Surface Science 544 (2003) 5–23

www.elsevier.com/locate/susc

A microkinetic model of the methanol oxidation over silver

A. Andreasen 1, H. Lynggaard, C. Stegelmann, P. Stoltze *

Department of Chemistry and Applied Engineering Science, Aalborg University, Niels Bohrs Vej 8, 6700 Esbjerg, Denmark
Received 14 May 2003; accepted for publication 5 August 2003

Abstract
A simple microkinetic model for the oxidation of methanol on silver based on surface science studies at UHV and
low temperatures has been formulated. The reaction mechanism is a simple Langmuir–Hinshelwood mechanism, with
one type of active oxygen and one route to formaldehyde and carbon dioxide, respectively. The model explains ob-
served reaction orders, selectivity, apparent activation enthalpies and the choice of industrial reaction conditions. More
interesting the model disproves the notion that the mechanism deduced from surface science in UHV cannot be
responsible for formaldehyde synthesis at industrial steady-state conditions. The present work therefore seriously
questions the prevailing models of formaldehyde synthesis in the literature. One of the reasons for this controversy is
that many of the models in the literature are derived from transient experiments exhibiting dynamic effects that are not
present at steady state under industrial conditions.

2003 Elsevier B.V. All rights reserved.

Keywords: Computer simulations; Models of surface kinetics; Equilibrium thermodynamics and statistical mechanics; Oxidation;
Catalysis; Silver; Alcohols; Oxygen

1. Introduction more water than hydrogen is produced [1,9].

Formaldehyde is only formed in the presence of
The partial oxidation of methanol to formal- oxygen [2]. Steam is added to increase selectivity
dehyde is an important industrial process due to and heat transport [1,3,4]. Traditionally the overall
the versatility of formaldehyde as an intermediate process is regarded as two parallel reactions: an
in chemical synthesis [1]. BASFÕs silver based oxidation (Eq. (1)) and a dehydrogenation (Eq.
process is carried out at
900 K and atmospheric (2)) [3–5].
pressure, the feed consist of a fuel-rich mixture of
methanol and air. At typical reaction conditions 1
CH3 OH þ O2
H2 CO þ H2 O ð1Þ
the selectivity is approximately 90% and the con- 2
version of oxygen approach 100%, and slightly
CH3 OH
H2 CO þ H2 ð2Þ

*
It has been proposed [1,3,4,6] that the selectivity
Corresponding author. Tel.: +45-79-12-76-63 (Office); fax:
+45-75-45-36-43 (Department).
towards formaldehyde is limited by the following
E-mail address: stoltze@aue.auc.dk (P. Stoltze). reactions:
1
Present address: Department of Materials Research, Risoe
National Laboratory, DK-4000 Roskilde, Denmark. 2CH3 OH þ 3O2
4H2 O þ 2CO2 ð3Þ

0039-6028/$ - see front matter
2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.susc.2003.08.007
6 A. Andreasen et al. / Surface Science 544 (2003) 5–23
H2 CO þ O2
CO2 þ H2 O ð4Þ Langmuir–Hinshelwood kinetic expression based
H2 CO
CO þ H2 ð5Þ
The formation of carbon dioxide is favored by low
temperatures [7–9]. The formation of carbon
monoxide is favored by high temperatures
(T > 900 K) [8], and is viewed as a pyrolytic gas
phase reaction [4,6].
Despite the fact that the formaldehyde synthesis
has existed for more than a century and a sub-
stantial research effort has been devoted to the
reaction, the mechanism remains controversial.
Based on UHV studies on Ag(1 1 0) at 200–300
K Wachs and Madix [10] proposed a simple
mechanism for methanol oxidation with only one
kind of active oxygen. According to this mecha-
nism the oxidation of methanol to formaldehyde
goes through a methoxy intermediate. The meth-
oxy intermediate is formed by a reaction between
methanol and atomic surface oxygen and decom-
pose to formaldehyde and hydrogen. The formal-
dehyde may be oxidized to carbon dioxide through
a formate intermediate i.e. a consecutive reaction
path (Eq. (4)). Wachs and Madix results shows no
indications of a direct reaction pathway from
methanol to carbon dioxide i.e. a parallel reaction
pathway (Eq. (3)) as proposed by some investiga-
tors [8,11]. The methoxy and formate intermedi-
ates have been identified and studied with in situ
Raman, HREELS, XPS, UPS, and TPR on both
single crystals and polycrystalline silver catalysts
[10,12–17]. Surface science and density functional
theory (DFT) calculations indicates that the mech-
anistic scheme of Wachs and Madix is general for
the oxidation of alcohols on silver and other metal
surfaces [12,18–20].
Bhattacharyya et al. [2] proposed a Mars–van
Krevelen mechanism [21] from differential reactor
data at 264–290 C. They observe positive reaction
orders with respect to both methanol and oxygen
and an inhibiting effect caused by water. Robb and
Harriott [11] have performed kinetic investigations
in a differential reactor at 420 C. The authors
observe approximately zero order with respect to
oxygen,
0.8 order with respect to methanol and
decreasing selectivity with increasing conversion.
On the basis of these experiments a modified
on adsorption on top of an oxide layer was pro-
posed. The selectivity was explained by total oxi-
dation of both methanol and formaldehyde.
During the last decade Schl€ ogl and coworkers
have studied the oxygen–silver system at high
temperatures (500–1000 K) and atmospheric pres-
sures in great detail (TPD, TPR, ISS, XPS, UPS,
in situ-XRD, STM, SEM, RHEED, etc.) [22–27].
The authors have identified three different atomic
oxygen species above room temperature, denoted
Oa , Ob , and Oc which desorbs in UHV at 600, 600–
850, and
900 K, respectively. Oa (XPS (O 1s) ¼
528.2 eV) is formed by dissociation of molecular
oxygen and is the well known chemisorbed surface
bound oxygen, also termed nucleophilic or ionic
oxygen in the literature [28–31]. Oa is the active
oxygen species in the mechanism proposed by
Wachs and Madix [10]. Ob (XPS (O 1s) ¼ 530.3 eV)
[26] is formed by dissolution of Oa in the bulk.
This formation is activated and occurs via inter-
stitial diffusion at a temperature above 450 K
[23,27]. Oc (XPS (O 1s) ¼ 529.0 eV) is embedded in
the uppermost layer of silver and is formed by the
segregation of Ob from the bulk to the surface via
interstitial diffusion above 580 K [23]. This for-
mation only occurs in the terminating closed
packed surface planes (Ag(1 1 1)) and leads to a
pronounced reconstruction and morphological
changes in silver catalysts [27]. Water induces
morphological changes presumable by formation
of subsurface OH by reaction with Oc [32–34]. Due
to reaction-induced restructuring of silver catalysts
and pronounced hysteresis Schl€ ogl et al. concluded
that formaldehyde synthesis is extremely structure-
sensitive [8]. However steady-state kinetics has not
been measured on different surface facets or par-
ticle sizes (1–5 nm), hence no conclusion on the
structure sensitivity can be made according to the
definition of structure sensitivity by Sajkowski and
Boudard [35]. The conclusion of the measurements
is a strong dependence on the preconditioning of
the catalyst in transient experiments. This is con-
sistent with the work of Wachs and coworkers [36]
who showed that preoxidized silver surfaces ini-
tially results in more active catalysts, but after sev-
eral hours the same steady-state is reached with or
without the preoxidation. Schl€ ogl and coworkers
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
also claim that Oa due to its desorption at 600 K
cannot be the dominant surface species involved in
partial oxidation of methanol at
900 K [26]. In-
stead they suggest that Oc is active at industrial
conditions. However this conjecture is only valid
in UHV and Oa could easily be the active species at
industrial conditions. Indeed it will be shown in
this work that Oa is present at industrial condi-
tions and most likely the active oxygen at all re-
action conditions. This is in agreement with the
work of Wachs and coworkers [36] who showed
that only one type of atomic surface oxygen (TPD
peak at
600 K) is active in the formaldehyde
synthesis.
In recent investigations Nagy et al. [8] and Nagy
and Mestl [9] observe an increasing selectivity to-
wards formaldehyde with increasing temperatures.
At low temperatures the reaction is inactive; it
ignites above 450 K. These observations are con-
sistent with the investigations of Gavrilin and
Popov [7], Leffert et al. [37], Qian et al. [38] and
Obraztsov et al. [39]. With this in mind, Gavrilin
and Popov [7] and Nagy et al. [8] suggest different
kinetics at high and low temperatures. Nagy et al.
also claim that different kinds of oxygen are re-
quired in order to explain their experiments [8]. At
low temperatures Oa is active in oxi-dehydroge-
nation of methanol towards formaldehyde. How-
ever due to the strong nucleophility of Oa this
reaction tends to go to complete oxidation prod-
ucts. At higher temperatures Oc is believed to
catalyze the selective dehydrogenation of methanol
to formaldehyde. Thus explaining increasing se-
lectivity with temperature. However the experi-
ments of Nagy et al. and Gavrilin and Popov was
carried out with 100% oxygen consumption and
kinetics can therefore not be deduced from these
experiments yet alone reaction mechanisms.
In our view the strongest evidence for the
participation of Oc in formaldehyde synthesis
was recently published by Muhler and coworkers
[40]. By using a temporal-analysis-of-products ap-
proach (TAP) Muhler et al. showed that Oc does
react with methanol to form formaldehyde with
high selectivity. Oa also formed formaldehyde but
with a lower selectivity. However the experiments
with Oa was conducted at lower temperatures than
the experiments with Oc and the lifetime of form-
7
aldehyde was significantly longer in these low
temperature experiments. Assuming CO2 is formed
by a consecutive reaction of formaldehyde a lower
selectivity is expected due to the longer lifetime of
formaldehyde. Decreasing the coverage of Oa led
to a significant increase in selectivity which seems
logical since more oxygen is needed to obtain total
oxidation products. Furthermore Oa also seems to
react more rapidly and more completely with
methanol than Oc and besides that the experiments
were transient and conducted without oxygen at-
mosphere. The experiments are therefore not
conclusive. It is very likely that the formation of
Oc is too slow to keep up with Oa at steady-state
conditions. A slow formation rate of Oc (slow
diffusion of Ob and Oc [27]) and hence a slow de-
sorption rate due to microscopic reversibility
seems to be the most reasonable explanation for
the high thermal stability of Oc . Schl€ ogl and co-
workers have shown that the amount of Oc de-
creased drastically in the presence of methanol
even in oxygen rich mixtures [26].
It is evident that there is a conflict between the
proposed mechanism deduced from UHV investi-
gations and the mechanisms deduced from studies
near industrial conditions. There can be two rea-
sons for this discrepancy. Either the reactions
proceeding at room temperature and UHV are
different from the reactions at industrial conditions
or the interpretation of the complex in situ ex-
periments at industrial conditions is flawed. We
believe the latter is the case because these experi-
ments includes transient effects that are not present
at industrial steady-state conditions. In this work
we will show that it is indeed possible to interpret
kinetic experiments performed at elevated pres-
sures and temperatures with a simple microkinetic
model consistent with the mechanism derived by
Wachs and Madix [10] and various surface science
experiments. Hence it is not necessary to invoke Oc
in order to explain formaldehyde synthesis. This
does not prove that Oc is not participating in the
reaction, but the effect is probably insignificant in
steady-state kinetics.
The approach of microkinetic modeling has
successfully been applied for a number of im-
portant industrial reactions [41–50], which proves
a close connection between surface science and
8 A. Andreasen et al. / Surface Science 544 (2003) 5–23

industrial catalysis. An important feature of a
microkinetic model is that thermodynamic and
kinetic parameters are physical meaningful and
consistent with experiment and/or theoretical
methods such as DFT and transition state theory
(TST). From a microkinetic model it is possible to
estimate surface coverages, reaction orders and
activation enthalpy during reaction conditions.
Hence the applicability of the model is not re-
stricted to a particular set of conditions, but can be
used under various conditions where simplified
models e.g. Power-Law expressions may break
down [44]. In short a microkinetic model should be
able to bridge the pressure-, temperature-, struc-
ture-, and reactor gaps between experimental and
theoretical surface science and industrial catalysis.
To the best of our knowledge this is the first
time microkinetic modeling has been applied to the
partial oxidation of methanol to formaldehyde on
silver catalysts.
2. Methods
The starting point of a microkinetic model is a
detailed reaction mechanism. The principle of
microscopic reversibility is applied to each ele-
mentary step, and the kinetics is described by
Arrhenius expressions. A statistical mechanical
description is used for all gas phase molecules and
adsorbates, giving a correct description of the de-
grees of freedom for each species. Furthermore
statistical thermodynamics ensures a correct de-
scription of the gas phase thermodynamics within
the ideal gas approximation. The pivot in statisti-
cal mechanics is the partition function from which
all thermodynamic information can be extracted
e.g. the equilibrium constants for each step in the
mechanism. Table 1 summarize the important
formulas used in the thermodynamic modeling.
2.1. Reaction mechanism
Our model is based on the Langmuir–Hinshel-
wood mechanism presented in Table 2. All the
elementary steps in Table 2 have have been ex-
tracted from the UHV work of Wachs and Madix
on Ag(1 1 0) [10]. It should be emphasized that O
participating in the mechanism in Table 2 corre-
sponds to the surface atomic oxygen denoted Oa
by Schl€ogl and coworkers. The reaction mecha-
nism has been kept as simple as possible by ig-
noring known elementary reactions involving
carbonate formation [51–53], OH formation [10],
HCOOH formation [38] and oxidative decompo-
sition of formate [54]. These elementary reactions
has been ignored for several reasons. First of all
the purpose of this work is not to develop a de-

Table 1
Statistical mechanical and thermodynamic functions applied in the microkinetic model
Type Partition function Enthalpy

 0
 1
Vibration mj hc mj hc
exp exp
Q 2k T P B1 2k T C
zvib ¼ j
B  Hvib ¼ jB
@2 hcmj þ
B C
mj hc mj hc A
1  exp 1  exp
kB T kB T

3
2pmkB T 2 kB T 5
Translation ztrans ¼ Htrans ¼ kB T
h2 p 2
kB T
Rotation (2D) zrot ¼ Hrot ¼ kB T
rhcB

3
1 kB T 2  p 12 3
Rotation (3D) zrot ¼ Hrot ¼ kB T
r hc ABC 2


Eg
Electronic zelec ¼ exp Helec ¼ Eg
kB T
Total z ¼ zvib ztrans zrot zelec H ¼ Hvib þ Htrans þ Hrot þ Helec
Eg is the ground state energy, h is PlanckÕs constant, kB is BoltzmannÕs constant, m is the molecular mass, p is the reference pressure
and c is the speed of light in vacuum. A, B and C are rotational constants, r is the symmetry number, mj are the vibrational frequencies.
 A. Andreasen et al. / Surface Science 544 (2003) 5–23 9

Table 2 limiting on copper [59]. Pepley et al. have pro-

Reaction mechanism used for the microkinetic model posed a kinetic model for methanol-steam re-
CH3 OH(g) + 
CH3 OH (step 1) forming on copper and concluded that in order to
O2 (g) + 
O2 (step 2) explain kinetic data with their model the decom-
O2 + 
2O (step 3)
2CH3 OH + O
2CH3 O + H2 O (step 4)
position of methoxy must be rate limiting [60].
CH3 O + 
H2 CO + H (slow) (step 5) DFT calculations of methoxy decomposition on
H2 CO
H2 CO(g) +  (step 6) copper confirm a high activation barrier for this
2H
H2 (g) + 2 (step 7) process. It is therefore reasonable to assume that
H2 O
H2 O(g) +  (step 8) methoxy decomposition is a slow process on silver.
H2 CO + O
H + HCOO (step 9)
HCOO + 
H + CO2 (slow) (step 10)
TPR studies of the decomposition of formic acid,
CO2
CO2 (g) +  (step 11) formaldehyde, methanol and methyl formate have
The  signifies a surface site and X  is an adsorbed specie.
shown that CO2 and H2 desorbs simultaneously
around 400 K [10,17,61,62]. The decompositions
of all the above species are believed to go through
tailed model in all respects but to demonstrate that the same formate intermediate [10,17]. This sug-
the reaction mechanism established at UHV can gests that the further oxidation of formaldehyde to
explain kinetic experiments near industrial condi- carbon dioxide is rate limited by the decomposi-
tions. Secondly, the experimental knowledge of the tion of formate (step 10). This conclusion is con-
ignored elementary reactions is too limited to ex- sistent with kinetic measurements of HCOOH
tract reliable kinetic and thermodynamic para- decomposition on different crystal facets of silver
meters. Thirdly, by ignoring these elementary [63–66]. Therefore its assumed in the following
reactions the numerics of the model is greatly that the decomposition of methoxy and formate is
simplified and an analytical solution can be ob- rate limiting while all the other elementary reac-
tained. The consequences of excluding these ele- tions are equilibrated. This assumption will be
mentary reactions will be discussed later in this justified later when the model successfully explain
section. But for now it should be stressed that in- kinetic data. It might seem strange that the uptake
cluding these elementary reactions will not change of oxygen is not rate limiting as its dissociative
the results and conclusions of this work. The sticking probability is very low (107 –105 ) [67,68].
model would merely contain more parameters and However the kinetic data of Robb and Harriott
maybe slightly different parameter values for those [11] indicates a zero order reaction in oxygen,
fitted to kinetic experiments. hence the uptake of oxygen cannot be rate limit-
An important feature of the mechanism by ing. This can be explained by the high oxygen
Wachs and Madix is that the formaldehyde syn- pressure used in kinetic experiments which com-
thesis cannot be divided into a dehydrogenation pensates for the low sticking.
and oxidation. Furthermore CO2 is only formed The following possible elementary reactions
by the further oxidation of formaldehyde through involving OH formation has been excluded from
a formate intermediate i.e. direct methanol com- the model:
bustion (Eq. (3)) is absent.
Wachs and Madix observe simultaneous de- O þ H
OHþ ð6Þ
sorption of different products in TPR experiments
2OH
H2 O þ O ð7Þ
when methanol is adsorbed on preoxidized
Ag(1 1 0) [10] and this desorption occur at higher
OH þ H
H2 Oþ ð8Þ
temperatures than the characteristic desorption
spectra of these products [55–58]. This indicates Therefore the simplified model does not describe
that the products evolves from the same rate lim- the selectivity towards water perfectly nor the
iting step namely the decomposition of the meth- existence of hydroxy species. At high selectivity
oxy intermediate (step 5). It has been shown that toward formaldehyde the mechanism predicts
the decomposition of methoxy (step 5) is rate the production of a 1:1 H2 :H2 O ratio. Including
10 A. Andreasen et al. / Surface Science 544 (2003) 5–23

Eqs. (6)–(8) will increase the formation of water by Surface science experiments have indicated that
decreasing the hydrogen production. Industrially the decomposition of formate could be promoted
only slightly more water than hydrogen is pro- by oxygen [54]:
duced which justifies the neglection of these reac- HCOO þ O
CO2 þ OH ð10Þ
tions at industrial conditions [1,3]. However at
other reaction conditions especially transient ki- This formate decomposition pathway could be
netic experiments this approximation is poor as significant at low temperatures where a large
will become apparent in Section 3. Furthermore coverage of O exists but will be of minor impor-
the formation of OH could serve as an oxygen tance at industrial conditions.
scavenger. Since the stability of OH is low (de- Applying the quasi-equilibrium approximation
sorbs at about room temperature) [57] and the [69] to the Langmuir–Hinshelwood mechanism in
oxygen uptake is assumed to be equilibrated this Table 2 assuming steps 5 and 10 as rate limiting,
scavenger effect would be negligible for steady- the rate and equilibrium equations can be derived
state kinetics above room temperature. According cf. Table 3.
to Wachs and Madix OH may react in a similar
manner as O [10]. 2.2. Model catalyst
The production of HCOOH and HCOOCH3
observed at some experimental conditions has When possible we have derived the model pa-
been ignored in the model [10,37,38]. These reac- rameters from experiments on Ag(1 1 1) because it
tions seems to be unimportant in steady-state ki-
netics. However Qian et al. [38] found that the
production of HCOOH increased to a significant Table 3
Rate and equilibrium equations for the kinetic model based on
level on an aged catalyst. The reason for this aging
reaction steps 1–11
effect is not clear. pCH3 OH
In this paper there will be no attempts to hCH3 OH ¼ K1 h
p
model the decomposition of formaldehyde to pO2
hO2 ¼ K2  h
carbonmonooxide (Eq. (5)) since this reaction is p
1

important only at very high temperatures [6,8, hO ¼ ðK3 hO2 h Þ2 

1 21
37]. Furthermore it is believed to be a pyrolytic hCH3 O ¼ hCH3 OH K4 hO
hH2 O
gas phase reaction [6] and should therefore not k5
r5 ¼ k5 hCH3 O h  hH2 CO hH
lead to site blockage of any kind on the catalyst. K5
1 pH2 CO
CO desorbs at low temperatures and reacts rap- hH2 CO ¼ h
K6 p
idly with Oa to form CO2 on silver hence CO 1 pH 2
1

should only be present in the absence of oxygen hH ¼ K7 2 2 h

p
[51]. 1 pH2 O
hH2 O ¼
 h
When adsorbed atomic oxygen and carbon di- K8 p 
oxide interacts on a silver surface, carbonate is hH CO hO
hHCOO ¼ K9 2
known to form [51–53]: hH
k10
r10 ¼ k10 hHCOO h 
 hCO2 hH
CO2 þ O
COþ
3 ð9Þ K10
1 pCO2
hCO2 ¼ h
This is considered to be a dead end in the mech- K11 p
anism and the most important effect is stabiliza- h ¼ 1  hCH3 OH  hO2  hO  hCH3 O
hH2 CO  hH  hH2 O  hHCOO  hCO2
tion of carbon dioxide. The carbonate will
compete with the other adsorbates in the con- Ki are the equilibrium constants calculated from the molecular
sumption of active sites, but we consider this effect partition functions of the intermediates, ki are the rate constants
assumed to be of the Arrhenius form, ri are the reaction rates of
to be negligible at high temperatures, and the
the rate limiting steps, hX  is the coverage of species X  , pi is the
phenomena is not included in the further model- partial pressure and p is the thermodynamic reference pres-
ing. sure.
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
is the thermodynamically stable facet and there-
fore supposed to be the most abundant facet on
the industrial catalyst. Nevertheless in some cases
the Ag(1 1 0) surface has been used due to the
absence of appropriate experiments on Ag(1 1 1).
The density of sites (d) is estimated from oxygen
chemisorption on Ag(1 1 1) [28] which results in a
saturation coverage of approx. 0.5 ML oxygen
atoms. This, combined with the fact that a
Ag(1 1 1) facet has 1.38 · 1015 surface atoms per
cm2 gives a site density of 6.9 · 1018 sites/m2 .
In general, microkinetic modeling is not sensi-
tive to the density or type of active sites as long as
the experiments on the model catalyst used to de-
rive the model parameters reflects what is hap-
pening on an industrial catalyst. This is due to an
inverse proportionality between the site density
and the prefactors of all the elementary reactions.
The best way to establish if a model catalyst
reflects an industrial catalyst is to compare steady-
state kinetics of the two systems. Unfortunately
such measurements has not been performed on
Ag(1 1 1) for methanol oxidation. Furthermore the
structure sensitivity of the reaction is unknown.
Formate decomposition has been studied on dif-
ferent Ag facets showing a significant change in
activation barrier with surface structure [63–66].
On the other hand the effect on activities was
canceled by a large compensation effect. Silver
reconstruct dramatically in the presence of oxygen
and/or water during the formation of Oc and
subsurface OH leading to faceting [27,32–34]. The
reconstruction seems to be decreased in the pres-
ence of methanol. It is possible that a more accu-
rate model could be produced by including a
dependence of reaction conditions on the site
density as in the case of methanol synthesis on
copper [70,71]. However, as a first approximation
we assume that the site density is constant. As it
will be shown in Section 2.3 it has been necessary
to fit some parameters to steady-state kinetics.
These parameters may therefore not reflect the
values on a particular facet but may be some kind
of average value of all the facets/steps/defects
present at industrial conditions. It should be em-
phasized that the purpose of this work is not to
show which facet is the active one or that the re-
action is structure insensitive but to demonstrate
11
that a simple mechanism deduced from surface
science is able to explain experiments at industrial
conditions.
2.3. Model parameters
The equilibrium constants appearing in the
microkinetic model can be calculated using statis-
tical thermodynamics [72] from parameters for gas
phase molecules and adsorbates. The central pa-
rameters are vibrational frequencies and ground
state energies. All parameters for gas phase mole-
cules can be extracted from the NIST database [73]
or equivalent. For the adsorbates vibrational fre-
quencies are determined from spectroscopic mea-
surements e.g. EELS, IR, Raman. Ground state
energies can be fitted from TPD experiments and
measurements of sticking coefficients. The method
used for parameter estimation will be explained
below and the results are summarized in Tables 4–
6. For more details on the method we refer to
other papers [74,75].
A first order desorption process in UHV for a
generic gas phase molecule A is described by the
following rate equation:
dhA k
¼  hA  ð11Þ
dt K
where k is the rate constant of adsorption, K is the
equilibrium constant and hA is the coverage of A.
The equilibrium constant can be calculated from
the partition function cf. Table 1:
zA
K¼ ð12Þ
zA
where zA is the partition function of the gas phase
and zA is the partition function of the adsorbate.
The equilibrium constants for adsorption and
surface reactions can be calculated in a similar
fashion.
The rate constant k is found by equating the
initial adsorption rate with the initial sticking rate:
r0 p
k ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð13Þ
d 2pmkB T
where r0 is the initial sticking coefficient, p is the
thermodynamic reference pressure, m is the mass
12 A. Andreasen et al. / Surface Science 544 (2003) 5–23

Table 4
Model parameters for gas phase molecules to the statistical thermodynamical treatment
Species Parameters
H2 B ¼ 60:848 cm1 , r ¼ 2, m1 ¼ 4405:3 cm1 , H ¼ 0 @ T ¼ 298 K [75]
O2 B ¼ 1:45 cm1 , r ¼ 2, m1 ¼ 1580 cm1 , H ¼ 0 @ T ¼ 298 K [73]

H2 O A ¼ 27:8847 cm1 , B ¼ 14:51181 cm1 , C ¼ 9:2806 cm1 , r ¼ 2, m1 ¼ 1594:7 cm1 , m2 ¼ 3651:1 cm1 ,
m3 ¼ 3755:9 cm1 , H ¼ 241:818 kJ/mol @ T ¼ 298 K [73].
CH3 OH A ¼ 4:2554 cm1 , B ¼ 0:823 cm1 , C ¼ 0:7928 cm1 , r ¼ 1, m1 ¼ 270 cm1 , m2 ¼ 1033 cm1 , m3 ¼ 1060 cm1 ,
m4 ¼ 1165 cm1 , m5 ¼ 1345 cm1 , m6 ¼ 1477ð2Þ cm1 , m7 ¼ 1455 cm1 , m8 ¼ 2844 cm1 , m9 ¼ 2960 cm1 ,
m10 ¼ 3000 cm1 , m11 ¼ 3681 cm1 , H ¼ 201:2 kJ/mol @ T ¼ 298 K [45,73].
H2 CO A ¼ 9:074 cm1 , B ¼ 1:2899 cm1 , C ¼ 1:129 cm1 , r ¼ 2, m1 ¼ 1163:5 cm1 , m2 ¼ 1247:4 cm1 ,
m3 ¼ 1500:6 cm1 , m4 ¼ 1746:07 cm1 , m5 ¼ 2766:4 cm1 , m6 ¼ 2843:4 cm1 , H ¼ 108:57 kJ/mol @ 298 K [73].
CO2 B ¼ 0:39038 cm1 , r ¼ 2, m1 ¼ 667:3ð2Þ cm1 , m2 ¼ 1384:26 cm1 , m3 ¼ 2349:49 cm1 ,
H ¼ 393:15 kJ/mol @ T ¼ 298 K [73].
A, B and C are the rotational constants, r is the symmetry number, mi are the vibrational frequencies and the degeneracy of a frequency
is enclosed in parentheses. H is the enthalpy of formation.

Table 5
Model parameters for adsorbates to the statistical thermodynamical treatment
Species Parameters
H m? ¼ 1121:0 cm1 , mk ¼ 927:5ð2Þ cm1 , H ¼ 25 kJ/mol [75].
O m1 ¼ 350 cm1 , m2 ¼ 508ð2Þ cm1 , H ¼ 63 kJ/mol @ T ¼ 298 K [81].
O2 m1 ¼ 50ð2Þ cm1 , m2 ¼ 300ð2Þ cm1 , m3 ¼ 675 cm1 , m4 ¼ 220 cm1 ,
H ¼ 44:5 kJ/mol @ T ¼ 298 K [81].
CH3 OH m? ¼ 29 cm1 , mk ¼ 35:51ð2Þ cm1 , mr ¼ 36:0ð3Þ cm1 , m1 ¼ 750 cm1 , m2 ¼ 820 cm1 ,
m3 ¼ 1030 cm1 , m4 ¼ 1150ð2Þ cm1 , m5 ¼ 1470ð3Þ cm1 , m6 ¼ 2860 cm1 , m7 ¼ 2970ð2Þ cm1 ,
m8 ¼ 3320 cm1 , H ¼ 241:6 kJ/mol @ T ¼ 298:15 K [12,45].
CH3 O m? ¼ 330 cm1 , mk ¼ 36:49ð2Þ cm1 , mr ¼ 360ð3Þ cm1 , m1 ¼ 1040 cm1 , m2 ¼ 1150ð2Þ cm1 ,
m3 ¼ 1450ð3Þ cm1 , m4 ¼ 2840 cm1 , m5 ¼ 2940ð2Þ cm1 , H ¼ 104 kJ/mol @ T ¼ 298:15 K [12,45].
H2 CO m? ¼ 400 cm1 , mk ¼ 37ð2Þ cm1 , mr ¼ 23ð3Þ cm1 , m1 ¼ 1476 cm1 , m2 ¼ 1694 cm1 , m3 ¼ 2839 cm1 ,
m4 ¼ 1218:15 cm1 , m5 ¼ 2839 cm1 , H ¼ 135 kJ/mol @ T ¼ 298:15 K [56,77].
H2 O m? ¼ 27:9 cm1 , mk ¼ 27:9ð2Þ cm1 , mr ¼ 740ð3Þ cm1 , m1 ¼ 1660 cm1 , m2 ¼ 3410ð2Þ cm1 ,
H ¼ 293:2 kJ/mol @ T ¼ 298:15 K [57,75].
CO2 m? ¼ 410 cm1 , mk ¼ 30:6ð2Þ cm1 , mr ¼ 12:71ð3Þ cm1 , m1 ¼ 605 cm1 , m2 ¼ 1365 cm1 , m3 ¼ 2350 cm1 ,
H ¼ 414:607 kJ/mol @ T ¼ 298:15 K [77,90].
HCOO m? ¼ 322 cm1 , mk ¼ 35:51ð2Þ cm1 , mr ¼ 400ð3Þ cm1 , m1 ¼ 770 cm1 , m2 ¼ 1340 cm1 , m3 ¼ 1640 cm1 ,
m4 ¼ 2900 cm1 , m5 ¼ 1050 cm1 , m6 ¼ 1377 cm1 , H ¼ 197 kJ/mol @ T ¼ 298:15 K [13,45,61].
mi are the vibrational frequencies and the degeneracy of a frequency is enclosed in parentheses. m? , mk and mr are the frustrated
translational orthogonal frequency, the frustrated translational parallel frequency and the frustrated rotational frequency, respectively.
H is the enthalpy of formation.

of A, kB is BoltzmannÕs constant and T is the

temperature. Substituting this into Eq. (11) the dhA r0 p
¼  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi hA ð14Þ
following equation is obtained: dt d 2pmkB T K
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
Table 6
Arrhenius parameters for the two slow steps determined from
optimization
Reaction, i Ai Hiz
1
5 10
4.2 · 10 s 60 kJ mol1
10 9.8 · 109 s1 77 kJ mol1
A is the prefactor and H z is the activation enthalpy.
Experimental TPD spectra are simulated by in-
tegrating equation (14) using numerical tech-
niques. The ground state energy comes into play
through the equilibrium constant and is opti-
mized until the experimental peak temperature is
reproduced.
It seems likely that the sticking coefficient of
methanol, formaldehyde and water is near unity
[56–58]. Assuming a sticking coefficient of unity,
implies that the activation barrier for the sticking
process is zero. These assumptions results in the
parameters presented in Table 5 which results in
TPD peaks at 167 K [58,76], 106 K [56], and 170 K
[57] for methanol, formaldehyde and water, re-
spectively which is consistent with experimental
values. The exact stability of these species are not
critical to the model, cf. Section 4.
The stability of hydrogen is deduced from an
experimental D2 desorption spectra by Wachs
and Madix [10] and with vibrational parameters
from Cu [77]. By assuming a desorption expo-
nential of 1013 and an activation barrier of 55.6
kJ/mol the peak temperature of 228 K is repro-
duced by simulation. Assuming a non-activated
sticking of hydrogen we come up with an H
enthalpy of )27.8 kJ/mol. The sticking of hy-
drogen on silver is probably activated as for Cu
[77], which results in a too high predicted value of
the enthalpy of H. We have also neglected the
isotope effect, which also leads to discrepancy
between the stability of H and D. It turns out
that the microkinetic model is rather insensitive
to the choice of H parameters. All parameters for
carbon dioxide are taken from Cu [77], because
experiments on Ag are insufficient. Both for hy-
drogen and carbon dioxide the stabilities will be
over predicted, but this is considered only to have
minor importance due to the small coverage of
these species cf. Fig. 7.
13
2.3.1. Adsorbed oxygen
The stability of dissociative adsorbed oxygen on
silver is controversial even though the oxygen/sil-
ver system has been studied extensively. On
Ag(1 1 1) the value referenced to by most people is
the one obtained by Campbell [67]. Campbell used
the first order Redhead equation [78], a desorption
preexponential of 1015 s1 and deduced a disso-
ciative adsorption heat of 170.5 kJ/mol O2 from
the TPD of dissociatively adsorbed oxygen. As
stated by Campbell [67] the Redhead analysis is a
simplification. The desorption is recombinative
(second order) and an attractive interaction exists
between dissociatively adsorbed oxygen. The at-
tractive interaction is observed by island forma-
tion at low coverage below 490 K in UHV and by
a very narrow TPD peak [30,31,67,68,79,80].
We have made a microkinetic model of oxygen
chemisorption on Ag(1 1 1) that (within experi-
mental uncertainties) describes molecular and
dissociative sticking, and molecular and recomb-
inative desorption [81]. The attractive interaction
between dissociatively adsorbed oxygen has been
modeled using the mean-field approximation. This
model resulted in a heat of adsorption of 126 kJ/
mol O2 and the attraction was found to be 15 kJ/
mol O2 .
In the following we will use the heat of ad-
sorption in the limit of low coverage assuming that
the attractive interactions should be ignored on
polycrystalline or real silver catalysts due to the
lack of long range order on such catalysts [82] and
the presence of other adsorbates.
2.3.2. Methoxy and formate
The methoxy and formate species are special in
the sense that there are no corresponding gas
phase molecules. The only information available
for these adsorbates are vibrational frequencies
and decomposition temperatures in TPR on
Ag(1 1 0). From the TPR experiments it is only
possible to establish the rate constants of methoxy
and formate decomposition. The parameters that
remains to be determined are ground state energies
and the specific decomposition Arrhenius para-
meters of methoxy and formate. Fig. 1 shows the
simulation of methoxy and formate decomposition
for different Arrhenius parameters but identical
14 A. Andreasen et al. / Surface Science 544 (2003) 5–23
0.06
0.05
Reaction rate [s ]
-1
0.04
0.03
0.02
0.01
0 These parameters are therefore not deduced
250 300 350 400
Temperature [K]
Fig. 1. Simulation of TPR experiments for methoxy (low
temperature peak) and formate decomposition (high tempera-
ture peak) using different sets of Arrhenius parameters. b ¼ 15
K/s and initial methoxy and formate coverage is 10%.
rate constants at the peak temperatures. Its is clear
from the figure that the specific Arrhenius pa-
rameters cannot be establish from such TPR ex-
periments.
The ground state energy of methoxy has been
determined by ensuring that the models heat of
reaction for decomposition of gas phase methanol
into adsorbed methoxy and H is consistent with
DFT calculations. The total energy calculation
code DACAPO [83] is used for the DFT calcula-
tions. A four layer 2 · 2 slab representing Ag(1 1 1),
periodically repeated in a super cell geometry with
six equivalent layers of vacuum between successive
slabs. The bottom layer is fixed while the other
layers are allowed to relax. Ionic cores are de-
scribed by ultra-soft pseudo-potentials [84], and
the Kohn–Sham one-electron valence states are
expanded in a basis of plane waves with a kinetic
energy below 25 Ry. The surface Brillouin zone is
sampled at 18 special k points. The exchange-
correlation energy and potential are described by
the generalized gradient approximation (GGA-
PW91) [85–87]. Methoxy and H is placed in the fcc
site found to be the best site for these species on
copper [20]. It was not our intention to conduct a
thorough DFT investigation of this reaction but
merely to get an estimate of the methoxy stability
on Ag(1 1 1).
The remaining parameters are fitted to the ki-
netic measurements of Robb and Harriott [11],
Leffert et al. [37], Nagy et al. [8] and Nagy and
10
10
11
Mestl [9] (see Figs. 2–5). The kinetic model is im-
10
10
12 plemented in a plug-flow reactor model and kinetic
experiments are simulated by integrating the de-
13
10
sign equations with an adaptive step size 4 order
Runge–Kutta algorithm. The optimization occurs
with a least-squares simulated annealing. The
calculation ensures that the correct TPR decom-
position temperatures of methoxy and formate are
reproduced, cf. Fig. 1.
450 explicitly from surface science but they reproduce
all known surface science experiments and have
physically realistic values according to DFT and
TST. Other choices of parameters could give simi-
lar results.
The prefactors of methoxy and formate de-
composition is in the region 1010 s1 cf. Table 6.
TST predicts a prefactor of about 1013 s1 for
surface reactions varying few orders of magnitude
dependent of the nature of the transition state
(mobile/immobile) [41,82]. Experimentally preex-
ponentials vary from 1010 –1016 s1 [82]. The values
of the prefactors in the model seems low compared
to TST, however within the range usually reported
in the literature and hence physically realistic.
Besides the mobility of the transition state the low
preexponentials could be a result of overestimating
0.04
Methanol Conversion [mol]
0.9 % Methanol
0.03 1.8 % Methanol
2.8 % Methanol
5.5 % Methanol
8.8 % Methanol
19.1 % Methanol
0.02
Simulations
0.01
0
0 0.05 0.1 0.15
Oxygen mole fraction
Fig. 2. Simulation of the experiments by Robb and Harriott at
695 K and 1.12 atm. Points are experimental measurements
from Ref. [11] and lines have been calculated with the micro-
kinetic model. The figure shows the conversion of methanol as a
function of the partial pressure of oxygen at different partial
pressures of methanol.
 A. Andreasen et al. / Surface Science 544 (2003) 5–23 15

1
Simulations
CH2O Set I H2CO
0.5
CO2 Set I CO2
0.8
CH2O Set II
0.4 CO2 Set II
0.6
Yield

Predicted yield
0.3

0.4
0.2

0.1 0.2

0
0 0.2 0.4 0.6 0.8 0
Conversion of CH3OH 0 0.2 0.4 0.6 0.8 1
Measured yield
Fig. 3. Simulation of the experiments by Robb and Harriott at Fig. 5. Simulation of 25 experiments by Leffert et al. [37]
695 K and 1.12 atm. Points are experimental measurements plotted on a predicted versus experimental plot.
from Ref. [11] and lines have been calculated with the micr-
okinetic model. The figure shows the yield of H2 CO and CO2 ,
respectively, as a function of the conversion of methanol. Set I
is data obtained with an aged catalyst of silver particles of
about 120 A
on alumina (sintered). Set II is data obtained with values reported on Ag(1 1 1) facets (67 kJ/mol) and

on alumina.
a fresh catalyst of silver particles of about 55 A Ag(1 1 0) facets (127 kJ/mol) [63].

1 3. Results

0.8 In this section our simulations of the kinetic

Selectivety / Conversion

0.6
0.4
0.2
0
500 600 700
Temperature [K]
Fig. 4. Simulation of the experiments by Nagy and Mestl.
Points are experimental measurements from Ref. [9] and lines
have been calculated with the microkinetic model. The figure
shows the selectivity towards H2 CO and CO2 , respectively, and
the conversion of methanol as a function of temperature.
the site density in the fitting process, inaccuracies
of the estimated vibrational frequencies for meth-
oxy and formate or perhaps a need for larger site
ensembles for methoxy and formate decomposi-
tion.
Interestingly the fitted activation barrier of
formate decomposition (77 kJ/mol) is between the
CH2O
experiments of Robb and Harriott [11], Nagy et al.
CO2 [8], Nagy and Mestl [9], and Leffert et al. [37] are
CH3OH presented using the parameters obtained in the
CH2O (mes.)
previous section.
CO2 (mes.)
CH3OH (mes.) Fig. 2 shows the simulation of the total rate of
methanol oxidation as a function of oxygen and
methanol pressures at 420 C as measured by
Robb and Harriott. Robb and Harriott only re-
800 900
ported a relative rate and analyzed their data ac-
cording to the differential reactor approximation
even though the conversion in some cases reached
30%. We have chosen to analyze their data ac-
cording to the integral reactor approximation and
converting the relative rates to absolute rates. This
transformation was done by calibrating the rela-
tive rates to a reported standard measurement and
approximating the density of sites from measure-
ments of particle sizes. However it turns out that
the trends deduced from the integral analysis is
identical to those obtained in the differential re-
actor analysis by Robb and Harriott. Fig. 2 shows
that our model correctly explains the reaction or-
ders of oxygen and methanol.
16 A. Andreasen et al. / Surface Science 544 (2003) 5–23

Fig. 3 shows the simulation of Robb and Har- 4. Discussion

riottÕs measurements of yield versus methanol
conversion at 420 C. The simulation clearly cap-
tures the correct trends, i.e. the selectivity towards
formaldehyde decreases with increasing conver-
sions of methanol. However the experimental se-
lectivity decreases much more rapidly with
methanol conversion than predicted by the model.
The reason for this discrepancy could be the use of
alumina as a support in the experiments of Robb
and Harriott. Experimental evidence shows that
alumina decomposes formaldehyde [14]. This hy-
pothesis has been confirmed by simulations in-
cluding formaldehyde decomposition on alumina
in the microkinetic model.
Nagy et al. investigated the influence of tem-
perature on selectivity by applying a low linear
heating rate (1 K/min) on a steady-state flow re-
actor while running the reaction. The oxygen
conversion was
100% at all conditions except at
very low temperatures (
500 K) [8,9]. Fig. 4 shows
the simulation of the experimental conversion of
methanol and selectivity towards formaldehyde
and CO2 , respectively. The model clearly captures
the trend that formaldehyde selectivity increases
with temperature and saturates at high tempera-
ture. The model also captures the trend that CO2
selectivity is high at low temperature but decrease
with increasing temperature. In order to fit the
experiment of Nagy et al. the inlet oxygen con-
centration was reduced with 15% otherwise total
methanol consumption was achieved at high tem-
peratures which is not observed experimentally.
The reason for this problem is the exclusion of
elementary reactions forming water from O and
H through OH in the model. Therefore the ox-
ygen consumption is too low for the model.
Leffert et al. [37] investigated the effect of tem-
In the previous section we have demonstrated
that a microkinetic model based on a mechanism
deduced from surface science is able to explain a
broad range of kinetic experiments. Even more
interesting some of these experiments has been
used as proof in the literature for a mechanism
different form the one observed in UHV. In this
section we will proceed with a more detailed
analysis in order to obtain a more thorough un-
derstanding of the model.
4.1. Coverages
As mentioned in Section 1 Schl€ ogl and co-
workers concluded that Oa cannot be active at
industrial conditions of 900 K since it desorbs at
600 K in UHV. However in the previous section
our microkinetic model explained experiments at
high temperatures using parameters of O repro-
ducing the experimental TPD peak temperature at
600 K in UHV. Fig. 6 shows the simulated cov-
erage of O as a function of temperature for dif-
ferent oxygen pressures, which clearly indicates
that O can be present at industrial conditions. At
UHV the coverage of O approach zero at 600 K
while a significant coverage is present at 900 K
at oxygen pressures above 1 kPa. Interestingly,
Scheffler and coworkers [88] found that surface
1
0.9 P = 0 Pa
P = 1 Pa
0.8 P = 10 Pa
P = 100 Pa
P = 1000 Pa
0.7
P = 10000 Pa
0.6 P = 100000 Pa
θO*

perature on methanol oxidation in a similar fash- 0.5

ion as Nagy et al. [8] and Nagy and Mestl [9]. 0.4
Furthermore the authors investigated the effect of 0.3
methanol and oxygen concentration on the yield 0.2
towards formaldehyde and CO2 , respectively. Fig. 0.1
5 shows the predicted yield of formaldehyde and
0
CO2 by the model versus experimental yields. As 400 500 600 700 800 900 1000 1100 1200
above it was necessary to reduce the inlet oxygen Temperature [K]
concentration by approx. 20% to explain the ex- Fig. 6. Simulated coverage of atomic oxygen on silver versus
periments. temperature at different oxygen pressures.
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
atomic oxygen is stable up to 830 K at atmo-
spheric pressure by using the approach of
atomistic ab initio (DFT) thermodynamics. Con-
sidering the systematic errors of DFT [88] the re-
sult of Scheffler and coworkers is very close to the
result obtained in this paper by the microkinetic
model. The conclusion of Schl€ ogl et al. that Oa
should be absent at 600 K is therefore unfounded.
Oa may be present at industrial conditions due to
its thermal stability, however it could be absent for
other reasons. For example, if a stable layer of Oc
is formed, it could block the sites for Oa uptake.
This might be the case in some of the transient
experiments conducted by Schl€ ogl and coworkers.
At steady-state industrial reaction conditions
blocking of Oa uptake is unlikely for several rea-
sons. First of all the formation of Oc is slow and
activated and can never exceed the rate of Oa up-
take since its formed from Oa . Secondly, the cov-
erage of Oc decreases significantly in the presence
of methanol and can therefore not block Oa up-
take in a reactive atmosphere. From the above
discussion it is clear that it is impossible to predict
the presence of intermediates from TPD tempera-
tures in UHV alone. Furthermore it is very im-
portant to note that an intermediate does not
necessarily need a large coverage in order to be
important.
Fig. 7 shows the calculated coverages at differ-
ent temperatures for the microkinetic model. The
coverages of all intermediates has been calculated
1 Table 7
* Apparent activation enthalpies Eiz for the oxidation of methanol
0.1 CH3OH*
O2 *
0.01 O*
CH3O*
H2O*
0.001
Coverage
H2CO*
H*
0.0001 HCOO*
CO2*
1e-05
1e-06
1e-07
400 500 600 700 800 900
Temperature [K]
Fig. 7. Initial coverages as a function of temperature at 1 atm,
calculated at typical reaction conditions.
17
at a gas composition of 32% CH3 OH, 4.5% O2 ,
11.6% H2 CO, 1.3% CO2 , 7.4% H2 O and 6.5% H2 .
This composition corresponds to an average gas
composition between the inlet and outlet of an
industrial reactor (without additional water in the
feed) and we will refer to this as typical reaction
conditions. The figure shows that O is the most
dominating adsorbate. The number of free sites
increases with the decrease of the coverage of O .
The coverage of all intermediates are small except
formate that becomes significant at low tempera-
tures.
4.2. Apparent activation enthalpies
The apparent activation enthalpy Ez for steps 5
and 10 can be derived analytically from the defi-
nition [43]:
d lnrþ
E z ¼ kB T 2 ð15Þ
dT
where rþ is the forward reaction rate. This gives
the results shown in Table 7 and it is observed that
a simple relation between Ez and the coverage of
surface species exist. Ez is a sum of the activation
enthalpy for the rate limiting step and a weighted
average of the desorption enthalpies for the in-
termediates. The sum of the activation enthalpy of
the rate limiting step is the activation barrier of the
elementary step itself plus the heat of formation/
adsorption of the reactants in the rate limiting
step. The average desorption enthalpy is formed
to formaldehyde (5) and further from formaldehyde to carbon
dioxide (10)
E5z ¼ H5z þ H1 þ 1=2H4 þ 1=2H8 þ 1=4H2 þ 1=4H3
2H1 hCH3 OH  2H2 hO2  ðH2 þ H3 ÞhO
ð2H1 þ H4 þ H8 þ 1=2H2 þ 1=2H3 ÞhCH3 O
þ2H6 hH2 CO þ H7 hH þ 2H8 hH2 O þ 2H11 hCO2
ð2H9  2H6 þ H2 þ H3 þ H7 ÞhHCOO
z z
E10 ¼ H10 þ H9  H6 þ 1=2H2 þ 1=2H3 þ 1=2H7
2H1 hCH3 OH  2H2 hO2  ðH2 þ H3 ÞhO
ð2H1 þ H4 þ H8 þ 1=2H2 þ 1=2H3 ÞhCH3 O
þ2H6 hH2 CO þ H7 hH þ 2H8 hH2 O þ 2H11 hCO2
ð2H9  2H6 þ H2 þ H3 þ H7 ÞhHCOO
Hiz is the activation enthalpy for the two slow steps in the
mechanism (5 and 10) and Hi is the reaction enthalpy for step i.
18 A. Andreasen et al. / Surface Science 544 (2003) 5–23

by multiplying the coverage of the intermediates increase in free sites cf. Fig. 7. Hence it is easy to
by twice the enthalpy of desorption of the inter- create two free sites for the rate limiting steps at
mediates through equilibrium steps. The factor of high temperatures. The activation enthalpy for the
two enters because the rate limiting step requires oxidation of methanol to formaldehyde is larger
two free sites. Ez is thus the sum of the activation than the activation enthalpy for CO2 formation at
enthalpy of the rate limiting step plus the average all temperatures even though the activation barrier
cost of creating two free sites on the surface. In the of step 5 is smaller than step 10 cf. Table 6. The
literature most people seem to be aware of the fact difference in apparent activation enthalpy of steps
that the apparent activation energy contain con- 5 and 10 is almost constant in the entire temper-
tributions from the activation barrier of rate lim- ature interval and correspond to the difference of
iting elementary step plus adsorption heats of the the sum of the activation enthalpy of the rate
adsorbates participating in the forward rate limi- limiting steps (steps 5 and 10) and the adsorption
ting step. However the coverage dependent terms heats.
are usually ignored in the literature. This is a se- In Fig. 9 calculated apparent activation enthal-
rious error in all systems that proceeds with a pies for the oxidation of methanol to formaldehyde
significant coverage of intermediates on the active are compared with experimentally observed values
sites. Coverage dependent terms are responsible from the work of Obraztsov et al. [39], Gavrilin
for the variations in apparent activation energies and Popov [7] and Bhattacharyya et al. [2]. The
with reaction conditions. apparent activation enthalpies of Qian et al. [38]
The expressions in Table 7 have been used to have been extracted from methanol conversions
calculate apparent activation enthalpies at typical versus temperature. Calculated values are obtained
reaction conditions as a function of temperature. from the differential reactor approximation ac-
Results of these calculations are shown in Fig. 8. cording to Eq. (15) (solid line) and according to
Large variations in the activation enthalpies are the integral reactor approach (dashed line) where
observed, as a consequence of changes in cover- rþ in Eq. (15) is substituted with the conversion of
ages with temperature. This is an illustrative ex-
ample of the difficulty in comparing activation
enthalpies obtained at different reaction condi- 200000
Bhattacharyya et al.
Apparent activation enthalpy [J/mol]

tions. Gavrillin et al.

Both activation enthalpies decrease as a func- 150000
Obraztsov et al.
Qian et al.
tion of temperature, which is primarily due to the Simulated (Differential)
Simulated (Integral)

100000
200000
Apparent activation enthalpy [J/mol]

Methanol to formaldehyde
Formaldehyde to carbondioxide 50000
150000

0
400 500 600 700 800 900
100000
Temperature [K]

Fig. 9. Apparent activation enthalpy for the oxidation of

50000 methanol to formaldehyde calculated at differential and integral
conditions and experimental observations from Refs. [2,7,
38,39]. Apparent activation enthalpies for differential condi-
0 tions are calculated at typical reaction conditions. Apparent
400 500 600 700 800 900
Temperature [K] activation enthalpies for integral conditions are calculated from
conversion of methanol in a plug reactor with a feed compo-
Fig. 8. Apparent activation enthalpies for the oxidation of sition of 9.82% CH3 OH and 14.3% O2 . The amount of catalyst
methanol to formaldehyde and further to carbon dioxide. and flow rate is chosen to give nearly complete conversion of
Calculations has been performed at typical reaction conditions. methanol at 900 K.
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
methanol. As shown the differential model quali-
tatively explains the trend in experimental obser-
vations i.e. a decrease in activation enthalpy as a
function of temperature. However the values seem
to be over estimated. The discrepancy between
calculated and measured values can be accounted
for as deviations from differential conditions i.e. a
substantial conversion in experimental observa-
tions. This is clearly shown by the activation en-
thalpies calculated for integral conditions giving a
much better description of the experimental ob-
servations.
In this context it is very important to mention,
that none of the experimental activation enthalpies
presented in Fig. 9 have been used in the optimi-
zation of parameters mentioned earlier.
4.3. Reaction orders
The reaction orders can be derived from the
following definition [43].
d lnðrþ Þ
ai ¼  
d ln ppi
For the formation of formaldehyde the following
reaction orders can be derived.
aCH3 OH ¼ 1  2hCH3 OH  2hCH3 O
aO2 ¼ 1=4  2hO2  hO  1=2hCH3 O  hH2 CO
aH2 CO ¼ 2hH2 CO  2hHCOO
aCO2 ¼ 2hCO2
aH2 O ¼ 1=2 þ hCH3 O  2hH2 O

As the coverages of all intermediates except O are
low according to Fig. 7, the model explains the
positive reaction order with respect to methanol
observed experimentally [2,11]. With respect to
oxygen Eq. (18) shows, that oxygen will have a
positive reaction order, if the coverage of oxygen,
methoxy and methanol are moderate. It can be
seen from Fig. 2 that the model captures the ob-
served trends in reaction order at 695 K very well
i.e. zero order in oxygen and nearly first order in
19
1
CH3OH
O2
Reaction order α i
H2CO
0.5
CO2
O2
0
-0.5
400 500 600 700 800 900
Temperature [K]
Fig. 10. Calculated reaction orders for the oxidation of meth-
anol to formaldehyde as a function of temperature at typical
reaction conditions.
methanol. Experimentally an inhibiting effect from
water [2] is observed. This observation is repro-
duced by the model, cf. Eq. (21). Calculated re-
action orders cf. Eqs. (17)–(21) at typical reaction
conditions are plotted in Fig. 10 as a function of
temperature. The reaction order of methanol,
ð16Þ
formaldehyde, water and carbon dioxide are in-
dependent of temperature and denotes values of 1,
0, )0.5 and 0, respectively. The reaction order of
oxygen varies between )0.75 at low temperatures
(high coverage of O ) to 0.25 at high temperatures
ð17Þ (low coverage of O ). Generally the quantitative
values of the reaction orders presented in Fig. 10
reproduce the qualitative features discussed above.
ð18Þ
4.4. Selectivity and industrial conditions
ð19Þ
The observed selectivity versus conversion trend
ð20Þ on Fig. 3 is due to the increasing amount of
formaldehyde formed at higher conversions. In-
ð21Þ
creasing the amount of formaldehyde leads to an
increased combustion of formaldehyde. This is a
classical effect of a consecutive reaction path.
Robb and Harriott [11] postulated that two routes
to CO2 formation was necessary to explain these
data. However our model shows that one oxida-
tion route for formaldehyde is enough to explain
the experiment. The reason for the different con-
clusion reached by Robb and Harriott is that they
used an aged catalyst for low conversion experi-
ments and a fresh catalyst for high conversions.
20 A. Andreasen et al. / Surface Science 544 (2003) 5–23
They state in their paper that a fresh catalyst has a
lower selectivity than an aged catalyst. This means
that they amplified the effect of conversion on se-
lectivity by using these different catalysts. The
difference between fresh and aged catalysts may be
explained by the use of alumina support.
The most important selectivity trend observed
experimentally and predicted by the model is an
increased selectivity with temperature. This effect
can be explained very simple by the microkinetic
model. As demonstrated in Fig. 8 the apparent
activation energy of formaldehyde synthesis is al-
ways larger than the apparent activation energy of
formaldehyde combustion. Hence the rate of
formaldehyde formation increases more rapidly
than formaldehyde combustion at increasing tem-
perature. In order to obtain a high selectivity the
industrial process should be conducted at high
temperatures. However the temperatures should
not be high enough to promote the pyrolytic gas
phase decomposition of formaldehyde. Further-
more a methanol rich feed should be used in the
process to ensure high selectivity by achieving
complete oxygen consumption. Otherwise the re-
maining oxygen would combust the formed
formaldehyde. This is consistent with the work by
Wachs and coworkers [36] who show that the se-
lectivity towards formaldehyde decreases from
92.3% to 69.1% when CH3 OH/O2 molar ratio de-
creases from 3.08 to 0.95 in a fixed bed reactor.
The rate of formaldehyde combustion does not
decrease with temperature it just does not increase
as rapidly as the rate for formaldehyde formation.
Another way to increase selectivity would be to
reduce the contact time in the reactor. However
the rate of these reactions are so fast that total
conversion is achieved very rapidly and this option
is therefore not feasible. Hence the microkinetic
model gives a very simple explanation for the
choice of industrial conditions.
Nagy and Mestl [9] observes that the reaction
ignites at approx. 450 K with a partial pressure of
methanol of 0.088 atm in the feed. Gavrilin and
Popov [7] has shown that the reaction ignites at
approx. 550 K with a partial pressure of methanol
of 0.435 atm. From Fig. 8 it is clear that the ap-
parent activation energy of both formaldehyde
and CO2 formation becomes very high below
500 K due to the blockage of free sites by O . This
site blockage therefore inhibits the reactions below
400–500 K dependent on the partial pressures of
oxygen and methanol.
It is informative to compare our modelÕs ex-
planation of the experiment depicted in Fig. 4 and
Nagy and MestlÕs own interpretation [8,9]. It
should be noted that Nagy and MestlÕs interpre-
tation was not supported by simulations. Our
model explains the trends in Fig. 4 by a high
production of CO2 at low temperatures due to a
low apparent activation energy of CO2 formation
leading to a low selectivity. Below 500 K the sta-
bility of O leads to site blockage and all methanol
oxidation is inhibited as mentioned above. At high
temperatures the selectivity increases due to a
higher apparent activation barrier for formalde-
hyde formation compared to CO2 formation.
However Nagy and Mestl explain the selectiv-
ity trends by the different temperature stability
of various oxygen species (Oa and Oc ) formed,
exhibiting different chemical reactivities. Oa par-
ticipates in an oxi-dehydrogenation path which
dominates at low temperatures and result in both
formaldehyde and complete oxidation products.
Oc participates in a direct dehydrogenation path
at high temperatures which exclusively leads to
formaldehyde. As evidence for this the authors
claim that Oc is dominating at high temperatures
and is able to react with methanol. However as
mentioned earlier the uptake of Oc is probable too
slow to keep up with the uptake of Oa . Further-
more the necessary reconstruction to form Oc
could be absent in the presence of methanol rich
mixtures [26]. Nagy and Mestl mention that the
selectivity increases dramatically above Oa de-
sorption temperature (580–620 K) indicating that
Oc plays a significant role at high temperatures.
This argument is questionable. First of all 620 K is
the desorption temperature of Oa in UHV. Sec-
ondly, the selectivity increases more rapidly below
the desorption temperature cf. Fig. 4. Nagy and
Mestl also postulate that a reaction-in-series model
in which formaldehyde is an intermediate product
for CO2 formation cannot explain the observed
increasing selectivity with temperature due to the
higher thermodynamic stability of CO2 . The au-
thors therefore conclude that the selectivity trends
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
is due to changing between two reaction paths i.e.
oxi-dehydrogenation and direct dehydrogenation.
However our simulations clearly demonstrate that
a reaction-in-series model is able to explain this
selectivity behavior (this conclusion is valid re-
gardless if one accept our model or not). The so-
lution to this apparent paradox is very simple: as
for all other partial oxidation reactions the selec-
tivity is determined by kinetics not thermody-
namics. All the oxygen is simple consumed rapidly
in the formation of formaldehyde leaving nothing
to oxidize formaldehyde further. Furthermore if
two very different reaction pathways operate at
low and high temperatures, respectively one would
intuitively expect some kind of evident bend in the
data at a temperature corresponding to the change
in mechanism. This is in contrast to the nice
smooth data reported in Fig. 4. With regard to the
above discussion we find our interpretation of
Nagy and MestlÕs experiments much more consis-
tent, simple and validated. It is well known that
it is impossible to deduce reaction mechanisms
explicitly from steady-state kinetics. It must be
considered even more dangerous to deduce mech-
anisms from kinetic experiments with total con-
version. The experiments of Nagy and Mestl led to
100% conversion of oxygen and can therefore not
be used to deduce the intrinsic kinetics of metha-
nol oxidation. It should be stressed that our model
is not deduced from Nagy and Mestl experiments
but their experiments are explained as a conse-
quence of our model.
It is well known that formaldehyde synthesis on
silver can be poisoned with electronegative sub-
stances such as halides and sulphur, and consid-
erable attention has been paid to their removal,
particular from process air [4]. It has been estab-
lished that electronegative substances such as
chloride blocks the uptake of O and significantly
lowers the rate of O chemisorption (Ref. [89] and
references therein). This poisoning is consistent
with our model because O is essential in our
model to the production of formaldehyde.
4.5. Critical parameters
The microkinetic model contains more than 100
parameters (see Tables 3–5). However it turns out
21
that only seven parameters are critical: the four
Arrhenius parameters of steps 5 and 10 and the
enthalpies of O , CH3 O and HCOO . It has not
been possible to determine the Arrhenius para-
meters and the enthalpies of formate explicitly. In
order to do so surface science experiments and/or
DFT calculations are needed probing the stability
of formate. In order to make a more complete
model including Eqs. (6)–(10) more critical pa-
rameters will have to be determined. We believe
that the experimental and theoretical knowledge of
this reaction does not justify such a model at
present. The present work has shown that the
surface mechanism proposed by Wachs and Madix
can explain industrial formaldehyde synthesis.
However this does not explicitly prove that this is
in fact the case. In order to validate the model or
discard it high quality steady-state kinetics pref-
erable close to industrial conditions are needed
badly. Such kinetic experiments are lacking at
present due to the difficulties in avoiding mass
transfer limitation, non-isothermity, deactivation,
and aging effects. Furthermore it would be very
interesting to investigate formaldehyde synthesis
on different single crystals especially Ag(1 1 1) to
determine the structure sensitivity of the reaction.
5. Conclusion
We have formulated a simple microkinetic
model for the oxidation of methanol on silver
based on surface science at UHV and low tem-
peratures. The reaction mechanism used in the
model is a simple Langmuir–Hinshelwood mech-
anism, with a statistical mechanical treatment of
gases and intermediates, having one type of active
oxygen and one route to formaldehyde and carbon
dioxide, respectively. The mechanism contains two
slow steps: methoxy decomposition to formalde-
hyde (step 5) and formate decomposition (step 10).
All other steps are assumed to be equilibrated.
Step 5 controls the rate of methanol oxidation to
formaldehyde, whereas steps 10 controls the
combustion of formaldehyde (selectivity). The
model has been used to simulate different ki-
netic experiments and it explains observed reaction
orders, selectivity, apparent activation enthalpies,
22 A. Andreasen et al. / Surface Science 544 (2003) 5–23
and the choice of industrial reaction conditions.
Interestingly the model explains experiments usu-
ally used to disprove that the mechanism deduced
at UHV is responsible for industrial formaldehyde
synthesis. The most important conclusions of this
work is that the mechanism of Wachs and Madix
can explain industrial formaldehyde synthesis and
that only one kind of oxygen (Oa ) needs to be
active in order to do so. This does not prove that
the mechanism of Wachs and Madix is correct at
industrial conditions but it is a likely and simple
explanation. Especially considering the recent ad-
vances in DFT and microkinetic modeling proving
a close connection between surface science and
industrial catalysis. We believe that the reason
other researchers find mechanisms involving sev-
eral kinds of atomic oxygen is due to the fact that
they deduce their mechanisms from transient ex-
periments where these other oxygen species be-
comes important due to dynamic effects, at these
conditions. It is plausible that subsurface oxygen
can participate in a sort of Mars–van Krevelen
mechanism replenishing the surface oxygen. How-
ever we emphasize that we believe that these effects
are unimportant at industrial steady-state condi-
tions. The microkinetic model is simple and in-
complete and only serves to demonstrate some of
the misunderstandings regarding formaldehyde
synthesis in the literature.
References
[1] C.N. Satterfield, Heterogeneous Catalysis, second ed.,
McGraw-Hill, 1991.
[2] S.K. Bhattacharyya, N.K. Nag, N.D. Ganguly, J. Catal. 23
(1971) 158–167.
[3] UllmannÕs Encyclopedia of Industrial Chemistry, fifth ed.,
vol. A11, VCH Verlagsgesellschaft, 1988.
[4] M.V. Twigg, Catalyst Handbook, Manson Publishing,
1996.
[5] Kirk-Othmer Encyclopedia of Chemical Technology, vol.
11, John Wiley & Sons, 1994.
[6] J.F. Griffiths, G. Skirrow, Oxidat. Combust. Rev. 3 (1968)
47–96.
[7] V.N. Gavrilin, B.I. Popov, Kinet. Catal. (USSR) 6 (1965)
799.
[8] A. Nagy, G. Mestl, T. R€ uhle, G. Weinberg, R. Sch€ ogl,
J. Catal. 179 (1998) 548–559.
[9] A. Nagy, G. Mestl, Appl. Catal. A 188 (1999) 337–353.
[10] I.E. Wachs, R.J. Madix, Surf. Sci. 76 (1978) 531–558.
[11] D.A. Robb, P. Harriott, J. Catal. 35 (1974) 176–183.
[12] Q. Dai, A.J. Gellmann, Surf. Sci. 257 (1991) 103–112.
[13] J. Wang, X. Xu, J. Deng, Y. Liao, B. Hong, Appl. Surf.
Sci. 120 (1997) 99–105.
[14] C.-F. Mao, M.A. Vannice, J. Catal. 154 (1995) 230–244.
[15] B. Xinhe, D. Jingfa, J. Catal. 99 (1986) 391–399.
[16] T.E. Felter, W.H. Weinberg, G.Y. Lastushkina, P.A.
Zhdan, G.K. Boreskov, J. Hrbek, Appl. Surf. Sci. 16
(1983) 351–364.
[17] M.A. Barteau, M. Bowker, R.J. Madix, Surf. Sci. 94 (1980)
303–322.
[18] I.E. Wachs, R.J. Madix, Appl. Surf. Sci. 1 (1978) 303–
328.
[19] M. Mavrikakis, M.A. Barteau, J. Mol. Catal. A 131 (1998)
135–147.
[20] J. Greeley, M. Mavrikakis, J. Catal. 208 (2002) 291–300.
[21] P. Mars, D. van Krevelen, Chem. Eng. Sci. Spec. Suppl. 3
(1954) 41.
[22] T. Schedel-Niedrig, X. Bao, M. Muhler, R. Schl€ ogl, Ber.
Bunsenges. Phys. Chem. 101 (7) (1997) 994–1006.
[23] X. Bao, M. Muhler, Th. Schedel-Niedrig, R. Schl€ ogl, Phys.
Rev. B 54 (3) (1996) 2249–2262.
[24] X. Bao, B. Pettinger, G. Ertl, R. Schl€ ogl, Ber. Bunsenges.
Phys. Chem. 97 (3) (1993) 322–325.
[25] X. Bao, J.V. Barth, G. Lehmpfuhl, R. Schuster, Y. Uchida,
R. Schl€ogl, G. Ertl, Surf. Sci. 284 (1993) 14–22.
[26] X. Bao, M. Muhler, B. Pettinger, R. Schl€ ogl, G. Ertl,
Catal. Lett. 22 (1993) 215–225.
[27] A.J. Nagy, G. Mestl, D. Herein, G. Weinberg, E. Kitzel-
mann, R. Schlogl, J. Catal. 182 (1999) 417–429.
[28] C.T. Campbell, J. Catal. 94 (1985) 436–444.
[29] V.I. Bukhtiyarov, V.V. Kaichev, I.P. Prosvirin, J. Chem.
Phys. 111 (5) (1999) 2169–2175.
[30] C.I. Carlisle, T. Fujimoto, W.S. Sim, D.A. King, Surf. Sci.
470 (2000) 15–31.
[31] C.I. Carlisle, D.A. King, M.-L. Bocquet, J. Cerda, P.
Sautet, Phys. Rev. Lett. 84 (2000) 3899.
[32] X. Bao, M. Muhler, B. Pettinger, Y. Uchida, G. Lehmp-
fuhl, R. Schl€ ogl, G. Ertl, Catal. Lett. 32 (1995) 171–
183.
[33] L. Leffert, J.G. van Ommen, J.R.H. Ross, J. Chem. Soc.
Faraday Trans. I 83 (10) (1987) 3161–3165.
[34] L. Leffert, J.G. van Ommen, J.R.H. Ross, Appl. Catal. 34
(1987) 329–339.
[35] D.J. Sajkowski, M. Boudart, Catal. Rev. Sci. Eng. 29 (4)
(1987) 325–360.
[36] C.-B. Wang, G. Deo, I.E. Wachs, J. Phys. Chem. B 103
(1999) 5645–5656.
[37] L. Leffert, J.G. van Ommen, J.R.H. Ross, Appl. Catal. 23
(1986) 385–402.
[38] M. Qian, M. Liauw, G. Emig, Appl. Catal. A 238 (2003)
211–222.
[39] A.E. Obraztsov, B.I. Popov, M.P. Shashalevich, L.P.
Taraban, E.A. Pronina, A.G. Shevchenko, Kinet. Catal.
(USSR) 12 (1971) 970.
[40] A.C. van Veen, O. Hinrichsen, M. Muhler, J. Catal. 210
(2002) 53–66.
 A. Andreasen et al. / Surface Science 544 (2003) 5–23
[41] J. Dumesic, D. Rudd, L. Aparicio, J. Rekoske, A. Trevi~ no,
The Microkinetics of Heterogeneous Catalysis, ACS, 1993.
[42] R.D. Cortright, J.A. Dumesic, Adv. Catal. 46 (2001) 161–
264.
[43] P. Stoltze, Phys. Scr. 36 (1987) 824–864.
[44] C.V. Ovesen, B.S. Clausen, B.S. Hammershøi, G. Steffen-
sen, T. Askgaard, I. Chorkendorff, J.K. Nørskov, P.B.
Rasmussen, P. Stoltze, P. Taylor, J. Catal. 158 (1996) 170–
180.
[45] T. Askgaard, J.K. Nørskov, C.V. Ovesen, P. Stoltze,
J. Catal. 156 (1995) 229–242.
[46] C.T. Campbell, Topics Catal. 1 (1995) 353.
[47] S.H. Oh, G.B. Fisher, J.E. Carpenter, D.W. Goodman,
J. Catal. 100 (1986) 360.
[48] V.P. Zhdanov, B. Kasemo, Surf. Sci. Rep. 20 (1986) 111.
[49] N. Waletzko, L.D. Schmidt, AIChE J. 34 (1988) 1146.
[50] D.A. Hickman, L.D. Schmidt, AIChE J. 39 (1993) 1164.
[51] C.T. Campbell, M.T. Paffett, Surf. Sci. 143 (1984) 517–535.
[52] M.A. Barteau, R.J. Madix, J. Chem. Phys. 74 (7) (1981)
4144–4149.
[53] I. Stensgaard, E. Laegsgaard, F. Besenbacher, J. Chem.
Phys. 103 (22) (1995) 9825–9831.
[54] A.G. Sault, R.J. Madix, Surf. Sci. 176 (1986) 415–424.
[55] M. Bowker, M.A. Barteau, R.J. Madix, Surf. Sci. 92 (1980)
528–548.
[56] L.E. Fleck, Z.C. Ying, M. Feehery, H.L. Dai, Surf. Sci. 296
(1993) 400–409.
[57] E.M. Stuve, R.J. Madix, B.A. Sexton, Surf. Sci. 111 (1981)
11–25.
[58] H.G. Jenniskens, P.W.F. Dorlandt, M.F. Kadodwala,
A.W. Kleyn, Surf. Sci. 357–358 (1996) 624–628.
[59] I.E. Wachs, R.J. Madix, J. Catal. 53 (2) (1978) 208–227.
[60] B.A. Pepley, J.C. Amphlett, L.M. Kearns, R.F. Mann,
Appl. Catal. A 179 (1999) 31–49.
[61] B.A. Sexton, R.J. Madix, Surf. Sci. 105 (1981) 177–195.
[62] E.M. Stuve, R.J. Madix, B.A. Sexton, Surf. Sci. 119 (1982)
279–290.
[63] H.M.C. Sosnovsky, J. Chem. Phys. 23 (8) (1955) 1486–
1490.
[64] H.M.C. Sosnovsky, J. Phys. Chem. Solids 10 (1959) 304–
310.
[65] J. Bagg, H. Jaeger, J.V. Sanders, J. Catal. 2 (1963) 449–464.
[66] H. Jaeger, J. Catal. 9 (1967) 237–250.
23
[67] C. Campbell, Surf. Sci. 157 (1985) 43–60.
[68] R.B. Grant, R.M. Lambert, Surf. Sci. 146 (1984) 256–
268.
[69] P. Stoltze, Prog. Surf. Sci. 65 (2000) 65–100.
[70] C.V. Ovesen, B.S. Clausen, J. Schiøtz, P. Stoltze, H.
Topsøe, J.K. Nørskov, J. Catal. 168 (1997) 7133.
[71] H. Topsøe, C.V. Ovesen, B.S. Clausen, N.-Y. Topsøe,
Stud. Surf. Sci. Catal. 109 (1997) 121.
[72] F. Reif, Fundamentals of Statistical and Thermal Physics,
McGraw-Hill, 1985.
[73] M.W. Chase Jr. (Ed.), NIST-JANAF Thermochemical
Tables, fourth ed., J. Phys. Chem. Ref. Data, Monograph
No. 9, Parts I and II.
[74] P.A. Taylor, P.B. Rasmussen, C.V. Ovesen, P. Stoltze, I.
Chorkendorff, Surf. Sci. 261 (1992) 191–206.
[75] C.V. Ovesen, P. Stoltze, J.K. Nørskov, C.T. Campbell,
J. Catal. 134 (1992) 445–446.
[76] R. Zhang, A.J. Gellman, J. Phys. Chem. 95 (1991) 7433–
7437.
[77] T. Askgaard, Catalytic reaction mechanisms, Ph.D. Thesis,
Physics Department, Technical University of Denmark,
1994.
[78] P.A. Redhead, Vacuum 12 (1962) 203.
[79] A. Raukema, D.A. Butler, F. Box, A.W. Kleyn, Surf. Sci.
347 (1996) 151–168.
[80] S.R. Bare, K. Griffiths, W.N. Lennard, H.T. Tang, Surf.
Sci. 342 (1995) 185–198.
[81] C. Stegelmann, P. Stoltze, in preparation.
[82] R.I. Masel, Principles of Adsorption and Reaction on Solid
Surfaces, John Wiley & Sons, 1996.
[83] B. Hammer, L.B. Hansen, J.K. Nørskov, Phys. Rev. B 59
(11) (1999) 7413–7421.
[84] D.H. Vanderbilt, Phys. Rev. B. 41 (1990) 7892.
[85] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77
(18) (1996) 3865–3868.
[86] J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson,
M.R. Pederson, D.J. Singh, C. Fiolhais, Phys. Rev. B
(1992) 6671.
[87] J.A. White, D.M. Bird, Phys. Rev. B 50 (1994) 4954.
[88] W.-X. Li, C. Stampfl, M. Scheffler, Phys. Rev. Lett. 90
(2003) 256102.
[89] C.T. Campbell, J. Catal. 99 (1986) 28–38.
[90] E.L. Force, A.T. Bell, J. Catal. 38 (1975) 440–460.