Nanofacet-resolved CO oxidation kinetics on alumina-supported Pd particles

Article abstract of DOI:10.1016/S0009-2614(02)00151-3

Employing a multi-molecular-beam approach, we have measured the angular distribution of CO₂ molecules formed during CO oxidation under steady-state conditions on well oriented and shaped nm-sized Pd crystallites grown on an ordered alumina film. The experiment is combined with kinetic Monte Carlo simulations based on a realistic structural model. The results obtained allow us (i) to differentiate between local reaction rates on the particle nanofacets and (ii) to conclude that oxygen diffusion on and between the (1 1 1) facets is rapid compared to reaction.

Full text of DOI:10.1016/S0009-2614(02)00151-3

1
8 March 2002
Chemical Physics Letters 354 (2002) 403–408
www.elsevier.com/locate/cplett
Nanofacet-resolved CO oxidation kinetics on
alumina-supported Pd particles
a
J. Hoffmann , S. Schauermann , J. Hartmann , V.P. Zhdanov , B. Kasemo ,
a
a
b,c
b
a,
a
J. Libuda *, H.-J. Freund
a
Fritz-Haber-Institut der Max-Planck-Gesellschaft, Abt. Chemische Physik, Faradayweg4-6, D-14195 Berlin, Germany
b
Department of Applied Physics, Chalmers University of Technology, S-412 96 G €o teborg, Sweden
Boreskov Institute of Catalysis, Russian Academy of Sciences, Novosibirsk 630090, Russia
c
Received 17 January 2002
Abstract
Employing a multi-molecular-beam approach, we have measured the angular distribution of CO
2
molecules formed
during CO oxidation under steady-state conditions on well oriented and shaped nm-sized Pd crystallites grown on an
ordered alumina film. The experiment is combined with kinetic Monte Carlo simulations based on a realistic structural
model. The results obtained allow us (i) to differentiate between local reaction rates on the particle nanofacets and (ii) to
conclude that oxygen diffusion on and between the (1 1 1) facets is rapid compared to reaction. Ó 2002 Elsevier Science
B.V. All rights reserved.
Heterogeneous catalysis is an interdisciplinary
field of natural sciences with immense practical
impact, as it forms the mainstay of chemical in-
dustry and environmental technologies [1]. In ac-
ademic studies, catalysis is usually explored on
single-crystal surfaces. In practice, however, cata-
lytic reactions often occur on very small (ꢀ10 nm)
crystalline metal particles deposited on the pore
walls of a more or less inactive support. The un-
derlying microscopic reaction kinetics on such
catalysts are still poorly understood, because of
the limited experimental information on surface
structures and processes on nm catalyst particles
under reaction conditions. Experimentally, this
problem can be addressed by using modern tech-
niques aiming at the preparation of supported
model catalysts, which consist of well controlled
arrays of metal particles deposited on a planar
oxide surface. Systems of this type have recently
been used in studies of the global reaction kinetics
[2–5]. Unfortunately, there are essentially no ex-
perimental tools, which may provide the corre-
sponding site-resolved kinetic information under
in situ conditions. Potentially, fast scanning tun-
neling microscopy [6] and/or photoemission elec-
tron microscopy [7] may be applied to model
catalysts, however, these methods remain limited
with respect to the accessible range of experimental
conditions and the kinetic information sup-
plied.
*
Corresponding author. Fax: +49-30-84134101.
E-mail address: libuda@fhi-berlin.mpg.de (J. Libuda).
0
009-2614/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved.
PII: S0009-2614(02)00151-3

404
J. Hoffmann et al. / Chemical Physics Letters 354 (2002) 403–408
In this Letter, we suggest a new approach to this
average, the particles contain about 2700 atoms
with 20% of them at the surface. About 20% of the
surface atoms are located in the top facet, the rest
belong to the side facets. Both the top and the side
facets have (1 1 1) geometry (the fraction of (1 0 0)
facets is small). Under reaction conditions applied
in this work, the particles remain stable for at least
two days after pre-stabilizing the sample by oxy-
gen exposure at elevated temperatures.
problem, which is based on measurements of the
angular distribution of the reaction products de-
sorbing from ordered and oriented nanoparticles,
performed by angular-resolved mass spectrometry.
This detection technique in combination with a
multi-molecular-beam approach allows us to con-
trol individually the impingement and reaction
rates on different crystallite facets. In this case, the
reactant angular distribution can provide a signa-
ture of the reactive sites, at which it has been
formed, and accordingly makes it possible to
clarify intimate details of elementary reaction
steps.
All the experiments were performed at the
Fritz-Haber-Institute (Berlin) in a special ultra-
high-vacuum apparatus [8] making it possible to
cross up to three molecular beams on the sample
surface (Fig. 1). The reaction products are ob-
served with a rotatable doubly differentially
pumped quadrupole mass spectrometer (QMS) in
an angular resolved mode or by a non-differen-
tially pumped QMS in an angular integrated
mode. Under reaction conditions, the adsorbate
coverages are controlled via in situ IR reflection
absorption spectroscopy.
In our experiments, the incidence angle of the
2
O beam, which is generated from a supersonic
expansion, is 35° with respect to the surface nor-
mal (Fig. 1b). From the beam and particle geom-
etry, it follows that for all Pd crystallites the top
facets and the side facets facing the oxygen beam
have similar inclination angles with respect to the
O
O
2
2
beam. Thus, they are exposed to similar direct
fluxes. The side facets on the opposite side of
particles are shaded from direct collisions with O
2
13
molecules. The second reactant, CO, is supplied
in a symmetric manner by superimposing two CO
beams (the use of beam sources instead of CO
exposure via the background pressure and iso-
topically labeled CO drastically reduces the noise
level in the measurements).
Under steady-state conditions, the global reac-
tion rate as a function of the fraction of CO in the
2
Here, we explore the CO–O reaction on Pd
particles (Fig. 2a) [9] grown on a well-ordered
alumina film prepared by oxidizing a NiAl(1 1 0)
single crystal [10]. The mean particle diameter and
height are about 5.5 and 1.7 nm, respectively. On
total impingement rate, xCO ¼ FCO=ðFCO þ FO2
Þ
(FCO and FO2 are the CO and O fluxes), typically
2
shows a maximum, dividing two characteristic re-
gimes (Fig. 3a) [11]. The catalyst is either primarily
Fig. 1. Schematic representation of (a) the molecular-beam experiment and (b) the experimental setup used in this work.

J. Hoffmann et al. / Chemical Physics Letters 354 (2002) 403–408
405
Fig. 2. (a) STM image (35 nm  25 nm) of the alumina-supported Pd particles and (b) the corresponding lattice model used in the
simulations (the arrows indicate the oxygen flux).
2
covered by adsorbed O at O excess or by CO at
CO excess (the left and right sides of Fig. 3, re-
spectively). In the angular resolved experiments
(
see e.g. Fig. 4a for 465 K), we probe the CO2
angular distributions corresponding to both reac-
tion regimes. It is observed that (i) the CO dis-
tribution is broader than a cosine distribution and
ii) the asymmetry of the distribution is small un-
2
(
der all conditions and close to the detection limit
of the experiment.
(i) To interpret the CO angular distribution, we
2
take into account that adsorption of this species on
the Pt-group metals is extremely weak, leading to
desorption on a very short time scale after for-
mation [12]. The potential energy surface govern-
2
ing the CO angular distribution depends on the
local surrounding of adsorbates. For CO oxida-
tion on Pd(1 1 1), the distribution has been found
to be close to the cosine law at low coverage [11a],
while at higher coverage, lateral adsorbate inter-
actions result in a CO flux strongly peaked along
2
the surface normal [11b]. We interpret the obser-
vation of the broadened angular distribution to be
2
due to CO formation on the side facets of the Pd
particles and desorption along the corresponding
facet normals.
(ii) The weak asymmetry observed for the an-
gular distributions indicates that despite the fact
Fig. 3. Angular-integrated steady-state reaction rates as a
function of the CO fraction in the total reactant flux at different
sample temperatures: (a) experimental data (for details, see
O
[
11]); (b) results of the simulations with p ¼ 1 and pcyc ¼ 0:01
dif
(each data point is obtained by averaging over 5000 MCS).
that there is no O supply on the shaded side of the
2

406
J. Hoffmann et al. / Chemical Physics Letters 354 (2002) 403–408
conclude that the nearly complete equilibration of
the chemisorbed oxygen requires rapid diffusion of
this species on the reaction time scale.
Taking into account that quantitative infor-
mation on oxygen diffusion on Pd single crystal
surfaces was lacking so far, we have scrutinized the
role of this process in the CO oxidation on Pd
particles by using Monte Carlo (MC) simulations
(see [14] for applications of this technique to nm
chemistry). The physics implemented into our
model is as follows: (i) Our earlier mean-field
analysis [11] indicates that the apparent Arrhenius
2
parameters for the CO -formation step on the
supported particle system used here are close to
those on the (1 1 1) face. Thus, we assume that CO
oxidation is dominated by the (1 1 1) facets. Spe-
cifically, the Pd particle is represented by a trun-
cated triangle, constructed from a hexagonal
lattice (Fig. 2b). Catalytically, all sites are consid-
ered to be equivalent. (ii) The nearest-neighbor
Fig. 4. (a) Angular distributions of CO
from the supported Pd crystallites at T
¼ 465 K under O
CO-rich reaction conditions (the corresponding total CO beam
2
molecules desorbing
s
2
- and
14
À2 À1
intensities are 3.2 and 6:4 Â 10 cm
s ; the O
À2 À1
2
beam inten-
14
sity is approximately equal to 4 Â 10 cm
s
in both cases).
(nn) O–O lateral interaction is well known to be
At polar angles between 10 and 65°, no data are available due
to shadowing of the supersonic beam by the rotatable mass
strongly repulsive. To mimic this effect, we pro-
hibit O adsorption in nn sites (consequently, O2
adsorption is considered to initially produce O
adatoms in next-nearest-neighbor (nnn) vacant
sites) and introduce a weak nnn O–O repulsion (8
kJ/mol). As a result, the model predicts the ex-
perimentally observed formation of the pð2 Â 2Þ
ÀO structure at saturation coverage of ’0.25 ML
[15]. (iii) At low coverage, CO adsorbs mainly in
three-fold hollow sites, but may occupy bridge and
on-top sites at higher coverages (see e.g. [16] and
references therein). For simplicity, we take into
account only one type of adsorption sites and al-
low CO adsorption only if not more than three CO
molecules are located in nn sites in order to re-
produce the experimental saturation coverages.
spectrometer. (b) Polar plot of the difference between the CO
2
angular distribution and the cosine distribution. (To reduce
drifts, the displayed data represent an average over both scan-
ning directions. The results have been corrected for the angular-
dependent acceptance function of the detector as described in
[
8].)
Pd crystallites, adsorbed oxygen is distributed
homogeneously over the full particle on the time
scale of reaction. Partly, this effect may be con-
nected with diffuse scattering of O
sion of O and CO via weakly bound precursor
2
and/or diffu-
2
2
states on the particles as well as on the support. A
possible contribution of these factors is however
expected to be minor (the average length of diffu-
sion in the precursor state is e.g. appreciably
smaller than the crystallite size). Moreover, steer-
ing effects due to an attractive molecule–surface
interaction potential might have to be taken into
account in a quantitative analysis of the exact flux
impinging on the facets [13]. However, we expect
steering not to be a dominating effect in the present
case, as the long range interaction potential is
relatively weak and the size of the particles is ra-
ther large on the length scale of the attractive
forces. Thus, in a qualitative discussion, we may
(iv) CO strongly inhibits O adsorption (see e.g.
2
[12]). To mimic this effect, we prohibit O2 ad-
sorption if at least one of the sites for adsorption
has more than two nn CO molecules (this rule is
however not applied to O diffusion). (v) For the
2
CO and O adsorption rate constants, we use the
CO
ad
experimental values (in particular, kad ꢁ k
þ
O2
À1
À6
kad ¼ 0:5 s at approximately 10 mbar total
pressure of the three beams). For CO desorption
and CO + O reaction, the Arrhenius parameters
are Edes ¼ 136 kJ/mol, mdes ¼ 10 s , Erea
15
À1
¼

J. Hoffmann et al. / Chemical Physics Letters 354 (2002) 403–408
407
7
À1
6
0 kJ=mol, and mrea ¼ 8 Â 10 s [11]. At 465 K,
In order to relate the MC time to the number of
À1
the corresponding rate constants are kdes ¼ 0:5 s
reaction attempts, we define one MC step as N
trials of a adsorption–reaction event (N is the
s
À1
and krea ¼ 15 s , respectively. For CO diffusion,
s
CO
dif
one has
1
E
¼ 17 kJ=mol and
D
ꢂ
¼ 2:2 Â
number of adsorption sites). This means that MC
time is calculated by multiplying the total number
À3 À1
2
0
cm s [17], and accordingly at 465 K the CO
10
À1
jump rate is about 10 s . Thus, CO diffusion is
very rapid compared to all other steps. To char-
acterize the ratio of the rates of O diffusion and
catalytic cycle to that of CO diffusion, we employ
s
of trials by pcyc=N . With this definition, the MC
and real time are interconnected as tMC ¼ ðkdes
þ
k Þt.
rea
First, we have calculated the angular-integrated
O
the dimensionless parameters pdif and pcyc, respec-
2
reaction kinetics with O adsorption on all facets
tively. To save computational time, a CO jump rate
is used, which is only 100 times higher than the
reaction rate, i.e., pcyc ¼ 0:01. This rate was found
to be sufficient in order to reproduce a homoge-
neous distribution of CO under all conditions.
With the specifications above, our MC algo-
rithm is as follows:
(Fig. 3b). The agreement between experiment and
simulation is found to be good at all rates of O
diffusion (cf. e.g. Fig. 3a,b). Minor deviations re-
main with respect to the temperature dependence,
which are related to the coverage dependencies of
the activation barriers (see e.g. [11]). It is of in-
terest to note that in agreement with our experi-
(
1) A random number q (q 6 1) is generated. If
q < pcyc, an adsorption–reaction trial is realized
item (2)). For q > pcyc, an adsorbate-diffusion
trial is performed (item (3)).
2) An adsorption–reaction attempt contains
mental observations no hysteresis in the CO
2
formation rate is found with the present set of
kinetic parameters (this effect is, however, possible
for other sets of parameters [18]).
(
(
To simulate the angular-resolved experiments,
we switch off the oxygen adsorption on one of
the side facets (Fig. 2b). As expected, the reac-
tion rate on this facet (Fig. 5a) and the occu-
pation of the shaded facet by oxygen (Fig. 5b–e)
increase with increasing O diffusion rate in this
case. Under oxygen-rich conditions, the rate on
this facet becomes indistinguishable from that on
several steps. (i) An adsorption site on the catalyst
particle is chosen at random. (ii) A new random
number q is generated. (iii) If the site selected is
vacant, CO or O
described above provided that q < pCO and
0
2
adsorption events are realized as
0
0
CO
CO ¼ k =ðkdes þ
ad
p
k
CO < q < pCO þ pO2
,
where
p
O2
ad
reaÞ and pO2 ¼ k =ðkdes þ kreaÞ are the normalized
O
adsorption probabilities. (iv) If the site selected is
occupied by CO, a CO desorption or reaction
the non-shaded facets for p ’ 1 (this value
dif
O
dif
À1
O
dif
corresponds to k ’ 1500 s , or to E ¼ 85 kJ/
0
0
event is realized for q < p and q > pdes, where
mol provided that the pre-exponential factor is
À1
des
À3
2
p
des ¼ kdes=ðkdes þ kreaÞ. For the CO reaction, one
‘normal’, D
ꢂ
¼ 10 cm s ). At the same diffu-
nn site is chosen at random, and the trial is fulfilled
if the latter site is occupied by O. (v) If the site is
occupied by O, the trial ends (note that the reac-
tion is executed in the CO branch (item (iv))).
sion rate and under CO-rich conditions, ap-
proximately 80% of the maximum rate is
obtained on the shaded facet. The difference
arises as a consequence of the changes on the
oxygen residence time, which depends on the
steady-state coverages. The above given value for
the diffusion barrier should be considered as an
experimental upper limit. So far, no other
quantitative experimental data are available for
oxygen diffusion on Pd single crystal surfaces.
The value is consistent with recent theoretical
investigations which provide an estimate of 54
kJ/mol for the diffusion barrier on Pd(1 1 1) [19].
However, it cannot be excluded that on the
nanoparticles in addition to the regular diffusion
(
3) For adsorbate diffusion, an adsorption site is
chosen at random. If the site is vacant, the trial
ends. Otherwise, a CO or O particle located on this
site tries to diffuse. In particular, an adjacent site is
randomly selected, and if the latter site is vacant
and located on the particle lattice, and there are no
spatial constraints (as described above), CO jumps
to the destination site with unit probability. Jumps
of O are executed with the probability pdif com-
bined with the standard Metropolis rule in order
to take into account the O–O lateral interactions.

408
J. Hoffmann et al. / Chemical Physics Letters 354 (2002) 403–408
obtain novel information on the rates of elemen-
tary reaction steps. In particular, we have derived
that under reaction conditions an upper limit for
the activation energy of oxygen diffusion on Pd
crystallites is about 85 kJ/mol.
Acknowledgements
The project has been funded by the Deutsche
Forschungsgemeinschaft and the Max-Planck
Society. The authors thank M. B €a umer and
M. Heemeier for the STM data used in this
work.
References
[1] J.M. Thomas, W.J. Thomas, Principles and Practice of
Heterogeneous Catalysis, VCH, Weinheim, 1997.
2] C.R. Henry, Surf. Sci. Rep. 31 (1998) 231.
3] H.-J. Freund, M. B €a umer, H. Kuhlenbeck, Surf. Sci. Rep.
[
[
3
4] J. Zhu, G.A. Somorjai, Nano Lett. 1 (2001) 8.
€
5] S. Johansson, L. Osterlund, B. Kasemo, J. Catal. 201
1 (1998) 231.
[
[
(
6] J. Wintterlin, Adv. Catal. 45 (2000) 131;
2001) 275.
[
[
J. Wintterlin et al., Science 278 (1997) 1931.
7] W. Huang, R. Zhai, X. Bao, Langmuir 17 (2001)
Fig. 5. (a) Steady-state reaction rate on the shaded facet at
different rates of oxygen diffusion for T
3
8] J. Libuda et al., Rev. Sci. Instrum. 71 (2000) 4395.
9] M. B €a umer, H.-J. Freund, Prog. Surf. Sci. 61 (1999) 127.
629.
s
¼ 465 K. For com-
[
[
parison, the reaction rate on the top (central) facet at high O
and CO mobility is added. (b)–(e) Snapshots of the lattice under
steady-state reaction conditions with no oxygen adsorption on
the shaded side (bottom) facet for xCO ¼ 0:29 and different rates
of oxygen surface diffusion. Oxygen atoms and CO molecules
are indicated by open and filled circles, respectively.
[
[
[
10] R.M. Jaeger et al., Surf. Sci. 259 (1991) 235;
J. Libuda et al., Surf. Sci. 318 (1994) 61.
11] (a) J. Libuda et al., J. Chem. Phys. 114 (2001) 4669;
(
b) J. Hoffmann et al., J. Catal. 204 (2001) 378.
12] (a) T. Engel, G. Ertl, Chem. Phys. Lett. 54 (1978) 95;
b) T. Matsushima, H. Asada, J. Chem. Phys. 85 (1986)
658.
(
1
barriers the contribution of steps and edges is
considerable.
[13] S. Van Dijken, L.C. Jorritsma, B. Poelsema, Phys. Rev.
Lett. 82 (1999) 4038.
[
[
[
[
[
14] V.P. Zhdanov, B. Kasemo, Surf. Sci. Rep. 39 (2000) 25.
15] H. Conrad et al., Surf. Sci. 65 (1977) 245.
16] T. Gießel et al., Surf. Sci. 406 (1998) 90.
17] M. Snabl et al., Surf. Sci. 385 (1997) L1016.
18] V.P. Zhdanov, B. Kasemo, Surf. Sci. 496 (2002) 251.
In summary, we have shown that a combination
of multiple molecular-beam techniques and angu-
lar-resolved mass spectroscopy makes it possible
to directly monitor reaction rates on the various
facets of the supported nanocrystallites and to
[19] K. Honkala, K. Laasonen, J. Chem. Phys. 115 (2001) 2297.