The steady-state of heterogeneous catalysis, studied by first-principles statistical mechanics
Extracted source text ยท arXiv cond-mat/0408080
\documentclass[prl,aps,twocolumn,showpacs,home,floats]{revtex4}
\usepackage{graphics}
\bibliographystyle{apsrev}
\begin{document}
\title{{\rm\small\hfill (Phys. Rev. Lett., accepted)}\\
The steady-state of heterogeneous catalysis,\\
studied by first-principles statistical mechanics}
\author{Karsten Reuter$^{1,2}$}
\author{Daan Frenkel$^{2}$}
\author{Matthias Scheffler$^{1}$}
\affiliation{$^1$ Fritz-Haber-Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, D-14195 Berlin, Germany}
\affiliation{$^2$ FOM Institute AMOLF, Kruislaan 407, SJ1098 Amsterdam, The Netherlands}
\begin{abstract}
The turn-over frequency of the catalytic oxidation of CO at
RuO$_2$(110) was calculated as function of temperature and
partial pressures using {\em ab initio} statistical mechanics.
The underlying energetics of the gas-phase molecules, dissociation,
adsorption, surface diffusion, surface chemical reactions, and
desorption were obtained by all-electron density-functional theory.
The resulting CO$_2$ formation rate [in the full ($T, p_{\rm CO},
p_{\rm O_2}$)-space], the movies displaying the atomic motion and
reactions over times scales from picoseconds to seconds, and the
statistical analyses provide insights into the concerted actions
ruling heterogeneous catalysis and open thermodynamic systems in
general.
\end{abstract}
\received{20 January 2004}
\pacs{82.65.+r, 68.43.Bc, 68.43.De, 82.20.Db}
% 82.65.+r Heterogeneous catalysis at surfaces, surf. chemistry
% 68.43.Bc Ab initio calculations of adsorbate structure and reactions
% 68.43.De Statistical mechanics of adsorbates
% 82.20.Db Transition state theory and statistical theories of rate
% constants
\maketitle
Under realistic conditions materials surfaces are in contact with a
rich environment \cite{Stampfl-SS500-2002}. Often, the resulting
surface composition and surface-actuated material function are
determined by equilibrium thermodynamics. A first-principles
description is then possible with the ``{\em ab initio} atomistic
thermodynamics'' approach, that has been successfully applied to
various systems (see e.g. Refs. \cite{Stampfl-SS500-2002,
Reuter-PRB-2002,Reuter-PRL/B-2003,Michaelides-2003,Norskov-2003}
and references therein). However, under many conditions
equilibrium thermodynamics does not provide the appropriate
description, and heterogeneous catalysis is a particularly interesting
and important example. Here one is dealing with an open system, i.e.,
a supply of gases or liquids comes into contact with a solid surface,
where a chemical reaction produces a new substance that is then
transported away. The entire concert of the various underlying,
interlinked atomistic processes is in this case determined by kinetics.
Still, the temperature and partial pressures of the reactants must be set
such that the system runs under {\em steady-state} conditions. Only
then the catalyst is not getting destroyed, but is stably enhancing
the rate of the desired chemical reaction. An {\em ab initio}
description of the full steady-state situation of heterogeneous
catalysis has not been achieved so far, and a microscopic
understanding of the competing and concerting actions of the various
atomistic processes is lacking.
The present paper describes the ``{\em ab initio} statistical
mechanics'' methodology appropriate for an open thermodynamic system
with a continuous conversion of chemicals $A$ and $B$ into $C$, using
the oxidation of CO at a RuO$_2$ model catalyst as example. We employ
density-functional theory (DFT) to obtain the energetics of all
relevant processes, as there are: motion of the gas-phase molecules,
dissociation, adsorption, surface diffusion, surface chemical
reactions, and desorption. These calculations use the all electron,
full-potential linear augmented plane wave (FP-LAPW) approach
\cite{wien,basisset}. The only notable approximations are the generalized
gradient approximation (GGA) for the exchange-correlation functional
\cite{PBE}, and the assumption that transition-state theory (TST)
\cite{Ruggerone-1997} is applicable. In fact, both approximations are well
justified for the present study, and we will particularly address the GGA below
when analyzing the results. A combination of DFT with TST, and subsequently
solving the statistical mechanics problem by the kinetic Monte Carlo (kMC)
approach \cite{Ruggerone-1997,Landau-2000}, has been employed before
(see, e.g. Refs. \cite{Ruggerone-1997,Ovesson-1999,Kratzer-2002,Neurock-2000}).
In distinction to (by now standard) ``empirical kMC'' calculations,
that use just a few {\em effective parameters}, which have only limited
(if any) microscopic meaning, these ``{\em ab initio} kMC'' calculations
include an extended set of elementary processes with full and direct
physical meaning. And describing an open system with a continuous
conversion of chemicals $A$ and $B$ into $C$ implies additional
complexity. For the example discussed below (the catalytic oxidation
of CO) we will solve the statistical mechanics problem, gaining
microscopic insight into the full dynamics from picoseconds to
microseconds and even to seconds. The reported results demonstrate
the new quality of and the novel insights gained by such description.
We will also compare the results to those obtained by ``constrained
thermodynamics'' \cite{Reuter-PRL/B-2003}. Although (as expected)
noticeable differences occur for certain environmental conditions, we
will confirm that the much simpler ``constrained thermodynamics''
approach can provide important guidance on where in $(T,p)$-space
high reaction rates are likely to be expected, and when results
obtained for certain $(T,p)$ conditions can be extrapolated to others.
The now obtained turn-over-frequencies (TOFs) of the catalytic oxidation
of CO are in unexpected and unprecedented agreement with experimental
results of Peden and Goodman \cite{Peden-1986} and Wang {\em et al.}
\cite{Wang-2002}. This is found to be a consequence of the fact that
the high reaction conditions are not just ruled by a singular chemical
reaction pathway. Instead, for the steady-state and high TOF conditions
it is necessary that various processes play together in a most efficient
manner. For the present example one crucial point is to realize an
optimum disordered and dynamic ``phase'' at the surface. Modest
errors due to the approximate treatment of the exchange-correlation
functional (e.g. the GGA) are then not crucial, as long as the trends
of the energetics of the various interplaying processes are described
correctly. In this sense, the frequently requested {\em chemical
accuracy} for the description of individual processes appears to be a
misleading concept. At least for the present system, a careful
combination of DFT and statistical mechanics is more important.
\begin{figure}[b!]
\scalebox{0.26}{\includegraphics{fig1.ps}}
\caption{\label{snapshots}
a) Perspective view of the RuO${}_2$(110) surface, illustrating the two
prominent adsorption sites in the rectangular surface unit cell. These
sites are labeled as br (bridge) and cus (coordinatively unsaturated) site,
and both are occupied in the example with oxygen atoms. This is the ``high
$p_{\rm O_2}$ termination'' \cite{Reuter-PRB-2002,Reuter-PRL/B-2003}.
b) Time evolution of the site-occupation numbers from the
thermodynamic surface termination shown in a) to the steady-state
of catalysis [cf. snapshot c)]. The temperature is $T= 600$\,K
and the CO and O$_2$ partial pressures are that of the optimum
reaction conditions: $p_{\rm CO} = 20$\,atm, $p_{\rm O_2} = 1$\,atm.
c) Snapshot (top view of the RuO$_2$(110) surface) from movies
displaying the dynamics of the surface under realistic catalytic
conditions. Though the full atomic structure is considered in the
calculations, for clarity we have marked here the substrate bridge
sites only by gray stripes and the cus sites by white stripes. Oxygen
adatoms are drawn as red circles and adsorbed CO molecules as blue
circles. Movies for various pressure conditions for runtimes of
500 ns, as well as 1 s can be found at
${\rm http:} \slash \slash$w3.rz-berlin.mpg.de$\slash\!\!\sim
$reuter$\slash$movies$\slash$movies.html.}
\end{figure}
\begin{table}
\caption{\label{table1}
DFT binding energies, $E_b$, for CO and O (with respect to (1/2)O$_2$)
at br and cus sites (cf. Fig. \ref{snapshots}a), diffusion energy
barriers, $\Delta E^{\rm b}_{\rm diff}$, to neighboring br and cus
sites, and reaction energy barriers, $\Delta E^{\rm b}_{\rm reac}$, of
neighboring species at br and cus sites. All values are in eV.}
\begin{ruledtabular}
\begin{tabular}{l|c@{\quad}|cc@{\quad}|cc}
& $E_b$ & \multicolumn{2}{c|}{$\Delta E^{\rm b}_{\rm diff}$} &
\multicolumn{2}{c}{$\Delta E^{\rm b}_{\rm reac}$} \\
& & to br & to cus & with CO$^{\rm br}$ &
with CO$^{\rm cus}$ \\ \hline
CO$^{\rm br}$ & -1.6 & 0.6 & 1.6 & - & - \\
CO$^{\rm cus}$ & -1.3 & 1.3 & 1.7 & - & - \\
\hline
O$^{\rm br}$ & -2.3 & 0.7 & 2.3 & 1.5 & 1.2 \\
O$^{\rm cus}$ & -1.0 & 1.0 & 1.6 & 0.8 & 0.9 \\
\end{tabular}
\end{ruledtabular}
\end{table}
The methodology will be used to study CO oxidation over RuO${}_2$(110)
(the surface is sketched in Fig. \ref{snapshots}a), as this system has
recently received considerable attention as a highly active model
catalyst (see Ref. \cite{Wang-2002} and references therein). In fact,
this was previously called the Ru catalyst, but recent experimental
and theoretical work has shown that at realistic O$_2$ pressure the
Ru(0001) surface is transformed into an epitaxial RuO$_2$(110) film
(see Ref. \cite{Reuter-CPL-2002} and references therein). It is by
now also established that this RuO$_2$(110) surface actuates the
catalytic reaction, and, although domain boundaries and steps are
present, their influence is not significant \cite{Jacobi-Ertl}.
The surface unit cell is rectangular and contains two adsorption
sites \cite{Reuter-PRL/B-2003}: bridge (br) and coordinatively
unsaturated (cus) sites (cf. Fig. \ref{snapshots}a). Either of them can
be empty or occupied by O or CO, and adsorbate diffusion can go br-to-br,
br-to-cus, cus-to-cus, or cus-to-br. Since this comprises the possibility
of missing O$^{\rm br}$ atoms, we note that our treatment implicitly
includes the effect of O surface vacancies. A total of 26
different elementary processes are possible on this lattice, and all
were carefully analyzed by DFT to obtain their pathways and
energy barriers \cite{Reuter-PRL/B-2003,long-version}. Table \ref{table1}
summarizes the adsorption energies and the diffusion and reaction
energy barriers used for the kMC study, as obtained from DFT
calculations with a $(1 \times 2)$ surface unit-cell. In a systematic
study of CO and O (co)adsorption at various coverages we found
adsorbate-adsorbate interaction to be always smaller than 150\,meV.
Thus, lateral interactions in this system are small (compared to
the other energies), and will therefore be neglected.
CO adsorption into vacant cus or bridge sites is non-dissociative, while
oxygen adsorption is dissociative and requires two vacant neighboring
sites, i.e. a br-br, cus-cus, or br-cus pair. The adsorption rate per
free site is given by the local sticking coefficient, $\tilde S$,
and the kinetic impingement: $\Gamma_{\rm ad} = {\tilde S} \, p /
\sqrt{2\pi m k_{\rm B} T}$. Here $m$ is the mass of the gas-phase
molecule, and $k_{\rm B}$ is the Boltzmann constant. The rate of the
time reversed process (desorption) then follows from the relation:
$\Gamma_{\rm des} / \Gamma_{\rm ad} = {\rm exp} \left( (F_{b} - \mu)/
(k_{\rm B}T) \right)$, where $F_{b}$ is the free energy of the
adsorbed species (approximated by $E_b$), and $\mu(T, p)$ is the
chemical potential of the gas-phase molecule \cite{Reuter-PRB-2002}.
The $\tilde{S}(T)$ are thus obtained from the calculated total energy surfaces
of desorption together with detailed balance. The only uncertainty
here arises from the vibrational properties of the transition state
which translates into uncertainties in the desorption rate by a factor
of 10 (at most 100). For the diffusion processes we use a
prefactor of $10^{12}$\,Hz, which has an uncertainty of a factor of 10.
We carefully checked that these uncertainties do not affect our below
reported results and conclusions.
%We emphasize that the resulting effects on the
%reaction rates and on our conclusions are marginal, e.g. changing the
%pressure scale in Fig. \ref{TOF} by only $2 \cdot 10^{-10}$\,atm.
Details of the employed new methodology and of the various test calculations
will be published elsewhere \cite{long-version}. Detailed balance
of the scenario was carefully checked by confirming that the earlier results obtained by ``atomistic thermodynamics'' \cite{Reuter-PRL/B-2003} are exactly
reproduced by the present statistical mechanics treatment, if surface reactions are not allowed to occur.
Kinetic Monte Carlo runs were performed for about 1000 different
$(T,p_{\rm CO},p_{\rm O_2})$ conditions covering the temperature
range $300\,{\rm K} < T < 800\,{\rm K}$ and partial pressures from
$10^{-10}$ to 10$^{3}$ atmospheres. Several calculations were
done on a system with $(30 \times 30)$ surface sites (450 bridge
plus 450 cus sites), but the vast amount of calculations was done
for a 400 sites system. The results for both system sizes were
identical. The kMC simulations were run until steady-state is
reached (cf. Fig. \ref{snapshots}b). Then the movies were recorded
(cf. Fig. \ref{snapshots}c) and the statistical analyses of the
frequencies of the various elementary processes performed. The
latter also gives the total TOFs of the CO$_2$ formation.
The results show that catalytic reaction conditions are only
established when both CO and O$_2$ partial pressures exceed certain
values below which the surface is in a thermodynamic equilibrium
phase of RuO${}_2$(110), characteristic for low oxygen pressures
and routinely observed under ultra-high vacuum (UHV) conditions:
All bridge sites are occupied by oxygen atoms and all cus sites are
empty. For higher pressures three surface conditions are worth
mentioning, which are e.g. obtained for $T = 600$\,K, $p_{\rm O_2}
= 1$\,atm, and $p_{\rm CO} = 10^{-2}, 20,$ and $10^{3}$\,atm.
$(i)$ At the low $p_{\rm CO}$, all bridge and all cus sites are
covered with oxygen atoms (cf. Fig. \ref{snapshots}a). This is
essentially the thermodynamic high $p_{\rm O_2}$ phase
\cite{Reuter-PRB-2002,Reuter-PRL/B-2003}. There is a noticeable
desorption/adsorption dynamics, i.e., about every 40\,$\mu$s two of
the 900 O adatoms of the $(30 \times 30)$ simulation cell desorb
(mainly from cus sites) as O$_2$. The resulting vacancies are then
rapidly filled again (within 1\,ns), most of the time with oxygen.
Only rarely CO adsorbs, and even if it does, it rather desorbs again
than initiating a reaction. The overall CO${}_2$ formation rate under
these conditions is with $0.9\!\cdot\!10^{12}$ cm$^{-2}$\,s$^{-1}$ very
low.
$(ii)$ For $p_{\rm CO} = 20$\,atm the situation corresponds to
what was previously suggested to be that of high chemical
activity \cite{Reuter-PRL/B-2003}, and this suggestion is now
confirmed. The time to reach steady-state is remarkably long, in particular when
compared to the pico second timescale of the underlying atomistic
processes (cf. Fig. \ref{snapshots}b). Finally, an interesting mix
of CO molecules and O atoms at bridge and cus sites is established
(a typical snapshot is shown in Fig. \ref{snapshots}c),
with average occupation numbers $N_{\rm CO}^{\rm br} = 0.11$,
$N_{\rm CO}^{\rm cus} = 0.70$, $N_{\rm O}^{\rm br} = 0.89$, and
$N_{\rm O}^{\rm cus} = 0.29$. The dynamics of this surface is extremely
fast. It is mainly due to CO desorption and adsorption. The rate of
CO$_2$ formation is a factor of 0.0004 lower, but truly significant, namely
$4.6 \cdot 10^{18}\,{\rm cm}^{-2}\,{\rm s}^{-1}$. It is dominated by
the reaction CO$^{\rm cus}+ {\rm O}^{\rm cus} \longrightarrow {\rm CO}_2$.
The CO$^{\rm br} + {\rm O}^{\rm cus} \longrightarrow {\rm CO}_2$ and
CO$^{\rm cus}+ {\rm O}^{\rm br} \longrightarrow {\rm CO}_2$ reactions
also contribute, though their rates are by a factor of 0.30 and 0.01
lower than the first one. The CO$^{\rm br} + {\rm O}^{\rm br}
\longrightarrow {\rm CO}_2$ reaction is insignificant. This result is
different from what one would expect from the reaction energy barriers
(cf. table \ref{table1}), which would give the lead to CO$^{\rm br}+
{\rm O}^{\rm cus}$, demonstrating the importance of the proper
mix of surface-site occupations.
$(iii)$ When the CO pressure is increased further, to $p_{\rm CO} =
10^3$\,atm, the surface becomes fully covered with CO. This result is
at variance with the ``constrained thermodynamics'' study in which
the CO$_2$ formation was forbidden. Now we find that catalysis is
practically poisoned by adsorbed CO. The encountered situation is in
fact close to that where the RuO$_2$ catalyst will be reduced by CO
to the pure Ru metal.
\begin{figure}[b!]
\scalebox{0.38}{\includegraphics{fig2.ps}}
\caption{\label{TOF}
a) Steady-state surface structures obtained by {\em ab initio} kMC
calculations at $T = 600$\,K and various CO and O$_2$ partial
pressures. In all non-white areas, the average site occupation
is dominated ($> 90$\,\%) by one species, i.e. either O, CO or
empty sites ($-$). b) Map of the corresponding turn-over frequencies
(TOFs) in ${\rm cm}^{-2}{\rm s}^{-1}$: White areas have a TOF\,$<
10^{11} {\rm cm}^{-2}{\rm s}^{-1}$, and each increasing gray level
represents one order of magnitude higher activity. Thus, the darkest
region corresponds to a TOF higher than $10^{17}\,{\rm cm}^{-2}{\rm s}^{-1}$.}
\end{figure}
\begin{figure}[t!]
\scalebox{0.38}{\includegraphics{fig3.ps}}
\caption{\label{Jacobi}
Rate of CO$_2$ formation at $T= 350$\,K. The experimental steady-state
results of Wang {\em et al.} \cite{Wang-2002} are presented as dotted lines,
and the theoretical results are shown as green, solid lines. Rates
given as function of $p_{\rm O_2}$ at $p_{\rm CO}= 10^{-10}$\, atm (left), and
as function of $p_{\rm CO}$ at $p_{\rm O_2}= 10^{-10}$\, atm (right).}
\end{figure}
Figure \ref{TOF} summarizes the results by showing the various
steady-state surface structures as well as a map of the TOFs in
($p_{\rm CO}$, $p_{\rm O_2}$)-space. The highest activity is found
to be in very good agreement with the early experimental results by
Peden and Goodman \cite{Peden-1986}, and occurs whenever the
environmental conditions lead to the dynamic coexistence ``phase''
described above under $(ii)$. Recently, Wang {\em et al.} \cite{Wang-2002}
have also measured the TOFs for the RuO$_2$(110) surface at
$T =350$\,K for various pressures, and in Fig. \ref{Jacobi} we compare
their results with ours. The agreement is again far better than what one
would have expected: The theoretical and experimental TOF values at the
optimum pressures are practically identical, and also the optimum
pressure conditions (when the ``dynamic phase'' is realized) agree
very well.
Because the errors due to the GGA are in the range of 0.1-0.3\,eV, this
good agreement between theory and experiment, seen in Fig. \ref{Jacobi},
may seem fortuitous. However, it is worth noting that the position of
the optimum catalytic efficiency in $(T,p)$-space and to some extent also
the value of the TOF are not determined by the energetics of a singular
process alone, but by the action of many players. Apparently,
the DFT-GGA calculations describe the differences between the various
surface processes better than the individual absolute values. In this
respect it is important to realize that a combination of different
calculations (employing different approximations) or of theory and
experiment could have spoiled the description. The consistent treatment
of all participating processes implies that the {\em optimum mix} of O
and CO at the surface is described well: The abundance of
CO$^{\rm cus}$-O$^{\rm cus}$ nearest neighbor pairs (as well as
CO$^{\rm br}$-O$^{\rm cus}$ nearest neighbor pairs) is apparently
playing a role of similar importance as the energy barriers. Particularly
at the optimum TOF conditions there is also an effective compensation, e.g.
when too high adsorption energies result in enhanced adsorption, but
also in reduced reaction barriers etc. Away from the optimum TOF
conditions such compensation effects become less effective, and modest
differences between the theoretical and experimental results arise.
Here DFT-GGA errors are more influential, and for the experimental
data we expect that contributions from surface imperfections may
play a bigger role.
In summary, we computed the surface kinetics of CO oxidation catalysis
at RuO$_2$(110). The TOFs are presented in $(T,p_{\rm CO},p_{\rm
O_2}$)-space, clearly identifying a narrow region of highest catalytic
activity. In this region kinetics builds an adsorbate composition
that is not found anywhere in the thermodynamic surface phase diagram.
The statistical analysis of the surface dynamics and of the various
processes reveals several surprising results. For example, the
chemical reaction with the most favorable energy barrier happens
a factor of 0.30 less frequently than the energetically second
favorable reaction.
The results also clarify how and when a bridging of the pressure gap
between UHV studies and realistic pressure conditions is possible. The
considered system has in fact two important components to this issue:
At first, an O$_2$-rich environment changes the material from Ru to
RuO$_2$. Thus, earlier high pressure studies on the Ru catalyst were
actually looking at RuO$_2$. Second, after RuO$_2$ has been formed it
is important to know how to set the pressure conditions correctly, as
also on RuO$_2$(110) there is a low-pressure surface phase (the orange
region in Fig. \ref{TOF}a), that has little in common with the
catalytically active situation. The obtained agreement between the
theoretical results and experimental data confirms furthermore (cf.
also Ref. \cite{Wang-2002}) that CO$_2$ is primarily formed from
{\em adsorbed} CO and O, and that the metal-oxide, once it was created,
does not play an active role, i.e., there is no indication of significant
bulk diffusion. Thus, the catalysis is explained in terms of a
Langmuir-Hinschlwood mechanism. Some contributions via the Eley-Rideal
mechanism (e.g. scattering of gas-phase CO at O$^{\rm cus}$) may
play a minor role, but this process requires further calculations and
experiments. With the advent of first-principles TOF-maps obtained from
DFT calculations (cf. Fig. \ref{TOF}), a more detailed data base is
in general becoming available for comparison with experiments, which will
eventually advance the microscopic understanding of catalysis.
\begin{references}
\bibitem{Stampfl-SS500-2002}
C. Stampfl {\em et al.}, Surf. Sci. {\bf 500}, 368 (2002).
\bibitem{Reuter-PRB-2002}
K. Reuter and M. Scheffler, Phys. Rev. B {\bf 65}, 035406 (2002).
\bibitem{Reuter-PRL/B-2003}
K. Reuter and M. Scheffler, Phys. Rev. Lett. {\bf 90}, 046103 (2003);
Phys. Rev. B {\bf 68}, 045407 (2003).
\bibitem{Michaelides-2003}
A. Michaelides {\em et al.}, Chem. Phys. Lett. {\bf 367}, 344 (2003).
\bibitem{Norskov-2003}
Z. {\L}odzianan and J.K. N{\o}rskov, J. Chem. Phys. {\bf 118}, 11179 (2003).
\bibitem{wien}
P. Blaha, K. Schwarz and J. Luitz,
{\em WIEN97}, Techn. Univ. Wien, Austria (1999); ISBN 3-9501031-0-4.
\bibitem{basisset}
The FP-LAPW approach is one of the most accurate DFT
methods for describing polyatomic systems. Basis set, super cell
setup, and the transition state location procedure by mapping relevant
reaction coordinates have been described before \cite{Reuter-PRL/B-2003}.
\bibitem{PBE}
J.P. Perdew, K. Burke, and M. Ernzerhof,
Phys. Rev. Lett. {\bf 77}, 3865 (1996).
\bibitem{Ruggerone-1997}
P. Ruggerone, C. Ratsch, and M. Scheffler,
In: The Chemical Physics of Solid Surfaces Vol. {\bf 8}, D.A. King, D.P. Woodruff (Eds.), Elsevier Science, Amsterdam (1997).
\bibitem{Landau-2000}
D.P. Landau and K. Binder,
{\em A guide to Monte Carlo Simulations in Statistical Physics},
Cambridge University Press, Cambridge (2000).
\bibitem{Ovesson-1999}
S. Ovesson, A. Bogicevic, and B.I. Lundqvist,
Phys. Rev. Lett. {\bf 83}, 2608 (1999).
\bibitem{Kratzer-2002}
P. Kratzer and M. Scheffler,
Phys. Rev. Lett. {\bf 88}, 036102 (2002).
\bibitem{Neurock-2000}
E.W. Hansen and M. Neurock, J. Catal. {\bf 196}, 241 (2000);
Surf. Sci. {\bf 464}, 91 (2000).
\bibitem{Peden-1986}
C.H.F. Peden and D.W. Goodman, J. Phys. Chem. {\bf 90}, 1360 (1986).
\bibitem{Wang-2002}
J. Wang {\em et al.}, J. Phys. Chem. B {\bf 106}, 3422 (2002).
\bibitem{Reuter-CPL-2002}
K. Reuter {\em et al.}, Chem. Phys. Lett. {\bf 352}, 311 (2002).
\bibitem{Jacobi-Ertl}
K. Jacobi and G. Ertl, private communication; H. Over and
M. Muhler, Prog. Surf. Sci. {\bf 72}, 3 (2003).
\bibitem{long-version}
K. Reuter {\em et al.}, to be published.
\end{references}
\end{document}