Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Chemical Engineering Science 54 (1999) 4295}4304

Monte-Carlo methods for simulating the catalytic oxidative


dehydrogenation of propane over VMgO catalyst
Yves Boudeville  *, Max Kolb , Amalia Pantazidis , Carlos Marquez-Alvarez ,
Claude Mirodatos , Vladimir Elokhin
Institut de Recherches sur la Catalyse, CNRS, 2, Avenue Albert Einstein, F-69626 Villeurbanne Ce& dex, France
Ecole Normale Supe& rieure, ENS/Lyon. Laboratoire de Chimie The& orique, 46 Alle& e d'Italie, F69364 Lyon Ce& dex 07, France
Boreskov Institute of Catalysis, Pr. Ak. Lavrentyeva, 5 630090 Novosibirsk, Russia

Abstract

The oxidative dehydrogenation of propane (ODHP) has been extensively studied by non-steady-state kinetic experiments and by
various experimental characterization techniques under conditions as close as possible to reaction conditions. It can be described in
terms of the Mars}Van Krevelen mechanism. This work reports on Monte-Carlo simulations of the ODHP over VMgO catalysts,
aimed at relating the main experimental insights obtained previously on the catalytic reaction to a theoretical description of the
working catalyst. The `theoretical "eld of catalysisa is explored by handling new Monte-Carlo models with a catalytic reaction graph,
2D pattern recognition of adsorbed molecules, 3D oxygen bulk di!usion and "nally the description of the selective, non-selective
routes of the ODHP process.  1999 Elsevier Science Ltd. All rights reserved.

Keywords: Monte-Carlo simulation; Oxidative dehydrogenation of propane; Pattern recognition; Mars}Van Krevelen mechanism; VMgO catalyst

1. Introduction pressures ratio of reagents as a macroscopic variable on


the microscopic atomic surface. Other more sophisti-
Monte-Carlo methods are involved in the description cated models (Boudeville et al., 1993) include Arrhenius
of non-linear dynamic e!ects in catalytic heterogeneous laws and chemical engineering constants to describe dif-
or homo-heterogeneous reactions. The simplest model- fusion- or reaction-controlled regimes in the catalytic
ling uses the Zi!, Gulari Barshad Monte-Carlo method reactor. At this level of modelling, temperature depend-
with one reaction A(surf)#B(surf)}Pt[1 0 0](single crys- ant non-linear e!ects can be investigated. The last level of
tal)PAB(gas) (Zi! et al., 1986). The Zi! scheme is a pure complexity can be described in terms of the chemical
mathematical one (everything that can happen happens; graph of reactions: not one catalytic reaction, but many,
it is a `boolean lawa, if two things can happen, each will not only reactions on the catalyst but also reactions in
be with probability of one half etc.) but the main features the gas phase.
of surface adsorption of reactants, desorption of prod- The `realisma of the reactor description was improved
ucts, reactions are treated with a surprisingly good suc- by using a `catalytic reaction compilera linked to
cess, since the mathematical topology of the chemical a Monte-Carlo program that simulates the whole het-
process is taken into account. This permits the investiga- ero-homogeneous catalytic reactor. This approach was
tion of two kinetic phase transitions by using the partial tested for the case of the oxidative coupling of methane
(Boudeville, unpublished). This model demonstrated the
necessity of layered catalysts to increase the selectivity in
ethane by more than 20%. This model uses a sub-set of
*Corresponding author. Isstitut de Recherches sur la Catalyse 17 reaction/di!usion equations with a catalysis data
CNRS, 2 Avenue Albert Einstein, F-69626 Villeurbanne Cedex, France. bank of 19 chemical species.
Tel.: #33-472-44-53-49; fax: #33-472-44-53-99.
Present adress: Societe TOTAL, Centre de Recherches, B.P. 27; Two types of catalytic mechanisms were investi-
F76700 Har#eur; France. gated up to now: the Langmuir}Hinshelwood (Kolb and
E-mail address: boudevil@catalyse.univ-lyon1.fr (Y. Boudeville) Boudeville, 1990) and Eley}Rideal mechanisms for

0009-2509/99/$ - see front matter  1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 9 9 ) 0 0 1 4 7 - 5
4296 Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304

CO oxidation or CH oxidative coupling reactions, Table 1


 Elementary reactions in the MC model 1
respectively.
The present paper reports our present e!orts to model
C H #[C]#2[O ] k P2[OH ]#[C]#C H ;
the Mars}van Krevelen mechanism which applies for   I  I  
[OH ]#[OH ] k P[O ]#[C]#H O;
most oxidation reactions. One model takes into account I I  I 
H O#[O ]#[C] k P[OH ]#[OH ];
 I  I I
the oxygen motion in the bulk oxide lattice (Fig. 1) O #2[C] k P2[O ];
  I
by a kind of solid-on-solid model used by Elokhim O #2[C]#2[O ] k P4 [O3 ];
 I  I
[O ] #[C] m P[C] #[OC] ;
(Resnyanskii et al., 1997) for surface corrugation under I G H I J%G H  G H I J
[OHC] #[C] m P[C] #[OH ] ;
catalytic reactions. The oxidative dehydrogenation of G H I J%G H  G H I I J
[O3 ] #[C] m P[C] #[O3 ] ;
propane over VMgO catalysts was chosen as a model I G H I J%G H  G H I I J
C H #[O3 ]#10 (O "O3 ) k P3[C]#8 [OH ]#3CO;
  I I I  I
reaction, since numerous and detailed mechanistic and C H #[O ]#8(O "O3 ) k P3[C]#6[OH ]#3CO;
  I I I  I
kinetic data are available (Kung et al., 1997; Pantazidis CO#[O ] k P[C]#CO ;
I  
and Mirodatos, 1996a,b; Pantazidis et al., 1998a,b). In
C subscript: chemical species linked to the surface.
this respect, a Monte-Carlo model was constructed to [C]: empty surface site for the Monte-Carlo simulation.
describe the ODHP process over an optimized V/MgO [O ]: lattice oxygen of the surface.
I
catalyst (14 wt%V). The latter can be described essential- [O3 ]: adsorbed oxygen on the surface.
I
ly as (VO )\ polymerized units supported on MgO (O "O3 ): any form of oxygen in the degradation (total oxidation to
 I I
CO, CO ) reactions
platelets, together with some large crystallites of 
[] #[] : surface migrations of oxygen species,
Mg V O (Pantazidis and Mirodatos, 1996a; Pantazidis G H I J
   k,l&i,j: the sites +k,l,, +i,j, are "rst neighbours.
et al., 1998b). Each reaction R is labelled by its kinetic constant } k P
G G
Each mobility M is labelled by a mobility constant } m P
H H
The symbols are also valid in Table 3
2. MC models

2.1. Model 1: Theoretical view of the elementary as necessary for the alkane reaction with oxygen (four for
catalytic steps C H , Table 2), in all the 2D space orientations, are
 
considered. If there is an exact topological match be-
A "rst approach of the bare surface mechanism, with- tween the grid and a pattern, i.e., if any carbon atom or
out bulk oxygen lattice motion, was carried out with hydrogen atom of the molecule can react with an oxygen
a system of 11 elementary steps occurring on the catalyst of the catalytic surface, and if this reaction is kinetically
surface only, this assumes that the 11 kinetic constants selected by the MC process, then the complete reaction
are known. The Monte-Carlo simulation is made on occurs on the catalytic site; if the match is not perfect for
a 225;225 grid of atoms representing the bare surface, any pattern, no reaction occurs.
adsorbed species and catalytic reaction products. The idea beneath these four patterns is the possibility
The pseudo-elementary steps of the catalytic process for the catalyst surface to give in one step all the neces-
are described in Table 1. sary reactants to the colliding molecule for the speci"ed
The theoretical graph of reactions in Table 1 matches reaction. This means that the sub-steps of one pseudo-
the main mechanistic features described in (Pantazidis elementary reaction (k for example) are orders of magni-

and Mirodatos, 1996b) and Pantazidis et al., 1998a). tude faster than the mobilities of the catalytic species.
They can be summarized as follows: These patterns have been selected because they corres-
(i) the selective propane-to-propene route occurs on pond to `ensemble of atomsa in the closest shape of
a surface site formed of lattice surface oxygen and anionic a circular site of the surface which are found to be
vacancy, necessary by experimentalists to develop such kind of
(ii) the non selective propane-to-CO degradation reactions (magnetism measurements).
V
route occurs on adsorbed electrophilic oxygen but also During the simulation, the spatial 2D behaviour of the
likely involves lattice oxygen in further steps, oxide catalyst can interactively be followed from the
(iii) the secondary propene-to CO route involves both initial contact between reactants and the catalytic surface
V
types of surface oxygen species, (Fig. 2), i.e., under unsteady state conditions until the
(iv) the site regeneration occurs by means of surface steady state is achieved, i.e. it implies that there is neither
dehydroxylation into water (from OH groups formed chaos nor oscillations (Qin Feng et al., 1998), by examin-
upon the previous steps) and lattice vacancies reoxida- ing the time evolution of the reaction rates. Thus, it is
tion by gaseous oxygen. observed in Fig. 3 that due to the still increasing surface
In this MC model 1, we use `pattern recognition tech- di!usion M of adsorbed electrophilic oxygen [O3 ], the
 I
niquesa to randomly decide if a catalytic site can undergo steady state is not yet reached after more than a thousand
degradation reactions leading to CO (the pattern is here of MC seconds. Fig. 4 gives the spatial 2D description
an ensemble of surface atoms). As many surface patterns of the catalytic surface with an interactive summary
Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304

Fig. 1. Left: Oxygen (blue) and vacancies (yellow) volume di!usion in bulk oxide; Upper Right: supposed reactive face (}cyan}"lattice surface oxygen atoms); Lower Right: supposed face of
O gas adsorption and dissociation on two neighbouring lattice vacancy sites (}yellow}).
4297


4298
Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304

Fig. 2. First view of the [1 0 0] surface catalyst at the time"1 for the Model 1; Ox: means lattice oxygen; Oh: Hydroxyl groups; Oo: adsorbed electrophilic oxygen species; Va: means
Vanadium induced oxygen vacancy; Pr: is used to label the surface sites where a Propene has been selectively made at one time of the interactive Monte-Carlo simulation.
Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304 4299

Table 2
Surface selected patterns for recognition of the reaction k (one

example)

Pattern 1 1 2 3 4 5
1 [.] [.] [.] [.] [.]
2 [.] [O ] [O ] [O ] [.]
I I I
3 [O ] [O ] [O3 ] [O ] [O ]
I I I I I
4 [.] [O ] [O ] [O ] [.]
I I I
5 [.] [.] [.] [.] [.]
Pattern EEWW.

Pattern 2 1 2 3 4 5
1 [.] [.] [O ] [.] [.]
I
2 [.] [O ] [O ] [O ] [.]
I I I
3 [.] [O ] [O3 ] [O ] [.]
I I I
4 [.] [O ] [O ] [O ] [.]
I I I
5 [.] [.] [O ] [.] [.]
I
Pattern NNSS.

Pattern 3 1 2 3 4 5
1 [.] [.] [.] [O ] [.]
I
2 [.] [.] [O ] [O ] [O ]
I I I Fig. 5. Time curves: Model 1. Variation in time of the di!erent surface
3 [.] [O ] [O3 ] [O ] [.]
I I I species coverages (between 0.0 and 1.0) and rate curves of reactants and
4 [O ] [O ] [O ] [.] [.]
I I I products. [A] a: Propane consumption; b: Propene rate of formation; c:
5 [.] [O ] [.] [.] [.]
I CO formation; d: CO . [B] e: [C] empty site coverage; f: [O ] site; g:
Pattern NESW.  I
[OH ] site; h: [O3 ] site coverage.
I I
Pattern 4 1 2 3 4 5
1 [.] [O ] [.] [.] [.]
I
2 [O ]
I
[O ]
I
[O ]
I
[.] [.] 225;225 atoms, is su$cient to have a small enough
3 [.] [O ] [O3 ] [O ] [.] random noise that concentration curves can be followed
I I I
4 [.] [.] [O ] [O ] [O ]
I I I without ambiguity. No physical meaning can be derived
5 [.] [.] [.] [O ] [.]
I from these curves since all the rate constants were taken
Pattern NWSE.
as 1 for this "rst MC model, a large part of these con-
stants being not available from any kinetic experiments.
A similar assumption was considered in the Zi!, Gulari,
Barshad ZGB model, where all the probability functions
are equal to 1.

2.2. Model 2: Toward an experimental description of


catalytic reactions

A second approach is now described where some of the


previous elementary reaction steps (Table 1) are lumped
in a way to be used with available kinetic coe$cients. The
later were determined from an experimental kinetic study
carried out on the selected VMgO catalyst (Pantazidis,
1996). This approach is less interesting for theoretical
purposes but presents the advantage that the predicted
surface and gaseous concentrations can be compared to
Fig. 3. Time curves: Model 1. Variation in time (Mcs"Monte-Carlo experimental data. The new set of equations is reported
second) of the di!erent reaction rates R between 1 and 1000 Mcs.
G in Table 3.
Like for the model 1, the model 2 is based on two main
mechanistic features:
of all the reaction parameters at the depicted MC time (i) the selectivity arises from the dynamic competition
(computer screen dumps). Fig. 5 gives the changes between the selective route leading to propene and the
with MC time of reaction rates (A) for the reactant non-selective one leading to CO and CO ,

(propane) and products (propene, CO, CO ) and in (ii) an island of a selected number of surface atoms is

surface concentrations (B) for surface oxygen and va- necessary to accommodate the catalytic reaction, as
cancy species. The size of the Monte-Carlo grid, deduced from the statement that at least two VO\

4300 Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304

Table 3 of random treatment is used if the selected lattice crystal-


The selected reactions for the `experimentala MC model 2 lographic site is occupied by an oxygen vacancy, but
conversely with the random search of neighbouring lat-
C H #[O ] k PC H #H O#[C];
  I     tice oxygen and actions dealing with adsorption and
O #2 [C] k P2 [O ];
  I dissociation of gaseous O on two adjacent vacancy sites
O #2 [C]#2 [O ] k P4 [O3 ];
 I  I 
C H #[O3 ]#6 (O "O3 ) k P3 CO#4 H O#7 [C]; on the selected adsorption face. The experimentalists
  I I I  
C H #[O3 ]#9 (O "O3 ) k P3 CO #4 H O#10 [C]; assumption here is that this face is more favourable to the
  I I I   
[O ] #[C] m P[C] #[O ] ;
I
I J%G H  G H I I J oxygen adsorption/dissociation process, while the "rst
[O3 ] #[C] m P[C] #[O3 ] ;
I G H I J%G H  G H I I J one is more favourable to the catalytic reaction. These
vacancies, propagating in the solid, meet then a face of
reoxidation by gaseous oxygen adsorption and dissocia-
tetrahedral units are required for dehydrogenating the tion (for example one among six [1 0 0] faces), and a cur-
propane molecule (Pantazidis and Mirodatos, 1996a,b). rent of lattice oxygen bulk di!usion is then involved,
Note that no hydroxyl OH adspecies appears in this balancing the vacancy current in the steady state. Only
model 2 (actually lumped in step 1), at variance with the two faces were considered here, one for reaction, the
complete model 1. other for oxygen adsorption, for theoretical simplicity
and because of the lack of full description of the catalytic
2.3. Model 3: Theoretical description of oxygen motion in sites on the VMgO catalyst. A more advanced descrip-
the bulk of the oxide; xrst Monte-Carlo approach of the tion of all the reactive sites, both for alkane and oxygen
Mars}Van Krevelen mechanism. activation, including an experimental measurement of
the rate constants, would certainly lead to a model 4,
As it has been done by Elokhin et al., 1997, it is now fully describing the ODHP process.
possible to apply the Monte-Carlo method to treat the At this point of the work, this theoretical model 3 can-
atom motion on a solid (surface corrugation under chem- not yet be connected to the model 1 of propane ele-
ical adsorption). On the basis of this surface corrugation mentary reactions because crystallographic data are
of the catalytic solid in the solid-on-solid (SOS) scheme, necessary and, a fortiori, to model 2 which needs experi-
a further step can be proposed: the oxygen motion in the mental bulk di!usion constants to work at the same level
oxide bulk. Like for the theoretical model 1 of a surface of approximation.
reaction on oxide, this model of bulk di!usion through
the solid remains purely theoretical, since, again, no
experimental data are available yet. Fig. 1 gives the 3. Results and discussion
spatial 3D description of the catalyst with an interactive
summary of all the parameters at the depicted MC time Table 4 lists the set of constants determined from the
(computer screen dump). kinetic analysis of the ODHP over VMgO catalyst,
First the 3D crystallographic structure of an oxide is which correspond to the set of equations proposed in
generated with 50,653 atoms. Then the solid is cut by Table 3 for the MC model 2.
a family of 14 planes to generate [1 1 1] and [1 0 0] faces The set of rate constants were obtained from a kinetic
of the oxide. In order to have a good insight into the solid study carried out on the selected VMgO catalyst (Panta-
di!usion mechanism, one face has been selected (for zidis, 1996) and are incorporated in the last model to
example one among eight [1 1 1] faces) to be catalytically trigger the probabilities of events in the MC process
active. On this face, the lattice oxygen vacancies, con- (Wolf et al., 1993). Actually, only the "ve k were deter-
secutive to catalytic surface reactions (see Table 1) can G
mined from this study, since no step of surface mobility
migrate throughout the bulk in a 3D random scheme. An
oxygen lattice site is randomly chosen. If there is an
Table 4
oxygen atom in it, a random search for a neighbouring Kinetics constants determined by a kinetic analysis of ODHP data over
3D vacancy is performed. If other actions (like a reaction) VMgO catalyst
are possible, all the actions are weighted by frequency
factors and the random number on the reduced scale of Steps Kinetic constants
actions [between 0 and less than or equal to 1] deter- (lmol/g s kPa)
mines the selected action. If the number is in the upper k 1.18E-02
zone corresponding to `no eventa, nothing is done in this 
k 9.36E-02

MC cycle (Fichthorn and Weinberg, 1991). In this article k 2.32E-02

we see a way of connecting the real time and the MC k

1.86E-03
time. This connection is not established in this present k 6.17E-03

m "k 1.18E-02
work because we deal only, for the moment, with the  
m "k 1.86E-03
qualitative features of reaction processes. The same kind  
Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304

Fig. 4. Model 1, spatial representation of the catalytic surface at time"1000 Mcs. The caption for surface species has been de"ned in Fig. 2. Number of Reaction 00 is the total number of
reactions that have occurred on the solid catalyst.
4301
4302 Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304
Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304 4303

vacancies species. As observed for the previous model


catalytic process: stable rates of conversion/production
and of surface concentrations. As for the reaction rates, it
can be derived from the MC model that most of the
propane is converted to propene, with a selectivity close
to 95% (Fig. 6) which is slightly more than the actual
experimental selectivity found around 75%. As for the
surface concentration, it is observed that the initially
prominent lattice oxygen species [O ] have been par-
I
tially replaced (i.e. oxidized) by adsorbed electrophilic
oxygen species [O3 ] while the concentration of anionic
I
vacancies remains at a smaller level. This would corres-
pond to a rather oxidised catalyst surface, as expected
under the reaction conditions. However, the concentra-
tion of adsorbed oxygen [O3 ] tends to be higher than
I
what was deduced from the experimental kinetic study
(Pantazidis and Mirodatos, 1996a,b; Pantazidis et al.,
1998a,b). Indeed such a high surface coverage with these
[O3 ] species would have favoured the total oxidation
Fig. 7. Model 2, Time curves. Variation in time corresponding to Fig. 6.
I
of propane into CO at the expense of the propene
One can see here that the absence of [O ] groups (lattice O\ oxygen V
VI formation, which was not observed experimentally. An-
species) prevents the selective formation of Propene. (The [OH ] has
I other weakness of the model 2 is the absence of surface
no value because the hydroxyl group is not used in these graphs of
[OH ] groups which were not considered for the
model 2). The position of the [O ] coverage (between 0.0 and 1.0)
VI I
shows that the catalyst is active and also selective } Fig. 6: Balance sake of reaction simpli"cation as previously stressed
[Propene/Propane]"100(6196619/6498644)"95% of selectivity. in Table 3. Indeed, the model will have to be improved
since a direct observation of the working surface by
was considered. Thus, the two mobility constants m and in situ infrared spectroscopy demonstrated the presence
 of surface OH groups under the ODHP reaction condi-
m required by the MC model 2 are assumed to be of the
 tions (5503C) (Pantazidis and Mirodatos, 1996b). For
same magnitude as k and k , respectively. Such an
  the oxygen cycle (reoxidation of anionic vacancies [C]
assumption is based on the general observation that the
time required for a catalytic cycle in an oxidation reac- into lattice oxygen [O ] (step 2 in Table 3) in the
I
tion is of the same order of magnitude as the time MC model 2, it can be speculated that in addition to
required for a surface oxygen species to move from one the surface reoxidation, a migration of lattice oxygen
atomic site to another one in order to regenerate the from the bulk to the surface could also contribute to
active site (this mobility corresponds to a surface di!u- this oxygen cycle, as expected in a Mars}van Krevelen
sion coe$cient around 10\ m/s). process. By ignoring this possibility in the model 2, this
These experimental data were incorporated in the could possibly explain the over concentration of
Monte-Carlo dynamic model, by using the real time adsorbed electrophilic oxygen as outlined above. Taking
pattern recognition model in order to see if the colliding into account the two phenomena model 2#model 3 can
molecule can undergo a reaction with adsorbed species then lead to the same kind of result, but by splitting the
on the catalytic site. k rate of reaction into two parts, one for each model

The spatial 2D behaviour of the oxide catalyst after process.
reaching the steady state is shown in Fig. 6. Fig. 7 gives
the changes with MC time of reaction rates (A) for the
reactant (propane) and products (propene, CO, CO ) and 4. Conclusion

in-surface concentrations (B) for surface oxygen and
Though signi"cant discrepancies were observed
䉳 between the experiment and MC kinetic data, a very
Fig. 6. Model 2, spatial representation of the catalytic surface at promising behaviour of the model towards a reliable
time"1000 Mcs. The surface is incompletely covered by electrophilic and detailed description of the working catalysts
oxygen (red) species that leads only to degradation products CO and was noted. However, the availability of complete
CO . The white patterns correspond to pattern recognition where and consistent kinetic constants from the kinetic experi-

a degradation reaction had just taken place. The catalyst is not dynam- ments which could signi"cantly improve the MC model
ically poisoned but in a state comparable to model 1. As for model 1,
1 (Fig. 5), a "rst period of unsteady-state regime is observed, but leading e$ciency is dubious for such a complex reaction.
rapidly to an equilibration of the Fig. 4, the [O3 ] coverage (red) is too Therefore, the next step in this theoretical treatment of
I
high. the ODHP could be the following one: The Monte-Carlo
4304 Y. Boudeville et al. / Chemical Engineering Science 54 (1999) 4295}4304

models can be fed by a set of k kinetic constants in References


G
the realistic range. The output of the models, with all the
Boudeville, Y., & Wolf, E. E. (1993). Surface Science, 297, L127.
physico-chemical features taken into account, can be the Fichthorn, K. A., & Weinberg, W. H. (1991). Journal of Chemical
time evolution of all the reactants, products, transient Physics, 95(2), 1090.
species, known or unknown, involved in the catalytic Kolb, M., & Boudeville, Y. (1990). Journal of Chemical Physics, 92, 3935.
reactions. The MC simulation acts then like an integrator Kung, H. H., & Kung, M. C. (1997). Applied Catalysis A, 157, 105.
of data and concepts. The interactive spatial "gures (vol- Pantazidis, A. (1996). Ph.D. thesis, Lyon.
Pantazidis, A., Buchholz, S., Zantho!, H. W., Schuurmann, Y. S.,
ume Fig. 1, surface Figs. 2, 4 and 6) are a natural way to & Mirodatos, C. (1998a). Catalysis Today, 40, 207}214.
follow with the best accuracy the evolution of catalyst Pantazidis, A., Burrows, A., Kiely, C. J., & Mirodatos, C. (1998b).
reactivity and it goes far beyond `mean "eld theorya Journal of Catalysis, in press.
models. The correct description of the islands of atoms, Pantazidis, A., & Mirodatos, C., (1996a). Heterogeneous hydrocarbon
ensemble that could constitute a catalytic site, can be oxidation, In B. Warren, & S. T. Oyama, ACS symposium series, vol.
638 (p. 207).
easily set-up by pattern recognition according to experi- Pantazidis, A., & Mirodatos, C. (1996b). 11th international congress on
mental assumptions. The known experimental curves on catalysis. In J. W. Hightower, W. N. Delglass, E. Iglesia, & A. T.
the variation of chemical species varying in the un- Bell, Studies in surface science and catalysis, vol. 101 (p. 1029).
steady-state regime can then be used to build up an Qin, F., Tagliabue, L., Piovesan, L., & Wolf, E. E. (1998). Chemical
objective function. Then the genetic algorithms can be Engineering Science, 53(5), 919.
Resnyanskii, E. D., Myshlyavtsev, A. V., Elokhin, V. I.,
applied to the output of a `population of k a, to have
G & Bal'zhinimaev, B. S. (1997). Chemical Physics Letters, 264, 174.
the best "t between the theoretical and experimental Zi!, R. M., Gulari, E., & Barshad, Y. (1986). Physical Review L, 56,
curves. 2553.

You might also like