Microkinetic analysis and mechanism of the water gas shift reaction over copper catalysts

Article abstract of DOI:10.1016/j.jcat.2011.03.008

Using microkinetic modeling, we have determined the closed catalytic cycle that describes the water gas shift reaction on copper catalysts. Eight elementary reactions constitute the cycle. Dissociation of water and the recombination of surface hydrogen re

Full text of DOI:10.1016/j.jcat.2011.03.008

Journal of Catalysis 281 (2011) 1–11
Contents lists available at ScienceDirect
Journal of Catalysis
journal homepage: www.elsevier.com/locate/jcat
Microkinetic analysis and mechanism of the water gas shift reaction
over copper catalysts
b
b
a
b
b
Rostam J. Madon a, , Drew Braden , Shampa Kandoi , Peter Nagel , Manos Mavrikakis , James A. Dumesic
⇑
a BASF Corporation, 25 Middlesex-Essex Turnpike, Iselin, NJ 08830, United States
b Department of Chemical and Biological Engineering, University of Wisconsin, Madison, WI 53706, United States
a r t i c l e i n f o
a b s t r a c t
Article history:
Using microkinetic modeling, we have determined the closed catalytic cycle that describes the water gas
shift reaction on copper catalysts. Eight elementary reactions constitute the cycle. Dissociation of water
and the recombination of surface hydrogen result in dihydrogen. While surface hydroxyl reacts with sur-
face CO to give a surface carboxyl intermediate, which, in turn, reacts with another surface hydroxyl to
give water and carbon dioxide to complete the cycle. The kinetic model further predicts that the most
abundant surface species is a formate species formed via the hydrogenation of carbon dioxide. This reac-
tion is not part of the catalytic cycle, and surface formate is a spectator species which increases substan-
tially at high pressure and low temperature. Adsorption–desorption of dihydrogen is equilibrated; while
water dissociation is kinetically significant with a high degree of irreversibility. And, depending on the
reaction conditions, carboxyl formation also becomes kinetically important. The model identifies the
binding energies of surface H, HCOO, and OH to be important parameters and allows us to explain the
difference in the site time yields of various copper catalysts.
Received 19 October 2010
Revised 8 March 2011
Accepted 9 March 2011
Available online 6 May 2011
Keywords:
Water gas shift
Copper catalyst
Microkinetic modeling
Mechanism
Copper–zinc oxide–alumina
Ó 2011 Elsevier Inc. All rights reserved.
1. Introduction
and co-workers [5–7], and Koryabkina et al. [8] exemplify the
experimental approach used to elucidate the LTS mechanism. Re-
Water gas shift is an important process step in the manufac-
ture of ammonia, on-purpose hydrogen, and is used wherever CO
needs to be converted to CO2 with concurrent production of
hydrogen. The reaction shown below is exothermic, and CO con-
version is limited in commercial units by thermodynamic
equilibrium.
cently, important theoretical studies [9–11] have added to our
understanding of this reaction. But to date, the sequence of ele-
mentary steps that make up the LTS catalytic cycle remains
unsettled. The debate centers around (a) a redox mechanism in
which Oꢁ is formed either via a two-step water dissociation or
OHꢁ disproportionation, and then reacts with COꢁ to form CO2
– where ꢁ is an active site, and (b) a formate mechanism where
HCOOꢁ serves as a reactive intermediate [12,13]. Recently, Gok-
hale et al. [11] using a DFT (Density Functional Theory) investi-
CO þ H2O ¢ CO2 þ H2
D
H298K ¼ ꢀ41 kJ molꢀ1
Due to this limitation, commercially, the shift process is carried
out in two steps: first, a high temperature water gas shift (HTS)
step at 623–673 K over an iron-based catalyst, followed by a low
temperature water gas shift (LTS) at 463–503 K over a Cu-based
catalyst. The latter step brings CO levels below 0.5 mol%. This paper
describes our attempts at understanding the reaction chemistry of
LTS over commercially relevant Cu–ZnO–Al2O3 catalysts.
Since the introduction of the Cu catalyzed shift reaction over
forty years ago, fundamental studies, both experimental and the-
oretical, have been carried out to understand this reaction. Since
the early work of Yurieva [1] who proposed an associative
mechanism and Grenoble et al. [2] who proposed a formic acid
intermediate, investigations by Ovesen et al. [3,4], Campbell
gation of the LTS reaction on Cu(1 1 1) proposed
a new
mechanism that involves a reactive surface carboxyl species,
ꢁCOOH.
The present study aims to resolve which elementary steps best
describe the catalytic cycle for the Cu-catalyzed LTS reaction. Our
analysis uses all proposed reaction steps that make chemical sense,
as per DFT calculations, and employs the microkinetic approach
described elsewhere [14]. With this approach, we obtain the steps
that determine the LTS catalytic cycle as well as the kinetic param-
eters that describe these steps. Furthermore, this exercise
identifies key parameters that affect LTS activity thus aiding exper-
imental catalyst research. Importantly, by addressing these param-
eters either by altering catalyst components or by varying the
preparation methodology, or both, one can obtain important
insights into the LTS catalytic chemistry.
⇑
Corresponding author. Fax: +1 732 205 5585.
E-mail address: rostam.madon@basf.com (R.J. Madon).
0021-9517/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcat.2011.03.008

2
R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
1040/(%D), gives us Cu particle diameter in Å. We thus report all
reaction rates as site time yields, Y sꢀ1
2. Experimental methods
.
2.1. Catalysts
3. Development of the microkinetic model
The water gas shift catalysts studied here are composed of Cu,
ZnO, and Al2O3. Traditionally, the catalyst is prepared [15,16] by
coprecipitating the nitrates of Cu, Zn, and Al with sodium carbon-
ate; at times sodium aluminate is also used. This precipitate is fil-
tered, washed to remove sodium, dried, and calcined to decompose
the carbonates. The catalyst may also be prepared with alumina
powder [17,18]. Our standard precipitation takes place in 90 min
at 333 K and at a constant pH of 7; the dried carbonate is calcined
at 673 K until all carbonate is decomposed. Because these prepara-
tions are well known, we will not discuss them further. In the Re-
sults Section, as a particular catalyst is used in the model, we will
note its composition and relevant properties.
3.1. The reaction sequence
Table 1 shows 18 elementary steps that are suggested from re-
sults of DFT calculations [11] to describe the LTS reaction. They
represent the various possible pathways: redox, formate-, and car-
boxyl-mediated mechanisms.
The first five reactions are common to all suggested mecha-
nisms: adsorption of reactants and desorption of products plus
the dissociation of steam. Reactions 6–8 describe the redox mech-
anism. Reactions 9–13 indicate a possible formate-mediated mech-
anism; Reaction 13 is important in that the bidentate formate is
more stable than the monodentate intermediate. And although
DFT analysis indicates that Reaction 12 is not an elementary step,
we include it here since it has been suggested as a key reaction in
the formate mechanism [5,12,13].
Finally, Reactions 14–18 represent the carboxyl mechanism.
The carboxyl species forms as a cis configuration with the hydro-
gen of the hydroxyl pointing away from the surface [11]. This spe-
cies transforms via Reaction 18 to the reactive trans species with
hydrogen pointing toward the surface. The trans carboxyl may
decompose (Reaction 15) and/or react with a surface hydroxyl spe-
cies (Reaction 16) or with a surface O (Reaction 17) to give the
product CO2. Initially, we include all these steps in our model
and allow microkinetic analysis to determine which steps consti-
tute the LTS catalytic cycle.
2.2. Reactivity measurement
We carried out experimental reaction kinetics studies in a
12.7 mm OD fixed bed stainless steel reactor. We used 0.6 g of
0.3–0.15 mm (50–100 mesh) sized catalyst mixed with ca. 10
times the volume of similarly sized inert corundum powder to re-
duce temperature gradients. About 400 mm of the reactor above
the catalyst bed acted as a preheater. Our control algorithm al-
lowed us to maintain the average catalyst temperature at the set
reaction temperature. We analyzed the reactor exit stream, includ-
ing steam, with an online Agilent microGC.
Before reaction, we reduced the catalyst in dilute hydrogen tak-
ing precautions that there were no exotherms. The catalyst tem-
perature was first raised to 443 K in He flowing at 500 cm3(STP)/
min. Three percent H2 in He was introduced at the same flow rate;
the concentration was then increased to 5% H2 and finally to 20%
H2. The catalyst temperature was then raised over 1 h to 473 K in
20% H2 in He, where it was kept for 2 h. The catalyst was flushed
in N2 before introducing reaction gases. Our standard reaction con-
ditions were as follows: 2.7 bars absolute pressure, 473 K temper-
ature, and an inlet feed composition of 36.7% H2, 16.7% N2, 8% CO,
5.3% CO2, and 33.3% steam. The feed composition translated to a
dry gas/steam molar ratio of 5. As we will see later, inclusion of
product CO2 in the inlet feed stream, as is practiced commercially,
influences the modeling approach. Under these experimental con-
ditions, the LTS reaction has a calculated equilibrium CO conver-
sion of 98.6% [19]. For our modeling studies, we also used data
obtained at other conditions: 30 bars total pressure, temperatures
ranging from 453 to 503 K, as well as, inlet feed composition of
36.7% H2, 16.7% N2, 2.7% CO, 10.6% CO2, and 33.3% steam. All exper-
iments were carried out at a range of CO conversions obtained by
varying catalyst weight-based gas hourly space-velocity (GHSV),
wet basis. And in all experimental sets, conditions were repeated
to ensure that there was no catalyst deactivation.
3.2. Parameterization of the model
3.2.1. Thermodynamic parameters
We parameterize the model in terms of either the forward or the
reverse rate constant ðki;for or ki;re Þ; these constants being related
v
to the thermodynamic equilibrium constant ðKi;eq
Þ
ki;for
¼ Ki;eq
ð1Þ
ki;re
v
Therefore, we first calculate Ki,eq for each of the 18 elementary steps
of the reaction sequence. Since
ꢀ
ꢁ
D
R
Sio
D
Hio
RT
Ki;eq ¼ exp
ꢀ
ð2Þ
Table 1
Elementary steps used in the microkinetic modeling of LTS.
Reaction step
Surface reactions
CO þ ꢁ $ COꢁ
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
2Hꢁ $ H2 þ 2ꢁ
H2O þ ꢁ $ H2Oꢁ
COꢁ2 $ CO2þꢁ
2.3. Cu dispersion measurement
H2Oꢁ þ ꢁ $ Hꢁ þ OHꢁ
OHꢁ þ ꢁ $ Oꢁ þ Hꢁ
COꢁ þ Oꢁ $ COꢁ2 þ ꢁ
We measured Cu dispersion (D) on fresh, reduced, catalyst sam-
ples using the N2O dissociative chemisorption method of Chinchen
et al. [20] on a Micromeritics Autochem 2920. The chemisorption
procedure involves bringing the reduced catalyst to 333 K in He,
switching to 2% N2O in He, and monitoring the N2 evolved with a
calibrated thermal conductivity detector or with a mass spectrom-
eter. For every dinitrogen evolved, one oxygen atom remains on
the surface bonded to two surface Cu atoms. Using a Cu atom sur-
face area of 0.68 nm2 per atom obtained from a surface packing of
1.47 ꢂ 1019 atoms mꢀ2, we calculate the percent dispersion (%D)
and copper surface area. Assuming spherical particles, the relation,
OHꢁ þ OHꢁ $ H2Oꢁ þ Oꢁ
COꢁ2 þ Hꢁ $ HCOOꢁ þ ꢁ
COꢁ2 þ H2Oꢁ $ HCOOꢁ þ OHꢁ
COꢁ2 þ OHꢁ $ HCOOꢁ þ Oꢁ
COꢁ þ OHꢁ $ HCOOꢁþꢁ
HCOOꢁ þ ꢁ $ HCOOꢁꢁ
COꢁ þ OHꢁ $ COOHcꢁis þ ꢁ
COOHꢁtrans þ ꢁ $ CO2ꢁ þ Hꢁ
COOHꢁtrans þ OHꢁ $ CO2ꢁ þ H2Oꢁ
COOHꢁꢁ þ Oꢁ $ CO2ꢁ þ OHꢁ
COOHcis $ COOHꢁtrans

R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
3
we need to obtain standard entropy and enthalpy changes ð
D
Soi and
all steps. The values of the fitted parameters are determined using
the Athena Visual Studio engineering software [24] that has the
solver and fitting routines built-in. The fitting procedure minimizes
the sum of the squares of the residuals. The goodness of fit is de-
rived by comparing predicted CO conversions obtained via the
model and CO conversions via experimentation, and calculating
the coefficient of determination, R2, the value of which approaches
unity as the fit improves.
Possible adjustable parameters in our model are the activation
energies of the 18 steps given in Table 1, and the binding energies
of the surface intermediates, except for the BEs of O and COOHcis
which, as suggested in Ref. [11], are fixed. Activation energies of
Reactions 1, 3, and 4 are all 0 kJ/mol since no bonds are broken.
Thus, there are 23 possible adjustable parameters shown in Table 2
with the DFT-calculated values.
D
Hoi Þ from the thermodynamic properties of all gaseous and surface
species, with R and T being the gas constant and temperature,
respectively.
First, we obtain or calculate the thermodynamic properties of
all species in the gas phase. Calculations of gas phase entropies
ðSogasÞ, not available in handbooks, have been discussed in detail
elsewhere [21,22]. We obtain gas phase enthalpies ðHogasÞ, corrected
for zero point energy (ZPE), from DFT calculations [11].
Next, we calculate the thermodynamic properties of all surface
intermediates by relating them to the thermodynamic properties
of gas phase species. We calculate the enthalpy of formation of a
surface species, H0i;surface, as
H0i;surface ¼ Hi0;gas þ BEi þ
D
ZPEi
ð3Þ
We make the initial parameter adjustments using a transient
CSTR approach because the numerical methods used are robust
when solving a series of simultaneous differential equations rather
than a combination of algebraic and differential equations as one
would solve in a plug flow reactor (PFR) or a CSTR operating at
steady state. Robust numerical solving methods are necessary be-
cause errors inherent in DFT-predicted parameter values may lead
to a poorly parameterized model and make converging on a solu-
tion difficult. A transient CSTR approach more easily adjusts the
sensitive parameters from their DFT values.
where BE is the binding energy of an intermediate i on Cu(1 1 1),
and ZPE is the difference between the ZPE correction of the spe-
cies on the surface and in the gas phase. Initial values of all BEs
and all ZPEs are obtained from DFT calculations [11]. The BE is an
important term that determines surface coverage of an
intermediate.
D
Finally, the standard entropy for each surface intermediate is
based on the local entropy of the corresponding gaseous species
and is calculated as shown in Ref. [21]. The local entropy is ob-
tained by subtracting the gaseous translational entropy from the
Briefly, material balances for adsorbed species may be written
as:
standard entropy Sgoas
.
3.2.2. Kinetic parameters
dSi
dt
¼ SRYi
ð5Þ
For the adsorption–desorption steps 1–4, we define the rate
constant in terms of adsorption, using collision theory to obtain
the preexponential factor [14,23]. We then use the equilibrium
constant to calculate the corresponding desorption rate constant.
For surface reactions, we choose to define rate constants in the
exothermic direction, where
where Si is the number of molecules of species i adsorbed on active
catalyst sites, SR is the total number of active catalyst sites, and Yi is
the site time yield per s for the formation of species i. Since for a
dSi
transient CSTR
is not zero, the differential equations for the ad-
dt
ꢀ
ꢁ
ꢀEi
sorbed surface species and gas phase molecules are solved as a
function of time during the simulation. The simulation starts with
a vacant catalyst surface and a reactor charged with reactants and
ends when the reactor reaches steady state operation, at which
ki ¼ A exp
ð4Þ
RT
A is the preexponential factor, Ei is the ZPE-corrected activation en-
ergy barrier in the exothermic direction. We obtain all preexponen-
tial factors via DFT calculations [11], and these are non-adjustable
parameters in our model. Once the forward rate constant is ob-
tained, we use the equilibrium constant to calculate the rate con-
stant in the endothermic direction. We parameterize the kinetics
of each surface step in terms of the ZPE-corrected activation
energies.
Importantly, by basing the kinetic model on thermodynamic
values for all reactants, intermediates, and observed products,
and by using Eq. (1), we ensure that all kinetic parameters of the
model are thermodynamically consistent.
Table 2
Binding energies and activation energies obtained from Ref. [11]and used as initial
parameters for the model.
Parameters
From DFT
Binding energy (kJ/mol)
1
2
3
4
5
6
7
8
CO
H
H2O
CO2
51.4
246.1
17.4
8.7
OH
COOHtrans
HCOO (monodentate)
HCOO (bidentate)
275.0
181.4
223.9
267.3
3.3. Modeling approach
Activation energy (kJ/mol)
H2 þ 2ꢁ $ 2Hꢁ
9
52.1
131.2
169.8
79.0
58.9
136.0
40.5
300.0
71.4
260.0
0.0
We use the microkinetic model to describe the experimentally
measured CO conversions. The inputs to the model are the inlet
flows of all gases, reaction temperature, total pressure, the total
number of active sites in the reactor, and the experimentally deter-
mined CO conversions. Here, we assume that the active sites corre-
spond to surface moles of Cu. We allow the reactor in the
simulation to operate as a transient continuous stirred tank reactor
(CSTR). We solve simultaneously 4 differential equations for the
gaseous flow rates together with 10 differential equations for the
fractional surface coverage of adsorbed intermediates, and one site
balance equation. The solution then yields the four exit gas flow
rates of H2, CO, CO2 and H2O, the coverage of the 10 surface inter-
mediates, and the forward and reverse rates and rate constants of
H2Oꢁþꢁ $ Hꢁ þ OHꢁ
10
11
12
13
14
15
16
17
18
19
20
21
22
23
OHꢁ þ ꢁ $ Oꢁ þ Hꢁ
COꢁ þ Oꢁ $ COꢁ2 þ ꢁ
COꢁ þ OHꢁ $ COOHcꢁis þ ꢁ
COOHꢁtrans þ ꢁ $ CO2ꢁ þ Hꢁ
COOHꢁtrans þ OHꢁ $ CO2ꢁ þ H2Oꢁ
COꢁ þ OHꢁ $ HCOOꢁ þ ꢁ
COꢁ2 þ Hꢁ $ HCOOꢁ þ ꢁ
COꢁ2 þ H2Oꢁ $ HCOOꢁ þ OHꢁ
OHꢁ þ OHꢁ $ H2Oꢁ þ Oꢁ
COꢁ2 þ OHꢁ $ HCOOꢁ þ Oꢁ
COOHꢁ þ Oꢁ $ COꢁ2 þ OHꢁ
HCOOꢁꢁ þ ꢁ $ HCOOꢁꢁ
132.0
60.0
9.7
COOHcis $ COOHꢁtrans
50.2

4
R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
dSi
dt
ZnO, 23% Al2O3, Cu dispersion of 6.2%, and Cu surface area of
point
equals zero; i.e., the composition of the reactor effluent
18.4 m2/g. And the second catalyst contained 21% Cu, 55% ZnO,
24% Al2O3, Cu dispersion of 7.1%, and Cu surface area of 9.7 m2/g.
At the standard experimental conditions, and 40% CO conversion,
the reaction rates were 0.13 and 0.12 sꢀ1, respectively. In accord
with the criterion discussed elsewhere [25,26], this identity of
rates for catalysts with the number of surface sites differing by
about a factor of two ensures that kinetic data are free from heat
and mass transfer limitations as well as catalyst by-passing and
poisoning.
and the catalyst surface coverage are constant with time, as re-
quired in a CSTR operating at steady state.
Material balances for gaseous species may be written as:
dFoi ut
dt
h
i
¼ K
Fiin ꢀ Foi ut þ SRYi
ð6Þ
v
oid
where Foi ut and Fiin are the outlet and inlet molar flow rates of species
i. The constant K
depends on the amount of void space in the
v
oid
reactor, and its value is taken to be unity. (This value only scales
the length of the transient, and it does not alter the steady state per-
formance of the reactor.)
4.1. The microkinetic model and catalytic cycle for the LTS reaction
We begin modeling with parameter values shown in Table 2.
The model adjusts the sensitive parameters, as a good fit is found
for the input data, at a given set of experimental conditions. This
new set of parameters then describes the kinetics of the LTS reac-
tion. For the model to be realistic and robust, the parameter set
should describe the LTS reaction over a range of conditions. Impor-
tantly, parameters obtained using a transient CSTR approach serve
as good initial values for simulations using the PFR approach.
Material balances for surface and gaseous species may be writ-
ten as follows for the PFR approach:
We obtained all data, in this section, using a catalyst containing
52% Cu, 32% ZnO, 16% Al2O3, with a Cu dispersion of 4.5%. First, to
obtain the catalytic cycle, we used our standard reaction condi-
tions and modeled CO conversion – GHSV results using the param-
eters in Table 2 as our initial input. Fig. 1 shows the result of the
simulation as a parity plot (R2 = 0.987), and Table 3 shows changes
in the initial parameter set to achieve the result. Remarkably, al-
most all our initial input parameters remain unchanged: there
are minor changes for the BEs of H, OH, and bidentate HCOO, a
6 kJ/mol and a 10 kJ/moꢁl change in the activation energies of steam
dissociation and COOHcis formation, respectively. The simulation
further demonstrates (Fig. 2) that the model predicts reaction rates
very well. Accordingly, we can now use these predictions to find
which elementary reactions constitute the closed catalytic cycle
and which reactions do not play a role in LTS chemistry.
dSi
dt
¼ 0
ð7Þ
dFi
dS
¼ Yi
ð8Þ
where S is the number of surface sites, equal to zero at the reactor
inlet and SR at the exit.
Table 4 gives the net rates derived by the model for of all the
reactions at 25% CO conversion. Although comparison at any con-
version is valid, it is always useful to compare far from the overall
However, elucidation and description of reaction kinetics phe-
nomena in terms of catalyst properties are most clearly accom-
plished in a CSTR compared to a PFR, because all of the catalyst
in a CSTR operates at the same well-mixed composition of reac-
tants and products. Whereas a gradient of compositions exists in
a PFR and the catalyst bed operates in this non-uniform environ-
ment. Therefore, discussions of how surface coverages on the cat-
alyst are affected by the presence of promoters or by changes in
binding energies of intermediates on the catalyst are clearer when
the catalyst is simulated in a CSTR compared to a PFR. Therefore we
have used a CSTR to probe such effects.
Errors associated with experimental work and with solver oper-
ation may result in small variations in key parameters; we, ad hoc,
allow variations in such parameters by 4 kJ/mol as experimental
conditions and catalysts are changed. However, we will address
variations for any key parameter as the need arises. Errors associ-
ated with DFT calculations for binding and activation energies are
about 10 kJ/mol.
Importantly, we determined that when starting within the
above acceptable DFT error limit, the same optimal parameter fits
were obtained. Thus, our study accurately captured the most prob-
able solution when we initialized within the acceptable DFT confi-
dence limits. Moreover, as shown later, our conclusions regarding
the reaction pathway for the water gas shift reaction did not
change even when we moved the parameters from DFT values by
much larger amounts.
70
60
50
40
30
20
10
0
0
10
20
30
40
50
60
70
% CO Conv — Observed
Fig. 1. Simulation results at standard reaction conditions: parity plot for % CO
conversion.
Table 3
Model derived parameter changes at standard reaction conditions.
Parameters
DFT (kJ/mol)
Model (kJ/mol)
4. Results
All binding energies same as DFT values EXCEPT
BEH
BEOH
BEHCOO (bidentate)
246.1
275
267.3
245.1
276.3
265.1
Before we describe our results and discuss the chemistry of the
LTS reaction, we need to show that the kinetic data are free from
transport artifacts. To that end, we made two catalysts ensuring
all preparative parameters were identical except for the Cu and
ZnO amounts, thus allowing us to change the Cu surface area by
about a factor of two. The first catalyst contained 46% Cu, 31%
All activation energies same as DFT values EXCEPT
E5 H2Oꢁ þ ꢁ $ Hꢁ þ OHꢁ
131.2
58.9
125
48.7
E14 COꢁ þ OHꢁ $ COOHcꢁis þ ꢁ

R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
5
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.08
The sequence of elementary steps that describe the reaction
mechanism for LTS on Cu is given in Table 5. The ratio of for-
ward/reverse rates indicates that Reactions 1–4 and 18 are in qua-
si-equilibrium. The ratio of net rate/forward rate (d) gives the
degree of irreversibility or distance from equilibrium; as d ? 0,
the reaction approaches equilibrium; and as d ? 1, the reaction be-
comes irreversible. As noted by Boudart [27] in terms of the chem-
ical affinity AF
Observed
Predicted
ꢀ
ꢁ
AF
RT
d ¼ 1 ꢀ exp
ꢀ
ð9Þ
This ratio d for step 5 is 0.88; thus, steam dissociation is the
most irreversible step in the sequence. Steps 14 and 16 are revers-
ible but not in quasi-equilibrium.
We next address another important result from the model that
bidentate formate is the most abundant surface species. The mech-
anism given in Table 5 does not account for surface formate species
that have also been discussed elsewhere [4,5,12,13]. Therefore we
need to identify which of the five reactions that involve surface for-
mate participate in the surface chemistry, albeit not as part of the
LTS catalytic cycle. As noted in Table 4, these reactions have very
low net rates. Reactions 10–12 also have extremely low forward
rates (Table 6), whereas Reactions 9 and 13 have higher forward
and reverse rates and are in quasi-equilibrium. Steps 9 and 13 ade-
quately describe the formation of the bidentate formate intermedi-
ate; the formate intermediate is not involved in the LTS reaction
directly and is a spectator species that affects the LTS reaction by
its surface coverage in accord with Refs. [4,11].
0
50
100
150
200
250
Wet Gas Hourly Space-Velocity, x103
Fig. 2. Simulation results for site time yields, standard reaction conditions.
Table 4
Net rates of elementary steps at standard reaction conditions, 25% CO conversion.
Surface reactions
Net rates (sꢀ1
)
CO þ ꢁ $ COꢁ
1.
2.
3.
4.
5.
6.
7.
8.
0.145
0.145
0.145
0.145
0.290
10ꢀ17
10ꢀ10
10ꢀ10
10ꢀ17
10ꢀ26
10ꢀ17
10ꢀ28
10ꢀ8
2Hꢁ $ H2 þ 2ꢁ
H2O þ ꢁ $ H2Oꢁ
COꢁ2 $ CO2 þ ꢁ
H2Oꢁ þ ꢁ $ Hꢁ þ OHꢁ
OHꢁ þ ꢁ $ Oꢁ þ Hꢁ
4.1.3. Model at varying pressure and temperature
Henceforth, we omit all unimportant reactions from the fitting
process and fit all data with only Reactions 1–5, 9, 13, 14, 16, 18,
and with the germane, reduced parameter set.
COꢁ þ Oꢁ $ COꢁ2 þ ꢁ
OHꢁ þ OHꢁ $ H2Oꢁ þ Oꢁ
COꢁ2 þ Hꢁ $ HCOOꢁ þ ꢁ
COꢁ2 þ H2Oꢁ $ HCOOꢁ þ OHꢁ
COꢁ2 þ OHꢁ $ HCOOꢁ þ Oꢁ
COꢁ þ OHꢁ $ HCOOꢁ þ ꢁ
HCOOꢁ þ ꢁ $ HCOOꢁꢁ
COꢁ þ OHꢁ $ COOHcꢁisþꢁ
COOHꢁtrans þ ꢁꢁ $ CO2ꢁ þ Hꢁ
COOHꢁtrans þ OHꢁ $ CO2ꢁ þ H2Oꢁ
COOHꢁꢁ þ Oꢁ $ CO2ꢁ þ OHꢁ
COOHcis $ COOHꢁtrans
9.
Using the same catalyst and the same reactant gas mixture as
above, we obtained data at 2.7 and 30 bars at 473 K, and over a
temperature range from 453 to 503 K at 2.7 bars. Figs. 3 and 4
show the parity plots for CO conversion for the pressure and tem-
perature sets, respectively, over a wide range of conversions. We
obtained excellent fits, R2 = 0.983 for the pressure set and
R2 = 0.984 for the temperature set, with the same model parame-
ters (Table 3) without any adjustment. This agreement substanti-
10.
11.
12.
13.
14.
15.
16.
17.
18.
0.145
10ꢀ6
0.145
10ꢀ11
0.145
Table 5
The catalytic cycle for LTS on copper.
thermodynamic equilibrium conversion of 98.6%. The experimen-
tal rate at 25% CO conversion is 0.148 sꢀ1. Since net rates of all
reactions in a catalytic cycle must be the same as the overall rate
of reaction, after correction for the stoichiometric number of a step,
Table 4 identifies the reaction steps that constitute the catalytic cy-
cle for the LTS reaction. Reactions 1–5, 14, 16, and 18 have the
same net rate as the overall experimentally observed value. Reac-
tion 5, the dissociation of water, has a stoichiometric number of 2;
i.e., two water molecules are dissociated in a cycle. And, although
the carboxyl intermediate can react via steps 15–17 to give the
product CO2, the model indicates that step 16 is the relevant step
in the mechanism.
The LTS mechanism
Forward/reverse
rate
Net/forward rate
(d)
1.
2.
3.
4.
1.0
1.0
1.0
1.0
8.3
2.2
1.1
1.0
10ꢀ7
10ꢀ4
10ꢀ8
10ꢀ7
0.88
0.54
0.06
10ꢀ3
CO þ ꢁ $ COꢁ
2Hꢁ $ H2 þ 2ꢁ
H2O þ ꢁ $ H2Oꢁ
COꢁ2 $ CO2 þ ꢁ
2H2Oꢁ þ 2ꢁ $ 2Hꢁ þ 2OHꢁ
COꢁ þ OHꢁ $ COOHcꢁis þ ꢁ
COOHꢁtrans þ OHꢁ $ CO2ꢁ þ H2Oꢁ
COOHꢁcis $ COOHtꢁrans
5.
14.
16.
18.
CO þ H2O $ CO2 þ H2
Reaction 7, the reaction of COꢁ with Oꢁ, and Reaction 8, the dis-
proportionation of surface hydroxyls, key reactions in the redox
mechanism, have very low rates of 10ꢀ10 sꢀ1. Importantly, even,
in a what-if simulation, when we force the activation energy of
step 7 from a DFT-calculated value of 79 to 10 kJ/mol, the net rate
of COꢁ oxidation is still an order of magnitude lower than the
experimental value. Furthermore, all steps containing the formate
intermediate also have extremely low net rates. Thus steps associ-
ated with the redox mechanism or formate-mediated mechanism
are not part of the LTS cycle.
Table 6
Forward reaction rates for formate-producing reactions (25% CO conversion).
Formate reactions
Forward rate, sꢀ1
10ꢀ3
COꢁ2 þ Hꢁ $ HCOOꢁ þ ꢁ
COꢁ2 þ H2Oꢁ $ HCOOꢁ þ OHꢁ
COꢁ2 þ OHꢁ $ HCOOꢁ þ Oꢁ
COꢁ þ OHꢁ $ HCOOꢁ þ ꢁ
HCOOꢁ þ ꢁ $ HCOOꢁꢁ
9.
10.
11.
12.
13.
10ꢀ27
10ꢀ18
10ꢀ28
10+7

6
R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
100
90
80
70
60
50
40
30
20
10
0
using a commercial Cu:ZnO:Al2O3 catalyst and experimental con-
ditions close to ours. Although we did not obtain reaction orders
for individual components, the near identity of site time yields at
2.7 and 30 bars pressure indicates a near zero-order dependence
on pressure, similar to the observations in Ref. [8].
473 K
4.1.4. Surface coverage
Fig. 5 shows surface coverages at 2.7 and 30 bars pressure. At
the lower pressure, the coverage is low with 8–10% HCOOꢁꢁ, 2%
Hꢁ, and 0.5% COꢁ. The rest of ꢁthe surface species, including the
important intermediate COOHtrans, have a fractional coverage of
10ꢀ5 or less. At higher pressure, there is a significant increase in
bidentate formate coverage, a small increase in COꢁ and no change
in Hꢁ. Fig. 6 shows the effect of temperature on coverage at 27% CO
conversion. There is a significant decrease in HCOOꢁꢁ as tempera-
ture increases, with no concurrent change in H coverage. Neither
temperature nor pressure alters coverage of other intermediates
to any determinant degree; only formate determines to what ex-
tent the reactive surface is crowded or not.
4.1.5. Plug flow analysis
So far we have generated a robust parameter set and ob-
tained all model results using the transient CSTR approach. With
0
20
40
60
80
100
% CO Conv — Observed
0.40
473 K
Fig. 3. Simulation results at 2.7 and 30 bars pressure, 473 K, feed composition:
0.35
0.30
0.25
36.7% H2, 16.7% N2, 8% CO, 5.3% CO2, 33.3% steam.
100
2.7 bars
HCOO**
90
80
70
60
50
40
30
20
10
0
0.20
0.15
0.10
CO*
H*
0.05
0
20
30
40
50
60
70
80
90
% CO Conversion
Fig. 5. Surface coverage at 2.7 (closed points) and 30 bars (open points) pressure,
473 K, feed composition: 36.7% H2, 16.7% N2, 8% CO, 5.3% CO2, 33.3% steam.
0.15
2.7 bars
0.12
0
20
40
60
80
100
% CO Conv — Observed
0.09
Fig. 4. Simulation results for a temperature range of 453–503 K, 2.7 bars, feed
HCOO**
composition: 36.7% H2, 16.7% N2, 8% CO, 5.3% CO2, 33.3% steam.
0.06
ates the validity of the mechanism over a wide range of tempera-
tures and pressures. The excellent fit over a 50 K temperature
range confirms that the activation energies obtained from both
the DFT calculations and the model are accurate. And the argu-
ments of reaction equilibrium and reversibility remain unchanged.
Experimentally, we obtained an overall apparent activation en-
ergy of 73 kJ/mol calculated at ca. 30% CO conversion and a tem-
perature range from 458 to 498 K. This value is close to the
activation energy of 79 kJ/mol obtained by Koryabkina et al. [8],
0.03
H*
0
460
465
470
475
480
485
490
495
500
Temperature, K
Fig. 6. Formate and hydrogen coverage versus temperature, 2.7 bars total pressure,
feed composition: 36.7% H2, 16.7% N2, 8% CO, 5.3% CO2, 33.3% steam.

R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
7
these parameters, we now fit the combined data of Figs. 3 and 4
using the PFR approach. We obtain an excellent fit (Fig. 7) of all
the data at two pressures and various temperatures with mini-
4.1.6. Variation in CO and CO2 levels
Here, we study the LTS reaction at 2.7 bars pressure, tempera-
ture ranging from 458 to 488 K, and inlet feed composition of
36.7% H2, 16.7% N2, 2.7% CO, 10.6% CO2, and 33.3% steam. Com-
pared to our standard conditions, we change the CO/CO2 ratio from
1.5 to 0.25. This low CO/CO2 ratio mimics commercial values for
example in ammonia synthesis plants. We obtain (Fig. 9) an excel-
lent fit for 64 data at four different temperatures, R2 = 0.982, using
the same parameter set as in Table 3 with a minimal increase of
3 kJ/mol for BEH and a decrease of 6 kJ/mol for BEOH. At the lower
CO/CO2 ratio, the reaction rate decreases significantly; Fig. 10, at
473 K and 2.7 bars, demonstrates how well the model describes
this change in rates.
mal changes of 3 kJ/mol or less for BEH, BEHCOO, and BEOH
.
Fig. 8 shows the surface coverage for HCOOꢁꢁ and Hꢁ. CSTR anal-
ysis gives surface coverage throughout the well-mixed reactor,
whereas, for the PFR, we plot coverage at the exit of the reactor.
The predicted results between CSTR and PFR approaches are
remarkably consistent and give credence to the robustness of
the model. Such consistency is expected in reactions where there
is modest inhibition of the rate by product species that are not
present in the feed stream. In our case, co-feeding CO2 at reactor
inlet ensures formation of bidentate formate throughout the PFR,
thus allowing facile transition from the CSTR approach to a PFR
approach. Furthermore, as noted earlier, the overall pressure
dependence for the LTS reaction is close to zero.
4.2. Modeling changes in LTS catalyst
4.2.1. Alkali as promoter
Our model indicates that bidentate formate is not only the
most abundant surface species but that it increases significantly
with increasing pressure and decreasing temperature. This result
is important since bidentate formate has been suggested [4,28]
to be an intermediate during methanol formation over Cu.
100
90
80
70
60
50
40
30
20
10
0
80
2.7 bars; 458 to 488 K
70
60
50
40
30
20
10
0
0
20
40
60
80
100
% CO Conv — Observed
0
20
40
60
80
Fig. 7. Simulation results using plug flow analysis with combined data from Figs. 4
% CO Conv — Observed
and 5.
Fig. 9. Simulation results for a temperature range of 458–488 K, 2.7 bars total
pressure, feed composition: 36.7% H2, 16.7% N2, 2.7% CO, 10.6% CO2, 33.3% steam.
0.5
0.16
0.14
HCOO**
PFR
CSTR
0.4
0.3
0.2
0.1
0
CO/CO2 = 1.5
0.12
0.1
0.08
0.06
CO/CO2 = 0.25
0.04
H*
0.02
0
PFR
CSTR
0
50
100
150
200
250
30
40
50
60
70
80
90
Wet Gas Hourly Space-Velocity, x103
% CO Conversion
Fig. 10. Site time yield comparisons. Experimental – closed symbols, model – open
symbols. Pressure 2.7 bars, Temperature 473 K. Diamond symbol: 36.7% H2, 16.7%
N2, 8% CO, 5.3% CO2, 33.3% steam. Square symbols: 36.7% H2, 16.7% N2, 2.7% CO,
10.6% CO2, 33.3% steam.
Fig. 8. Formate and hydrogen coverage using transient CSTR (closed points) and
plug flow (open points) analyses at 30 bars total pressure, 473 K, feed composition:
36.7% H2, 16.7% N2, 8% CO, 5.3% CO2, 33.3% steam.

8
R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
Methanol, even in ppm levels, is an unwanted byproduct during
commercial LTS operation and is particularly prevalent at high
pressures used commercially. However, addition of alkali to the
Cu catalyst [18,29] mitigates methanol formation. We therefore
used a catalyst containing 49% Cu, 33% ZnO, 17% Al2O3, and 1%
K2O plus Cs2O, Cu dispersion of 4.5%, for LTS at 30 bars pressure,
473 K, and standard feed composition to obtain %CO conversion
versus GHSV data. We were unable to model these data with
parameters from Table 3. Only after the value of BEH was re-
duced from ꢀ245 kJ/mol to ꢀ234 kJ/mol and the value of BEHCOO
was reduced from ꢀ265 kJ/mol to ꢀ220 kJ/mol, were we able to
obtain an excellent fit with R2 = 0.963. These changes resulted in
a small decrease in H coverage from 2% to 0.5% and a substantial
decrease in HCOOꢁꢁ from 36% to 0.005%. Thus, the presence of al-
kali appears to destabilize the formation of bidentate formate,
and we infer that the reduction in methanol formation is due
to such destabilization of a key intermediate. The model teaches
us that formate species in LTS is not only a spectator species, but
it probably plays a role in byproduct methanol formation. How-
ever, we were unable to observe ppm levels of methanol, so the
connection between formate intermediate and methanol product
is by analogy with previous work.
0.18
0.16
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0
0
50
100
150
200
250
300
Wet Gas Hourly Space-Velocity, x103
Fig. 11. Site time yield comparisons for Catalysts 1–4 (Table 7). Experimental –
closed symbols, Model – open symbols. Standard reaction conditions.
0.18
0.16
0.14
0.12
0.1
4.2.2. Varying catalyst site time yield
In the previous section, we varied the reaction rate by changing
the CO/CO2 ratio, and the model was able to describe the data well.
Here, we investigate four catalysts with varying activities. Table 7
gives catalyst composition, Cu dispersion, and site time yields at ca.
35% CO conversion. Catalysts 1 and 2 are from a single batch of cat-
alyst made with our standard method as noted in Section 2.1. We
calcined Catalyst 1 at 673 K for 2 h to completely decompose the
carbonate, whereas Catalyst 2 was calcined at a lower temperature,
523 K for 2 h, and a substantial amount of carbonate remained be-
fore the catalyst was reduced. We made Catalysts 3 and 4 in the
same way as Catalyst 1, except that neodymium nitrate was added
to the copper and zinc nitrate before precipitation for Catalyst 3,
and yttrium nitrate was used for Catalyst 4.
0.08
0.06
0.04
0.02
0
-240
-242
-244
-246
-248
-250
Fig. 11 shows the model description of the rate data for all four
catalysts. Remarkably, the model describes the data for all catalysts
with only one parameter, BEH, needing to be changed for each cat-
alyst as shown in Fig. 12. The systematic decrease in BEH from
ꢀ249 kJ/mol for the most active catalyst to ꢀ241 kJ/mol for the
least active catalyst suggests that BEH is a key sensitive parameter
that defines surface behavior during LTS. Fig. 13 shows a system-
atic decrease in Hꢁ coverage as catalyst activity decreases;
HCOOꢁꢁ and COꢁ coverages remain invariant with catalyst activity.
H Binding Energy, kJ/mol
Fig. 12. Model prediction: systematic decrease in H binding energy as activity
decreases for Catalysts 1–4 (Table 7) at 35% CO conversion. Standard reaction
conditions.
ꢀ
ꢁ
ki dY
XRC;i
¼
ð10Þ
Y
dki
K
i;eq;kj–i
where ki is the rate constant for step i, Ki;eq is the equilibrium con-
stant for this step, and Y is the rate of the overall reaction. Thus, we
calculate XRC;i for each step i by computing the effect on the net rate
Y of increasing the forward and reverse rate constants for a given
step i while maintaining Ki;eq and k for all other steps constant.
For all equilibrated steps, XRC;i is 0. Table 8 gives XRC;i at 25% CO con-
version for steps 5, 14, and 16 at our standard conditions, and at the
condition where the CO/CO2 ratio was changed from 1.5 to 0.25. At
the standard conditions, XRC;5 is 0.79 while the other non-equili-
brated steps have much lower values. These values are directionally
the same as the degree of irreversibility given in Table 5. However,
when we change the molar composition of CO and CO2, XRC;5 de-
creases to 0.6 while XRC;14 increases to 0.2, indicating that the ki-
4.3. Sensitivity analysis
Here, we use Campbell’s formalism for the degree of rate con-
trol [30], where the degree of rate control XRC;i for an elementary
step i is defined as
Table 7
Catalyst compositions, Cu dispersions, and site time yields at standard reaction
conditions.
netic importance of step 14 has increased. Interestingly, Table 8
Catalyst Cu
ZnO
Al2O3
Additive
(wt%)
Cu
Y at 35% CO
(sꢀ1
P
(wt%) (wt%) (wt%)
dispersion conversion
(%)
shows that iXRC;i – 1. We explore this further in Fig. 14, where
P
P
)
iXRC;i is plotted versus %CO conversion. iXRC;i ¼ 1 only as conver-
1
2
3
4
46
46
45
45
29
29
24
24
25
25
26
26
–
–
10
7
8
0.16
sion tends to zero.
0.097
0.058
0.047
Recently, Stegelmann et al. [31] evaluated the degree of ther-
modynamic control of surface intermediates, thus examining
how stabilization of surface intermediates affected reaction rates.
Nd2O3 = 5
Y2O3 = 5
10

R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
9
0.030
0.025
0.020
0.015
0.010
0.005
0
states. In our case, altering BEH, BEHCOO, and BEOH may change sur-
face coverages, but the previously determined kinetic importance
of individual reaction steps will remain unchanged, because con-
clusions from Campbell’s degree of rate control are controlled by
locations in energy of transition states, and not binding energies
of reaction intermediates.
5. Discussion
The 8 step closed catalytic cycle given in Table 5 depicts the
mechanism of the low temperature water gas shift reaction on cop-
per. Hydrogen formation occurs by the dissociation of water and
the recombination of surface hydrogen. Surface hydroxyl reacts
with COꢁ to give a cis surface carboxyl intermediate that resides
in equilibrium with its trans form. The latter species reacts with
a surface hydroxyl to give carbon dioxide and water to complete
the cycle. Our approach, to initially use 18 reaction steps, including
formate reactions and reactions associated with the redox mecha-
nism, allows us to avoid bias toward a certain mechanism. For
example, the reaction of surface CO and OH leads to a carboxyl
intermediate rather than a surface formate, since the latter reac-
tion is energetically very demanding. Our model using the carboxyl
mechanism describes all our data over a wide range of conditions
and, more importantly, permits us to explain observations result-
ing from catalyst modifications. Thus, our study corroborates and
amplifies the original work of Gokhale et al. [11].
1
2
3
4
Catalyst
Fig. 13. Model prediction: surface H coverage decreases as activity decreases for
Catalysts 1–4 (Table 7) at 35% CO conversion. Standard reaction conditions.
All adsorption–desorption steps are in quasi-equilibrium. And
although there is no single exergic step, the dissociation of water,
with the ratio d of 0.88, is the furthest from equilibrium (Table 5).
This ratio is a good measure of the impact of a reaction step, in our
case water dissociation, on the reaction rate, and agrees with the
analysis using Campbell’s degree of rate control. Interestingly,
Campbell and Daube [5], while discussing the redox and for-
mate-mediated mechanisms, showed that for either of these mech-
anisms, water dissociation would be the rate-determining step.
The activation energy they developed for water dissociation from
a surface science investigation on Cu (1 1 1) was 113 kJ/mol. The
DFT-calculated value is 131 kJ/mol [11]. And our model, using
the DFT value as an adjustable parameter, gave an activation en-
ergy of 125 kJ/mol, close to the experimental value of Campbell
and Daube. Besides water dissociation, the reversible formation
of the carboxyl intermediate, with d equal to 0.54, is also non-
equilibrated with an activation energy of 49 kJ/mol. Of all DFT-de-
rived activation energies on Cu(1 1 1) [11] that we used, the values
for these two reactions were the only activation energies that the
model needed to adjust. This behavior plus the fact that the values
of BE we used were either the same or very close to those calcu-
lated on Cu(1 1 1) [11] suggests that the Cu particles in Cu:ZnO:A-
l2O3 which are greater than 10 nm, behave like Cu(1 1 1). This
observation is in accord with the experimentally derived conclu-
sion of Campbell and Daube [5].
Interestingly, the endothermic desorption of dihydrogen with a
high activation energy of 98 kJ/mol [11] and the corresponding dis-
sociative adsorption is equilibrated; whereas, the reaction of COꢁ
with OHꢁ with Ef,14 of only 49 kJ/mol to form COOHꢁcis is a non-
equilibrated reversible step with possible kinetic significance. In-
deed, as we change the inlet gas ratio so that CO is decreased from
8% to 2.7% and CO2 is increased from 5.3% to 10.6%, the reaction
rate decreases by ca. factor of 3 (Fig. 10). Concomitantly, d for
the water dissociation step decreases to 0.62 while that for the
COOHꢁ formation step increases to 0.8, indicating that this step is
now more irreversible and, perhaps, more kinetically significant.
We cite this example of change, because it shows that variations
in reaction conditions can subtly alter the active surface which
we observe, via our model, kinetically.
Table 8
Campbell’s degree of rate control for non-equilibrated steps calculated at 25% CO
conversion.
P
Experimental Condition
XRC;5
XRC;14
XRC;16
XRC;i
1
2
0.786
0.612
0.036
0.206
0.0019
0.0006
0.824
0.818
Experimental condition: 2.7 bars, 473 K; Feed 1: 36.7% H2, 16.7% N2, 8% CO, 5.3%
CO2, 33.3% steam; Feed 2: 36.7% H2, 16.7% N2, 2.7% CO, 10.6% CO2, 33.3% steam.
1.00
0.95
0.90
0.85
0.80
0.75
0.70
0.65
0.60
0.55
0.50
0
10
20
30
40
50
60
% CO Conv — Observed
Fig. 14. Total Campbell’s degree of rate control versus % CO conversion.
Such an analysis is conducted by adjusting the free energy of a spe-
cific surface intermediate, for example, by changing its binding en-
ergy, while keeping the free energies of all other surface
intermediates constant, as well as constant energies for transition

10
R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
Under our conditions, increasing CO2 levels by a factor of 2 in
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Model
the feed does not increase HCOOꢁꢁ. Bidentate formate coverage is
increased mainly by increasing pressure and decreasing tempera-
ture. Importantly, formate coverage is reduced significantly in
the presence of alkali promoter. Kinetically, the model indicates a
Catalyst 1
Catalyst 2
Catalyst 3
Catalyst 4
Cu (111)
ꢁꢁ
large reduction in BEHCOO , implying an alkali-assisted destabiliza-
tion of this intermediate.
A microkinetic model not only addresses important mechanistic
questions, but also helps determine how to enhance kinetically sig-
nificant steps of a catalytic cycle. The 4 catalysts in Table 7 are a
case in point. All 4 catalysts show results similar to those given
in Table 5, with Reactions 1–4 and 18 being in quasi-equilibrium,
and with steam dissociation having the same degree of irreversibil-
ity of ca. 0.88 for all the catalysts. The only adjustable parameter
that differentiates the kinetics for the catalysts is BEH; thus steps
containing Hꢁ are necessarily affected.
-235
-240
-245 -250
-255
-260
-265 -270
-275
H Binding Energy, kJ/mol
Reaction 2 is in quasi-equilibrium, and the equilibrium constant
for Reaction 2, K2,eq, increases with an increase in hydrogen binding
energy concomitant with an increase in Hꢁ. However, the surface
coverage of adsorbed H atoms remains low for all cases, and the in-
crease in activity caused by changing BEH must be caused by an-
other effect. In this respect, the value of ki,for for Reaction 5
increases (Fig. 15) as catalyst activity increases; with BEH influenc-
ing K5,eq as shown below
Fig. 16. Model derived volcano plot demonstrating effect of H binding energy on
site time yield. Catalysts 1–4 (Table 7) and Cu(1 1 1) indicated on the plot. 30% CO
conversion. Standard reaction conditions.
does not play as important a role for enhancing catalytic activity.
This analysis demonstrates that facility in steam dissociation is
important for high catalytic activity.
Importantly, the model allows us to ascertain to what extent
site time yields can be increased. Using the model, we obtain site
time yields at varying BEH while also changing total flow rates to
ensure constant 30% CO conversion. This analysis results in a vol-
cano plot (Fig. 16). The figure also includes model results for the
4 catalysts as well as for Cu(1 1 1), all at 30% conversion. As BEH in-
creases to -256 kJ/mol, Y increases, but with further increase in
BEH, Y decreases: an excellent example of the Sabatier Principle.
Catalyst 1 and 2 are prepared identically except for the calcination
temperature, and their results bracket the result with Cu(1 1 1).
This result agrees with the suggestion [32,33] that subtle but
important chemical promotion by ZnO on Cu is possible. Yttrium
oxide and Nb2O3 have a deleterious effect on the LTS rate, indicat-
ing that strong interaction of Cu with such oxides is possible,
although, in this case, a negative one. Noting the work of Nakam-
ura et al. [6], where they found the more open Cu(1 1 0) surface
to be more active than Cu(1 1 1), we speculate that in order to
achieve the maximum rate of Fig. 16, the Cu catalyst surface should
be made to mimic Cu(1 1 0).
ꢀ
ꢁ
ꢀBEH
K5;eq / exp
ð11Þ
RT
But because this step is endothermic, the model was parameterized
in terms of the reverse rate constant and the equilibrium constant:
ꢀ
ꢁ
ꢀBEH
k5;for ¼ k5;re K5;eq / exp
ð12Þ
v
RT
For the other non-equilibrated reaction steps, 14 and 16, the
forward and reverse rate constants are the same for all 4 catalysts,
indicating that formation and reaction of the carboxyl intermediate
10
9
The value of BEH, as noted from Eqs. (3) and (11), perturbs the
activation barrier for steam dissociation, which is an endothermic
step, by changing the enthalpy of formation of surface H. A similar
perturbation of the activation barrier can also occur if we keep BEH
constant and perturb the inherent activation energy of steam dis-
sociation. In this case, as this activation barrier is decreased, the
net rate will increase until its kinetic significance decreases rela-
tive to another step or diffusion limitations are reached.
8
7
6
1
-
,s
3
5
4
3
2
1
0
10
x
6. Concluding Remarks
Eight elementary reactions that constitute a closed catalytic cy-
cle adequately describe the mechanism of the water gas shift reac-
tion on Cu. Adsorbed hydrogen atoms formed from the dissociation
of water desorb as dihydrogen, while OHꢁ reacts with surface CO to
form a carboxyl intermediate which in turn reacts with another
OH⁄ to give CO2. Besides these 8 reactions, two equilibrated reac-
tions, which are not part of the LTS catalytic cycle, produce the ob-
served bidentate formate species. This surface formate is
a
1
2
3
4
spectator species, and although it does not participate directly in
the shift reaction, its high surface coverage, especially at high pres-
sures and low temperatures, can influence the LTS reaction by
blocking surface sites. This mechanistic description concurs with
Catalyst
Fig. 15. Forward rate constant for water dissociation for Catalysts 1–4 (Table 7).
Standard reaction conditions.

R.J. Madon et al. / Journal of Catalysis 281 (2011) 1–11
[2] D.C. Grenoble, M.M. Estadt, D.F. Ollis, J. Catal. 67 (1981) 90.
11
the earlier work of Gokhale et al. [11]. Indeed, the robustness of our
model is due in part to the use of DFT-calculated values for binding
energies, activation energies, and preexponential factors we ob-
tained from Ref. [11]. And although our model differs from previ-
ous models in the literature, several aspects of our work are in
step with previous observations of, for example, Ovesen et al. [4]
regarding formate being a spectator species, and Campbell and
Daube [5] regarding the kinetic importance of the water dissocia-
tion step and the kinetic resemblance of Cu–ZnO–Al2O3 catalysts
to Cu(1 1 1).
The kinetic significance of each reaction step was obtained by
noting its degree of irreversibility (d) and by determining Camp-
bell’s degree of rate control. Both water dissociation and carboxyl
formation have d values approaching 1, and as conditions change,
one or the other step becomes more kinetically significant. Com-
plexity in the LTS surface chemistry arises due to possible changes
in the binding energies of adsorbed components. For the catalysts
and experimental conditions investigated here, we found the val-
ues of BEH, BEOH, and BEHCOO, to be important, with BEH being
the most sensitive. Under other conditions and with other Cu-
based catalysts, binding energies and even activation energies
may change to a greater extent; however, the mechanism and
model we have described should hold.
[3] C.V. Ovesen, P. Stoltze, J.K. Nørskov, C.T. Campbell, J. Catal. 134 (1992) 445.
[4] C.V. Ovesen, B.S. Clausen, B.S. Hammershøi, G. Steffensen, T. Askgaard, I.
Chorkendorff, J.K. Nørskov, P.B. Rasmussen, P. Stoltze, P. Taylor, J. Catal. 158
(1996) 170.
[5] C.T. Campbell, K.A. Daube, J. Catal. 104 (1987) 109.
[6] J. Nakamura, J.M. Campbell, C.T. Campbell, J. Chem. Soc. Faraday Trans. 86
(1990) 2725.
[7] J. Yoshihara, S. Parker, A. Schafer, C.T. Campbell, Catal. Lett. 31 (1995) 313.
[8] N.A. Koryabkina, A.A. Phatak, W.F. Ruettinger, R.J. Farrauto, F.H. Ribeiro, J. Catal.
217 (2003) 233.
[9] P. Liu, J.A. Rodriguez, J. Chem. Phys. 126 (2007) 164705.
[10] J.L.C. Fajin, M.N.D.S. Cordeiro, F. Illas, J.R.B. Gomes, J. Catal. 268 (2009) 131.
[11] A.A. Gokhale, J.A. Dumesic, M. Mavrikakis, J. Am. Chem. Soc. 130 (2008) 1402.
[12] T. van Herwijnen, W. A de Jong, J. Catal. 63 (1980) 83.
[13] T. van Herwijnen, W. A de Jong, J. Catal. 63 (1980) 94.
[14] J.A. Dumesic, D.F. Rudd, L.M. Aparicio, J.E. Rekoske, A.A. Treviño, The
Microkinetics of Heterogeneous Catalysis, American Chemical Society,
Washington, DC, 1993.
[15] F. Schüth, K. Unger, in: G. Ertl, H. Knözinger, J. Weitkamp (Eds.), Handbook of
Heterogeneous Catalysis, vol. 1, 1997, p. 72.
[16] J.-L. Li, T. Inui, Appl. Catal. A 137 (1996) 105.
[17] R.E. Reitmeier, H.W. Fleming, US Patent 3,388,972 (1968) (to Catalysts and
Chemicals Inc.).
[18] Y. Cai, S. Davies, J. Wagner, US Patent 6,627,572 B1 (2003) (to Süd-Chemie
Inc.).
[19] L. Lloyd, D.E. Ridler, M.V. Twigg, in: M.V. Twigg (Ed.), Catalyst Handbook,
second ed., Wolfe Publishing Ltd., England, 1989.
[20] G.C. Chinchen, C.M. Hay, H.D. Vandervell, K.C. Waugh, J. Catal. 103 (1987) 79.
[21] S. Kandoi, J. Greeley, M.A. Sanchez-Castillo, S.T. Evans, A.A. Gokhale, J.A.
Dumesic, M. Mavrikakis, Topics Catal. 37 (2006) 17.
[22] M.A. Sanchez-Castillo, R.J. Madon, J.A. Dumesic, J. Phys. Chem. 109 (2005)
2164.
Acknowledgments
[23] I. Chorkendorff, J.W. Niemantsverdriet, Concepts of Modern Catalysis and
Kinetics, Wiley-VCH GmbH & Co. KGaA, Weinheim, 2003.
[24] M. Caracotsios, Athena Visual Studio, www.athenavisual.com.
[25] R.J. Madon, M. Boudart, I&EC Fundamentals 21 (1982) 438.
[26] M.A. Vannice, Kinetics of Catalytic Reactions, Springer, 2005, p. 77.
[27] M. Boudart, Topics Catal. 14 (2001) 181.
[28] Y. Yang, C.A. Mims, R.S. Disselkamp, C.H.F. Peden, C.T. Campbell, Topics Catal.
52 (2009) 1440.
[29] K. Gründler, M. Irgang, N. Neth, M.J. Sprague, G. Zirker, European Patent 0 205
130 B1 (1992) (to BASF).
We thank the management at BASF Corporation for their sup-
port for carrying out this research. We gratefully acknowledge
funding from the Department of Energy (Basic Energy Sciences)
and BASF Corporation. We thank Dr. Ivan Petrovic, and Ms. Beth
Nartowicz (BASF Corporation) for setting up the N2O chemisorp-
tion method, and Ms. Nartowicz also for carrying out all the chemi-
sorption measurements. And finally, we thank Mr. David Disantis
(BASF Corporation) for his assistance in obtaining kinetic data.
[30] C.T. Campbell, Topics Catal. 1 (1994) 353.
[31] C. Stegelmann, A. Andreasen, C.T. Campbell, J. Am. Chem. Soc. 131 (2009) 8077.
[32] N.-Y. Topsøe, H. Topsøe, J. Mol. Catal. A 141 (1999) 95.
[33] P. L Hansen, J.B. Wagner, S. Helveg, J.R. Rostrup-Nielsen, B.S. Clausen, H.
Topsøe, Science 295 (2002) 2053.
References
[1] T.M. Yurieva, Kinet. Katal. 10 (1969) 862.