Ozone promoted carbon monoxide oxidation on platinum/γ-alumina catalyst

Article abstract of DOI:10.1016/j.jcat.2006.01.002

CO oxidation with oxygen and ozone-oxygen mixtures was studied over a platinum/alumina monolith catalyst. Temperature ramp experiments were combined with mean-field simulations to study the reaction mechanisms. In the absence of ozone, a slow CO oxidation reaction was observed at low temperatures. The rate of this slow reaction was proportional to the square root of the oxygen pressure and independent of the CO concentration. At higher temperatures, the three-step Langmuir-Hinshelwood reaction mechanism dominated the CO oxidations. When some of the oxygen was exchanged for ozone, rapid oxidation of CO by ozone was observed. The suggested explanation was an Eley-Rideal mechanism, in which colliding ozone reacts with adsorbed CO on the platinum surface. When this additional reaction step was included in the model, the simulation results indicated a reduction in the bulk CO pressure. The experimentally observed ozone promotion of CO oxidation was thus attributed to a decrease in CO surface self-poisoning.

Full text of DOI:10.1016/j.jcat.2006.01.002

Journal of Catalysis 238 (2006) 321–329
www.elsevier.com/locate/jcat
Ozone promoted carbon monoxide oxidation on platinum/γ -alumina catalyst
Martin Petersson a,b,c, David Jonsson a,b, Hans Persson a,c,d, Neil Cruise a,e, Bengt Andersson a,b,∗
a
Competence Centre for Catalysis, Chalmers University of Technology, SE-412 96 Göteborg, Sweden
b
Chemical Reaction Engineering, Chalmers University of Technology, SE-412 96 Göteborg, Sweden
c
Volvo Technology AB, Chalmers Science Park, SE-412 88 Göteborg, Sweden
d
Department of Applied Physics, Chalmers University of Technology, SE-412 96 Göteborg, Sweden
e
Perstorp Specialty Chemicals AB, Process and Catalyst, SE-284 80 Perstorp, Sweden
Received 2 September 2005; revised 20 December 2005; accepted 3 January 2006
Available online 20 January 2006
Abstract
CO oxidation with oxygen and ozone–oxygen mixtures was studied over a platinum/alumina monolith catalyst. Temperature ramp experiments
were combined with mean-field simulations to study the reaction mechanisms. In the absence of ozone, a slow CO oxidation reaction was
observed at low temperatures. The rate of this slow reaction was proportional to the square root of the oxygen pressure and independent of
the CO concentration. At higher temperatures, the three-step Langmuir–Hinshelwood reaction mechanism dominated the CO oxidations. When
some of the oxygen was exchanged for ozone, rapid oxidation of CO by ozone was observed. The suggested explanation was an Eley–Rideal
mechanism, in which colliding ozone reacts with adsorbed CO on the platinum surface. When this additional reaction step was included in the
model, the simulation results indicated a reduction in the bulk CO pressure. The experimentally observed ozone promotion of CO oxidation was
thus attributed to a decrease in CO surface self-poisoning.
© 2006 Elsevier Inc. All rights reserved.
Keywords: Catalysis; Low-temperature activity; CO oxidation; Ozone; Platinum; Pt/Al O ; Mean-field modeling; Simulations
2
3
1. Introduction
shown to describe the reaction kinetics at atmospheric pressure
if the lateral repulsion between nearest-neighbor CO–CO is ex-
plicitly taken into account [4].
The total oxidation of CO has been widely studied due to its
importance in emission control. It is also a simple reaction be-
tween diatomic species and thus is well suited for theoretical
studies. One of the first studies was presented by Langmuir [1],
who suggested an Eley–Rideal type of reaction between ad-
sorbed oxygen and colliding CO. Later works [2,3] showed that
the reactions follow the three-step Langmuir–Hinshelwood re-
action scheme under UHV conditions:
Because CO and oxygen are competing for the same adsorp-
tion sites in the foregoing three-step Langmuir–Hinshelwood
model, the reaction might be self-poisoned under some gas
compositions. CO has a high initial sticking (in the order of 0.8)
compared with the low initial oxygen sticking (usually reported
as <0.1) [5]. Furthermore, oxygen requires two adjacent free
sites, whereas CO requires just one site. As a result, the fore-
going model predicts that the platinum surface will be almost
completely CO-covered at low temperatures unless the oxygen
content in the gas is much higher than the CO concentration.
Therefore, virtually no reactions will occur unless the catalyst
temperature is increased and the CO starts to desorb. CO des-
orption will provide access to the surface for oxygen, resulting
in rapid oxidation of CO. The model thus predicts two stable
states for CO oxidation, one state with high CO coverage and
a low rate of oxidation and the other with high oxygen cover-
COg + S ꢀ CO–S,
(O2)g + 2S → 2O–S,
CO–S + O–S → (CO2)g + 2S,
where the subscripts “g” refer to a gaseous species and S corre-
(1)
(2)
(3)
sponds to a platinum site. The same reaction scheme has been
*
Corresponding author. Fax: +46 31 772 3035.
E-mail address: bengta@chemeng.chalmers.se (B. Andersson).
0021-9517/$ – see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcat.2006.01.002

322
M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
age and a high rate of oxidation. The bistable kinetics of the CO
oxidation reaction has been studied previously [6–8].
Recently, an alternative reaction path for CO oxidation on
platinum was suggested by Bourane and Bianchi [9–13]. Ac-
cording to this mechanism, the dissociative adsorption of oxy-
gen occurs on a second site (denoted by S2) without the com-
petition of CO, followed by a consecutive reaction with ad-
sorbed CO on the ordinary platinum sites (denoted by S1). The
heat of sorption for this oxygen species was found to be low
(30 kJ/mol), which is why the specie was termed “weakly ad-
sorbed” oxygen:
Table 1
Pt/Al O monolith sample properties
2
3
Property
Value
Unit
Diameter
Length
20
40
mm
mm
g
Sample weight
Washcoat weight
Platinum content
BET surface area
Noble metal dispersion
6.68
1.12
1.0
22.5
10.7
g
wt%
2
m /g
%
2. Experimental
COg + S1 ꢀ CO–S1,
(4)
(5)
(6)
2.1. Catalyst preparation and characterization
(O2)g + S2 ꢀ 2O–S2,
A cylindrical core (diameter, 20 mm; length, 50 mm) was
cut out of a commercial corderite monolith (Corning Glass;
400 cpsi, 6.5-mil wall thickness). This core was coated with γ -
alumina (16.8 wt%) and calcined at 873 K for 1 h in air. It was
then impregnated with a platinum ammonium nitrate solution
(Johnson Matthey) to achieve a platinum content of 1 wt% of
the washcoat, using the incipient wetness impregnation method,
and finally calcined at 823 K for 90 min in air.
This catalyst sample was trimmed off to 40 mm. The ex-
cess material was used to analyze total surface area and no-
ble metal dispersion using a Multisorb ASAP 2010 instrument
(Micromeretics). The total surface area was determined from
five nitrogen adsorption–desorption measurements at relative
nitrogen pressures of 0.06–0.21, assuming a 0.162-nm2 cross-
sectional area of the nitrogen adsorbate. The noble metal dis-
persion was determined from the irreversible CO adsorption at
307 K. The irreversible CO adsorption was determined as the
difference between two repeated CO adsorption measurements
at five different CO pressures. Each surface platinum atom was
assumed to adsorb 0.7 CO molecules [31]. The basic character-
istics of this catalyst are given in Table 1.
CO–S1 + O–S2 → (CO2)g + S1 + S2.
The adsorption equilibrium of the weakly adsorbed oxygen
[Eq. (5)] explains the observed initial reaction order of 1/2 for
oxygen in CO excess [11], where the conventional three-step
Langmuir–Hinshelwood model would predict a reaction order
of zero for oxygen. At catalyst ignition, Bourane and Bianchi
suggested that the reduced platinum surface is rapidly converted
into an oxygen-covered surface with different kinetics, thus ex-
plaining the observed bistable kinetics of CO oxidation [13].
The self-poisoning of CO oxidation has been well estab-
lished in many industrial applications. Several techniques have
been evaluated for decreasing CO blockage of the platinum sur-
face. The techniques evaluated thus far include the addition of
various promoters, which will allow oxygen to spillover to the
reactive sites [14–18] or create new adsorption sites for oxygen
at the noble metal surface [19,20]. Other approaches include
optimal positioning of the noble metal [21], periodic variations
in exhaust gas composition [22,23], and photo-enhanced oxida-
tion [24].
2.2. Flow reactor experiments
One alternative approach suggested in the patent literature is
to add ozone to the exhaust gas [25]. CO oxidation in the pres-
ence of ozone has been studied over MnO2 [26], CeO2 [27],
Co2O3 [28], and Au/Fe2O3 [29]. It has also been used to pro-
mote the total oxidation of volatile organic compounds over
noble metal catalysts [30]. As far as we know, however, the
present study is the first scientific study of CO oxidation over
supported platinum catalysts in the presence of ozone.
In the current study, oxidation of CO was studied in oxy-
gen and in ozone–oxygen mixtures to investigate the promo-
tional effect of ozone on the CO oxidation reaction. To study
the reaction mechanisms of CO oxidation with oxygen and
with ozone, flow reactor experiments were combined with
mean-field modeling. In the model, a conventional three-step
Langmuir–Hinshelwood mechanism for CO oxidation in oxy-
gen was combined with the noncompeting model of Bourane
and Bianchi to describe the observed CO oxidation even at al-
most complete CO coverage. Reaction steps for ozone reaction
with CO and ozone decomposition are also included.
CO oxidation experiments were conducted in a vertically-
mounted metal tube reactor. The reactor tube had a diameter
of 20 mm, but had a flow restrictor that decreases the effective
sample diameter to 17.5 mm. The gas entering the reactor was
heated by a capillary preheater, and the reactor wall was heated
by metallic heating coils. The temperature was measured by
thermocouples before and after the catalyst and at the reactor
wall; separate temperature control units were used to regulate
the temperature in the capillary preheater and the tube reactor.
A sintered quartz frits was mounted between the preheater and
the catalyst sample to distribute the gas evenly over the reactor
cross-section.
Three separate mass flow controllers (Brooks Instrument
models 5850E and 5850S) were used to mix the reactor feed
stream from argon (AGA gas, 99.996%), CO (AGA gas, 99%),
and oxygen (AGA gas, 99.95%). The oxygen gas stream
was flown through a high-voltage discharge ozone generator
(Thermo Electron model 100B) before the gases was mixed.

M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
323
During the experiments without ozone, the ozone generator
was turned off. All tubing between the ozone generator and the
preheater was made of Teflon, and the amount of tubing was
minimized to minimize ozone decomposition. The CO concen-
tration in the gas leaving the reactor was determined with an
infrared analyzer (Thermo Environmental Instruments model
48C).
CO conversion was determined under constant heating
ramps from 350 to 555 K at 10 K/min. In all experiments, the
overall gas flow was 3207 mL/min, corresponding to a space
velocity of 20,000 h−1. The CO concentration was varied be-
tween 0.1 and 0.5%, and the oxygen concentration was varied
between 1 and 5%.
To determine the ozone concentration reaching the catalyst,
empty reactor measurements were conducted. The ozone con-
centration was recorded using a UV analyzer (Dasibi Enviro-
mental model 1000-AH). Due to the limited maximum mea-
surement range of the UV analyzer, the sample gas was diluted
with air at dilution ratios of 25.9 and 44.7 to give analyzable
concentrations. These experiments were then repeated with CO
present in the feed, to determine whether the homogeneous gas-
phase reaction between ozone and CO could affect the results.
Furthermore, an alumina-coated monolith sample, prepared as
above with no impregnation of noble metal, was used to deter-
mine whether any reaction occurred on the support.
the reactor became significant. No ozone was detected from the
empty reactor above 530 K.
To determine the CO oxidation in any competing reactions,
measurements were conducted over a γ -alumina-coated sam-
ple with and without ozone. The measurements were con-
ducted with 0.1 CO and 5% oxygen at constant temperature
points. Without ozone, no CO oxidation could be detected be-
low 425 K, whereas up to 8% conversion could be detected at
435 K. In the presence of ozone, some CO oxidation could be
detected at 375 K and almost 17% conversion was detected at
435 K.
3.2. Measured carbon monoxide oxidation during heating
ramps
The temperature ramp experiments are presented in Figs.
2A–C. The oxidation of CO by oxygen was increased by in-
creased oxygen concentrations and by decreased CO concen-
trations. The observed dependence of the CO concentration
indicates that the oxidation reaction was self-poisoned by CO.
3. Results and discussion
3.1. Measured ozone concentration and blank sample tests
Fig. 1 shows the concentration of ozone in the gas leaving
the empty reactor. The ozone concentration was found to be
weakly dependent on the oxygen concentration. Measurements
in the gas stream entering the reactor indicated that a minor part
of the ozone was decomposed in the reactor at 323 K (<1%
conversion). At higher temperatures, ozone decomposition in
Fig. 2. Measured CO concentrations during ramp experiments (10 K/min) with
(A) 0.5, (B) 0.3, and (C) 0.1% CO. Squares, triangles and circles corresponds
to 1, 3, and 5% oxygen. Empty and filled symbols represent measurements with
and without ozone, respectively.
Fig. 1. Measured ozone concentration at 323 K with varying oxygen concentra-
tions.

324
M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
It was also found that ozone has a promotional effect on the
CO oxidation. This effect was more pronounced at low CO
concentrations, but seemed to be independent of the oxygen
concentration. Finally, at temperatures well below catalyst light
off, some CO oxidation was found in the presence of ozone.
This low-temperature reaction was increased with increasing
oxygen concentration and decreasing CO concentration.
Comparing these results with the blank alumina oxidation
experiments presented in Section 3.1 shows that the rate of CO
oxidation on the blank sample was small. At 425 K with 0.1%
CO, 5% oxygen, and no ozone, the platinum-coated sample had
about 70% conversion, compared with the 8% measured with
the blank sample. The same findings were seen in the presence
of ozone, with complete oxidation measured over the platinum-
containing sample and only 17% conversion over the blank
sample.
One important finding shown in Figs. 2A–C is the initial
slow decrease in CO concentration in the absence of ozone. For
the experiment with 0.5 CO and 5% oxygen (Fig. 2A, black
squares), CO concentration began to decline around 400 K, but
catalyst light-off did not occur until the inlet temperature ex-
ceeded 475 K. A similar slow initial CO oxidation reaction was
observed for all the other temperature ramp experiments both
with and without ozone.
To study this slow reaction in detail, the CO oxidation rate
was calculated from the low conversion points (<20% conver-
sion) without ozone. The low conversion made it possible to
approximate the reactor as one continuous stirred tank. Quasi-
steady state was assumed, because the observed reaction rate
was approximately 0.5/s and the collision rate of CO and oxy-
gen on a Pt site was on the order of 105–107/s for the current
inlet gas pressures. The results from these calculations are given
in Fig. 3A. It was found that rate of oxidation was determined
by the oxygen concentration and that the rate was independent
of the CO concentration. Investigating the oxygen concentra-
tion dependence, the rate was found to be proportional to the
square root of the oxygen pressure, as shown in Fig. 3B. This
slow CO oxidation thus appears to follow the model suggested
by Bourane and Bianchi, which is described in Eqs. (4)–(6).
Fig. 3. (A) Approximate CO oxidation rate during ramp experiments
(10 K/min) at conversions less than 20%. Filled squares corresponds to 0.1%
CO and 5% oxygen, filled triangles 0.3% CO and 5% oxygen, filled circles
0.5% CO and 5% oxygen, cross 0.1% CO and 3% oxygen, stars 0.3% CO
and 3% oxygen, filled diamonds 0.5% CO and 3% oxygen, open squares 0.1%
CO and 1% oxygen, open triangles 0.3% CO and 1% oxygen, and open cir-
cles 0.5% CO and 1% oxygen. (B) Approximate CO oxidation rate divided
with the square root of the inlet oxygen concentration during ramp experiments
(10 K/min) at conversions less than 20%. Annotations as above.
r
7
CO–S1 + O–S2 →(CO2)g + S1 + S2,
(11)
3.3. Model for CO oxidation with oxygen
where S1 and S2 correspond to two different platinum sites.
The three-step Langmuir–Hinshelwood mechanism [Eqs. (7)–
(9)] was used to explain the CO self-poisoning observed at
elevated conversions, whereas the model proposed by Bourane
and Bianchi [Eqs. (7) and (10)–(11)] was essential to describe
the slow CO oxidation discussed in Section 3.2.
The CO oxidation reaction was described by the conven-
tional three-step Langmuir–Hinshelwood mechanism, com-
bined with the noncompeting mechanism suggested by Bourane
and Bianchi:
r1
The surface coverages of reactants and bulk gas concen-
trations were expressed according to the standard mean-field
model. The rate expressions and parameter values used in the
simulation are given in Table 2. The collision rate of CO and
oxygen on site S1 was calculated according to kinetic gas the-
ory with a Pt site size of 8 Å2 [32]. The sticking coefficient
for oxygen on site S1 was set proportional to P00, that is, the
probability that a pair of nearest neighbor sites is vacant. This
COg + S1 ꢀCO–S1,
(7)
r2
r3
(O2)g + 2S1 →2O–S1,
(8)
(9)
r4
CO–S1 + O–S1 →(CO2)g + 2S1,
r5
probability was taken from the quasi-chemical approximation
(O2)g + S2 ꢀ2O–S2,
(10)
rep
r6
[33] with a CO–CO lateral interaction of E
(see Table 2)
CO

M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
325
Table 2
Kinetic parameters for simulation of CO oxidation on Pt/Al O . All rates are expressed per site S , except r which is expressed per weight of washcoat. Parameters
2
3
1
9
with confidence intervals were fitted in the current study
Process and rate expression
Parameter
Value
0.84
95% conf.
Ref. [6]
Unit
(0)
s
CO adsorption on site S :
1
CO
(0)
CO
2
CO
a
CO
r = s · (1 − θ ) · k
p
CO,w
1
d
ν
15
−1
CO desorption from site S :
1.25 × 10
180
Ref. [6]
s
1
CO
ꢀ
ꢀ
ꢁ
ꢁ
d,0
d,0
CO
d
CO
r = ν exp − E (1 − αθ ) /RT · θ
E
Ref. [37]
0.0011
kJ/mol
2
CO
CO
CO
α
0.5254
(0)
Oxygen adsorption on site S :
s
0.02
Ref. [30]
1
O ,S
2
1
ꢀ
ꢁ
rep
CO
rep
CO
(0)
a
r = s
P
θ
, θ 1 , E
· k
p
E
26.34
0.11
kJ/mol
3
00 CO O,S
O ,w
2
O ,S
2
O ,S
1
2
1
14
−1
CO reacting with oxygen from site S :
ν
1.65 × 10
100.9
0.5
Ref. [6]
Ref. [6]
Ref. [30]
Ref. [9]
s
1
4
ꢀ
ꢁ
0
r1,0
r = ν exp −E (1 − βθ )/RT · θ
θ
E
β
E
kJ/mol
4
4
CO
CO O,S
4
1
d
Oxygen equilibrium on site S :
30
kJ/mol
2
O ,S
2
2
ꢂ
ꢃ
d
3
E
(h
)
1
plank
O ,S
2
2
K
=
exp
O ,S
2
2
3/2 5/2
RT
k (2πm k )
T
B
B
O
Oxygen desorption from2site S :
2
ꢀ
ꢁ
d
2
r = (k T/h
)exp −E
2 /RT · θ
2
B
6
plank
O ,S
O,S
2
2
CO reacting with oxygen from site S :
E
n
67.66
5.49 × 10
0.36
3.0 × 10
kJ/mol
mol/m
7
−4
−5
3
r = (k T/h
)exp(−E /RT ) · θ · (n /n ) · θ
7
B
7
S
S
S
plank
CO
O,S
2
2
1
2
CO reacting with ozone
a
r = k
p
θ
O ,w CO
3
8
O ,S
3
1
O
decomposition on site S :
ν
9
1.95
0.16
mol/(Pa kg s)
3
3
r = ν p
9
9 O ,w
3
and all other lateral interactions assumed to be negligible, as
described previously [34]. The rate constant for adsorption of
oxygen on site S2 was calculated from the adsorption equilib-
rium constant and the rate constant for desorption.
drop in catalyst outlet temperature was observed at high tem-
peratures, attributed to the radiation heat transfer between the
monolith and the reactor lid. The finite area view factors for
this heat transfer were taken from Ref. [38]. The reactor lid was
modeled as one thermal mass that interacted with the mono-
lith only through radiation heat transfer. The heat capacity of
the lid was fitted to 0.432 0.021 J/K. The radiation heat
transfer was low compared with the conducted heat transfer
to the reactor wall. The radial heat loss coefficient was fitted
to 13.08 0.66 W/(m2 K). The heat conductivity and specific
heat capacity of the monolith [39] and the reaction enthalpy
for CO oxidation [34] were as reported previously. The ki-
netic and heat transfer parameters were fitted to the measured
CO instrument signal and the catalyst outlet temperature from
the ramp experiments without ozone, using lsqnonlin in Mat-
lab.
The transient response of the monolith catalyst was mod-
eled as a single monolith channel comprising of 30 continu-
ous stirred tank reactors. Mass transport between the gas bulk
and the catalyst wall was described with the film model [35].
The mass transfer coefficient was calculated from a Sherwood
number for monolith channels [36] with an asymptotic value
of 2.98 [37]. The diffusivities and viscosities were calculated
according to the Chapman–Enskog equations [35]. Pore trans-
port limitations were not included in the model, because Weisz
modulus calculations indicated that these limitations would not
significantly affect the results. The tank series model for the
monolith was coupled to a model describing the response time
of the CO instrument. This CO instrument model consisted of
one time lag and one continuous stirred tank. The time lag (15
s) and tank volume (1.49 0.03 dm3) were fitted to step-change
experiments at low temperature (not shown), using lsqnonlin in
Matlab.
The heat transfer coefficient, used in the bulk gas mass bal-
ance, was determined by setting the Nusselt number equal to
the Sherwood number. The heat conductivity of the gas was de-
termined using the Chapman–Enskog equation [35]. The heat
balance for the catalyst wall included heat convection from the
bulk gas, heat conduction between neighboring tanks, radiation
heat transfer from reactor inlet and outlet to each tank, radial
heat losses, and CO oxidation enthalpy. Radiation heat transfer
between the tanks in the reactor was neglected because it was
low compared with heat conduction in the monolith. A slight
3.4. Simulation results for CO oxidation with oxygen
The modeled instrument signals are compared with the mea-
sured signal in Fig. 4. Good agreements between the results
were found for all CO and oxygen concentration combinations.
If the model suggested by Bourane and Bianchi is removed
from the CO oxidation model, the deviation between the sim-
ulation and the measured results becomes significant before
catalyst light-off, as shown in Fig. 5.
The detailed simulation results for the model with the re-
action path suggested by Bourane and Bianchi for 0.1% CO
and 5% oxygen and 0.5% CO and 1% oxygen are shown in
Fig. 6. At low temperatures, the coverage of CO on site S1 was
close to unity, resulting in very low oxygen coverage at site S1

326
M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
Fig. 4. Measured and simulated CO instrument signals during ramp experiments
(10 K/min) without ozone and with (A) 0.5, (B) 0.3, and (C) 0.1% CO. Simu-
lations include the model proposed by Bourane and Bianchi. Squares, triangles
and circles corresponds to measurements with 1, 3, and 5% oxygen. Dashed,
dotted and solid lines correspond to simulations with 1, 3 and 5% oxygen.
Fig. 5. Measured and simulated CO instrument signals without the model pro-
posed by Bourane and Bianchi during ramp experiments (10 K/min) without
ozone and with (A) 0.5, (B) 0.3, and (C) 0.1% CO. The parameters from Ta-
ble 2 were used in the simulation, but with the number of S sites (n 2 ) set to
2
S
zero. Squares, triangles and circles corresponds to measurements with 1, 3 and
5% oxygen. Dashed, dotted and solid lines correspond to simulations with 1, 3,
and 5% oxygen.
and consequently a low rate for the surface reaction on site S1
[Eq. (9)]. The overall CO oxidation rate was instead controlled
by the reaction with oxygen on site S2 [Eqs. (10) and (11)].
The slow rate of this surface reaction [Eq. (11)] resulted in oxy-
gen coverage close to equilibrium on site S2. This coverage was
thus proportional to the square root of the oxygen pressure, as
is seen when comparing Figs. 6G and H. As the temperature
was increased, CO adsorbed on S1 sites began to desorb, caus-
ing the rate of oxygen adsorption on S1 sites to increase and
the rate of reaction between CO and oxygen on site S1 to be-
come increasingly important. When the CO coverage decreased
below a certain threshold level, a rapid transformation from
a CO-covered surface to an oxygen-covered surface occurred.
The rapid transformation shows that the CO oxidation reaction
was controlled by the adsorption/desorption processes [Eqs. (7)
and (8)] rather than surface kinetics [Eq. (9)]. The threshold
level was dependent on the CO and oxygen concentrations,
and the transformation started from the end of the catalyst and
migrated toward the inlet. At high temperatures, the bulk gas
concentration leveled out to a finite level in the first couple of
tanks, indicating that the rate of oxidation became external mass
transfer-limited.
The three-step Langmuir–Hinshelwood part of the model
above is closely related to the model presented by Rinnemo
et al. [34]. In the present model, the desorption energy of CO
at low coverage was higher than in the model suggested by
Rinnemo (180 kJ/mol compared with 146 kJ/mol). The re-
ported CO desorption energies for platinum single crystals with
low CO coverage range from 138 [2] to 180 kJ/mol [40],
with reports of energies up to 220 kJ/mol for supported plat-
inum/alumina catalysts [41]. Previous CO modeling studies
performed by our group found CO desorption energies of
180 kJ/mol [21,42]. The increase in low-coverage CO desorp-
tion energy also required a change in CO cover dependence of
the CO desorption energy and CO–CO lateral interaction.
The oxygen adsorption and reaction model on site S2 was
based on the model of Bourane and Bianchi. To combine it
with the three-step Langmuir–Hinshelwood model, the model

M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
327
with colliding ozone molecules. The alternative explanation
that ozone dissociates and reacts with adsorbed CO is less likely
because of the low number of free platinum sites and the obser-
vation that the apparent stoichiometric coefficient was <1.
No ozone decomposition occurred over the empty reactor
at 323 K, but significant ozone conversion was detected at ele-
vated temperatures. Comparing the residence time in the reactor
with the lifetime of ozone in air at different temperatures [43]
demonstrated that the observed ozone conversion can be at-
tributed to a homogeneous gas-phase reaction. But this reaction
was negligible compared with the heterogeneous reaction on
alumina, because no ozone could be detected over an alumina-
coated monolith, indicating that ozone decomposition was rapid
even on the alumina surface. Therefore, the measured outlet
ozone concentration presented in Fig. 1 was used as the ozone
inlet concentration in the ozone simulations.
To model the observed reactions, it was thus sufficient to
complement the reaction mechanism for CO oxidation in the
absence of ozone [Eqs. (7)–(11)] with one Eley–Rideal reaction
between ozone and adsorbed CO and one ozone decomposition
step on the alumina surface:
r8
(O3)g + CO–S1 →(O2)g + (CO2)g + S1,
(12)
(13)
3
r9
(O3)g → (O2)g,
S3
2
where S3 corresponds to a site on the alumina wash coat. No
ozone decomposition reaction on platinum was included, be-
cause of the low number of free platinum sites observed in the
CO oxidation calculations with oxygen, which is why this re-
action was assumed to be negligible. The rate expressions for
the two reactions are given in Table 2. The rate constant for
ozone colliding and reacting with adsorbed CO [Eq. (12)] was
calculated according to kinetic gas theory, with a reaction prob-
ability of 1. The ozone decomposition rate constant (ν9) was
fitted to the CO instrument signal during the ramp experiments
with ozone, using lsqnonlin in Matlab. It was assumed that the
decomposition of ozone on site S3 [Eq. (13)] was not activated.
The decomposition rate was expressed per mass of wash coat
rather than per site, because the site density was not yet estab-
lished. The value of ν9 is given in Table 2. The ozone reactions
were implemented in the mass and heat transfer model from
Section 3.4. The Lennard–Jones parameters needed to deter-
mine ozone diffusivity were estimated from the critical temper-
ature and volume of ozone [35]. The reaction enthalpy of ozone
was as specified in the literature [44].
Fig. 6. Results from simulations of temperature ramps (10 K/min) with 0.1%
CO and 5% oxygen (A, C, E, and G), and with 0.5% CO and 1% oxygen (B,
D, F, and H). Simulations include the model proposed by Bourane and Bianchi.
A and B show bulk gas pressure of CO, C and D show CO coverage, E and F
show O coverage on site S , and G and H show O coverage on site S . Solid,
1
2
dotted and dashed lines correspond to tank 1, 15, and 30 respectively.
of Bourane and Bianchi was reformulated to explicitly ac-
count for the transient coverage of oxygen on site S2. This
required that the number of S2 sites be determined. The num-
ber of S2 sites (nS ) was found to account for 0.16% of the
2
total number of surface platinum sites. The low number of
S2 sites suggests that these sites were edge, corner, or defect
sites on the platinum crystals. The activation energy of the sur-
face reaction was within the range determined by Bourane and
Bianchi [9].
3.5. Model for CO oxidation with ozone
Based on the results presented in Figs. 2A–C, addition of
ozone to the reaction caused a low-temperature CO oxidation
reaction on the platinum sites. The ratio of oxidized CO to
feed ozone concentration at low temperature was between 0.5
and 1, suggesting that the dominating reaction between CO and
ozone occurred with a stoichiometric coefficient of 1. Further-
more, the foregoing simulations show that the platinum surface
was almost completely covered with CO at these temperatures,
which is why the most likely explanation for the observed reac-
tion is an Eley–Rideal reaction, in which adsorbed CO reacts
3.6. Simulation results for CO oxidation with oxygen–ozone
mixtures
Fig. 7 shows a comparison between the simulated CO in-
strument signals and the measured signals. The agreement be-
tween the results was good for all CO and oxygen concentration
combinations. Weizs modulus calculations indicate that pore
transport limitations will be negligible for all reactants. The
detailed simulation results for 0.1% CO and 5% oxygen and
ozone, and for 0.5% CO and 1% oxygen and ozone are shown

328
M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
Fig. 8. Results from simulations of temperature ramps (10 K/min) with 0.1%
CO, 5% oxygen and ozone (A, C, E, and G), and with 0.5% CO, 1% oxygen
and ozone (B, D, F, and H). A and B show bulk gas pressure of CO, C and D
show CO coverage, E and F show O coverage on site S , and G and H show O
1
coverage on site S . Solid, dotted, and dashed lines correspond to tank 1, 15,
and 30 respectively.
2
Fig. 7. Measured and simulated CO instrument signals during ramp experiments
(10 K/min) with ozone and with (A) 0.5, (B) 0.3, and (C) 0.1% CO. Squares,
triangles and circles corresponds to measurements with 1, 3, and 5% oxygen.
Dashed, dotted and solid lines correspond to simulations with 1, 3, and 5%
oxygen.
reactions with oxygen. This conclusion is supported by the
observation that the transformation from CO-covered surfaces
to oxygen-covered surfaces began at the outlet of the catalyst
and migrated toward the catalyst inlet. Nonetheless, the de-
crease in CO pressure resulted in a somewhat earlier catalyst
light-off compared with the experiments without ozone, and
the overall oxidation reaction thus can be considered ozone-
promoted.
The high rate of reaction between CO and ozone suggests
that it is possible to achieve complete CO removal at room tem-
perature, given a sufficient supply of ozone. It should also be
possible to use ozone to force a catalyst from the CO-covered
surfaces to the oxygen-covered surfaces under bistable reac-
tion conditions, even though that effect was not observed in
the current study. One drawback of this CO abatement tech-
nique is the high energy consumption associated with ozone
production. It may be possible to reduce the ozone requirement
by pulsing the ozone supply. It has been shown that catalyst
light-off will occur at lower average oxygen to CO ratios un-
der oxygen pulses compared with stationary oxygen concen-
trations [23], and we predict a similar effect for the current
system.
in Fig. 8. The results shown in Figs. 8A and B indicate that the
reaction between CO and ozone occurred primarily in the first
couple of tanks, and that 90% of the ozone was decomposed af-
ter 8 mm. While the surface was CO covered, the dominating
reaction path for ozone was the reaction with CO on site S1. Be-
fore light-off, 71–76% of the ozone was decomposed through
this reaction. The very rapid reaction between CO and ozone
makes it difficult to study the kinetic parameters for this reac-
tion. Even though we have achieved good agreement with the
experiments solely by adjusting the rate of ozone decomposi-
tion, we would suggest that future studies of the CO + ozone
reaction be conducted at temperatures well below room tem-
perature. This is beyond the scope of the current study, how-
ever.
The CO coverage (Figs. 8C and D) was not significantly
affected by the reaction between CO and ozone at low tem-
peratures. The S1 sites that were emptied in that reaction thus
were quickly CO-covered once again. But the reaction did re-
sult in a decrease in bulk CO pressure, which led to a decrease
in CO self-poisoning and consequently a higher rate for the

M. Petersson et al. / Journal of Catalysis 238 (2006) 321–329
329
4. Conclusion
[5] V.P. Zhdanov, B. Kasemo, Surf. Sci. Rep. 39 (2000) 25.
[6] M. Bär, Ch. Zülike, M. Eiswirth, G. Ertl, J. Chem. Phys. 96 (1992) 8595.
[7] P.A. Carlsson, V.P. Zhdanov, B. Kasemo, Appl. Surf. Sci. 239 (2005) 424.
[8] M. Berdau, G.G. Yelenin, A. Karpowicz, M. Ehsasi, K. Christmann, J.H.
Block, J. Chem. Phys. 110 (1999) 11551.
[9] A. Bourane, D. Bianchi, J. Catal. 202 (2001) 34.
[10] A. Bourane, D. Bianchi, J. Catal. 209 (2002) 114.
The findings of our study of the reaction mechanism of CO
oxidation over a platinum/alumina monolith catalyst comparing
temperature ramp experiments with mean-field simulations can
be summarized as follows:
[11] A. Bourane, D. Bianchi, J. Catal. 209 (2002) 126.
[12] A. Bourane, D. Bianchi, J. Catal. 220 (2003) 3.
[13] A. Bourane, D. Bianchi, J. Catal. 222 (2004) 499.
[14] A. Holmgren, D. Duprez, B. Andersson, J. Catal. 182 (1999) 441.
[15] S. Johansson, L. Östlund, B. Kasemo, J. Catal. 201 (2001) 275.
[16] R.H. Nibbelke, A.J.L. Nievergeld, J.H.B.J. Hoebink, G.B. Marin, Appl.
Catal. B 19 (1998) 245.
[17] J.M.A. Harmsen, J.H.B.J. Hoebink, J.C. Schouten, Stud. Surf. Sci.
Catal. 133 (2001) 349.
[18] J.M.A. Harmsen, J.H.B.J. Hoebink, J.C. Schouten, Catal. Lett. 71 (2001)
81.
[19] M. Kolodziejczyk, R.E.R. Colen, M. Berdau, B. Delmon, Surf. Sci. 375
(1997) 235.
[20] I.N. Yakovkin, V.I. Chernyi, A.G. Naumovets, Surf. Sci. 442 (1999) 81.
[21] A. Drewsen, A. Helmersson, M. Skoglundh, B. Andersson, Chem. Eng.
Sci. 55 (2000) 4939.
1. The three-step Langmuir–Hinshelwood model, with lat-
eral CO–CO interactions explicitly included via the quasi-
chemical approximation, was not sufficient to describe the
observed slow CO oxidation at low temperatures.
2. The rate of oxidation for this reaction was found to be pro-
portional to the square root of the oxygen concentration
and independent of the CO concentration, thus showing the
same reaction order as reported by Bourane and Bianchi
[9–13].
3. As the temperature was increased, the dominating oxida-
tion process was changed to the conventional three-step
Langmuir–Hinshelwood reaction.
4. The experimental results were well fitted when using a
simulation model that combined the Bourane and Bianchi
mechanism with the three-step Langmuir–Hinshelwood
mechanism.
[22] M. Skoglundh, P. Thormählen, E. Fridell, F. Hajbolouri, E. Jobson, Chem.
Eng. Sci. 54 (1999) 4559.
[23] P.A. Carlsson, P. Thormählen, M. Skoglundh, H. Persson, E. Fridell, E.
Jobson, B. Andersson, Top. Catal. 16/17 (2001) 343.
[24] S. Ljungström, D. Chakarov, J. Bergeld, S. Johansson, B. Kasemo, Top.
Catal. 16/17 (2001) 433.
[25] R.N. Miller, R.P. Caren, J.A. Ekchian, US Patent 5,806,305 (1996), to
Lockheed Martin Corporation.
[26] A. Naydenov, D. Mehandjiev, Dokl. Akad. Nauk. 46 (1993) 49.
[27] A. Naydenov, R. Stoyanova, D. Mehandjiev, J. Mol. Catal. A. 98 (1995)
9.
[28] A. Naydenov, D. Mehandjiev, Dokl. Akad. Nauk. 50 (1997) 65.
[29] Z. Hao, D. Cheng, Y. Guo, Y. Liang, Appl. Catal. B. 33 (2001) 217.
[30] A. Gervasini, C.L. Bianchi, V. Ragaini, ACS Symp. Ser. 552 (1994) 353.
[31] P. Lööf, B. Kasemo, S. Anderssson, A. Frestad, J. Catal. 130 (1991) 181.
[32] J.R. Andersson, Structure of Metallic Catalysts, Academic Press, London,
1975, p. 296.
[33] V.P. Zhdanov, Elementary Physicochemical Processes on Solid Surfaces,
Plenum, New York, 1991, p. 109.
[34] M. Rinnemo, D. Kulginov, S. Johansson, K.L. Wong, V.P. Zhdanov, B.
Kasemo, Surf. Sci. 376 (1997) 297.
5. Exchanging some of the oxygen for ozone, a rapid reac-
tion between CO and ozone was observed on the platinum
surface. To model this reaction, the CO oxidation model de-
scribed above was combined with an Eley–Rideal reaction
between ozone and adsorbed CO and an ozone decompo-
sition step on the alumina surface. The simulation results
indicated that ozone addition caused a reduction in bulk
CO concentration, and consequently a decrease in CO self-
poisoning and a lower catalyst light-off temperature. Ozone
was thus found to have a promotional effect on the CO ox-
idation reaction.
Acknowledgments
[35] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, J. Wiley
and Sons, New York, 1960.
[36] R.D. Hawthorn, AIChE Symp. Ser. 70 (137) (1974) 428.
[37] E. Tronconi, P. Forzatti, AIChE J. 38 (1992) 201.
[38] R. Siegel, J. Howell, Thermal Radiation Heat Transfer, fourth ed., Taylor
& Francis, New York, 2002, p. 844.
[39] U. Adler, H. Bauer, A. Cypra, F. Dinkler, P. Künne, J. Röder, Bosch Au-
tomotive Handbook, third ed., Robert Bosch GmbH, Stuttgart, 1993, p.
207.
[40] Y.Y. Yeo, L. Vattuone, D.A. King, J. Chem. Phys. 106 (1997) 392.
[41] A. Bourane, D. Bianchi, J. Catal. 218 (2003) 447.
[42] P.-A. Carlsson, M. Skoglundh, P. Thormählen, B. Andersson, Top.
Catal. 30/31 (2004) 375.
The authors thank Dr. Per-Anders Carlsson for providing
the original Matlab simulation code. Financial support was
provided by the Swedish Research Council for Engineering
Sciences (grant 285-99-291). The work reported herein was
conducted at the Competence Centre for Catalysis, which is
supported by the Swedish Energy Agency and member compa-
nies: AB Volvo, Saab Automobile AB, Johnson Matthey CSD,
Perstorp AB, AVL–MTC AB, Albemarle Catalysts, and the
Swedish Space Administration.
References
[43] P. Hunter, S.T. Oyama, Control of Volatile Organic Compound Emissions:
Conventional and Emerging Technologies, John Wiley & Sons, New York,
2000, p. 182.
[44] P.W. Atkins, Physical Chemistry, fourth ed., Oxford University Press, Ox-
ford, 1990, p. 944.
[1] I. Langmuir, Trans. Faraday Soc. 17 (1922) 621.
[2] T. Engel, G. Ertl, Adv. Catal. 28 (1979) 1.
[3] G. Ertl, Adv. Catal. 37 (1990) 213.
[4] V.P. Zhdanov, B. Kasemo, Appl. Surf. Sci. 74 (1994) 147.