FINE / SPECIALTY CHEMICALS

FEDERICO GALVANIN, ENHONG CAO, NOOR AL-RIFAI, VIVEK DUA, ASTERIOS GAVRIILIDIS*
*Corresponding author
Department of Chemical Engineering, University College London,
Torrington Place, London WC1E 7JE, United Kingdom

Federico Galvanin

Optimal design of experiments

for the identification of kinetic models of methanol

oxidation over silver catalyst
KEYWORDS: methanol oxidation, silver, microreaction technology, model-based design of experiments, model discrimination,
kinetic modeling.

Abstract

Partial oxidation of methanol to formaldehyde on silver catalyst represents an important industrial process
due to the versatility of formaldehyde as an intermediate in chemical synthesis. The development of
kinetic models is essential for a quantitative description of the changes in concentration of the chemical species involved in the
process due to reaction as well as for process design and optimisation purposes. Microreactor platforms represent effective tools for
the quick development of reliable kinetic models. However, the development and identification of kinetic models is strictly related to
the execution of informative experiments, allowing either elucidation of the complex reaction pathways involved in the oxidation
process or providing a precise estimation of the kinetic parameters for each candidate model. In this work a model-based design of
experiments (MBDoE) procedure is proposed where experiments are optimally designed for both discriminating among competing
models and for improving the estimation of kinetic parameters. The proposed methodology allows the most influential reaction
pathways to be elucidated and provides a sequence of optimally informative experiments showing the key role of temperature in the
kinetic model identification procedure.

INTRODUCTION
The partial oxidation of methanol to formaldehyde over a
silver catalyst represents a reaction of high industrial
significance due to the importance of formaldehyde as a
precursor to the production of urea, formaldehyde resin,
melamine resin, phenol formaldehyde resin,
polyoxymethylene plastics, 1,4-butanediol, and methylene
diphenyl diisocyanate. Under industrial conditions the silver
catalyst process is usually carried out at atmospheric pressure
and high temperatures (T = 853-923 K); if steam is introduced
with H2O/CH3OH = 0.67 and CH3OH/O2 = 2.4-2.5 a high
selectivity can be reached (~ 90%) (1). The overall oxidation
process is usually regarded as an oxidative dehydrogenation,
i.e. a combination of methanol oxidation (CH3OH + 1/2O2 ->
CH2O + H2O ) and dehydrogenation (CH3CH -> CH2O + H2).
The main by-products are H2, H2O, CO2, CO, and methyl
formate/formic acid.
Throughout the years, numerous research efforts have been
devoted to this reaction system to understand the catalytic
role of silver and the possible reaction mechanisms occurring
on the catalyst surface (2,3). Methanol oxidative
dehydrogenation on silver is strongly related to the presence
of chemisorbed oxygen and it is driven by the different
activity of oxygen species which are present on the catalyst
surface and in the bulk (4). The accurate quantitative
description of the concentration of the chemical species

Chimica Oggi - Chemistry Today - vol. 33(3) May/June 2015

involved in the process is related to the availability of i) a

reliable kinetic mechanism, defining the mathematical
structure of the kinetic model (i.e. the set of constitutive
equations); ii) the precise estimation of the set of kinetic
parameters for the model. However, despite the industrial
significance of this system, only a few attempts to model the
kinetics have been reported in the scientific community (3,5,6)
and a full understanding of the reaction on silver under
industrial reaction conditions has yet to be established.
Recently, a microkinetic model of methanol oxidation on
silver, based on a Langmuir-Hinshelwood mechanism, has
been proposed (5) in order to explain surface science
experiments and kinetic experiments at industrially relevant
conditions, applying physically realistic parameters. However,
a simplification of the model was required in order to be
applied for reactor engineering purposes (6), given the high
number of kinetic parameters and the complexity of the rate
expressions.
Microreactor platforms represent an ideal system for the study
and determination of intrinsic reaction kinetics for strongly
exothermic, endothermic, and fast catalytic reactions. With
careful experimental design, microreactors allow these
reactions to be performed isothermally and in the absence of
mass transfer limitations, ensuring the rapid manipulation of
reaction conditions and the precise control of the
hydrodynamic environment (7). However, the quantity,
quality, and speed of information generated by these systems

51

are affected by the experimental conditions realised during

the trials. Design of experiments (DoE) (8) methodologies may
help for designing preliminary screening and exploratory
experiments, but these techniques are based on empirical
models which do not imply a fundamental understanding of
the chemical system. Conversely, model-based design of
experiments (MBDoE) techniques, based on the
phenomenological model of a system, have been proposed
for designing a set of experiments yielding the most
informative data to be used for the development of
fundamental models. These techniques have been
successfully adopted in a wide range of applications (9)
including the investigation of chemical kinetics in catalytic
systems (10,11) and microfluidic devices (12). According to
MBDoE, experiments can be specifically designed for i)
maximising the difference between model predictions,
allowing a clear discrimination between candidate kinetic
models (13); ii) improving the precision in parameter
estimation for each candidate model (14). These two MBDoE
objectives usually represent distinct sequential steps to be
carried out in the conventional model building procedure
(15). Some studies have tried to achieve these two goals
simultaneously (16,17) by using a multi-objective approach
where the objective functions were evaluated on a grid of
experimental conditions, but no multi-objective optimisation
was involved.
In this paper, a MBDoE procedure based on a multi-objective
optimisation is proposed where experiments are optimally
designed for improving the estimation of kinetic parameters
as well as the simultaneous discrimination among competing
kinetic models. MBDoE methodologies are exploited to design
a sequence of experiments to be carried out in a specifically
designed silicon-glass microreactor (18).

OPTIMAL EXPERIMENTAL DESIGN PROCEDURE FOR THE

IDENTIFICATION OF KINETIC MODELS
A sketch of the suggested procedure for the optimal design
of experiments for the development of kinetic models of
oxidative dehydrogenation of methanol on silver is shown in
Figure 1. Insights from surface science and chemistry lead to
the formulation of candidate kinetic mechanisms (Step 1)
including the potential reactions taking place on the catalyst
surface. Kinetic mechanisms represent the basis for the
formulation of kinetic models (Step 2). Model reduction
strategies based on the evaluation of the rate determining
step (RDS) are used for the development of structurally
identifiable models (19) (i.e. models where kinetic parameters
can be uniquely determined from experimental data) with
chemically consistent reaction rate expressions (10,20).
Available historical data are then used for a preliminary
model discrimination between candidate kinetic models
(Step 3) based on a-posteriori statistics obtained after
parameter estimation (21), allowing for a preliminary model
selection and an initial estimation of kinetic parameters. Once
a set of candidate kinetic models is defined, model-based
design of experiment (MBDoE) techniques are applied for
planning a sequence of optimally informative experiments
allowing for the simultaneous discrimination among
candidate models and for improving the precision of
parameter estimation (Step 4). MBDoE optimisation computes
both the experimental conditions yielding the minimum
expected variability in the kinetic parameters and the ones
providing the maximum discrimination power in terms of

52

model predictions. The optimally designed experimental

conditions are implemented in the microreactor system (Step
5), providing new experimental data to be used for
parameter estimation (Step 6). The iteration of steps 4) to 6)
represents the core of the MBDoE identification procedure,
leading to the detection of the best model structure
representing the system (i.e. inadequate models are
rejected), elucidating the most plausible kinetic mechanism
and ensuring at the same time a precise estimation of the
kinetic parameters. The MBDoE procedure will stop when a
model is found adequate to represent the system (from lackof-fit 2 tests based on a-posteriori statistics (21)) and a
statistically satisfactory estimation of model parameters
(based on t-test (14)) is achieved.

Figure 1. MBDoE procedure for simultaneous model

discrimination and improvement of parameter precision.

Formulation of competitive kinetic models of methanol

oxidation on silver
The set of reactions involved in each proposed model is
shown in Table 1. The simplified model proposed by
Andreasen (6) is used as a reference model (Model 1) and
two additional simplified kinetic models of increasing level of
complexity (Model 2, Model 3) are considered. According to
Model 1, methanol oxidative dehydrogenation (reaction 1)
and formaldehyde partial oxidation (reaction 2) constitute
the base (global) mechanism. Model 2 includes reactions 1-2
and the total oxidation reactions for both CH3OH and CH2O
(reactions 3 and 4 respectively). Model 3 includes the total
oxidation reactions 3,4 for CH3OH and CH2O (like Model 2)
but global methanol oxidation (reaction 1 used in Model 1
and 2) is split into a dehydrogenation reaction (reaction 5)

Chimica Oggi - Chemistry Today - vol. 33(3) May/June 2015

and a selective oxidation step (reaction 6). Hydrogen

oxidation (reaction 7) has also been included in each
mechanism. This reaction is known to occur only at higher
temperatures (22), and it has been primarily considered to
represent the low hydrogen concentrations observed in the
experiments.

Table 1. Set of reactions involved in the proposed kinetic models.

The microchannel reactor is modelled as a dynamic plug flow

reactor (PFR) in the form:

Eq. 1

where ci is the species concentration, rj and ij are the

reaction rate and the stoichiometric coefficient of the i-th
species in the j-th reaction respectively, z is the axial
coordinate, uz is the flow velocity in the z-direction and t the
integration time.
The microreactor was fabricated by photolithography and
deep reacting ion etching. The reactor channel had a
rectangular cross section with 8 mm width and 0.12 mm
depth. The silver film catalyst was deposited on the
channel floor by sputtering and was 12.5 mm long and
0.25 mm thick. In order to minimise heat losses, the
microreactor assembly was insulated with ceramic
insulation material. Even though some heat loss was
present from the top of the reactor, the temperature at
the catalyst location was constant. Mass transfer
resistances can be neglected due to the small
microchannel depth. The reactor and experimental setup
have been described in detail in a previous publication
(18).
The reaction was carried out at 725-826K and operation
pressure ~1.6 atm. The residence time at reaction
temperature was kept at ~6.5-7 ms to avoid formaldehyde
decomposition. The effect of O2, CH3OH and H 2O
concentrations was studied by varying each of these
components (one at a time) and keeping the total flow
constant (25.8-26.5 mL/min) using helium. The inlet and the
effluent were analysed online using a ThermoQuest Trace
gas chromatograph (GC) to determine the concentration
of CO2, CH3OH and CH 2O, while the non-condensable
gases were analysed using a Shimadzu GC to determine

Chimica Oggi - Chemistry Today - vol. 33(3) May/June 2015

O 2 and H2. The GC analyses were assumed to be

corrupted by Gaussian noise with zero mean and a
standard deviation of 1% on the reading. A molar
composition of the inlet gas mixture CH3OH=10.4%, and
ratios CH3OH/O 2=2.25 and CH 3OH/H 2O=1.32 were used as
a standard case. Experimental reproducibility was
checked by performing the standard run at 783 K
between each set of experiments. The variables which
can be manipulated during the experiments are: inlet
composition of reactants in terms of molar fractions (y0 =
[yCH3OH, yO2, yH2O]), temperature (T), total pressure (P) and
volumetric flow rate (F).
Preliminary model discrimination
Preliminary data from Nexp= 21 one-factor-at-a-time
(OFAT) preliminary experiments from the microreactor
system were available for a first discrimination among the
proposed competitive models (Model 1, Model 2 and
Model 3) where the effect of temperature (T) and feed
composition (CH3OH, O 2 and H2O molar fraction y 0) on
final products (CH 3OH, O 2, H2O, CH2O, H2, CO2) was
investigated:
1. Experiments E1-5: T varied from 725 to 826 K (inlet
composition fixed at y CH3OH=0.10, y O2=0.04, y H2O=0.07);
2. Experiments E6-9: T varied from 725 to 826 K (inlet
composition fixed at y CH3OH=0.15, y O2=0.06, y H2O=0.11);
3. Experiments E10-21: T kept at 733 K, variable y CH3OH
(range 0.07-0.14, E10 to E14; y O2=0.04, y H2O=0.08),
variable y O2 (range 0.03-0.10, E15 to E17; variable
y CH3OH=0.10, y H2O=0.08) and variable yH2O (range 0.020.22, E18 to E21; y CH3OH=0.10, y O2=0.04).
Data fitting results (Figure 2) shows the importance of
including the total oxidation reactions in the model
formulation for achieving a good representation of
oxygen concentration (Figure 2a). Model 3, including
distinct dehydrogenation and oxidation pathways,
provides the best performance in terms of fitting
formaldehyde data (Figure 2b).
Model-based Design of Experiments (MBDoE)
Eq. 1 and the power-law reaction rate expressions r j
represent a system of differential and algebraic equations
where the set of unknown kinetic model parameters
(Arrhenius kinetic constants Ai and Ea,i) have to be
estimated in the most precise and accurate way. The
variables optimised by MBDoE are:
1. Composition of reactants in terms of molar fractions
y 0: methanol (y CH3OH = 0.07-0.14), oxygen (y O2 = 0.030.10) and water (y H2O = 0.02-0.22);
2. Temperature T (725 K < T < 826 K).

The ranges of operability (shown in parenthesis above)
represent the currently investigated design space D,
where F and P have been kept constant to F = 26.5
mL/min, P = 1.6 atm. MBDoE techniques for improving
parameter estimation (MBDoE-PE) aim at decreasing
the parameter uncertainty region predicted by each
model through the solution of the optimisation
problem:

Eq. 2

53

In Eq. 3 MBDoE-MD is the discriminating power, y,i is the

standard deviation of the i-th reading, M,i and N,i are the i-th
predicted responses of model M and N, while Pi is the relative
probability of the i-th model to be the true model, as
computed after maximum likelihood parameter estimation
(13) from the following expression
Eq. 4

where SSWRi is the sum of squared weighted residuals for the i-th
model, and SSWRT is the total sum of weighted residuals for the
full set of candidate models. The probability Pi reflects the
confidence on the i-th model in such a way that MBDoE-based
model discrimination will preferentially operate on the best
candidate models representing the system. After the preliminary
parameter estimation P1 = 0.26, P2 = 0.34 and P3 = 0.40, reflecting
the superiority of Model 3 on fitting the available experimental
data.

(a)

MBDoE RESULTS AND DISCUSSION

(b)
Figure 2. Preliminary model discrimination results for Model 1,
2 and 3: molar fractions as a function of temperature for
selected species: (a) methanol and oxygen; (b)
formaldehyde. Concentrations from experiments E1-E5 are
shown, including error bars.

The design optimisation given by Eq. 2 is carried out by

computing the n-dimensional experiment design vector ,
including all the variables which can be optimised during
an experiment, i.e. = [y0, T]T. In Eq. 2 is the
N-dimensional vector of model parameters, V and H are
the variance-covariance matrix of model parameters and
the global Fisher Information Matrix (FIM), which is the sum
of FIMs of candidate models. According to Eq. 2, the
experiment is designed so as to minimise a measurement
function of V, representing the chosen design criterion
and the objective function to be optimised for improving
parameter precision ( = MBDoE-PE) The most common
design criteria are the A-, D-, E-optimal criteria, minimising
the trace, the determinant and the maximum eigenvalue
of V respectively, or its singular values (SV-optimal design)
(14). In this work the A-optimal design criterion was used.
However, the discussion that follows remains unchanged if
D- and E-optimal design criteria are considered (See
Section MBDoE Results and Discussion).
MBDoE for model discrimination (MBDoE-MD) is formulated
with the purpose of maximising the discriminating power
(i.e. the relative difference between the predictions of
candidate models):

Eq. 3

54

In order to show the potential of MBDoE techniques, a simulation

study is presented here to compute the optimal experimental
conditions to be used for both discriminating among competing
models and improving the estimation of kinetic parameters. The
optimal experimental conditions provided by MBDoE
optimisation for improving parameter precision (MBDoE-PE, Eqs.
2) and for model discrimination (MBDoE-MD, Eq. 3) are,
respectively:
1. MBDoE-PE: y0 = [yCH3OH, yO2, yH2O] = [0.14, 0.10, 0.22],
T = 732 K;
2. MBDoE-MD: y0 = [yCH3OH, yO2, yH2O] = [0.14, 0.10, 0.22] ,
T = 826 K.
As can be observed from Figure 3a, the optimally designed
conditions are far from the ones investigated by OFAT
experiments in terms of inlet concentrations of reactants. In
particular, the optimal operating region suggested by MBDoE
(for both model discrimination and improving parameter
precision) is the one at high concentrations of reactants. An
A-optimal design criterion was used for MBDoE-PE, but
optimisation results obtained from a D-optimal and E-optimal
design were very similar to the ones obtained from the A-optimal
(i.e. at the extremities of the design space). The difference was
only of 3% on the optimal temperature value found from MBDoEPE, whilst no difference on the solution was observed concerning
the initial concentration of reactants. Reaction temperature T
becomes a key factor for model discrimination: 1) at low T the
confidence in the estimation of model parameters can be
increased (MBDoE-PE); 2) at high T a clear distinction between
model predictions along the microreactor can be realised
(MBDoE-MD). Figure 3b shows the Pareto curve showing the
trade-off solutions (points 1 to 4) between MBDoE-PE and
MBDoE-MD in terms of objective functions (MBDoE-PE and MBDoEMD, Eq. 2 and Eq. 3 respectively) according to an -constraint
multi-objective optimisation approach (23), where intermediate
solutions 2 (T = 765 K) and 3 (T = 796 K) have been computed.
Figure 3c shows the impact of the MBDoE-MD design on the
prediction of formaldehyde molar fraction along the reactor for
candidate kinetic models. The maximum discriminating power is
obtained at the outlet of the reactor (point 2 in the figure, solid
lines), where Model 1 formaldehyde molar fraction can be
clearly distinguished from Model 2 and 3. However, note how a

Chimica Oggi - Chemistry Today - vol. 33(3) May/June 2015

net discrimination between Model 2 and 3 can only be

obtained at point 1. For obtaining this condition at the outlet of
the reactor, higher volumetric flow rate F has to be used
(dashed lines). Figure 3d shows the effect of MBDoE-PE design
on the parameter estimation confidence of activation energies
Ea,2 (formaldehyde partial oxidation) and Ea,3 (methanol total
oxidation) of Model 3.

(a)

(b)

Based on the solutions of the multi-objective MBDoE

optimisation, a sequence of steady-state experiments (EXP1-4) is
designed for improving the estimation of model parameters and
the simultaneous discrimination among candidate models
(MBDoE-SIM) minimising the experimental effort required for
model identification. Suggested experiments in terms of
temperature profile are given in Figure 4a. The first experiment
(EXP1) starts at the MBDoE-PE planned experimental conditions
(y0 = [0.14, 0.10, 0.22], T1 = 732 K) and the temperature is
increased to T2 = 765 K until, after a transient phase 1 (in which
the concentrations need to be stabilised) a new steady state is
reached (EXP2). After a second transient (2), the same
approach is then used from EXP3 (T3 = 796 K) to EXP4, where the
temperature is increased until MBDoE-MD conditions (y0 = [0.14,
0.10, 0.22], T1 = 826 K) are realised. As it becomes apparent from
Figure 4b, this experiment allows a very clear distinction between
model predictions in terms of formaldehyde selectivity. At y0, the
increase in temperature designed by MBDoE-SIM is expected to
increase the selectivity for Model 3, while a decrease in
selectivity will be observed for Model 1 and 2.

CONCLUSION

(c)

(d)
Figure 3. Experiments suggested by MBDoE for improving
parameter precision and simultaneous model discrimination.
(a) MBDoE optimal point and OFAT experiments in the space
of reactants concentration. (b) Pareto plot for the
simultaneous solution of the multi-objective optimisation
problems given by Eq. 2 and Eq. 3. In the plot, points 1 to 4
are distinct solutions of the multi-objective optimisation
problem given by the simultaneous solution of Eq. 2 and Eq. 3.
(c) MBDoE-MD planned experiment: formaldehyde molar
fraction profiles along the microreactor for candidate kinetic
models at nominal flow rate (F = 26.5 mL/min, solid lines) and
at increased flow rate (F = 106 mL/min, dashed lines). (d)
MBDoE-PE planned experiment: impact on estimation
precision for parameters E a,2 and E a,3 of Model 3.

Chimica Oggi - Chemistry Today - vol. 33(3) May/June 2015

A MBDoE procedure has been proposed for the optimal

design of experiments in microreactor platforms showing the
best experimental conditions to be used for a precise
estimation of the set of kinetic parameters of candidate
kinetic models and for elucidating the different reaction
pathways of proposed kinetic models based on historical
data. A preliminary discrimination of simplified kinetic models
of methanol oxidation on silver has been carried out,
underlining a better representation of experimental results
when dehydrogenation and a selective oxidation step are
included in the model formulation. Furthermore, MBDoE results
demonstrated the important role of temperature in the model
discrimination task for this system. At high concentrations of
the reactants, low temperature experiments provide the
maximum confidence for the estimation of kinetic
parameters, while high temperature ones support model
discrimination. The management of this design variable allows
designing a highly informative sequence of steady-state
experiments providing at the same time a high discriminating
power and an improvement of parametric precision.

ACKNOWLEDGEMENTS
Funding from EPSRC grant (EP/J017833/1) is gratefully
acknowledged.

REFERENCES
1.

Reuss, G., Disteldorf, W., Gamer, A. O., Albrecht, H. In: Ullmanns

Encyclopedia of Industrial Chemistry, Wiley-VCH, Wien, 15, 737-768 (2012).

55

6.

7.

8.
9.

10.

(a)
11.

12.

13.

14.
15.

(b)

16.
17.
18.

Figure 4. MBDoE results for improving parameter precision

and simultaneous model discrimination (MBDoE-SIM). (a)
Temperature profile for the sequence of steady-state
experiments EXP1-4. (b) Formaldehyde selectivity for
candidate models realised from the sequence of steadystate experiments EXP1-4.

2.
3.
4.
5.

Wachs, I. E., Madix, R. The oxidation of methanol on a silver (110)

catalyst. J. Surf. Sci. 76,531-539 (1978).
Robb, D. A., Harriott, P. The kinetics of methanol oxidation on a
supported silver catalyst. J. Catal. 35:176-183 (1974).
Montoya, A., Haynes, B. S. Methanol and Methoxide Decomposition
on Silver J. Phys. Chem. 111, 9867-9876 (2007).
Andreasen A, Lynggaard H, Stagelman C, Stolze P. A microkinetic
model of the methanol oxidation over silver. Surf. Sci. 544,5-12 (2003).

56

19.

20.

21.
22.

23.

Andreasen, A., Lynggaard, H., Stagelman, C., Stolze, P. A

simplified kinetic model of methanol oxidation on silver, Applied
Cat. 189, 267-275 (2005).
Al-Rifai , N., Cao, E., Dua, V., Gavriilidis, A. Microreactor
technology aided catalytic process design, Curr. Opin. Chem.
Eng. 2, 1-8 (2013).
Montgomery, D. Design and analysis of experiments. Hoboken,
NJ: John Wiley & Sons (2013).
Franceschini, G., Macchietto, S. Model-based design of
experiments for parameter precision: state of the art, Chem. Eng.
Sci. 63, 4846-4872 (2008).
Prasad, V., Vlachos, D. G. Multiscale model and informaticsbased optimal design of experiments: application to the catalytic
decomposition of ammonia on ruthenium. Ind. Eng. Chem. Res.
47, 6555-6567 (2008).
Schneberger, J. C., Arellano-Garcia, H., Wozny, G., Krkel, S.,
Thielert, H. Model based experimental analysis of a fixed bed
reactor for catalytic SO2 oxidation. Ind. Eng. Chem. Res. 48, 51655176 (2009).
Reizman, B. J., Jensen, K. F., An automated continuous-flow
platform for the estimation of multistep reaction kinetics, Org.
Process Des. Dev. 16, 1770-1782 (2012).
Schwaab, M., Silva, F. M., Queipo, C. A., Barreto Jr., A. G., Pinto, J.
C. A new approach for sequential experimental design for model
discrimination, Chem. Eng. Sci. 61, 5791-5806 (2006).
Galvanin, F., Bezzo, F., Macchietto, S., Model-based design of
parallel experiments, Ind. Eng. Chem. Res., 46, 871-877 (2007).
Asprey, S. P., Macchietto, S. Statistical tools for optimal dynamic
model building, Computers Chem. Eng. 24, 1261-1267 (2000).
Petrov, L. A., Eliyas, A. E., Maximov, C. S., Ind. Eng. Chem. Res., 30,
636-645 (1991).
Akiti, T. T., Birchumshaw, L. D., Eckert, R. E., Can. J. Chem. Eng., 75.
238-244 (1997).
Cao, E., Gavriilidis, A. Oxidative dehydrogenation of methanol in
a microstructured reactor, Catal. Today 110, 154-163 (2005).
Galvanin, F., Ballan, C.C., Barolo, M., Bezzo, F. A general modelbased design of experiments approach to achieve practical
identifiability of pharmacokinetic and pharmacodynamic models.
J. Pharmacokinet. Pharmacodyn. 40, 451467 (2013).
Deshmukh, S. R., Vlachos, D. G. A reduced mechanism for
methane and one-step rate expressions for fuel-lean catalytic
combustion of small alkanes on noble metals. Combustion and
Flame 149, 366383 (2007).
Stewart, W. E., Shon, Y., Box, G. E. P. Discrimination and goodness
of fit of multiresponse mechanistic models. AlChE J. 44, 1404-12
(1998).
H. Schubert, U. Tegtmeyer, R. Schlgl, On the mechanism of the
selective oxidation of methanol over elemental silver, Catal. Lett.
28, 383-390 (1998).
Mavrotas, G. Effective implementation of the -constraint method
in multi-objective mathematical programming problems, Appl.
Math. Comp. 213, 455-465 (2009).

Chimica Oggi - Chemistry Today - vol. 33(3) May/June 2015