gen atoms from the platinum oxide following the
Mars–Van Krevelen mechanism [27]. The bista-
bility in the reaction rate showed striking similarity
to the reaction kinetics on polycrystalline Pt and
Pd wires [14,15]. Based on this similarity in mac-
roscopic behavior, we expect comparable results
for the Pd(1 0 0) surface, which is considered to be
easier to oxidize than Pt(1 1 0).
2
. Experimental
We have used a combined micro-flow reactor
and STM setup, which is described in detail in Ref.
28], to study the oxidation of CO on a Pd(1 0 0)
Most experiments were performed at a total
pressure of 1.25 bar. Argon gas was used as an
inert gas to dilute the CO-flow and vary the CO-
pressure, while keeping the oxygen partial pressure
and the total pressure constant. We have used a
quadrupole mass spectrometer (QMS) to simulta-
neously monitor a variety of gases leaking from
the reactor into the UHV system. For reasons of
[
surface. Briefly, the STM is combined with the
small volume micro-flow reactor (volume ¼ 0.5
ml), in such a way that only the tip of the STM is
inside the reactor. The other components of the
STM, e.g. the scan piezo element, are separated
from the reactor volume by a small, flexible viton
O-ring, which provides a gas tight seal, but is
flexible enough to allow the scanning motion. The
reactor-STM combination is integrated in an
ultrahigh vacuum setup, which allows us to use
standard surface science tools for preparation and
analysis before high-pressure experiments in the
reactor-STM, without first exposing the sample to
air. We should note that the surface structure and
coverage can change drastically during cooling
down and when the reactor is evacuated after a
high-pressure experiment and transferred to the
UHV section of our setup, as we have observed for
2 2
clarity only the O , the CO and the CO pressures
are reproduced here. Since we kept the total
pressure constant, the absolute pressures PO2 and
PCO could be calibrated using a pure O -flow and a
2
pure CO-flow. The CO pressure was calibrated
indirectly by measuring the variations in the CO-
pressure during the reaction. Next to the reactant
2
2
gases O and CO, we detected small traces of
impurities in the gas flow (estimated value <100
ppm) with masses corresponding to water and
nitrogen, but we have found no indication that
these qualitatively affected the results presented
here.

232
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
3
. Results
t ¼ 0 s the CO-pressure was slowly decreased and
the CO signal decreased accordingly, until at
2
The results of a typical experiment are sum-
t ¼ 1612 s there was a sudden step up in the CO
2
marized in Fig. 1. In the lower panel of Fig. 1 a
selection of STM images from an STM movie [40]
is displayed, showing the surface structure during
the catalytic oxidation of CO on Pd(1 0 0). The
upper panel depicts the partial pressures of the
pressure (indicated by the first arrow). The step in
the CO -pressure occurred during the recording of
image (B), which was scanned from bottom to top
2
in 43.3 s. The dashed line indicates where the
reaction rate switched. Simultaneously with the
switch in the CO production the formation of a
reactant gases CO and O
product CO , as measured by the mass spectro-
2
, and the reaction
2
2
large number of small islands or clusters was ob-
meter simultaneously with the STM movie. The
measured Ar-signal is not shown. The experiment
served in Fig. 1(B). These islands had heights equal
to the monatomic step height of Pd(1 0 0). After
this, substantial roughness developed on the sur-
2
shown here started in a flow of CO and O at a
total pressure of 1.25 bar and 408 K. Image (A)
shows the Pd(1 0 0) surface, consisting of flat ter-
races separated by monatomic steps. Starting at
face (images (C)–(E)) and the CO pressure de-
2
creased proportional to the CO-pressure. The
initially small clusters developed to a polycrystal-
2 2
Fig. 1. (Upper panel) Mass spectrometer signals of the reactant gases O and CO, and the reaction product CO . These partial
pressures were measured in the UHV-system via a small leakdirectly from the reactor. The labels (A)–(H) refer to the STM images in
the lower panel, which were recorded simultaneously. The arrows markwhere the reaction rate switched from Rlow to Rhigh and vice
versa. (Lower panel) STM images (140 nm · 140 nm) from an STM movie [40] of the Pd(1 0 0) surface at a temperature of 408 K in a
n 2
5.0 ml /min gas flow mixture of CO, O and Ar (not-depicted) at a total pressure of 1.25 bar. Images (A) and (B) (the part of (B) below
the dashed line) show the metallic surface consisting of flat terraces separated by monatomic steps. Images (B) (part above the dashed
line), (C), (D) and (E) show the evolution of the oxidic surface. Images (F), (G) and (H) again show the metallic surface. After the
removal of the oxide a number of Pd adatom and vacancy islands were present. They decayed with time and the flat Pd(1 0 0) surface
was restored again. The grey-scales correspond to height ranges of: (A)–(D) 1.28 nm; (E) 1.65 nm; (F)–(G) 1.0 nm (V
¼ 0:2 nA).
t
¼ 112 mV,
I
t

B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
233
line, granular structure, as can be observed in
image (C). Images (D) and (E) show that the
imaging quality decreased. When the CO-pressure
was increased again at t ¼ 4700 s (indicated by the
second arrow) the roughness was removed several
Pd adatom and vacancy islands were left behind at
the surface. These islands decayed fast to restore
the flat surface, as can be observed in images (F)–
stage as the metallic Pd(1 0 0) surface. There was a
modest CO production, indicated by the partial
pressure of CO of 7 mbar, which suggests that
both oxygen and CO were present in significant
quantities on the surface in image (A). The oxygen
pressure was significantly higher than the CO-
pressure. The surface must have been covered
mainly with oxygen, as is reflected by the decrease
2
2
(
H). When we increased the CO-pressure (at
t ¼ 4700 s) there was a small peakin the CO
pressure, after which the CO -signal slowly in-
creased again, in a weakresponse to the CO-
2
in the CO -pressure when we decreased the CO-
2
pressure by diluting the CO flow with argon
(Section 4.4).When the CO-pressure was reduced,
no structural change was observed on the surface
until t ¼ 1612 s.
2
pressure. At t ¼ 6700 s we stopped the oxygen flow
and increased the CO-flow. The oxygen pressure
slowly dropped and, in switching from the O -rich
4.2. The oxide surface (B)–(E)
2
to the CO-rich flow, the CO production showed a
2
maximum at t ¼ 7010 s. After that, the CO
2
-
At t ¼ 1612 s there was an instantaneous step up
pressure dropped to its base level around t ¼ 9000
2
in the CO production rate by a factor of 1.5. This
s. Images (G) and (H) show no essential structure
difference in switching from a high pressure of O
step in reactivity occurred when the decreasing CO-
pressure dropped below a value of 50 mbar.
Coinciding with the change in reactivity, small is-
lands or clusters appeared on the Pd(1 0 0) surface
as can be seen in image (B) in the part above the
dashed line. These islands had a height equal to the
monatomic step height of Pd(1 0 0) (dstep ¼ 0:195
nm). Such islands were also formed when we ex-
posed the Pd(1 0 0) surface to a high pressure of
oxygen only, as can be seen in Fig. 2. A few large
adatom islands were present initially (Fig. 2(a)),
resulting from a previous oxidation–reduction
2
to a high pressure of CO. The characteristic
switching of the reaction rate and the surface
structure were reproduced in all experiments that
we have performed. Experiments at a range of
temperatures (T ¼ 378–443 K) showed that at
higher temperatures the switching of the reaction
rate is more pronounced and occurs already at
higher CO-pressures. Fig. 1 represents a typical
result from our switching experiments in CO oxi-
dation on the Pd(1 0 0) surface. We will interpret
these results step by step and provide additional
results to support our interpretation.
cycle. At a high pressure of O the adatom islands
2
could still be observed, but in addition a large
number of small clusters had appeared (Fig. 2(b)).
The formation of the small islands has been
found to accompany the initial stage of the oxi-
dation of Pd(1 0 0) also in low-pressure experi-
ments [20]. Since the areal density of Pd atoms in a
monolayer of PdO is lower (20%) than the density
of Pd atoms in a layer of the metal lattice, excess
Pd atoms are expelled from the oxide layer,
forming small oxide islands on top. This has also
been observed in low-pressure oxidation experi-
ments on Pd(1 1 1). In these experiments, after the
initial stage, bulkoxidation started and all struc-
tural order was lost [18]. Accordingly, we conclude
that in image (B) of Fig. 1 the surface oxidized to
form PdO, and that this caused the observed in-
crease in reactivity.
4
. Discussion
We start our discussion with a detailed inter-
pretation of the stages illustrated by the STM
images of Fig. 1.
4
.1. The metal surface (A)
The experiment reproduced in Fig. 1 started
2
with a flow of a mixture of CO and O . Image 1(A)
shows that in the initial gas mixture, the palladium
surface had the same structure as a freshly pre-
pared clean Pd(1 0 0) surface in UHV. In analogy
with the Pt(1 1 0) surface [25], we interpret this

234
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
Fig. 3. STM image (276 nm · 216 nm · 8.99 nm, depicted
differentiated as if illuminated from the left side) of the oxidized
surface after having been exposed to the reaction for a duration
of 3558 s. Prior to the oxidation the metallic surface was already
rough. The oxide had a grain-like appearance. (PO2 ¼ 1:2 bar,
P
CO ¼ 0:05 bar, f ¼ 4:0 ml
n
/min, T ¼ 433 K), (V ¼ 100 mV,
t
I
t
¼ 0:2 nA).
Fig. 2. (a) STM image (98 nm · 54 nm) of the metallic Pd(1 0 0)
surface in an Ar-flow. Several large adatom island were present
and a number of bright but unidentified spots could be ob-
served. (b) STM image of the same area as in (a), but now at an
oxygen pressure, which was slowly increased to 0.5 bar. The
with a variety of grain orientations, but further
structural information of the oxide could not be
obtained. Our STM observations are consistent
with recent SXRD measurements of the formation
of a poorly ordered bulkoxide PdO film at atmo-
spheric pressures and elevated temperatures [24].
After the surface had oxidized and the reaction
rate had switched up in Fig. 1 at t ¼ 1612 s, the
2
high O -pressure induced the formation of a large number of
small clusters, which is characteristic for the initial stage of the
oxidation of Pd(1 0 0) at low pressures [20]. (Ptot ¼ 0:5 bar,
f ¼ 4:0 ml
n
/min, T ¼ 443 K), (V
t
¼ 199 mV, I ¼ 0:2 nA).
t
We also found that after image (B) the surface
became increasingly rough, with no recognizable
lattice features. Initially a granular structure could
be observed (image (C)), but on a timescale of
several minutes the images became increasingly
unclear. This was in part because the surface was
dynamic and became rough, and to some extent
because the resolution of the STM tip had de-
creased. Fig. 3, which is not part of the series of
Fig. 1 but had been formed in the same way as the
oxide in Fig. 1, shows a larger-scale image of the
surface at a later stage of oxidation. The oxide in
Fig. 3 had been formed on a relatively rough
metallic Pd(1 0 0) surface, which was the result of a
previous oxidation–reduction cycle. Fig. 3 illus-
trates the granular character of the PdO film. The
grains had a size ranging from 4 to 16 nm, and the
image suggests that they form a polycrystalline film
2
CO pressure decreased proportional to the CO-
pressure, as can be seen in Fig. 4(a) (oxide branch).
The reaction on the metallic surface (metal branch)
is slightly more complex and will be discussed in
Section 4.3. The slope of the oxide branch did not
depend noticeably on the oxygen pressure, sample
temperature, or the ÔstructureÕ and age of the oxide.
This is illustrated by Fig. 4(b), which shows the
uncalibrated mass spectrometer signals of CO
2
versus CO, for the reaction on the oxide, from all
CO oxidation experiments that we have performed
on the Pd(1 0 0) surface. They all obeyed essentially
the same rate equation:
R ¼ kPCO
;
ð1Þ
where the reaction rate constant k was the same
under all conditions. The dashed line indicates that
k is also equal to the experimentally determined

B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
235
case a further increase in oxygen pressure would
have necessarily resulted in a decrease of the CO
adsorption and, consequently, a decrease of the
reaction rate.
We have performed experiments over a limited
temperature range (ꢁ10%). Whereas on the metal-
lic surface there was a significant temperature de-
pendence of the reaction rate, we have found no
temperature dependence for the rate on the oxide.
Following transition state theory, one might think
that the reaction on the oxide surface was not
activated. However, it seems more probable that
the reaction itself was not the rate limiting step and
that a non-intrinsic process determined the CO2
production rate on the linear branch. This is also
suggested by the fact that the slope of the linear
branch of Pd(1 0 0) was more or less identical to
that of Pt(1 1 0). We propose that in all cases the
reaction on the oxide was so efficient that gas
diffusion of CO in the reactor was the rate limit-
(
a)
2
ing process, so that the overall rate of CO pro-
duction was not sensitive to the precise reactivity
of the oxide. We note, however, that the estimated
turnover-rate (10 molecules/site/s) was orders of
magnitude lower than values commonly found for
diffusion limited processes [30].
In Ref. [25] we have shown that on Pt(1 1 0) the
oxidation itself did not lead to the formation of
roughness and we could conclude that the rough-
ness was reaction induced, implying that a Mars–
Van Krevelen mechanism was active. Unlike the
oxidation of Pt(1 1 0), the oxidation process on
Pd(1 0 0) already introduced roughness. This made
a quantitative analysis of roughness versus amount
(
b)
Fig. 4. (a) The CO -pressure of Fig. 1 versus the CO-pressure.
At low CO-pressures the reaction rate shows two branches. One
branch corresponds to the reaction on the metallic surface
2
(
metal branch). The other branch corresponds to the reaction
on the oxide and shows a linear dependence of the CO -pro-
duction on the CO-pressure (oxide branch). (b) The raw mass
spectrometer signals for mass amu 44 (CO ) versus mass amu
8 (CO) for the reaction on the oxide. By plotting only the raw
2
2
2
data the linear dependence was not sensitive to potential dif-
ferences in the calibration between the different data sets. The
data are plotted for experiments performed at temperatures in
the range of 408–443 K at total pressures of 0.5 and 1.25 bar:
2
of CO produced less meaningful for palladium
than for platinum [26]. Still, the observation that
the reaction made the oxide surface increasingly
rough and the fact that the reaction did not follow
Langmuir–Hinshelwood kinetics strongly suggest
that the oxide was a reaction intermediate and that
also here the Mars–Van Krevelen scenario was
(
ꢂ
) T ¼ 433 K, P ¼ 1:25 bar; ( ) T ¼ 408 K, P ¼ 1:25 bar; (j)
ꢃ
T ¼ 443 K, P ¼ 0:5 bar; (Ã) T ¼ 443 K, P ¼ 0:5 ꢀ 1:0 bar. All
data fall on a single linear branch. The dashed line indicates the
slope for the Pt(1 1 0) data of Ref. [25].
rate constant for the oxide branch for CO oxida-
tion on Pt(1 1 0) [25]. The fact that the reaction
rate on the oxide surface did not depend on the
oxygen pressure indicates that the oxide was al-
ways ÔsaturatedÕ with reactive oxygen. This means
that on the oxide the traditional Langmuir–Hin-
shelwood type of competition between the oxygen
and CO adsorption was not at play, since in that
1
the active reaction mechanism. The Mars–Van
Krevelen mechanism has been proposed also for
1
The involvement of bulkdiffusion, which is an essential
element in Mars–Van Krevelen, does not become clear from
our experiments.

236
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
oxidation of volatile organic compounds (VOCs)
on PdO/Al [31].
of Fig. 1), showing the Pd(1 0 0) surface immedi-
ately after reduction to the metallic state. Fig. 5(a)
shows the reduced surface after having stayed in
the oxide phase for 3500 s at a temperature of 378
K, during the catalytic reaction. On each terrace a
large number of square adatom and vacancy is-
lands could be observed. They were all one atomic
layer high or deep. The square shapes reflect the
square symmetry of the Pd(1 0 0) surface. Most
islands decayed by ripening, but the islands were
also mobile and some of them coalesced. The
surface in Fig. 5(b), which had been subjected to
the oxidation reaction over a period of 2104 s at a
higher temperature of 433 K, was affected even
more. The islands had heights up to ten atomic
layers (see cross-section). They appeared as small
crystallites with a truncated pyramid shape and
well defined {1 1 1} faces. Also the pits in the
surface had this characteristic shape. The crystal-
lites slowly decayed by subsequent removal of
their top layers and a decrease in their lateral size.
Whereas before the switch to the oxide the steps
were straight and they were running in the [0 1 1]
direction, after the reduction to the metallic state,
the steps showed both straight and curved sec-
tions. The density of step edge had increased
considerably. Clearly a substantial transport of Pd
2
O
3
4
.3. Back to the metal surface (F)–(H)
When we increased the CO-pressure again at
t ¼ 4700 s (Fig. 1), the oxide was reacted away at a
slightly lower CO-pressure than that at which it
had been formed. Immediately after the reduction
of the oxide, palladium adatom and vacancy is-
lands could be observed. These islands decayed
(
images (F)–(H)) by ripening and coalescence to
restore the flat surface again. No further structural
changes were observed in switching from the
oxygen-rich flow (images (F) and (G)) to the CO-
rich flow (image (H)), thereby crossing the char-
acteristic maximum in the CO
2
production at
t ¼ 7010 s. This shows that the surface in images
pffiffi pffiffiffi
structure because in switching to the CO-rich flow,
(
F) and (G) did not have the ( 5 Â 5)R27ꢁ
pffi ffiffi pffiffi
the reduction of the ( 5 Â 5)R27ꢁ would have
led to the formation of vacancy islands in image
(H), since this oxide structure has a lower Pd-atom
density than the metallic surface.
The extent to which the initially flat surface is
affected by the oxidation and reaction is further
illustrated by Fig. 5 (Fig. 5 is not part of the series
Fig. 5. (a) STM image (140 nm · 140 nm · 1.38 nm) of the metallic surface immediately after the reduction of the oxide in 1.25 bar CO.
The surface had been in the oxidic phase for 3500 s at a temperature of 378 K. The process of oxidation, reaction, and reduction, left
behind a large number of adatom and vacancy islands on the surface of one atomic layer in height or depth (V
b) The reduced surface (140 nm · 140 nm · 4.20 nm) in 1.25 bar CO, after having been in the oxidic phase at 433 K for 2104 s. Beside
adatom and vacancy islands, which are several atomic layers high or deep, the result of what appears as step etching can be observed.
Below the image, a cross-section is shown, along the line indicated in the image (V
t
¼ 79 mV, I ¼ 0:2 nA).
t
(
t
¼ 101 mV, I
-exposure of the oxide at PH2 ¼ 0:5 bar and T ¼ 443 K. Adatom islands and a large
¼ 199 mV, I ¼ 0:2 nA).
t
¼ 0:2 nA). (c) The reduced surface
(
280 nm · 280 nm · 3.97 nm), obtained by H
2
number of curved step edges were observed (V
t
t

B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
237
pffiffiffiffiffiffiffi
P P
atoms had occurred in the process of oxidation,
the CO oxidation reaction, or in both, the Mars–
Van Krevelen reaction mechanism involving both
oxidation and reduction. These observations are
reminiscent of roughening induced by cycles of
electrochemical oxidation and reduction, obtained
on Pd(1 1 1) in a solution of sulphuric acid with an
in situ electrochemical STM [32].
The oxide on Pd(1 0 0) could also be removed
by exposing the surface to a hydrogen gas flow, as
is shown in Fig. 5(c). After the reduction, steps and
several large adatom islands with rounded and
curved shapes, could be observed.
c
O2 CO
R ¼ k
3
;
Á2
ð3Þ
À
pffiffiffiffiffiffiffi
c
P
pffiffiffi
þ PCO
O2
.
k1
k4
k5
where c ¼ pffiffiffi
accounts for the ratios of the
rate constants for adsorption and desorption for
k2
oxygen (k
pend on the temperature. Fig. 6(a) depicts the
partial pressures of CO, O and CO , which were
1
and k
2
) and CO (k
4
5
and k ), which de-
2
2
4.4. Langmuir–Hinshelwood kinetics on the metal
surface
2
After the small peakin the CO production,
which coincided with the removal of the oxide
(
second arrow in Fig. 1 at t ¼ 4700 s), the CO
2
production, now on the metallic surface, increased
again when we further increased the CO pressure.
At t ¼ 7010 s we stopped the oxygen flow and
switched to a CO-rich flow. This resulted in a
2
maximum in the CO production around the point
where the oxygen and CO pressures were equal.
Such a maximum is characteristic for Langmuir–
Hinshelwood (LH) kinetics, where the reaction
rate depends directly on the surface coverages of
the reactants, as has been found for example in
low-pressure molecular beam measurements of CO
oxidation on Pd(1 1 1) [33,34]. If we assume ran-
dom adsorption, there should be a maximum
probability for reaction at equal coverages of the
Fig. 6. (a) Partial pressures of CO, O
experiment performed at a temperature of 433 K, a total
pressure of 1.25 bar, and a flow of 4.0 ml /min. During the
experiment we switched from a CO-rich to an O -rich flow and
vice versa. The arrow marks the period where the oxide was
present. At t ¼ 797 and 4650 s maxima in the CO -production
were observed for the reaction on the metallic surface. (b)
Surface coverages for oxygen (h ) and CO (hCO), and their
product h CO, which is proportional to the reaction rate. The
coverages were calculated from Eq. (3) using c ¼ 1 as the rate
parameter and using the experimental O and CO-pressures as
2 2
and CO from an
n
2
reactants, h
by the rates of adsorption and desorption for O
O
¼ hCO. The coverages are determined
2
2
and CO, resulting in the following expression for
the reaction rate:
O
O
h
dNCO2
2
R ¼
¼ k
3
h
O CO
h
dt
input. The model reproduces the qualitative behavior of the
reaction rate on the metallic surface quite well. This strongly
suggests that on the metal surface the reaction followed Lang-
muir–Hinshelwood kinetics. The model also shows that slightly
before the oxidation and after the reduction the surface was
mainly covered with chemisorbed oxygen. The fact that the
reaction rate on the oxide was higher indicates that the chem-
isorbed oxygen on the metal surface blocked CO-adsorption
and thereby inhibited the reaction.
pffiffiffiffiffiffiffiffiffiffiffipffiffiffiffi
k
k P k k P
1
5
O2
2 4 CO
¼
k
3
:
ð2Þ
À pffiffiffiffiffiffiffiffiffi ffiffi pffiffiffiffi
pffiffiffiffiÁ
k
1
2
k
5
k
1
P
O
þ
k
2
k
4
PCO þ k
5
2
When we assume that there are no empty sites at
these high pressures, the equation for the reaction
rate reduces to:

238
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
recorded during an experiment in which we swit-
ched from a CO-rich to an O -rich flow and back
In other words, the CO-adsorption was limiting
the reaction rate for 797 < t < 1418 s.
2
to a CO-rich flow. From t ¼ 800 s on, the experi-
ment and the results were comparable to those in
Fig. 1. For the period of 1418 < t < 3503 s, at low
CO-pressures, the surface was in the oxidic state
and showed the increased reactivity. When the
CO-pressure was increased, the oxide was reacted
When the surface oxidized, the blocking of the
CO-adsorption by the chemisorbed oxygen was
lifted. This partly explains why the oxide had a
higher reactivity. In the form of the oxide, the
concentration of oxygen atoms at the surface
could be at its maximum, without blocking the
reaction with CO. A second reason for the higher
reaction rate on the oxide is that the oxygen atoms
in a chemisorbed layer on Pd(1 0 0) are more
strongly bound than those in a surface oxide. The
heat of formation of PdO ()25.65 kcal/mol) is less
negative than the heat of adsorption of oxygen on
clean Pd(1 0 0) ()50.2 kcal/mol) [18] and therefore
the PdO is energetically less stable. This explains
why the oxide takes part in the reaction so easily
and why it is more reactive than a chemisorbed
oxygen overlayer [1,35]. It is only by the virtue of
the high oxygen gas pressure that the palladium
oxide surface layer is thermodynamically stable
[36].
away and the CO
s). After that, the CO-pressure was further in-
creased. Again a maximum in the CO production
was observed when we stopped the O -flow at
2
production dropped (t ¼ 3503
2
2
t ¼ 4650 s and switched to a CO-rich flow. Before
t ¼ 800 s we first switched from a CO-flow to an
O -rich flow. This also resulted in a maximum in
2
the reaction rate close to where the CO and O2
pressures crossed at t ¼ 797 s. The CO
on the metallic surface, which is proportional to
CO, could be reproduced qualitatively by Eq.
3) in combination with the experimental values
for the CO and O partial pressures PCO and P 2 , as
can be seen in Fig. 6(b).
2
production
h h
O
(
2
O
Next to the reaction rate, proportional to
h h , the surface coverages of oxygen (h ) and
carbon monoxide (hCO), which were calculated
from the LH-model, are plotted in Fig. 6(b). In the
initial CO-flow the surface was completely covered
When the CO-pressure was increased again, the
relatively unstable oxide was reacted away and the
surface switched backto the metallic state, mainly
covered with oxygen. Upon further increase of the
CO-pressure, the CO-coverage increased, leading
to a higher reaction rate. When the oxygen flow
was stopped and the oxygen and CO coverages
crossed at h ¼ h ¼ 0:5, the reaction rate passed
O
CO
O
with CO, until at t ¼ 479 s the O
2
-flow was started.
The CO-coverage decreased and the O-coverage
increased. At t ¼ 797 s the coverages were equal,
O
CO
h
¼ h ¼ 0:5, and the reaction rate was at its
through a maximum again, after which it de-
creased together with the O-coverage.
CO
O
maximum (h h ¼ 0:25). After that, the reaction
CO
O
rate dropped together with the CO-coverage and
the surface was mainly covered with oxygen. At an
4.5. Bistability and oscillations
oxygen coverage h
O
¼ 0:9 at t ¼ 1418 s the surface
oxidized and the reaction rate switched up by a
factor 1.5. Obviously, the oxidation is not de-
scribed by the LH-model. From this switch in
reactivity we can draw an important conclusion
about the reaction slightly before the switch. The
reaction rate switched up for a decreasing CO-
pressure, which proves that also before the switch,
when the surface was in the metallic state, there
was a sufficiently high supply of CO in the gas
phase for a reaction rate of at least a factor 1.5
higher (the value after the switch). Apparently, the
high oxygen coverage was blocking the CO-
adsorption, thereby partly inhibiting the reaction.
Fig. 7 shows the CO production plotted versus
2
the CO-pressure. During the cycle also the O and
2
the Ar pressure were varied, to keep the total
pressure constant. This makes Fig. 7(a) projection
of a three dimensional plot, PCO2 ðP ; P Þ. Start-
O2
CO
ing from a high CO-pressure of 1.25 bar, the CO
production initially increased with decreasing CO-
2
pressure and increasing O -pressure, after which it
2
decreased when the surface became mainly covered
with chemisorbed oxygen, which blocked the CO-
adsorption. When the surface switched to the
oxide, for values of the CO-pressure below 0.16
bar, the blocking was lifted and the CO produc-
2

B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
239
reaction rate branches made the system bistable,
which could lead to oscillations between the two
stable branches.
Such spontaneous oscillations were indeed ob-
served as can be seen in Fig. 8(a). At t ¼ 436 s we
started to decrease the CO-pressure, which re-
sulted in the switch to the oxide at t ¼ 591 s. After
that, the CO pressure decreased proportional to
2
the CO-pressure. Around t ¼ 2000 s we started to
slowly increase the CO-pressure and at t ¼ 2311 s
spontaneous oscillations in the reaction rate oc-
curred. With increasing CO-pressure their ampli-
tude increased, while the frequency decreased. The
oscillations stopped when we decreased the CO-
pressure again at t ¼ 3380 s. The CO-pressure
2
Fig. 7. The CO pressure versus the CO-pressure (data from
Fig. 6). Since the total pressure was constant (Ptot ¼ 1:25 bar)
the oxygen pressure also varied. The arrows indicate the
direction in which the CO-pressure was varied. On the high CO-
pressure side, the surface was mainly covered with CO, and the
reaction rate was low. At intermediate CO-pressures, the CO
2
(
represented by the signal of 12 amu in this case)
production showed a maximum around PCO ¼ 0:8 bar, follow-
ing Langmuir–Hinshelwood kinetics. For lower CO-pressures,
the surface was mainly covered with oxygen, leading to a lower
also showed oscillations, albeit smaller in relative
2
value for the CO -production. For PCO < 0:16 bar the surface
was in the oxidic state with the corresponding reaction rate
branch which depended linearly on the CO-pressure. When the
CO-pressure was increased again, the oxide was removed and
the surface switched backto the metallic state at
P
CO ¼ 0:11
bar. For increasing CO-pressure the reaction rate maximum
occurred at a higher CO-pressure than for decreasing CO-
pressure, possibly resulting from hysteresis in the LH-kinetics.
There was significant overlap (0:11 < PCO < 0:16 bar) between
the reaction branches of the metal surface and that of the oxide.
tion increased stepwise by a factor 1.8. After that,
2
the CO -production decreased proportional to the
CO-pressure following the linear oxide branch.
When the CO-pressure was increased again above
0
mainly oxygen-covered surface and the CO -pro-
.11 bar, the surface switched backto the metallic,
2
duction increased for higher CO-pressures. The
maximum in the Langmuir–Hinshelwood kinetics,
coming from the low CO-pressure side was shifted
towards a higher CO-pressure (1.0 bar) with re-
spect to maximum for decreasing CO-pressure (0.8
bar), suggesting that there was also hysteresis in
the LH-kinetics. Both Figs. 6 and 7 show several
spikes in the reaction rate of the oxide phase. It
seems as if the reaction rate switched several times
spontaneously from the high (oxide) branch to the
low (metallic) branch and backagain. In the CO-
pressure range of 0:11 < PCO < 0:16 bar in Fig. 7,
Fig. 8. (a) Partial pressures of CO, O
(
2
and CO
2
at T ¼ 443 K
Ar in the mixture is not shown). The O
2
-pressure was kept
more or less constant and the CO-pressure was first decreased
in order to switch to the oxide, which occurred at t ¼ 590 s.
After t ¼ 2000 s the CO-pressure was slowly increased. During
the subsequent, slow increase in CO-pressure, oscillations in the
2
CO and the CO-pressures could be observed. When the CO-
pressure was decreased again at t ¼ 3380 s, the oscillations
disappeared. (b) The oscillations reflected the spontaneous
switching between the metallic and oxidic reaction branches.
and in Fig. 4(a) there was hysteresis in the CO
production. The overlap of the metallic and oxidic
2

240
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
magnitude, which were in anti-phase with the CO
oscillations. This illustrates once more that indeed
CO reacted to CO and that the CO-pressure was
2
would correspond to the mainly oxygen-covered
metallic surface, with the high oxygen coverage
not blocking the CO-adsorption significantly. The
switching between the two branches would then
proceed via the formation of an oxide, which
Turner et al. considered to be catalytically non-
active. If, on the other hand, the bistability and the
oscillations on polycrystalline wires of Pd, ob-
served by Turner et al., had the same origin as the
bistability in our observations on Pd(1 0 0), namely
the switching between Pd and PdO, then the data
of Turner et al. imply that PdO is the active phase
in CO oxidation at temperatures above 503 K and
for PCO=PO2 -ratios up to 2.0. This suggests that
under realistic conditions, in the catalytic con-
verter of a car, PdO can be stable. In addition, real
technical catalysts consist of small nano-particles,
which tend to oxidize more easily than single-
crystal surfaces [37–39]. We conclude that palla-
dium oxide can play an important role in auto-
motive catalysis and that it might even be a key
element of the actual catalyst. Nevertheless, the
processes occurring on practical three-way cata-
lysts, involving the oxidation of CO and hydro-
2
considerably affected by variations in the reaction
rate. Fig. 8(b) shows that the oscillations were
indeed the switching between the oxidic and
metallic branches [5]. In our experiments the
spontaneous oscillations were only observed when
we increased the CO-pressure on the oxidic reac-
tion branch and not when we decreased the CO-
pressure.
On the Pt(1 1 0) surface we have never observed
spontaneous oscillations. In this respect, there were
two major differences between the results for
Pt(1 1 0) and Pd(1 0 0). The first one was that
whereas Pd(1 0 0) switched between an oxide and an
oxygen-covered metallic surface, the Pt(1 1 0) sur-
face switched between an oxide and a metal surface,
which was probably mainly covered with CO. We
emphasize, however, that the reaction kinetics on
the metallic Pt(1 1 0) surface were not fully under-
stood. The second difference was the hysteresis. For
Pt(1 1 0) we observed ÔclockwiseÕ hysteresis, i.e. the
oxide was formed at a lower CO-pressure than that
at which it was removed, and for the Pd(1 0 0)
surface the hysteresis was Ôanti-clockwiseÕ. We do
not know what precisely determines the ÔdirectionÕ
of the hysteresis, but the Ôanti-clockwiseÕ behavior
on Pd(1 0 0) forms an essential condition for oscil-
latory behavior on this surface. Possibly the reac-
tion-induced formation of the roughness on PdO
has a destabilizing effect on the oxide, which in turn
may be responsible for the Ôanti-clockwiseÕ orien-
tation of the hysteresis. The ÔclockwiseÕ hysteresis
on Pt(1 1 0) naturally tends to drive that surface out
of the bistability region [26].
Our observations and interpretations are in
complete disagreement with the well-established
Turner–Sales–Maple model [14,15], which inter-
prets the oscillations as the switching between two
stable solutions of the Langmuir–Hinshelwood
equations for the surface coverages of CO and O,
which are coupled in a non-linear way through the
reaction. In that model, the low reaction rate
branch would correspond to the CO-covered
metallic surface. The high CO-coverage would in-
hibit the reaction by blocking the adsorption of
oxygen. The reaction branch with the higher rate
x
carbons as well as the reduction of NO , combined
with catalyst particle size effects, sintering, and
promotor effects, make the real system so complex
that our results for a single reaction on a simple
model catalyst may not be applicable directly.
5. Conclusions
We have studied the oxidation of CO over a
Pd(1 0 0) model catalyst, using in situ STM and
simultaneous gas analysis. This allowed us to di-
rectly relate the structure of the Pd(1 0 0) surface to
the reaction kinetics. We have observed two dif-
ferent structures, i.e. an adsorbate-covered metal-
lic surface and an oxide, corresponding to different
reaction rates and kinetics. For high PO2 =PCO ratios
the surface oxidized and the reaction rate in-
creased stepwise. In the oxidic state the surface
had a significantly higher reactivity. The oxidation
of the surface introduced roughness and the CO
oxidation reaction on the oxide was accompanied
by a steady increase in surface roughness. The
reaction rate on the oxide depended linearly on the

B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
241
[
3] M. Lyubovsky, L. Pfefferle, A. Datye, J. Bravo, T. Nelson,
J. Catal. 187 (1999) 275.
CO-pressure. From a comparison of this depen-
dence for different temperatures, oxygen pressures,
and even sample materials (Pt(1 1 0)), we suggest
that the rate was limited by a non-intrinsic process.
Langmuir–Hinshelwood reaction kinetics were not
observed on the oxide surface. Analogous to our
findings for CO oxidation on platinum oxide
[
[
4] G. Veser, A. Wright, R. Caretta, Catal. Lett. 58 (1999) 199.
5] X. Zhang, C.S.-M. Lee, D.M.P. Mingos, D.O. Hayward,
Appl. Catal. A: Gen. 250 (2003) 183.
[6] H. Over, Y.D. Kim, A.P. Seitonen, S. Wendt, E. Lundgren,
M. Schmid, P. Varga, A. Morgante, G. Ertl, Science 287
(
2000) 1474.
[
[
[
7] S. Ladas, R. Imbihl, G. Ertl, Surf. Sci. 219 (1989) 88.
8] M.R. Bassett, R. Imbihl, J. Chem. Phys. 93 (1990) 811.
9] N. Hartmann, R. Imbihl, W. Vogel, Catal. Lett. 28 (1994)
373.
formed under O -rich conditions on Pt(1 1 0), we
2
propose a Mars–Van Krevelen mechanism for this
reaction on the PdO. By contrast, the reaction
kinetics on the metallic surface could be mod-
eled satisfactorily by a Langmuir–Hinshelwood
mechanism. Depending on the gas pressures the
reaction rate was either limited by CO- or by oxy-
gen-inhibition. For roughly equal pressures of CO
[
[
[
[
10] H. Tanaka, J. Yoshinobu, M. Kawai, Surf. Sci. 327 (1995)
L505.
11] V. Bondzie, P. Kleban, D.A. Brown, J. Vac. Sci. Technol.
A 11 (1993) 1946.
12] V.A. Bondzie, P. Kleban, D.J. Dwyer, Surf. Sci. 347 (1996)
319.
13] V.A. Bondzie, P.H. Kleban, D.J. Dwyer, Surf. Sci. 465
and O we observed a maximum in the rate on the
2
(
2000) 266.
metallic surface. For high PO2 =PCO ratios there
existed a CO-pressure range where both the oxy-
gen-inhibited reaction branch on the metallic sur-
face and the reaction branch of the oxide surface
were stable. We have observed that this bistability
could lead to oscillations in the reaction rate,
caused by spontaneous switching between the
oxygen-covered metal surface an the palladium
oxide.
Our results are in disagreement with widely
accepted models for oscillatory reactions at
atmospheric pressures. Our findings suggest an
alternative origin of oscillatory behavior in oxi-
dation reactions on Pt-group metals and indicate
that palladium oxide might be the more active
phase in automotive catalysis.
[
[
14] J.E. Turner, B.C. Sales, M.B. Maple, Surf. Sci. 103 (1981) 54.
15] J.E. Turner, B.C. Sales, M.B. Maple, Surf. Sci. 109 (1981)
5
91.
[16] G.A. Somorjai, Introduction to Surface Chemistry and
Catalysis, John Wiley and Sons, New York, 1994.
[
17] M.M. SlinÕko, N.J. Jaeger, P. Svensson, J. Catal. 118
(
1989) 349.
[
[
18] G. Zheng, E.I. Altman, Surf. Sci. 462 (2000) 151.
19] E. Lundgren, G. Kresse, C. Klein, M. Borg, J.N. Andersen,
M.D. Santis, Y. Gauthier, C. Konvicka, M. Schmid, P.
Varga, Phys. Rev. Lett. 88 (2002) 246103.
[
[
20] G. Zheng, E.I. Altman, Surf. Sci. 504 (2002) 253.
21] M. Saidy, O.L. Warren, P.A. Thiel, K.A.R. Mitchell, Surf.
Sci. 494 (2001) L799.
[22] M. Todorova et al., Surf. Sci. 541 (2003) 101.
[23] G. Zheng, E.I. Altman, J. Phys. Chem. B 106 (2002) 1048.
[
24] E. Lundgren, J. Gustafson, A. Mikkelsen, J.N. Andersen,
A. Stierle, H. Dosch, M. Todorova, J. Rogal, K. Reuter,
M. Scheffler, Phys. Rev. Lett. 92 (2004) 046101.
[
[
[
[
25] B.L.M. Hendriksen, J.W.M. Frenken, Phys. Rev. Lett. 89
(
2002) 046101.
26] B.L.M. Hendriksen, M.D. Ackermann, J.W.M. Frenken,
submitted for publication.
27] P. Mars, D.W. van Krevelen, Chem. Eng. Sci. 3 (1954) 41
Acknowledgements
This workis supported by the Dutch Technol-
ogy Foundation (STW) and the Dutch Organiza-
tion for Scientific Research (NWO).
(
Spec. Suppl.).
28] B.L.M. Hendriksen, G.S. Verhoeven, L. Crama, A. Geluk,
E. de Kuyper, J.W.M. Frenken, P.B. Rasmussen, H.
Zeijlemaker, H.G. Ficke, W. Barsingerhorn, Rev. Sci.
Instrum., submitted for publication.
[
[
29] S.C. Bobaru, B.L.M. Hendriksen, M.D. Ackermann,
J.W.M. Frenken, in preparation.
30] G.J. Kramer, private communication.
References
[
[
1] M. Ziauddin, G. Veser, L.D. Schmidt, Catal. Lett. 46
1997) 159.
[31] E.M. Cordi, J.L. Falcorner, J. Catal. 162 (1996) 104.
[32] K. Sashikata, N. Furuya, K. Itaya, J. Vac. Sci. Technol. B
9 (1991) 457.
(
2] A.K. Datye, J. Bravo, T.R. Nelson, P. Atanasova, M.
Lyubovsky, L. Pfefferle, Appl. Catal. A: Gen. 198 (2000)
[33] T. Engel, G. Ertl, J. Chem. Phys. 69 (1978) 1267.
[34] T. Engel, G. Ertl, Adv. Catal. 28 (1979) 1.
179.

242
B.L.M. Hendriksen et al. / Surface Science 552 (2004) 229–242
[
[
35] J.G. McCarty, Catal. Today 26 (1995) 283.
36] M. Todorova, W.X. Li, M.V. Ganduglia-Pirovano, C.
Stampfl, K. Reuter, M. Scheffler, Phys. Rev. Lett. 89 (2002)
[38] E.H. Voogt, A.J.M. Mens, O.L.J. Gijzeman, J.W. Geus,
Catal. Today 47 (1999) 321.
[39] C.-B. Wang, C.-T. Yeh, J. Catal. 178 (1998) 450.
[40] Selected STM movies can be viewed on <http://www.phys-
ics.leidenuniv.nl/sections/cm/ip/>.
096103.
[
37] E.H. Voogt, Ph.D. thesis, Universiteit Utrecht, 1997.