Structure and electronic properties of Ca-Doped CeO2 and implications on catalytic activity: An experimental and theoretical study

Article abstract of DOI:10.1021/jp983323e

Doping CeO₂ with for example, Ca gives an enhanced reactivity toward reduction of SO₂ by CO, and total combustion of methane. Theoretical modeling using static minimizations and molecular dynamics (MD) simulations of the doped (110) face in combination with ab initio quantum chemical cluster models shows large effects on the Ce(IV)/Ce(III) balance due to the doping. Computed oxygen-to-cerium charge-transfer energies are strongly reduced as a result of the introduction of defects and oxygen vacancies, but not sufficiently to explain the observed reactivities. The structures resulting from the MD simulations for both the doped and undoped material are in good agreement with recent experimental pulsed neutron scattering results.

Full text of DOI:10.1021/jp983323e

J. Phys. Chem. B 1999, 103, 7627-7636
7627
Structure and Electronic Properties of Ca-Doped CeO2 and Implications on Catalytic
Activity: An Experimental and Theoretical Study
Stefano de Carolis, Jos e´ Luis Pascual, and Lars G. M. Pettersson*
FYSIKUM, Stockholm UniVersity, Box 6730, S-113 85 Stockholm, Sweden
Micael Baudin, Mark W o´ jcik, and Kersti Hermansson
Inorganic Chemistry, The Ångstr o¨ m Laboratory, Uppsala UniVersity, Box 538, S-751 21 Uppsala, Sweden.
Anders E. C. Palmqvist and Mamoun Muhammed
Department of Materials Chemistry, Royal Institute of TechnologysKTH, S-100 44 Stockholm, Sweden
ReceiVed: August 7, 1998; In Final Form: May 6, 1999
2 2
Doping CeO with for example, Ca gives an enhanced reactivity toward reduction of SO by CO, and total
combustion of methane. Theoretical modeling using static minimizations and molecular dynamics (MD)
simulations of the doped (110) face in combination with ab initio quantum chemical cluster models shows
large effects on the Ce(IV)/Ce(III) balance due to the doping. Computed oxygen-to-cerium charge-transfer
energies are strongly reduced as a result of the introduction of defects and oxygen vacancies, but not sufficiently
to explain the observed reactivities. The structures resulting from the MD simulations for both the doped and
undoped material are in good agreement with recent experimental pulsed neutron scattering results.
I. Introduction
vacancy depends on the association energy between the vacancy
and its surrounding. Furthermore, the local structures of the
One of the most important groups of materials for heteroge-
neous catalysis is that of the metal oxides. Understanding the
mechanisms of catalyzed reactions over metal oxide surfaces
and the influence of different material characteristics on the
performance of the catalyst is therefore of central importance
for future rational catalyst development. Cerium oxide (CeO2)
is commercially an important part of both oil-refining catalysts
and automotive exhaust gas converters. It is also used as an
oxidation catalyst for combustion of carbonaceous deposits in
diesel exhaust gas particulate traps and as an oxygen and
electron-transfer agent in catalysts for the ammoxidation of
surface, as well as that of the bulk, of the doped samples are
believed to depend on several properties of the dopant, such as
its oxidation state, electronegativity, and size. The introduction
of dopants in the CeO2 lattice can thus be expected to affect
both the oxygen-transfer and the electron-transfer characteristics
of CeO2.
For a more detailed understanding of the effects of dopants
and the origin of the enhanced (or reduced) reactivity of these
modified materials we believe that experiment need to be
complemented by theoretical/computational studies. An atomic
level description of the activity of a material as complex as
doped CeO2 requires consideration of the dopant effects on both
the geometrical and electronic structures, especially for the
surface layers. The particular approach taken in the present work
is therefore to perform both static (energy minimization) and
dynamical (including temperature effects) simulations of the
crystal structure and then build quantum chemical embedded
cluster models of different sites resulting from the simulations;
the structures obtained from the molecular dynamics simulations
are fed directly into the quantum chemical modeling program.
Thus, the structural effects induced by the doping, such as the
local geometry around vacancies and defects, are obtained from
molecular dynamics (MD) simulations of a slab model exposing
the desired crystal face, while the effects on the electronic
structure of the material, mainly the availability of Ce(III)
relative to Ce(IV), are investigated using fully quantum chemical
methods. The structural results from the MD simulations on
the bulk materials are supported by results from pulsed neutron
scattering studies.
1
propylene in the production of acrylonitrile. The ability of CeO2
to store oxygen is utilized in the three-way car catalysts (TWC)
for increasing the width of the lambda window, i.e., to increase
the working range of redox conditions in which the catalyst is
able to simultaneously convert NOx, CO and unburned hydro-
2
carbons to harmless products. Its function in this application
is, however, possibly more complicated than that of a simple
oxygen storage component.1
In principle, it should be possible to engineer, at the atomic
level, active sites on a metal oxide surface to give a catalyst
desired selectivity and activity for certain reactions; such an
experimentally based molecular approach for enhancing the
reactivity of cerium oxide by introduction of dopants has also
3
recently been suggested. In this approach, cation dopants with
oxidation states lower than +IV are incorporated in the lattice
of CeO2 leading to the formation of oxygen vacancies. The
chemical and physical properties of these depend on the type
of dopants introduced, where for example, the size of the
vacancies can be expected to depend much on the size of the
dopant; the concentration of oxygen vacancies depends on the
oxidation state of the dopant, and the mobility of the oxygen
Due to its technological importance, CeO2 has been the
subject of a rather large number of earlier investigations where
4
-8
static simulation techniques have been applied.
Since CeO2
*
Corresponding author.
is a well-known oxygen storage material much of the interest
1
0.1021/jp983323e CCC: $18.00 © 1999 American Chemical Society
Published on Web 08/19/1999

7
628 J. Phys. Chem. B, Vol. 103, No. 36, 1999
de Carolis et al.
has focused on determinations of oxygen transport and diffusion
rates, but the charge-transfer properties and concomitant reduc-
tion of Ce(IV) to Ce(III) have also been of great interest. In
two recent papers Balducci et al.7,8 discuss the properties of
solid solutions of CeO2 -ZrO2 and find that the introduction
of zirconia into the ceria lattice strongly lowers the reduction
energy of Ce(IV); the resulting values are still positive, however.
The importance and usefulness of the experimental approach
to use dopants for oxygen vacancy engineering has been verified
experimentally for CeO2 for three types of reactions. First, the
He. The composition of outlet gas was analyzed with a Balzers
mass spectrometer model QMG421C, and the CO and SO2 gases
were mixed just before the reactor inlet using two Brooks 5850E
mass-flow controllers. The catalyst samples were pretreated in
situ in a flow of air (600 °C, 30 min), after which they were
cooled to room temperature before exposure to the reactant gas
at a space velocity of 10 000 mL gas/h‚g catalyst, and a
temperature increase rate of 10 °C/min.
B. Atomistic Simulations. In this section we will describe
the different aspects of the atomistic simulations of the bulk
oxygen storage capacity (OSC) was found to increase by 20-
and surface models of CeO , both doped and undoped. Both
static and dynamic simulations were performed on the basis of
2
9
4
0% by introducing dopants such as Ca, Nd, Pb, etc. Second,
the light-off temperature for the reduction of SO2 by CO to
form elementary sulfur and CO2 was lowered by 70 °C by
doping CeO2 with 10% Ca2 as elaborately discussed in ref
the same set of potential parameters that were taken from the
literature. The pure CeO bulk lattice has the well-known fluorite
2
+
4+
structure, in which each Ce cation is surrounded by eight
equivalent O ions forming the corners of a cube, and with
each O coordinated to four Ce . The experimental lattice
2
-
1
0. Third, the activity for total oxidation of methane was also
2
-
4+
found to increase by doping with Ca, Mn, and Nd, whereas
Pb-doping was found to decrease the activity of the material
parameter is 5.41 Å leading to a Ce-O distance in the bulk of
1
1
13
for this reaction. Several other metal oxides are also expected
to be useful candidates as host matrixes for oxygen vacancy
engineering, e.g., oxides of Bi, Mo, Nb, Pr, Ti, V, W, and Zr
to name a few.
2.343 Å. The (110) terminated CeO surface-layer consists
2
4
+
2-
of 6-coordinated Ce ions and 3-coordinated O ions, and all
planes parallel to the surface are charge neutral.
The programs used for the static simulations were GULP
for the bulk structures while the MARVIN code was used for
determining the relaxed surface structures. The MARVIN code
was used to determine the relative stabilities of different faces
14
1
5
The present paper aims at illuminating the effects on the
structural and electronic properties of cerium oxide upon
2+
introduction of Ca ions, and on the implications of these
effects on the catalytic activity for redox reactions, in a
combined theoretical and experimental study. The theoretical
treatment is based on a molecular dynamics simulation of the
ion positions in a slab model of the surface, followed by a
calculation of the electronic properties for a small cluster model
embedded in the remaining periodic MD slab. The main result
is that, as a result of the modified structure and coordination
around the Ce ions as obtained from the molecular dynamics
simulations, we find a substantially reduced charge-transfer
excitation energy in the doped CeO2 lattice. In other words, we
find a much lowered cost for reducing Ce(IV) to Ce(III) after
doping with Ca. This is in agreement with what has been
reported earlier,7 but the resulting values for the energy cost
of the CeO crystal; the most interesting of these was then
2
studied in the subsequent molecular dynamics simulations
presented here. The MD simulations were performed using an
1
6
in-house MD code.
1
. Interatomic Potentials. The interatomic potential param-
eters used in both the static energy minimizations and the MD
5
simulations were those used by Sayle et al. in an earlier static
2-
2-
2+
2-
modeling investigation of CeO2. The O -O and Ca -O
parameters in ref 5 were previously described by Lewis and
17
4+
2-
Catlow, and the Ce -O interaction in ref 5 is a modifica-
tion of the potential by Butler et al. In the current paper the
18
non-Coulombic cation-cation interactions were, as in the
,8
19
previous investigations, set to zero. The shell model was used
(endothermicity), computed using high-level ab initio quantum
to describe the polarization energy. In this model the polarizable
ion is assigned two charges: one charge on the core (nucleus)
and one charge on a displaceable shell which is linked to the
core through a harmonic spring. In addition we have also, in
the molecular dynamics simulations, assigned a fraction of the
ion-mass to the shells of the ions, as first introduced by Mitchell
chemical techniques, are still rather high. Thus, additional
energy-lowering contributions must be considered in order to
completely understand the observed reactivities.
The layout of the paper is as follows. In the Methods part
we describe in order the experiments, the static and dynamic
atomistic simulations, and finally the quantum-chemical calcula-
tions. The same sequence is followed in the Results section.
The results are tied together in the Discussion and Summary
sections.
20
and Fincham.
The Coulombic interaction was in all cases evaluated using
21-23
the Ewald
summation technique for 2-dimensional (slab
models of surfaces) or 3-dimensional (bulk models) periodic
systems, excluding the interaction between a core and its own
shell.
II. Methods
The quality of the interaction potentials for the Ca-doped and
undoped CeO2 system was investigated by a comparison with
experimental diffraction results. Thus, the crystallographic lattice
parameter for the pure bulk crystal at 300 K obtained from the
MD simulation was 5.42 Å, in very good agreement with the
A. Experimental. Nanophase cerium oxide solid solutions
were prepared by coprecipitation of Ce and dopants in the form
of oxalates followed by washing, drying, and calcination at 600
°
C for 4.5 h in air. Details on the synthesis and characterization
3,10
are given elsewhere. Chemical composition of the products
was determined with an ARL 3520B inductively coupled plasma
atomic emission spectrophotometer (ICP-AES). Experimental
total neutron correlation functions T(r) were obtained through
pulsed neutron scattering studies at the LAD instrument at the
ISIS facility in the United Kingdom and details of these
13
experimental value of 5.41 Å. Analysis of the pair distribution
function, g(r), for the Ce-O distances from the 300 K MD
simulation leads to an average r (Ce-O) of 2.33 Å to be
13
compared to the 2.34 Å from the experimental structure. Upon
2
+
1
2.5% doping with Ca ions the simulated cell expands by
12
1
2
2% in accordance with what has been found experimentally.
experiments and their interpretation are given elsewhere.
Catalytic activities were evaluated with a Micromeritics TPD/
TPR 2900 Chemisorption Analyzer using a nearly stoichiometric
reactant gas mixture containing 3.20% CO and 1.65% SO2 in
A comparison between the pulsed neutron scattering data,
T(r), and the MD results, G(r), (Figure 1), demonstrates a very
good agreement, which will make a detailed analysis of the

Ca-Doped CeO2 and Implications on Catalytic Activity
J. Phys. Chem. B, Vol. 103, No. 36, 1999 7629
scheme,25,26 ensuring a translationally invariant Hamiltonian.
2
7,28
The Nos e´ -Hoover formalism
was used for the constant
temperature control. The Ewald summation technique for
2
1-23,29
2
-dimensional periodic systems
has been implemented
for the slab simulations. Details about the MD program have
1
6
been given elsewhere.
All systems were equilibrated for 1.0 ps with temperature
scaling invoked every tenth step, followed by production runs
of 2 ps for the doped and undoped bulk and slab at 10 K, 8.5
ps for the undoped bulk and slab systems at 300 K and 24.5 ps
for the doped bulk and slab systems at 300 K. All simulations
used a time-step of 0.1875 fs. Such a short time-step was
necessary to ensure the constancy of the Hamiltonian in these
systems, where the ionic charges and interatomic forces are
large; in particular the large charges in the shell model generate
forces of a magnitude that make it necessary to use a small
time step. The equilibration times of 1.0 ps appear to be
sufficient for all our runs. For the 10 K runs, which essentially
are energy minimizations, no changes were observed in
thermodynamic, structural or dynamical parameters after the
equilibration runs. The equilibrations at 300 K were started from
previous simulations. Moreover, the cell volume and potential
energy were monitored throughout the equilibrations and
subsequent production runs, and were found to have stabilized
well before the start of the production runs. Moreover, the
positional order was monitored via a modification of Verlet’s
translational order parameter, which had stabilized already after
half the equilibration time (0.5 ps). During the remaining
equilibration time and during the subsequent production runs,
all these quantities displayed appropriate fluctuations and no
drifts. The resulting structures were then fed directly into the
ab initio embedding program for the electronic structure
calculations.
3
. System Geometries. Within the MARVIN code the static
simulation cell is repeated infinitely in two dimensions, x and
y, with the free surface perpendicular to the z direction. The
atoms in the “surface region” are allowed to relax while those
in the “bulk region” are kept fixed. Here we studied the relative
stabilities of the undoped (111), (110), (211), and (310) surfaces,
and for reasons discussed in the Results section, we have chosen
to concentrate on the (110) surface in the present study.
Figure 1. Total neutron correlation functions for Ca-doped and
undoped CeO . (a) Experimental T(r) curves from pulsed neutron
2
scattering. (b) Molecular dynamics simulated total and partial pair
distribution functions G(r).
The MARVIN code was also used to investigate the interac-
tion between dopant and vacancy for the (110) plane. The
experimental investigation was done at 10% doping which was
simulated by introducing one dopant-vacancy pair per two
crystallographic unit cells, resulting in a doping level of 12.5%.
Two different combinations of dopants and vacancies are then
possible within this expanded cell; first, with the created oxygen
vacancy next to the dopant and second, with the vacancy as far
as possible from the dopant. The first situation can be seen as
a neutral bound pair, while the second as an unbound pair.
structural peaks meaningful. In particular, the broadening of the
peaks upon doping, the dopant-induced structures in the region
3
4
-3.5 Å, and the shape of the peaks in the region from 3.5 to
.5 Å are all well-reproduced by the simulation. Furthermore,
the simulations allow a breakdown and detailed analysis of the
individual ion contributions to the radial distribution functions
which is invaluable for the analysis of the experimental data;
the full, detailed analysis of the structure will be reported
12
elsewhere. We here conclude that the potential parameters and
simulation procedures employed in this paper appear to give a
realistic description of the bulk lattice structures for both the
Ca-doped and undoped systems.
MD simulations of the undoped CeO2 (110) slab were
performed at different temperatures. Since the (111) face shows
the highest stability and furthermore has been shown to be
3
0
inefficient for oxygen exchange, we here selected to study
only the (110) surface; the (211) and (310) faces could in
principle also be of interest, but were excluded in the present
work in order to perform a more complete simulation study of
the (110) face. The pure CeO2 (110) surface system in the MD
simulations was described as a slab, 12 atomic layers thick and
periodic in two dimensions (x and y), with faces perpendicular
to z and -z. The x, y, and z directions are equivalent to the (0
0 1), (1-1 0) and (1 1 0) directions in the CeO2 bulk crystal.
2. MD Simulations. Four oxide systems were investigated by
MD simulations: two bulk systems (doped and undoped) and
two slab systems (doped and undoped). The geometries of these
systems will be discussed below. The MD simulations were
performed with a constant-stress, constant-temperature molecular
dynamics program for systems periodic in three and two
dimensions, with dynamically variable lattice vectors (lengths
and angle). The equations of motion for the box are handled
by the Cleveland modification24 of the Raman-Parrinello

7630 J. Phys. Chem. B, Vol. 103, No. 36, 1999
de Carolis et al.
Figure 2. Definition of different embedded cluster surface models of doped CeO
2
in the full surface environment; atoms contained in the model
cluster are highlighted. Cluster models 5A (right), 5B (center), and 5C (left) have five-coordinated central cerium ions, while clusters 6B (right),
6
A (center), and 6C (left) represent fully coordinated surface cerium ions.
3
The MD box volume was 16 × 15 × 23 Å and contained
altogether 432 ions. All ions were allowed to move in the MD
simulations.
trostatic interactions with the surrounding crystal. Thus, the
short-range Coulomb (incomplete screening), exchange, and
orthogonality interactions, together with the major relativistic
effects (Darwin and Mass-velocity potentials) are included in
the description.
The doped slab system in the MD simulations was doped
with 1/8 Ca2 ions, as in the static simulations, and was then
charge-balanced with oxygen vacancies. The initial configuration
was taken from the ordered dopant-vacancy distribution found
to give the lowest surface energy in the static calculations
+
In the different embedded cluster calculations, ions out to a
certain distance from any cluster ion are represented by total-
ion model potentials; this distance was always larger than 13
Å to ensure a large enough buffering region between the cluster
ions and the unscreened point-charges. To represent the bulk
and (110) surface of undoped ceria, we have chosen embedded
(
namely “the bound pair”; note, however, that the “unbound
pair” gave a slightly lower bulk energy). The dopants and the
corresponding oxygen vacancies were introduced in such a way
as to preserve the zero net dipole moment across the surface
region; otherwise the total energy increases with increasing
thickness of the slab and the surface energy is not defined.
MD simulations for the 3-dimensional CeO2 bulk crystal (pure
and doped) were performed to provide a reference for com-
parison with the undoped and doped slab systems. The MD-
box for the undoped bulk system consisted of 324 ions and its
1
2-
2-
8-
(
(
CeO8)
CeO8)
and (CeO6)
clusters, respectively; here, e.g.,
1
denotes a cluster model with one cerium and eight
oxygens. For the Ca-doped ceria crystal, we have chosen
8-
6-
(
CeO6) and (CeO5) clusters for the (110) surface (see Figure
2
for the definition of the different embedded cluster models),
1
2-
10-
while (CeO8)
and (CeO7)
were constructed for the bulk
3
situation. It should be noted that each model is charge neutral
when both the central cluster and the embedding are considered
together. The details about the basis sets used in the calculation
of the electronic structure of the clusters and of the potentials
used to represent the lattices are given in the appendix.
The ECPAIMP program38 has been used to evaluate the
volume was approximately 16 × 16 × 16 Å . The doped bulk
system was doped with 1/9 Ca2+ ions, and was then charge-
balanced with oxygen vacancies.
C. Ab Initio Cluster Calculations. In the present work, we
have calculated the energies and several electronic properties
of cerium ions in pure and calcium-doped cerium oxide crystals
based on the structures obtained either from the pair-potential
AIMP (core and embedding) potential integrals and to generate
the one-electron integrals over the Madelung potential. The
implementation is such that the unit cell from, for example, the
MD simulations can be read directly into the program and the
potential from the fully 2-D periodic structure used for the
calculations. The one-electron integrals can then be replaced in
a number of different programs, and two of these programs,
31
minimizations or from the MD simulations. It has been shown
that relatively small and also nonstoichiometric clusters, if
properly embedded, can provide a good model of the crystal.
In the description of strongly ionic systems, an embedding of
the cluster into a lattice of point-charges will lead to an artificial
polarization of the anions in the cluster.31-34 This may be
avoided by a proper embedding scheme using effective core
39
the ones used here, are the direct-SCF code DISCO and the
35-37
potentials (ECP) based on the frozen ionic charge distribution
4
0
Stockholm program package.
to describe the neighbors nearest to the cluster. In the present
work we have used the AIMP embedding35,36 together with a
The relative stability of Ce(III) and Ce(IV) in the different
lattice positions in the undoped and doped structures was
investigated by explicitly performing the required oxygen-
cerium charge-transfer excitation within the embedded cluster
models. For this purpose we have performed SCF and Modified
representation of the crystal potential of the different lattices
based on structures obtained either from the static or MD
simulations. The crystal potential was included through evalu-
ation of all integrals over the explicit Ewald2
1-23
sums over
41
the infinite crystal based on the simulation cell, always using
the nominal ionic charge for the ions. The embedding scheme
is based on total-ion model potentials which include an
approximate description of all quantum mechanical and elec-
Coupled Pair Functional (MCPF) calculations on the different
clusters mentioned above. Some additional calculations to
investigate the electron affinity of cerium ions in different
positions were furthermore performed.

Ca-Doped CeO2 and Implications on Catalytic Activity
J. Phys. Chem. B, Vol. 103, No. 36, 1999 7631
2
TABLE 1: Experimental Characterization of Different
Cerium Oxide Samples Studied: Oxygen Deficiency (δ)
Measured at Room Temperature, Unit Cell Parameter (a in
TABLE 2: Surface Energies, E (in J/m ), of Pure CeO
2
surface
E before relaxation
E after relaxation
2
Å), BET Area (S in m /g), and Light-Off Temperatures
(111)
(110)
(211)
1.72
3.60
6.85
1.22
1.65
1.84
2.54
Corresponding to 50% Conversion of CO (T50,CO) and SO
2
(T
50,SO2), Respectively (in °C)
(310)
11.59
sample
composition
δ
a
S
T
50,CO
T
50,SO2
undoped
Ca-doped Ce0.90Ca0.10
CeO2-δ
0.00 5.412 29.5
O2-δ 0.09 5.416 29.3
490
430
505
435
can only be understood when reasonable models for both the
doped and undoped surface structures exist. From pure static
energy minimizations we found, in agreement with other
4
-6
groups,
that the most stable surface is (111), followed by
(
110), (211), and (310), see Table 2. Most stable often means
3
0
least reactive, and it has been shown that the (111) surface
does not release oxygen easily. Thus, in the following we focus
our interest on the (110) surface, which has a sufficiently low
surface energy that it can be expected to be exposed on the
small CeO2 particles.
The resulting surface energy for the relaxed (110) surface
2
computed with the MARVIN program is 1.65 J/m . This is about
half of the surface energy of the unrelaxed surface, which
indicates the importance of including these effects in the
modeling; it can be expected that, in a theoretical study, the
unrelaxed surface should exhibit an artificially increased reactiv-
4
3
ity over that of the fully relaxed surface.
Since the Ca(+II) dopant has a lower valency than the
Ce(+IV) of the original material, each dopant has to be charge
compensated through the creation of an oxygen vacancy. For a
given level of doping this introduces questions about how the
dopants and vacancies should be distributed initially in the
simulation cell; for gases and liquids, final distributions can be
expected to be reasonably independent of the initial guess, but
this may not be the case for a high-melting-point oxidic material
such as the present. Using the MARVIN code the interaction
between dopant and vacancy was first investigated: the
2
Figure 3. Experimentally measured SO
reactor outlet as a function of catalyst temperature over undoped and
Ca-doped CeO . Circles (O) and crosses (x) indicate the results for the
2 2
, CO, and CO flow rates in
2
undoped and doped samples, respectively.
III Results
A. Experimental. Chemical compositions of the experimen-
tally studied samples are given in Table 1. XRD patterns
collected on a conventional X-ray diffractometer of the samples
revealed that the Ca-doped sample retains the fluorite structure
with a slightly increased unit cell parameter as compared to
the undoped CeO2. The use of the very intense pulsed neutron
beam showed, however, the presence of a small fraction of
CaCO3 in that sample in agreement with X-ray photoelectron
spectroscopy (XPS) results.1 The oxygen deficiency values,
δ, in Ce1-xCaxO2-δ, agree well with the expected values
considering the oxidation state of the dopant.
calculated surface energy of the Ca-doped (111) CeO is 0.56
2
2
J/m for the bound pair system and 0.67 J/m for the unbound,
but this is due to a somewhat greater bulk stability of the
unbound pair system. The energy difference, 0.02 eV/unit cell,
between the two situations is very small, indicating only a very
slight interaction energy between dopant and vacancy. However,
to draw conclusions on the mobility of oxygens one would also
need an estimate of the barrier to migration.
2,42
Doping of the material could cause a segregation of the dopant
ions resulting in different concentrations at the surface and in
the bulk. This was investigated by obtaining the surface energies
for the (110) crystal with the dopants and vacancies introduced
only in one layer and varying the distance from the surface of
this layer. The thus obtained surface energy as a function of
the distance of the introduced defect layer from the surface of
CeO2 showed it to be energetically more favorable for the dopant
layer to be at the surface. However, not surprisingly, the MD
Figure 3 shows the ability of the materials to catalyze the
reduction of SO2 by CO described in reaction 1 below:
0
SO + 2CO f S + 2CO
(1)
2
2
In this reaction, elementary sulfur and CO2 are formed. The
sulfur was condensed in a cooling trap, whereas the concentra-
tions of SO2, CO, and CO2 were quantified by mass spectrom-
eter. In Figure 3 the flow rates of the gases in the reactor outlet
are presented as a function of catalyst temperature. It is clear
that both the on-set and the light-off temperatures were much
lower for the Ca-doped sample compared to those of the
undoped CeO2 sample. The light-off temperatures corresponding
to 50% conversion, T50, are given in Table 1, and both samples
2
+
simulations show no tendency for Ca migration for any
feasible simulation time. The surface energy is also dependent
on the concentration of defects at the surface, with a rapidly
increasing surface energy (for the model with dopants and
vacancies in one layer only) as the concentration increases;
2
values of 28.4, 8.2, 4.5, and 1.6 J/m (note that in this case
showed lower values of T50,CO than of T50,SO , which indicates
much higher surface formation energies are obtained due to the
inhomogeneity in the dopant concentrations) were found for
concentrations of 50, 12.5, 5.5, and 0%, respectively. Thus, as
the dopant concentration is increased, it is likely to prevent
further segregation and no significant segregation will be
expected.
2
that the first step in the mechanism of reaction 1 is the oxidation
of CO by oxygen from the catalyst surface. Then, as the surface
becomes sufficiently reduced, at a slightly higher temperature,
it reduces the SO2, thus completing the catalytic cycle.
B. Atomistic Simulations. 1. Low-Temperature Surface
Systems (0 and 10 K). The impact of doping on the catalyst
reactions and the reason and mechanism for the increased OSC
2+
Figure 4 shows the relaxed slab structure of the Ca -doped
slab after 1.9 ps of simulation at 10 K. The figure is a projection

7632 J. Phys. Chem. B, Vol. 103, No. 36, 1999
de Carolis et al.
pair. This is to be expected, since the introduced oxygen
vacancies alter the potential energy surface for the other ions
in close vicinity. Furthermore, the introduction of DV pairs in
the second layer seems to suppress the oscillatory interplanar
4
+
relaxation of the Ce ions and enhance this behavior for the
2
-
O
ions, while the introduction of DV pairs in the outermost
layer does not affect these oscillations in the relaxation.
Both in the bulk and in the slab system the relaxation effects
in the close vicinity of the DV pairs involve the displacement
of ions, mainly oxygen ions, from their crystallographic
equilibrium positions by as much as 0.7 Å. In the slab system
some oxygen ions also migrate to other crystallographic sites
or, for the outermost layer, to new sites off the crystallographi-
cally determined positions. The mechanism of these ionic
relaxations/migrations depends on the ion’s distance from the
surface. For oxygen ions located deeper than the outermost layer,
the migration only occurs along the x-direction (crystallographic
(001) or (00 1h ) directions) from an occupied oxygen site to an
oxygen vacancy (Type I). For oxygen ions in the outermost
layer of the slab, the migrations also occur in the x-direction,
but the ions stay between two equivalent crystallographic sites,
somewhat displaced in the z-direction (Type II) due to the higher
degree of translational freedom that the ion possesses near the
surface. Moreover, at the surface we can also observe migrations
in the yz-direction (Type III) (crystallographic (010) or (0 1h 0)
direction), or more specifically, if the dopant is located in the
outermost layer the associated oxygen vacancy maintains its
position in the outermost layer, but if the dopant resides in the
second layer the associated oxygen vacancy migrates out to the
2+
Figure 4. The 12.5% Ca2+-doped CeO
.9 ps at 10 K. The figure is a projection of all ions onto the plane of
the paper, which is the xz plane. The slab surfaces are normal to the z
(110) slab after relaxation for
topmost layer, keeping the Ca dopant in the second layer fully
coordinated. This indicates that oxygen vacancies are more
stable at the surface than in the bulk, in agreement with an earlier
investigation of isolated defect formation energies when doping
2
1
direction.
3
+
2+
2+ 45
with Rh , Pd , and Pt
and the recent investigation of
of all ions onto the plane of the paper, which is the xz-plane. It
can be seen that the relaxation introduces a rather large overall
distortion from the ideal CeO2 structure, but, more importantly,
the distortion is larger at the surface than in the middle of the
slab.
8
doping with zirconia by Balducci et al. In the present
investigation we find that this turns out to be true even at higher
defect concentrations and in the close vicinity of a Ca dopant,
2+
both when the dopant is located in the uppermost layer and
when located in the second layer.
The MD simulations at 10 K show that, without dopant-
vacancy (DV) pairs in the slab the relaxation consists of a rather
small rumpling in the four outermost layers, while the rest of
the slab remains rather bulklike. The rumpling mainly involves
the oscillatory interplanar relaxation of Ce4 ions by -10%,
2
. Temperature Effects on the CeO2 (110) Surface. The
structure of the oxide substrate at different temperatures, both
at an atomic scale and at a time-resolved level, is of great
interest. In particular, a closer analysis of the dynamics at the
surface of the slab in terms of mean-square and instantaneous
displacements of the ions could answer questions such as
whether reactive sites could become available, i.e. the surface
activated, through large-amplitude motions and jumps of the
ions at the surface, and whether it would be possible to have a
reduction of barrier heights for specific reactions as an effect
of surface ion mobility.
+
+
9.4%, -2.6%, and +2.6% in the first, second, third, and fourth
layers, respectively, while the oxygen ions relax less than 1%,
and in the opposite direction compared with the cerium ions
(
see Figure 5). In a parallel investigation on a large number of
44
low-index cuts of CeO2, Vyas et al. find a rather large
dependence on the potential for the relaxations of the outermost
layers as obtained from static energy minimizations (corre-
sponding to our 10 K simulations). For the most stable, (111),
plane the largest relaxation effects were found for the oxygen
ions, while the cerium ions were found to relax to a much
smaller extent leading to the opposite rumpling compared to
our present investigation of the (110) plane. Results for the
relaxation of the (110) plane were not reported in ref 44 making
a detailed comparison difficult. However, for the higher-index
planes, e.g., the (331) surface, larger distortions and oscillations
for both types of ions are reported, showing that the details of
the relaxation effects at the surface depend both on the specific
potential and on the specific plane chosen for study.
Increasing the temperature up to 300 K does not alter the
oscillatory interplanar distance behavior for the relaxation,
described above, in the pure and doped CeO2 slab. The number
of migration events, on the other hand, appears to increase with
temperature. During 24.4 ps of simulation time at 300 K, two
migrations of Type I were observed in the doped bulk system,
and five events of Type II in the doped slab system. The Ca2
2
+
doping of bulk CeO results in higher atomic mean-square
4
+
displacements (msd’s). The msd’s for Ce increase from
approximately 0.0022 Å (average of x, y and z directions) in
2
2
the undoped ceria to 0.0026 Å when doped, and the msd’s for
2
-
2
2
In the presence of DV pairs the relaxation effects are much
more pronounced, especially in the close vicinity of the DV
O
increase from 0.0039 Å in the undoped bulk to 0.0047 Å
2
for doped ceria. The dopant has an msd of 0.0043 Å .

Ca-Doped CeO2 and Implications on Catalytic Activity
J. Phys. Chem. B, Vol. 103, No. 36, 1999 7633
Figure 5. Average distances between adjacent planes in the x, y, and z directions in the slab (definitions in the text). For comparison, the corresponding
interplanar distances in the bulk are given, where “Bulk 110” corresponds to the value for the y and z directions (equivalent distances) and “Bulk
0
01” corresponds to the value for the x direction.
When the above msd’s are studied as a function of depth in
calculations on the embedded ions. The lowest charge transfer
state found was from the 2p level of one of the oxygens in the
cluster model (localized excitation) to Ce(4f), which results in
an increased screening of the Ce valence and outer core orbitals
making the Ce(III) ion about 5% larger than the Ce(IV) ion.
The values computed are expected to provide an upper bound
to the energy required to change the charge states since several
energy-lowering effects are not included. In particular, polariza-
tion effects in the lattice, which will be present for the C-T
state but not for the ground state, could be expected to lower
the undoped slab system at 300 K (Figure 6) one can observe
msd’s higher than the bulk values down to about 6 Å from the
surface, while the msd’s in a 4 Å thick region in the middle of
the slab are essentially the same as in the bulk. A very similar
situation appears for the doped systems. The increase is largest
in the x and z directions, and for z follows the oscillatory
interplanar distance behavior observed in the relaxation (see
above). Upon doping, the effects of increased atomic msd’s are
suppressed for the cerium ions and enhanced for the oxygen
ions. In the outermost layer, some oxygen ions have a higher
msd than the average in the same layer, which can be explained
by oxygen ions that have migrated, according to a Type II
mechanism, to a new, less strongly bound site.
C. Ab Initio Calculations. Ca-doped CeO2 has experimen-
tally been found to have greatly increased oxygen ion conduc-
tivity compared to undoped CeO2.46 This may be connected with
the higher catalytic activity for reaction 1 found for the Ca-
doped sample. The concentration of oxygen vacancies is higher
for Ca-doped CeO2, which should contribute to a higher oxygen
ion mobility and a higher concentration of active sites for, e.g.,
SO2 reduction. In addition to the oxygen transfer processes,
involved in the reduction of SO2 by CO, there must however
be elementary steps involving charge-transfer processes. Hence,
the electronic properties, e.g., electron affinities or availability
of charge-transfer (C-T) excitations, of the materials are likely
of importance for the catalytic activity of these materials.
4
7
the energy by at least 1 eV. This will however be canceled
by the correlation energy contribution which is about 2 eV larger
4
+
2-
3+
-
for the Ce O
pair than for the Ce O combination.
However, since the size of the cerium ion increases, some
geometrical effects may be expected, where the structure around
the larger Ce(III) ion expands to accommodate it; this can only
reduce the CT excitation energy. In an effort to estimate the
magnitude of this effect a limited geometry optimization was
performed of the oxygen positions in the embedded cluster.
Since the optimization now is done at a different level of theory
compared to what was used to generate the embedding potential
it is essential to first establish the optimum geometry of the
quantum mechanically described embedded cluster. The geom-
etry of the oxygens around the embedded cerium(IV) ion was
thus optimized giving very small differences from the geometry
determined from the simulations; this is a gratifying result
showing the stability of the chosen combination of techniques.
Changing the electronic structure of the cerium to Ce(III) by
considering the triplet state of the cluster and redoing the
optimization again resulted in only minor modifications with
only a small resulting effect on the CT energy, less than 0.1
eV.
1
. Ce(IV) to Ce(III) Excitations (C-T). As a consequence of
the obtained structural model for the doping, the cerium ions
are not all fully coordinated; this might be expected to lead to
more easily reduced species. We have computed the energy
required to transfer an electron from O2 to Ce for each of
the three different five- and six-coordinated cerium ions in the
doped (110) surface and have compared those with the corre-
sponding values for the bulk and the undoped surface.
-
4+
The results from this study are reported in Table 3. The
doping and the modified coordination around the ceria is seen
to lead to a substantial reduction (by over 2 eV) of the CT
excitation energy. The electron affinity computed for the Ce
ions in the different clusters (last column of the same table)
The calculations have been performed at the Hartree-Fock
level with an estimate of effects of correlation from MCPF

7634 J. Phys. Chem. B, Vol. 103, No. 36, 1999
de Carolis et al.
oxygen vacancies efficiently exposes highly unsaturated cat-
4
8,49
ions.
Characteristic of cerium oxide is that the cerium ion
may have two oxidation states, whereby these ions can
participate in reactions that involve changes in the oxidation
4
+
state of the cation. By substituting part of the Ce in the CeO2
structure by other cations, the properties of the material are
affected, e.g., the concentration and mobility of oxygen vacan-
cies are increased, and several electronic properties are greatly
altered. It is hence not certain which of these contribute the
most to an increase in the catalytic activity of the material.
The static and the MD simulations of the calcium-doped ceria
show that the created oxygen vacancies prefer to be located
away from the top layer for the (111) surface, while for the
(110) surface the vacancies were more stable at the surface.
These exposed oxygen vacancies could of course be of
importance and a reason for the increased OSC. In addition,
they may well have a central role in the catalytic efficiency of
the material. For example, a first step in reaction 1 can be
assumed to be the adsorption of a CO molecule on the surface,
followed by the reaction with an oxygen atom by this CO
forming CO2; this would leave behind an oxygen vacancy and
two electrons, i.e., a reduced surface. The electrons can be
assumed to be localized to two adjacent Ce ions reducing them
3
+
+
to the Ce state or be associated with the vacancy as an Fs
5
0
center and one Ce(III). In a following step, an SO2 molecule
adsorbs on the oxygen vacant site and the two electrons are
utilized for the first step of the reduction of SO2, which then
dissociates into O and SO. The SO can then in a similar
fashion be further reduced by two more electrons from an
additional oxygen vacancy forming elementary sulfur and an
oxidized surface. The efficiency of a catalyst following this
scheme would be favored by the availability of low-lying CT
excitations for the transfer of charge from the oxygens to the
cerium in order to facilitate the oxidation of the CO. In the
second step, however, the Ce ions, or Ce plus Fs center,
should have a low affinity for this electron in order to facilitate
the reduction of the SO2 molecule. There should hence be an
optimum electron affinity of the Ce ions for which redox
reactions such as the one in reaction 1 occur at the highest rate,
assuming all other conditions being constant.
2
-
Figure 6. Mean-square displacements (msd’s) as a function of depth
in the x, y, and z directions for Ce4+ and O in the undoped CeO
2-
2
(110) slab at 300 K. The bulk msd’s in the corresponding directions
are given for comparison. Each point on the slab curves is the average
of 9 values, each computed as the average msd during a 0.94 ps long
time interval (the total simulation time was 8.46 ps). Each error bar is
the standard deviation of these 9 values; no drift was observed. The
outermost point corresponds to a depth z ) 0.5 Å, i.e., it contains all
atoms between z ) 0 and z ) 0.5 Å.
3
+
3+
+
TABLE 3. Computed Charge-Transfer Energies and
Electron Affinities (eV), for Different Ce Sites in the Top
a
Layer of Ca-Doped (110) CeO
2
cluster type
n-fold Ce
CT
EA
5
5
5
6
6
6
7
A-(110)
B-(110)
C-(110)
A-(110)
B-(110)
C-(110)
-bulk
5
5
5
6
6
6
7
2.09
3.34
2.27
2.40
2.63
2.55
-0.41
0.70
Speculating in a possible explanation and a mechanism for
the increased OSC could thus be that the reduction and the
oxidation balance of cerium, Ce(+III)/Ce(+IV), has become
more accessible, and knowing that this balance is essential for
the OSC, this could be the key. The proposed mechanism is
thus: two Ce(+IV) are reduced to two Ce(+III) or the two
-0.05
-2.10
-0.58
-1.37
-2.54
a
+
The 6-fold coordinated Ce in pure (110) has a computed C-T
electrons are distributed over one cerium and the Fs center,
energy of 4.18 eV and E A ) -1.14 eV. The different clusters are
through the reaction of a surface oxygen with e.g. the CO
molecule in the SO2 desulfurization reaction, creating an oxygen
vacancy and a reduced surface. Then, as the surface has been
reduced, it has the ability to pick up oxygen (from various
oxygen-containing molecules). The way this can be achieved
denominated according to Figure 2.
shows a rather similar picture. In the undoped (110) CeO2
surface, the 6-coordinated cerium has a calculated electron
affinity of -1.14 eV, i.e., it prefers to remain as Ce(IV), while
doping results in a less negative electron affinity for all the
3+
is by the donation of electrons from the Ce to the oxygen-
containing molecule, which is looking to be reduced and releases
5
-coordinated and for one of the 6-coordinated ceriums. This
2-
an O to the surface.
would imply that, at least the 5-coordinated ceriums should be
more easily reduced from +IV to +III and this could be one
contributing factor in explaining the higher catalytic activity of
the doped compared to the undoped sample.
From the computed CT energies and electron affinities it is
clear that the Ce(IV)/Ce(III) balance has been affected. How-
3+
-
ever, we still find the Ce O species to be higher in energy
for all structures considered in the present work, even though
the energy difference has become substantially reduced. Fur-
thermore, we have not been able to find any site with a hydrogen
affinity large enough to even approach the requirements for
hydrogen abstraction from methane, which is another reaction
proceeding more easily over the doped CeO2. It is clear that
IV. Discussion
The reactivity of metal oxides in general is largely dependent
on the amount and character of coordinatively unsaturated
exposed ions at the surface and especially the presence of

Ca-Doped CeO2 and Implications on Catalytic Activity
J. Phys. Chem. B, Vol. 103, No. 36, 1999 7635
the internal charge distribution in the crystal is determined by
the Madelung potential; if this becomes weaker through an
expansion of the lattice or a distortion of the lattice structure
the CT energy should go down. Thus, the present work suggests
that additional distortions could be important in order to explain
the observed reactivities. One such possibility could be through
larger-amplitude thermal motions at the surface where, e.g., an
anion moves out of the stabilizing crystal potential and thus
becomes more reactive. The combination of the MD simulations
with quantum chemistry allows for a direct calculation of the
electronic structure of representative, or particularly interesting,
instantaneous geometrical structures and, indeed, preliminary
studies of particular such structures do indicate an even lower
CT energy in some such cases and, as a result, a significant
reactivity toward, e.g., hydrogen abstraction and abstraction of
lattice oxygen using CO as in the desulfurization reaction. It
is clear that the combination of experimental and different
theoretical approaches involving both temperature and electronic
structure determinations is a prerequisite for the development
of a reliable and realistic picture of the complicated surface
processes going on in the complex oxide materials studied in
the present paper.
evaluation of pulsed neutron data, respectively. This work has
been financially supported by the Materials Consortium of
Clusters and Ultrafine Particles, the Swedish Research Council
for Engineering Sciences (TFR), and the Swedish Natural
Science Research Council (NFR). A.E.C.P. thanks the European
Union for an Access Grant under the TMR program for his
travel to ISIS.
Appendix: Cluster Basis Set
In all the cluster calculations, the basis set for the oxygen
atoms is a (9s 5p) primitive basis taken from ref 51, augmented
with one s and one p function (these extra functions have been
taken from ref 32 and two d polarization functions (exponents
R1 ) 0.5657, R2 ) 0.2828), so that the final contraction is
[5s4p2d]. For the study of charge-transfer excitations only one
5
0
d function (R ) 0.5657) was used.
For the cerium atom, an ab initio core model potential
5
2
(
AIMP) representation developed for the present work has
been used. These are based on a [Kr]-core AIMP (CG-
5
5,56
3
quasirelativistic
) and a (13s10p6d6f) valence basis set
optimized for the H state of the cerium neutral atom, corre-
2
0
2
sponding to the electronic configuration ([Xe]4f 5d 6s ), the
SCF ground state. This valence basis set, that is used to describe
the 4d 4f 5s 5p 6s valence of the cerium atoms, has been finally
contracted to [5s4p4d2f].
V. Summary and Conclusions
The present work combines molecular dynamics simulations
of both doped and undoped CeO2 with quantum chemical studies
of the electronic structure of embedded cluster models built from
the MD structures. The two theoretical techniques are applied
in an effort to understand the enhanced reactivity for e.g. SO2
decomposition and total combustion of methane that is found
experimentally upon doping of CeO2. Several interesting aspects
are studied which may bear upon the origin of this reactivity:
in particular, the finding of coordinatively unsaturated cerium
ions at all depths in the doped crystal leads to interesting
possibilities for CT reactions where the Ce(IV) may accept an
electron from a neighboring oxygen anion or some other electron
donor to become Ce(III).
The techniques involved are shown to give consistent results,
where in particular the agreement between the experimental
radial distribution functions from pulsed neutron scattering and
those obtained from the molecular dynamics simulations is very
striking. Furthermore, the experimentally observed bulk expan-
sion upon doping is also reproduced by the simulations. Thus,
the present theoretical description seems to give a very reliable
picture of the structural effects on the CeO2 lattice upon doping.
The structures generated through the molecular dynamics
simulations are directly fed into the quantum chemical modeling
calculations. Here the effects on the electronic structure from
the doping are studied using embedded cluster models based
on the MD results. The average, low-temperature geometries
obtained from the MD simulations are found to be close to
optimal also for the quantum chemical calculations, lending
further support to the chosen combination of techniques. The
computed CT energies leading to a Ce(III) species are found to
be significantly reduced, but still insufficient to by themselves
explain the observed reactivity of the material; the energy cost
is still of the order of 2-3 eV which is too high to lead to
exothermic or thermoneutral reactions. It is suggested that
additional temperature-controlled distortions could be important
for a complete understanding of chemical reactions involving
the doped and undoped CeO2.
A. Embedding Potentials and Computational Techniques.
3
5,36
The AIMP embedding approach
is based on a division of
the wave function for the crystal into a local part describing
the cluster and the external wave function. External ionic wave
functions, suitable to generate the embedding model potentials,
were obtained from self-consistent embedded ion (SCEI)
3
5,36
calculations.
Briefly, SCF wave functions appropriate for
4
+
2-
the embedded Ce and O ions are found and then used to
generate the corresponding total-ion model potentials. These give
an approximate description of the short-range Coulomb (in-
complete screening), exchange, and orthogonality interactions,
together with the major relativistic effects (Darwin and mass-
velocity potentials), obtained within the Cowan-Griffin ap-
5
7
proximation.
The SCEI calculations have been performed using the perfect
crystal structures taken from ref 13. The basis set used for the
Ce embedded ion calculations has been developed for this
work: a (11s10p6d) valence basis set together with a [Kr]-core
1
4+
CG-AIMP, optimized for the S state of Ce , totally uncon-
2
-
tracted; the basis set for the O ion is the O(53/5) basis set
from the compilation of Huzinaga and co-workers, augmented
with a p function describing the anion and totally uncontracted.
58
59
References and Notes
(1) Kilbourn, B. T. Cerium: A guide to its role in chemical technology;
Molycorp, Inc.: Fairfield, NJ, 1992.
(2) Kreuzer, T.; Lox, E. S.; Lindner, D.; Leyrer, J. Catal. Today 1996,
9, 17.
2
(3) Persson, A. E.; Zhang, Y.; Muhammed, M. Catalyst materials for
high-temperature processes, ceramic transactions 73. American Ceramic
Society, Westerville, OH, 1997; p 85.
(4) Sayle, T. X. T.; Parker, S. C.; Catlow, C. R. A. J. Chem. Soc.,
Chem. Commun. 1992, 977.
(
5) Sayle, T. X. T.; Parker, S. C.; Catlow, C. R. A. Surf. Sci. 1994,
3
16, 329.
(6) Conesa, J. C. Surf. Sci. 1995, 339, 337.
(7) Balducci, G.; Ka sˇ par, J.; Fornasiero, P.; Graziani, M.; Saiful Islam,
M. J. Phys. Chem. B 1997, 101, 1750.
8) Balducci, G.; Ka sˇ par, J.; Fornasiero, P.; Graziani, M.; Saiful Islam,
M. J. Phys. Chem. B 1998, 102, 557.
9) Zhang, Y.; Andersson, S.; Muhammed, M. Appl. Catal., B: 1995,
6, 325.
(
Acknowledgment. The authors thank Dr. Marcus Zwinkels
and Dr. Alex Hannon for help with catalytic reactor studies and
(

7636 J. Phys. Chem. B, Vol. 103, No. 36, 1999
de Carolis et al.
(
10) Palmqvist, A. E. C.; Zwinkels, M. F. M.; Zhang, Y.; J a¨ rås, S. G.;
Muhammed, M. Nanostruct. Mater. 1997, 8, 801.
11) Palmqvist, A. E. C.; Johansson, E. M.; J a¨ rås, S. G.; Muhammed,
M. Catal. Lett. 1998, 56, 69.
12) Palmqvist, A. E. C.; Berastegui, P.; Eriksson, S.; Hannon, A. C.;
Furenlid, L. R.; Baudin, M.; W o´ jcik, M.; Hermansson, K. To be published.
(37) J. A. Mejias; J. Fern a´ ndez Sanz. J. Chem. Phys. 1995, 102, 327.
(38) ECPAIMP is an integral program for ECP and AIMP calculations
written by L. G. M. Pettersson, L. Seijo, and M. A. Nygren.
(39) Alml o¨ f, J.; Faegri, K., Jr.; Feyereisen, M.; Korsell, K. DISCO, a
direct SCF and MP2 code.
(40) STOCKHOLM is a general purpose quantum chemical set of
programs written by P. E. M. Siegbahn, M. R. A. Blomberg, L. G. M.
Pettersson, B. O. Roos, and J. Alml o¨ f.
(
(
(
(
13) Wyckoff, R. W. G. Crystal Structures; Wiley: New York, 1963.
14) General Utility Lattice Program (GULP); developed by J. D. Gale,
Royal Institution of Great Britain/Imperial College, 1992-1995.
15) Gay, D. H.; Rohl, A. L. J. Chem. Soc., Faraday Trans. 1995, 91,
25.
(41) Chong, D. P.; Langhoff, S. R. J. Chem. Phys. 1986, 84, 5606.
(42) Palmqvist, A. E. C.; Wirde, M.; Gelius, U.; Muhammed, M. To be
published.
(43) Nygren, M. A.; Pettersson, L. G. M.; Freitag, A.; Staemmler, V.;
Gay, D. H.; Rohl, A. L. J. Phys. Chem. 1996, 100, 294.
(44) Vyas, S.; Grimes, R. W.; Gay, D. H.; Rohl, A. L. J. Chem. Soc.,
Faraday Trans. 1998, 94, 427.
(
9
3
1
(16) Baudin, M.; W o´ jcik, M.; Hermansson, K. Surf. Sci. 1997, 375,
74.
(17) Lewis, G. V.; Catlow, C. R. A. J. Phys. C.: Solid State Phys. 1985,
8, 1149.
(
18) Butler, V.; Catlow, C. R. A.; Fender, B. E. F.; Harding, J. H. Solid
(45) Sayle, T. X. T.; Parker, S. C.; Catlow, C. R. A. J. Phys. Chem.
1994, 98, 13625.
State Ionics 1983, 8, 109.
(
(
19) Dick, B. G.; Overhauser, A. W. Phys. ReV. 1958, 112, 90.
(46) Yashiro, H.; Ohuchi, T.; Eguchi, K.; Arai, H. J. Mater. Sci. 1988,
23, 1036.
20) Mitchell, P. J.; Fincham, D. J. Phys.: Condens. Matter 1993, 5,
1
031.
(47) Janssen, G. J. M.; Nieuwpoort, W. C. Phys. ReV. B 1988, 38, 3449.
(48) Wise, H.; Oudar, J. Materials Concepts in Surface ReactiVity and
Catalysis; Academic Press: San Diego, 1990.
(49) van Santen, R. A.; Niemantsverdriet, J. W. Chemical Kinetics and
Catalysis; Plenum Press: New York, 1995.
(50) Triguero, L.; de Carolis, S.; Nygren, M. A.; Pettersson, L. G. M.
To be published.
(51) Dunning, T. H., Jr. J. Chem. Phys. 1970, 53, 2823.
(52) Huzinaga, S.; Barandiar a´ n, Z.; Seijo, L.; Klobukowsky, M. J. Chem.
Phys. 1987, 86, 2132.
(
(
(
(
(
(
(
(
(
21) Ewald, P. P. Ann. Phys. 1921, 64, 253.
22) Parry, D. E. Surf. Sci. 1921, 49, 433.
23) Parry, D. E. Surf. Sci. 1976, 54, 195.
24) Cleveland, C. L. J. Chem. Phys. 1988, 89, 4987.
25) Parinello, M.; Raman, A. Phys. ReV. Lett. 1980, 45, 1196.
26) Parinello, M.; Raman, A. J. Appl. Phys. 1981, 52, 7182.
27) Nos e´ , S. J. Chem. Phys. 1984, 81, 511.
28) Hoover, W. G. Phys. ReV. A 1985, 31, 1695.
29) Heyes, D. M.; Barber, M.; Clarke, J. H. R. J. Chem. Soc., Faraday
Trans. 2 1977, 10, 1485.
30) Cordatos, H.; Bunluesin, T.; Vohs, J. M.; Gorte, R. J. J. Phys. Chem.
996, 100, 785.
31) Pascual, J. L.; Pettersson, L. G. M. Chem. Phys. Lett. 1997, 270,
51.
32) Nygren, M. A.; Pettersson, L. G. M.; Barandiar a´ n, Z.; Seijo, L. J.
Chem. Phys. 1994, 100, 2010.
(53) Seijo, L.; Barandiar a´ n, Z.; Huzinaga, S. J. Chem. Phys. 1989, 91,
7011.
(54) Barandiar a´ n, Z.; Seijo, L.; Huzinaga, S. J. Chem. Phys. 1991, 94,
3762.
(55) Barandiar a´ n, Z.; Seijo, L. Can. J. Chem. 1992, 70, 409.
(56) Barandiar a´ n, Z.; Seijo, L.; Huzinaga, S. J. Chem. Phys. 1990, 93,
5843.
(57) Cowan, R. D.; Griffin, D. C. J. Opt. Soc. Am. 1976, 66, 1010.
(58) Andzelm, J.; Klobukowsky, M.; Radzio-Andzelm, E.; Sakai, Y.;
Tatewaki, H. Gaussian basis sets for molecular calculations; Elsevier:
Amsterdam, 1984.
(
1
(
3
(
(33) M. P o¨ hlchen; Staemmler, V. J. Chem. Phys. 1993, 97, 2583.
(34) A. M. Ferrari; G. Pacchioni. Int. J. Quantum Chem. 1994, 58, 241.
(35) Barandiar a´ n, Z.; Seijo, L. J. Chem. Phys. 1988, 89, 5739.
(36) Barandiar a´ n, Z.; Seijo, L. In Studies in Physical and Theoretical
Chemistry: Vol. 77(B), Computational Chemistry: Structure, Interactions
and ReactiVity; Fraga, S. Ed.; Elsevier: Amsterdam, 1992; p 435.
(59) Dunning, T. H., Jr.; Hay, P. J. In Modern Theoretical Chemistry;
Schaeffer, H. F., III, Ed.; Plenum: New York.