You are on page 1of 10

Viewpoint

pubs.acs.org/acscatalysis

The Degree of Rate Control: A Powerful Tool for Catalysis Research


Charles T. Campbell*
Department of Chemistry, University of Washington, Seattle, Washington 98195-1700, United States

ABSTRACT: The “degree of rate control” (DRC) is a mathematical approach for analyzing multistep reaction mechanisms that
has proven very useful in catalysis research. It identifies the “rate-controlling transition states and intermediates” (i.e., those
whose DRCs are large in magnitude). Even in mechanisms with over 30 intermediates and transition states, these are generally
just a few distinct chemical species whose energies, if they could be independently changed, would achieve a faster net reaction
rate to the product of interest. For example, when there is a single “rate-determining step”, the DRC for its transition state (TS)
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

is 1, which means (by definition) that if this TS’s energy could be decreased by kBT (where kB is Boltzmann’s constant and T is
temperature), the net rate would increase by a factor of e. Because the (relative) energies of these key adsorbed intermediates and
transition states can be adjusted by modifying the catalyst or solvent, or even a reactant’s molecular structure, the DRC values
provide important ideas for catalyst improvement. The species with large DRCs are also the ones whose energetics must be most
accurately measured or calculated to achieve an accurate kinetic model for any reaction mechanism. A tutorial on DRC analysis,
Downloaded via 114.30.78.204 on April 14, 2022 at 11:03:53 (UTC).

the calculation of DRCs, and examples of the applications of DRCs in catalysis research is presented here. Applications of DRC
analysis include the following: clarifying reaction kinetics, improving the accuracy of computational models, improving reaction
conditions, improving choice of oxidant in selective oxidation, incorporation in algorithms which calculate net reaction rates of
multistep mechanisms without solving the differential equations involved, and high-throughput computational screening of
catalyst materials. Because DRC values can be determined experimentally, a full microkinetic model is not required to take
advantage of DRC analysis.

1. INTRODUCTION where r is the net reaction rate to the product of interest, and
In analyzing chemical kinetics involving multistep reaction the partial derivative is taken holding constant the rate
constants, kj, for all other steps j ≠ i and the equilibrium
mechanisms, discussion of the rate-determining step (RDS) has
constant, Ki, for step i (and all other steps too, since their
proven to be very powerful. When there is a single RDS and
forward and reverse rate constants are held fixed). Note that
one knows which step it is, one can often derive a simple rate
keeping Ki constant means that the forward and reverse rate
law for the net reaction’s rate involving only a few kinetic and
constants for step i, ki and k−i, both must be varied by equal
thermodynamic parameters, a common practice in textbooks on factors so that their ratio remains constant. The reaction
kinetics. Furthermore, knowing the RDS gives fruitful ideas for conditions (e.g., temperature, concentrations) are also held
modifying the reaction conditions or catalyst to achieve a better constant here.
rate. The degree of rate control (DRC) is a tool for analyzing By this definition above, XRC,i equals the relative increase in
reaction mechanisms and kinetics which serves a very similar the net rate per relative increase in the rate constant for step i
purpose to determining the RDS, but in a much broader and (differentially). The larger the numeric value of XRC,i is for a
more general way. given step, the bigger is the influence of its rate constant on the
I conceived of and introduced the degree of rate control overall reaction rate r. A positive value indicates that increasing
initially in a paper in 1993, in response to a question by a ki will increase the net rate r; such steps are termed rate-limiting
referee of an invited review paper I had submitted to the steps (RLS). A negative value indicates the opposite; such steps
proceedings of a symposium held in honor of Haldor Topsoe’s are termed inhibition steps. It turns out that for typical
80th birthday.1 I remain eternally grateful to that referee, who textbook cases of reaction mechanisms where there is a single
wanted to know how I had determined which step was the RDS, XRC,i = 1 for the RDS and zero for all other steps.2
RDS, since I had mentioned the RDS for several catalytic To my knowledge, no other method had been developed
reactions in that paper. In response, I added a paragraph in which so unambiguously defines the rate-determining step.
which I tried to articulate what had been going on in my mind However, it is much more powerful than just doing that, as
for a number of years when analyzing net reaction kinetics in shown below. Without my knowledge until recently, a quantity
the context of some reasonably good knowledge of the with the same definition as XRC,i had already been introduced
elementary-step kinetics involved. To explain this, I introduced by K. J. Laidler in the Journal of Chemical Education in 19883
the “degree of rate control for elementary step i,” XRC,i, which I (which he attributed there to a personal communication from
defined as1 W. J. Ray), where he called it instead the “control function” (or
“control factor”) for step i, “CFi.” To my knowledge, neither
ki ⎛ ∂r ⎞ ⎛ ∂ln r ⎞
XRC, i = ⎜ ⎟ =⎜ ⎟
r ⎝ ∂ki ⎠ ⎝ ∂ln ki ⎠k Received: January 12, 2017
kj≠i , Ki j≠i , Ki (1) Published: March 3, 2017

© 2017 American Chemical Society 2770 DOI: 10.1021/acscatal.7b00115


ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

concept was used to a significant extent, probably due to the


fact that its real applications awaited modern computational
power for theoretical rate predictions. While eq 1 looks much
like equations used in standard differential sensitivity analyses
by engineers to ascertain the sensitivity of outputs of a model to
one input parameter, it also represents quite an advance
compared to that, since the entities that are held constant in
this partial derivative ensure thermodynamic and kinetic
consistency. As we show next, these are held constant because
XRC,i is a conceptualization of the effect of changing the
potential energy surface upon which the reaction proceeds, so
that DRC has a deeper physical meaning than any such usual
“sensitivity” for a parameter in a model.
Within transition state theory, the rate constant ki is given
by4
ki = (kBT /h)exp( −ΔGi 0,TS/RT ) (2) Figure 1. Schematic standard-state free-energy diagram for the
reaction AC(g) + B(g) → AB(g) + C(g), showing an incremental
where h is Planck’s constant, ΔGi0,TS
is the difference in
decrease in the standard-state free energy of one transition state TS3,
standard-state molar free energy between the transition state as implied by the definition of the degree of rate control for that step
and the reactants for step i, R is the universal gas constant, and (XRC,TS3) in eq 5. Also shown is the change implied by the definition
T is the absolute temperature. (Sometimes a transmission of the degree of rate control for one reaction intermediate, AB*
coefficient κ is added in front, and for second-order reactions (XRC,AB*), as defined in eq 6, ref 2. Note that such reaction energy
this must be divided by the standard-state concentration.) diagrams require element balance at each step, so that the same
Taking the logarithm of both sides gives number of atoms of each element must be present at each energy
minimum and maximum as in the reactants and products. Thus, some
ln ki = ln(kBT /h) − (ΔGi 0,TS/RT ) (3) species are necessarily carried along as spectators in individual steps, as
is C* in the step across TS3.
Taking the derivative of both sides (holding temperature
constant) gives
already pointed this out in a 2001 paper where I showed the
d(ln ki) = d(ln(kBT /h)) − d(ΔGi 0,TS/RT ) advantages of DRC analysis compared to analysis of DeDonder
Relations for identifying the RDS.5 There, I wrote that the
= −d(ΔGi 0,TS/RT ) (4)
mathematical operation associated with finding the DRC for a
Thus, eq 1, which defines XRC,i, can be rewritten in terms of step “is analogous to finding an additive which somehow
the standard-state free energy of activation for step i (ΔGi0,TS), magically stabilizes the transition state to that one elementary
and the standard-state free energy for its transition state, Gi0,TS, step without affecting the energies of any other adsorbed
as species.”
When the DRC is defined in this way as the change in net
ki ⎛ ∂r ⎞ ⎛ ∂ln r ⎞ rate due to the decrease in energy of one of the maxima in the
XRC, i = ⎜ ⎟ =⎜ ⎟ reaction energy diagram, it suggests that one can extend this
r ⎝ ∂ki ⎠ ⎝ ∂ln ki ⎠k
kj≠i , Ki j≠i , Ki concept to the minima in reaction energy diagrams too. Two
⎛ ⎞ ⎛ ⎞ engineers from Denmark, Drs. Carsten Stegelmann and Anders
⎜ ∂ln r ⎟ ⎜ ∂ln r ⎟ Andreasen, realized this and asked for my help in developing a
=⎜ ⎟ =⎜ ⎟ degree of rate control that applies instead to intermediates
⎜ ∂ −ΔGi ⎟ ⎜ ∂ −Gi ⎟
0,TS 0,TS

⎝ (
RT ) ⎠k
j≠i , Ki
⎝ ( )
RT ⎠G0,TS , G 0
j≠i m (5)
rather than transition states. We thus developed an equivalent
definition for the degree of rate control of any intermediate, n,
in a reaction mechanism, XTRC,n, as:2
The partial derivative in the final equality here is taken
holding the Gibbs free energy of all other transition states j ≠ i ⎛ ⎞ ⎛ ⎞
and all intermediates m constant, which is the same as holding 1 ⎜ ∂r ⎟ ⎜ ∂ln r ⎟
the rate constants, kj, of all other steps j ≠ i and the equilibrium X TRC, n = ⎜ ⎟ =⎜ ⎟
r ⎜ ∂ −Gn0 ⎟
0
⎜ ∂ −Gn ⎟
constant, Kj, of all steps j constant. This is best understood by ⎝ ( )
RT ⎠G 0
m ≠ n , Gi
0,TS ⎝ ( )
RT ⎠G 0
m ≠ n , Gi
0,TS
inspection of the standard-state free-energy diagram for a
typical reaction shown in Figure 1, recognizing that this (6)
requires holding the equilibrium constant for all steps i and j where the partial derivative is taken holding constant the
constant, and that the equilibrium constant for any step n, Kn, standard-state free energy of all other intermediates m ≠ n
equals exp(−ΔGn0,rxn/RT), where ΔGn0,rxn is the difference in (Gm0), all reactants, all products, and all transition states i
standard-state molar free energy between the products and the (Gi0,TS). Since the free energy of intermediates is inherently a
reactants for step n. This final equality in eq 5 shows that XRC,i thermodynamic quantity, whereas that for a transition state is a
can be equivalently defined as the relative increase in the net kinetic quantity, we initially referred to this is as the degree of
rate to the product of interest due to a (differential) decrease in thermodynamic rate control of intermediate n, hence the added
the standard-state free energy for the transition state for step i “T” in the subscript XTRC,n. By definition, it is the relative
(divided by RT), while holding constant the standard-state free increase in the net rate to the product of interest per
energy for all other transition states and all intermediates. I had (differential) decrease in the standard-state free energy for
2771 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

intermediate n (divided by RT), while holding constant the where U is the molar internal energy, P is the standard pressure
standard-state free energy for all other intermediates and all (1 bar), and V is molar volume. If one modifies the catalyst or
transition states. We did not realize at that time that a very solvent in some way, it is usually the case that the molar
similar extension of my degree of rate control concept had entropy and molar volume of all species i change so little that
already been made by Kozuch and Shaik in 2006,6 who TΔΔSi0 and P(ΔΔVi0) are negligible compared to ΔΔUi0, in
introduced the “degree of turnover frequency control of each which case the change in Gi0 due to this modification is
intermediate” specifically defined only for analyzing multistep essentially given by ΔGi0 = ΔUi0, where Δ now refers to
catalytic mechanisms, so that the intermediates they consider differences induced by the modification. Because calculations
are always bound to a catalyst (see also ref 7). Kozuch and (like density functional theory, DFT) and experiments allow us
Shaik also had presented many useful applications of this to estimate the effect of such modifications on ΔUi0, eq 7
concept, as well as ideas that evolve from it.8−10 Our XTRC,n provides a route to quantitatively estimate the effects of such
concept is more general than this, in that it is not intended to modifications on the net rate and selectivity, if one knows DRCi
be limited to catalytic reactions and is generally appropriate for before modification (see below).
any multistep reaction mechanism with any type of The degree of rate control for a transition state (or
intermediates, whether bound to a catalyst or not. elementary step) answers the question: Suppose one could
Because of the extreme similarity between the definitions change the catalyst with a modifier that differentially lowered
above for XRC,i and XTRC,n, we now most commonly express the energy (i.e., standard-state free energy) of the transition
these same concepts in terms of a more general definition for state for that one elementary step, without changing any other
the degree of rate control that applies to any species i, whether such energy. By what relative amount would this increase the
it be a transition state or intermediate. We define this net rate, per unit change in energy (in units of RT)? Similarly,
generalized degree of rate control for species i (DRCi or the degree of rate control of an intermediate answers the
simply Xi) as closely related question: Suppose one could introduce a catalyst
or catalyst modifier that differentially lowered the energy (i.e.,
⎛ ⎞ ⎛ ⎞ standard-state free energy) of one intermediate, without
1 ⎜ ∂r ⎟ ⎜ ∂ln r ⎟
changing any other such energy. By what relative amount
DRCi = Xi = ⎜ ⎟ =⎜ ⎟
r ⎜ ∂ −Gi0 ⎟ ⎜ ∂ −Gi
0
⎟ would this increase the net rate?
⎝ RT ( ) ⎠G0
j≠i
⎝ RT( ) ⎠G0
j≠i
2. AN EXAMPLE OF DRC ANALYSIS OF A SIMPLE
⎛ ⎞ MECHANISM WITH AN ANALYTICAL RATE
⎜ −∂ln r ⎟ EXPRESSION
=⎜ 0

⎜ ∂ Gi ⎟
⎝ RT ( ) ⎠G0
j≠i (7)
To demonstrate the use of eq 7, consider a simple Langmuir−
Hinshelwood mechanism with one rate-determining step
(RDS) and all other steps quasi-equilibrated:
where the partial derivative is now taken holding constant the
standard-state free energy of all other species (intermediates, A + * ⇌ A* (fast to equilibrium) (reaction 1)
transition states, reactants, and products), j. Its value describes
B + * ⇌ B* (fast to equilibrium) (reaction 2)
the relative increase in net rate due to the (differential)
stabilization of the standard-state free energy for species i,
A* + B* ⇌ AB* + * (RDS) (reaction 3)
holding all the other species’ energies constant. Thus, eq 7
probes the importance of one species’ free energy in the full AB* ⇌ AB + * (fast to equilibrium) (reaction 4)
standard-state free-energy surface for the full reaction. It gives
the same value for the DRCi (or Xi) as XRC,i in eq 5 and XTRC,i where * denotes a free surface site, X is a gas phase species, and
in eq 6. Although the final expression in eq 7 is mathematically X* is a surface-bound intermediate. By applying the quasi-
equivalent to the middle expression, the final expression is equilibrium approximation under conditions where step 3 is
recommended for practical implementation. This is because it is irreversible and the surface is almost saturated with A*, we have
impossible to stabilize the transition state for elementary steps shown that this gives a simple rate law for the rate to make AB2:
that have no real standard-state free-energy barrier between K p
reactants and products (as is common in molecular adsorption r = k3 2 B
and desorption, for example) without also stabilizing either the K1 pA (8)
reactant or the product. That is, it is physically impossible to
perform the mathematical operation as written in the middle where pX is the partial pressure of X. Using the above
expression. In such cases, one must instead destabilize the expressions for the rate constant for step i (ki = (kBT/h)
transition state (i.e., introduce a small free-energy barrier) and exp(−ΔGi0,TS/RT)), and the equilibrium constants for step n
change the sign on the resulting change in relative rate, as done (Kn), equals exp(−ΔGn0,rxn/RT)), and remembering that the
in the final expression here. gaseous reactants are taken as the reference of zero free energy,
The beauty of eq 7 is thus that it directly connects the net this can be written as
rate with individual energies that one can hope to control to
advantage. If we take the reactants as the reference state of zero
standard-state free energy and enthalpy, each standard-state
(molar) free energy Gi0 breaks up into a sum of an enthalpy
contribution (ΔHi0) and an entropy contribution (−TΔSi0) as
Gi0 = ΔGi0 = ΔHi0 − TΔSi0, where Δ refers to differences Performing the partial derivative in eq 7 using this rate
relative to the reactants. Note that ΔHi0 = ΔUi0 + P(ΔVi0), expression to calculate the degrees of rate control gives that
2772 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

DRCTS3 = 1, DRCA* = −2 and DRCi = 0 for all other transition vast majority of steps. For example, Agarwal, Sanchez-Castillo,
states and adsorbates. The RDS has a DRC of +1 as expected. Madon, Dumesic et al.13 showed that in 2-methylhexane
The DRC of A* is −2, which is minus 1 times the number of conversion over USY zeolite for ∼30% conversion at 773 K,
times GA*0 enters the rate eq (eqs 8 and 9) via rate constants only 5 of the 36 steps in the simplified mechanism have DRCs
and equilibrium constants (= 2 here) times the fractional that exceed 0.06, with one step most dominant (DRC near 0.5).
coverage of sites by A* (= ∼1.0 here). In general, we f ind that They also analyzed a 371-step model for conversion of
the DRC for adsorbed intermediates is some small negative integer isobutane over USY zeolite, and they found only 20 steps
times the f ractional coverage of sites by that adsorbate, so that it is that had DRCs that exceeded 0.1 even when their individual-
essentially zero for adsorbates at low coverage and a small negative product DRCs were summed over all the seven different
integer for adsorbates that saturate the surface.2,11 It is not yet products.14
clear how to predict the value of that integer when there is not The degree of rate control of transition states (for the rate to
a single RDS. produce any one product or to consume any one reactant)
This analysis reveals that the key to catalyst improvement seems to be conserved when summed over all steps i (i.e., all
and design is to stabilize key transition states (those with large transition states) in the mechanism:
positive DRCs) without stabilizing key intermediates (those
with large negative DRCs) too much, or to destabilize those ∑ XRC,i= ∑ DRCi = 1
key intermediates without destabilizing related key transition steps TSs (10)
states too much. This is similar to the well-known Sabatier
Principle, as has been shown beautifully by Dumesic.12 Baranski has proven this rule for parallel (series) reaction
However, DRCs are more quantitative in their applicability, mechanisms,15 and Dumesic has proven this rule more
and they are more general because they can include multiple generally for any reaction scheme that leads to a single net
species of each type. One could also improve the catalysts by reaction.12 This has also been found in a wide range of studies
destabilizing the transition states of inhibition steps (those with where microkinetic models were analyzed for their DRCs (for
large negative DRCs). example, see refs 2,11,13−24), including cases where branching
While we have calculated the DRCs above using rigorous leads to more than one net reaction and selectivity becomes an
math to take the partial derivatives of an analytical rate issue.
expression, in practice the DRCs are usually determined instead In collaboration with Per Stoltze’s group at Aalborg
numerically, by using a computer and finite-difference methods University and N.C. Schiødt at Haldor Topsøe, we performed
to take the numerical derivative of a rate that is typically a study of the selective oxidation of ethylene with O2 to
determined from a numerical solution to the systems of ethylene epoxide over metallic Ag catalysts which provides an
differential equations defined by the microkinetic model and all excellent example of the power of DRC analysis for such
of its parameters. branching reactions.23 In this case, the alternate reaction
It is important to recognize that DRCs can also be estimated produces the thermodynamically favored product, CO2 + H2O.
experimentally. This is easiest to see for the DRCs for We used a modified version of the mechanism first put forth by
adsorbates, which are just proportional to their fractional Linic and Barteau,16 who identified the key intermediate as an
coverages, which can be measured by a wide variety of oxametallacycle (*CH2CH2O*) based on DFT calculations. In
experimental methods. However, it is also common to our model, the kinetic parameters for the 17 elementary steps
determine the rate-determining step (RDS) for a reaction in forward and reverse were determined experimentally, some
from a variety of experimental measurements (e.g., reaction by fitting this microkinetic model to the net reaction rate data.
orders in reactant concentrations, comparing the activation The degrees of rate control are shown in Figure 2 versus
energy of the net reaction to that for isolated elementary steps, temperature. This common intermediate is produced via steps
etc.). The transition state for the RDS has a DRC of 1, leaving 2 and 5 (that are rate-controlling for both ethylene
the DRCs for the other steps near zero (unless the mechanism consumption and ethylene epoxide production) before the
has branching). There are many situations, however, where branching steps 7 and 8 that determine selectivity. Step 7 is
there is not a single RDS, most commonly when two steps have oxametallacycle ring closure, leading to the desired epoxide,
large DRCs that sum to nearly 1. This is more challenging whereas step 8 is H migration on theh oxametallacycle, which
experimentally but is sometimes seen by increasing or leads to complete combustion products. Note that the DRCs
decreasing a reactant concentration or temperature, which for these branching steps mirror each other exactly, with
can tune the system between conditions where one of these opposite signs. The negative DRC for step 8 indicates that it
steps is the RDS, and then the other becomes the RDS, with a inhibits epoxide production (by consuming the oxametallacycle
mixture of two strongly rate-controlling steps in between (and to make CO2 before it can make the epoxide). This occurs
often a rate maximum). whenever the selectivity is determined in a branching
competition that occurs after the steps that control the rate
3. PROPERTIES OF THE DEGREE OF RATE CONTROL of reactant consumption. That this occurs here after the steps
(DRCI) that control the rate of ethylene consumption is consistent with
We have shown that for previously analyzed mechanisms at our earlier experiments on Ag(110), which showed that the
reaction conditions where there is a single rate-determining activation energies and reaction orders in both reactants for the
step (e.g., in numerous textbook examples where an analytical rates to both products varied with the coverage of chlorine
rate expression was derived for a mechanism), the degree of adatoms in very similar ways, over the same range of chlorine
rate control is equal to one for the transition state for that step, coverage where selectivity improved dramatically up to
and zero for all other steps,2 as also shown for one example 80%.25,26 In more complex mechanisms, the DRCs for such
above. There are many studies of multistep mechanisms which branching steps do not always mirror each other with opposite
have shown that the degree of rate control is nearly zero for the signs as in this simple example.
2773 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

sum to a number exceeding 1 here only because CO* and C*


occupy different types of sites in this microkinetic model.

4. THE DEGREE OF SELECTIVITY CONTROL (XSC,i)


We previously extended the definition of the degree of rate
control in eq 1 to similarly define the degree of selectivity
control for step i, whereby the net rate r was just replaced
throughout eq 1 with the selectivity to the desired product P
from the most valuable reactant R, S = rP/rR, where rP is the
rate of production of P and rR is the rate of consumption of R
(scaled by the stoichiometric ratio of P:R).27 We now extend
the definition of the more generalized degree of rate control for
species i in eq 7 to define the degree of selectivity control for
species i (DSCi) as follows:

Figure 2. Degrees of rate control for the rate of producing ethylene ⎛ ⎞ ⎛ ⎞ ⎛ ⎞


epoxide for the rate-controlling elementary steps (transition states) 1 ⎜⎜ ∂S ⎟⎟ ⎜ ∂ln S ⎟
⎜ ⎟
⎜ −∂ln S ⎟
⎜ ⎟
DSCi = = =
versus temperature at industrial reaction conditions (pO2 = pC2H4 = 100 S ⎜⎜ ⎜⎛ −Gi0 ⎟⎞ ⎟⎟ ⎜⎜ ⎜⎛ −Gi0 ⎟⎞ ⎟⎟ ⎜⎜ ⎜⎛ Gi0 ⎟⎞ ⎟⎟
∂ RT ∂ RT ∂ RT
⎝ ⎝ ⎠⎠ 0 ⎝ ⎝ ⎠⎠ 0 ⎝ ⎝ ⎠ ⎠G0
kPa). Values for the 12 other elementary steps are very close to zero, G j≠i G j≠i j≠i
so they are not shown. The dashed curves for branching steps 7 and 8 ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
would be at zero too if the degree of rate control for the rate of ⎜ −∂ln(r /r ) ⎟ ⎜ −∂ln r ⎟ ⎜ −∂ln r ⎟
consumption of ethylene were plotted here instead. Step 13 is the RDS =⎜ P R ⎟
=⎜ P⎟
−⎜ R⎟
⎜⎜ ⎛ G0 ⎞ ⎟ ⎜⎜ ⎜⎛ Gi0 ⎟⎞ ⎟⎟ ⎜⎜ ⎜⎛ Gi0 ⎟⎞ ⎟⎟
for a separate parallel pathway for ethylene combustion that makes a ∂⎜ RTi ⎟ ⎟ ∂ RT ∂ RT
⎝ ⎝ ⎠ ⎠ ⎝ ⎝ ⎠ ⎠G0 ⎝ ⎝ ⎠ ⎠G0
very minor contribution to the rate of combustion and whose nature is Gj0≠ i j≠i j≠i
poorly understood. Modified from ref 23. = DRCi ,P − DRCi ,R (11)

where DRCi,P and DRCi,R are the degees of rate control of


species i for the rates of making P and consuming R,
As noted above, the DRC for adsorbed intermediates is
respectively. The value of DSCi (= DRCi,P − DRCi,R) describes
generally a negative integer times the fractional coverage of sites
the relative increase in net selectivity to P from R due to the
by that adsorbate, so that it is essentially zero for adsorbates at
(differential) stabilization of the standard-state free energy for
low coverage and a large negative integer for adsorbates that
species i (either one transition state or one stable adsorbed
saturate the surface,2,11 unless there is a branching competition.
intermediate) holding all other species’ energies constant.
Table 1 shows an example set of DRC values for steady-state
methane steam reforming over Rh(211) computed using a
microkinetic model based on DFT energies.11 Note that the 5. USING THE DEGREES OF RATE CONTROL TO
DRCs are nearly zero for all but four species. The DRCs for the IMPROVE MICROKINETIC MODELS
TSs sum up to 1.0 as expected. The DRCs for adsorbates are Thanks to successes of density functional theory with periodic
equal to their factional coverages times − 1. These coverages boundary conditions (DFT) in calculating relatively accurate
reaction energy diagrams for catalytic reactions on surfaces,
Table 1. Degrees of Rate Control for All Intermediates and microkinetic models for reaction mechanisms involving many
Transition States Computed for Methane Steam Reforming elementary steps have become very common in the past 20
over Rh(211)a years.2,11,13−24,28−43 Because these highly efficient computa-
tional codes for calculating the energies of adsorbed
species Xi species Xi
intermediates and transition states still have some problems
intermediates transition states with energy accuracy,44 one would prefer to use more accurate
CO*s −0.76 C−O* 0.87 (and computationally more expensive and time-demanding)
C*f −0.54 H−CH3* 0.11 quantum mechanical methods when possible. Greeley’s group
CH*f −0.01 H−OH* 0.01 has shown that DRCs provide a highly efficient way to
H*h −0.01 CO-H* 0.00 approach such improvements in energy accuracy. By first
O*s 0.00 H−C* 0.00 calculating the DRCs from a microkinetic model based on a
OH*s 0.00 H−CH* 0.00 lower-accuracy but much faster version of DFT, they identified
CH3*t 0.00 H−CH2* 0.00 the few adsorbates and adsorbed transition states with large
CH2*t 0.00 H−CO* 0.00 DRCs. Since these are the only species whose energies affect
HCO*s 0.00 C−OH* 0.00 the net rate, they then recalculated their energies with a more
COH*s 0.00 O−CH* 0.00 accurate but much more computationally demanding method,
H−H* 0.00 HSE06.45,46 This clever approach has great potential for
O−H* 0.00 markedly improving rate predictions based on DFT with only
moderate increases in computational costs.
a
Reaction conditions are T = 773 K, PCO = 0.050 bar, PH2 = 0.15 bar, The degree of rate control is similarly a powerful aid to the
PCH4 = PH2O = 0.40 bar. Superscripts on adsorbates correspond to development of more accurate microkinetic models based on
different types of sites. Dashes in transition states correspond to the experimental measurements of elementary-step energetics,
bonds that are being broken/formed in the adsorbed species. From ref since its values identify which intermediate and transition
11. state energies are most important to measure with high
2774 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

accuracy. It offers many opportunities for theory and SUNCAT-Center/catmap), and it includes a tool for DRC
experiment to synergistically interact. analysis of microkinetic models. The code is used quite
extensively in academia and industry. It is now being modified
6. A NOTE REGARDING STANDARD STATE to also include methods for calculating standard-state entropies
CONCENTRATIONS AND ENTROPIES of adsorbates (and adsorbed transition states) using the
hindered translator approximation mentioned above.
Note that the definition of DRC in eq 7 refers to “standard
state” free energies. Indeed, knowing the free energies (or 8. A NOTE REGARDING BEP RELATIONS AND OTHER
energies and entropies) of the standard states for all the SCALING RELATIONS
adsorbates and transition states is really required for micro-
kinetic modeling, whether or not DRC analysis is involved. One would like to alter a catalyst or the solvent with the aim of
Here we use the usual definition of standard state in changing the energy of a specific transition state or intermediate
thermodynamics, so that for each species, a specific reference to achieve a higher rate or selectivity, chosen based on
condition must be defined, which includes in its choice a knowledge of the degrees of rate control. However, one should
specific standard-state concentration value (which affects the recognize that stabilizing a transition state will generally also
entropy). For example, for ideal gases in mixtures, that is stabilize the reactant and/or product of that step, and vice
typically taken to be the pure gas at one bar pressure and the versa. These energies (or free energies) are connected through
same temperature. The standard state of a solute in a liquid the Brønsted-Evans−Polanyi (BEP) relations24 (or linear free-
solution is often taken to be unit mole fraction if the solution energy relations49,50), which state that the change in the
resembles an ideal solution, but for other solutions where the activation (free) energy for an elementary step (in this case, due
solute−solvent interactions are quite different than solute− to changing the material) are proportional to the change in its
solute interactions, it is better to use some ideally dilute net reaction (free) energy, with a slope between zero and unity.
solution reference state, a hypothetical state where the solute’s Depending on the values of this slope and the values of the
concentration is either unit mole fraction or one molar DRCs, one can find cases where real gains can be achieved in
concentration, but where its energy is the same as if it were the net reaction rate or selectivity, but BEP relations generally
ideally dilute (surrounded by solvent only). For adsorbates, the make it more difficult. For example, whenever there is a single
situation can be rather complicated. They are often treated with rate-determining transition state and a single rate-controlling
the 2D ideal gas (or free translator) model or the 2D ideal adsorbate, changing the catalyst to make it less noble will
lattice gas (or localized oscillator) model. These represent two stabilize this transition (increasing the rate) while simulta-
limiting behaviors of the real situation for adsorbates on solid neously stabilizing this adsorbate (decreasing the rate). This
surfaces, which are more accurately described as hindered trade-off suggests an optimum catalyst that is neither too
translators since there are periodic energy barriers for adsorbate aggressive nor too noble, just as expected based on the Sabatier
translation parallel to the surface.4,47 We recently described a Principle.24
simple extension of the ideal 2D gas model to a more realistic Bligaard, Norskøv, and co-workers have discovered BEP and
ideal hindered translator model, which enables easy but much many other linear scaling relations among the energies of
more accurate estimations of the partition function and entropy adsorbed intermediates and transition states as estimated with
of adsorbates.47 Depending on the mechanism, it is sometimes DFT calculations, which even enabled them to estimate net
necessary to intermix these different adsorbate models (ideal catalytic reaction rates on different materials using only two
2D gas, ideal 2D lattice gas and ideal hindered translator) descriptors for each material (typically its adsorption energies
within the same equilibrium constant or rate constant to C, O, N, or H atoms).24,28,29,51−56 This was possible due to
expression. This requires great care to ensure that consistent the fact that the energy of every adsorbate and transition state
choices of standard-state concentrations are used throughout. in the mechanism could be estimated as a linear combination of
We recently derived expressions for equilibrium constants and these two descriptors.
rate constants within transition state theory (TST) using both When such linear scaling relations are used in microkinetic
partition functions and standard-state entropies, in a self- models, it means that there are fewer independent variables in
consistent way that can treat all three models simultaneously.4 rate predictions for different materials than the energies of each
That formalism enables the mixing of these models, using adsorbate and transition state on each material, since many of
activities instead of concentrations to do so. We showed there these are coupled by their scaling relations. Nørskov, Bligaard,
that a standard state for 2D (and 1D) ideal gases defined such and Kleis 55 therefore introduced an extension of the
that their translational entropy is 2/3 (or 1/3) that for the generalized DRC concept to define the “degree of catalyst
corresponding ideal 3D gas offers intuitive advantages for control for descriptor i,” XCC,i, in terms of the materials’ mutually
estimating equilibrium and rate constants for adsorbates. This independent underlying descriptors, where the definition of
standard-state concentration of the ideal 2D gas is approx- XCC,i has the exact same form as eqs 5, 6, and 7, except now the
imately the 2/3 power of the standard state concentration of Gi0 in the denominators in these equations is the value of
the corresponding 3D ideal gas (i.e., the concentration at the descriptor i (energy of descriptor i).
standard pressure of 1 bar at that temperature).4 This concept of the “degree of catalyst control” makes it
quickly obvious that when materials strictly follow such linear
scaling relations, there must be serious constraints on our
7. WEB-ACCESSIBLE, OPEN-SOURCE CODE FOR DRC ability to find better catalysts. Bligaard et al.56 pointed out that
ANALYSIS AND ESTIMATION OF ENTROPY “the mere existence of linear scaling relationships within a
Medford, Shi, Hoffmann, Lausche, Fitzgibbon, and Bligaard in certain class of materials severely limits our ability to find better
Jens Nørskov’s group have developed a powerful, open-source catalysts within that class. The activity of even the best material
catalytic (and electro-catalytic) kinetics code called “Cat- is inevitably constrained, as dictated by the scaling relation-
MAP”.48 It is accessible for free online (https://github.com/ ships.” This was recognized previously by Nørskov’s group,
2775 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

who further pointed out the advantages of finding new classes 2O2 + 3H 2O + 7n−C4 H10 → 7C4 H 9OH + 3H 2
of materials that break the limitations of linear scaling for
conventional classes of materials, and even offered suggestions Thus, ∼43% of the oxygen is supplied by water, and only
on how one might “break the scaling relations.”57−59 57% by O2, clearly demonstrating that one can sometimes
Phillip Christopher’s group has included BEP and other achieve high selectivity in oxidation reactions by using a mixture
linear scaling relationships in DRC analysis, and they developed of H2O with O2 as the oxidant. This idea may be applicable in
an approach to identify the relevant rate- or selectivity- many selective oxidations. Indeed, a recent paper reviews other
controlling steps that are tunable within the conf ines of such studies which indicate that the catalytic activity and selectivity
of selective oxidation reactions may be tuned by judiciously
linear relations between kinetic parameters.60 They also extended
controlling the water content during the reaction.63 These
that approach to the degree of selectivity control.61
results demonstrate that the careful analysis of DRCs is a
powerful way to optimize catalytic process conditions and even
9. APPLICATIONS OF DRC ANALYSIS to improve upon the choices of reactants.
Analyses of DRCs have been used on many occasions to gain
insights into reaction mechanisms and ways to improve 10. DUMESIC’S METHOD USING DRCS WITHIN HIS
catalysts (including electrocatalysts) or catalytic processes. Its MAXIMUM RATE ANALYSIS TO BYPASS SOLVING
power in such applications has been discussed in some detail by DIFFERENTIAL EQUATIONS TO CALCULATE
Nørskov, Bligaard, and Kleis.55 DRC analysis is even included RATES IN MICROKINETIC MODELING
in some undergraduate Chemical Engineering classes, as DRCs have also found use in analytical solutions to kinetic
pioneered by James Dumesic at the University of Wisconsin. equations. Motagamwala and Dumesic64 have shown that the
In 2011, Kozuch and Shaik8 developed an “Energy Span defining equations for DRCs can be combined with simple
Model” based on DRCs in a paper which has already been cited equations for the maximum rate of each step to provide
over 250 times, mainly in the context of homogeneous catalysis. excellent estimates of the net catalytic reaction rate, thus
It neglects entropic differences between species, which is easier bypassing the need to solve the many coupled differential
to do when all the species are in liquid solution and the choice equations that define the full microkinetic model. They showed
of standard-state concentration can be the same. Care should that the steady-state kinetics of a chemical reaction can be
be taken when applying it to adsorbed species where very analyzed analytically in terms of proposed reaction schemes
different statistical mechanical models are often used to composed of a series of steps by calculating the “maximum
describe different classes of adsorbates (see above), and these rates” of the constituent steps, rmax,i (the rate assuming that the
can have large standard-state entropy differences that arise only TS for step i is equilibrated with the gas-phase reactants). They
from differences in choices of standard-state concentrations. derived analytical expressions in terms of rmax,i to calculate
The analysis of DRCs has also been used to improve reaction DRCs for each step to determine the extent to which each step
conditions to get better yields and selectivity over the same controls the rate of the net reaction. The values of rmax,i can be
catalyst material. For example, together with Rachel Getman’s used to estimate the rate of the net reaction, thus making it
group, we used DRC analysis to discover reaction conditions possible to estimate the net reaction rate without solving the
that dramatically improve the regio-selectivity to 1-butanol in a differential equations as usual. This approach gives the exact
computational model for n-butane oxidation over model MOF- rate for reaction mechanisms where the stoichiometric
encapsulated Ag3Pd catalysts.62 Metal nanoparticles encapsu- coefficients of the constituent steps are equal to unity, and
lated within metal organic frameworks (MOFs) offer steric the most abundant adsorbed species are in quasi-equilibrium
restrictions near the catalytic metal that can improve selectivity, with the gas phase and can be used as an approximation with
much like in enzymes. We first developed a microkinetic model good accuracy for many other types of reaction mechanisms.
for the oxidation of n-butane to 1-butanol with O2 over a model This allows one to identify the best mechanism among several
for MOF-encapsulated Ag3Pd nanoparticles. The model possible mechanisms, which can then be followed by a more
consisted of a Ag3Pd(111) surface decorated with a 2-atom- detailed analysis of the full microkinetic model of the best
thick ring of (immobile) helium atoms that creates an artificial mechanism to determine the surface coverages and DRCs of
pore of similar size to that in common MOFs, which sterically adsorbed species and the DRCs of the elementary steps.
constrains the adsorbed reaction intermediates. The kinetic
parameters for the elementary steps were calculated on the 11. A METHOD FOR FAST COMPUTATIONAL
basis of energies from DFT. The microkinetic model was SCREENING OF CATALYSTS BASED ON DRCS
analyzed at 423 K in a mixture of butane and O2 to determine Because of the rapid evolution of computational methods that
the dominant pathways and which species (adsorbed predict reaction rates for different catalyst materials with ever-
intermediates and transition states in the reaction mechanism) increasing speed and accuracy,28 it is likely that the computa-
have energies that most sensitively affect the reaction rates to tional screening of materials for catalytic activity and/or
the different products, using DRC analysis. This analysis selectivity will play an ever increasing role in catalysis research.
provided evidence which suggested to (1) add water to produce There have been many successes already in high-throughput
more OH*, thus inhibiting certain undesired steps which computational screening to accelerate the discovery of new
produce OH*, and (2) eliminate most of the O2 pressure to catalyst materials.33,34,37,52,65−67 The most successful and
reduce the O* coverage, thus also inhibiting these steps. When commonly used method for computational screening of solid
combined with an increase in butane pressure, these changes in catalysts has been the descriptor-based, linear-scaling approach
H2O and O2 pressure led to dramatic improvements in both the developed by Nørskov, Bligaard, and co-workers,24,28,29,51−54
1-butanol selectivity (from 0 to 95%) and the rate (to 2 referred to below as the “NB” (for Nørskov−Bligaard) method.
molecules site−1 s−1).62 At these optimum conditions, the net It uses linear scaling relationships (including BEP relations) to
reaction was found to be approximately:62 estimate the energies of all adsorbed intermediates and
2776 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

transition states based on a few key “descriptors”, typically the To assess the accuracy of eq 15, we compared its predictions
binding energy of the catalyst surface to two selected atoms (C, to those of the NB method for the relative rates of the (211)
O, or N). surfaces of 12 late transition metals for methane steam
We recently introduced a new method capable of high- reforming (at T = 773 K, PCO = 0.050 bar, PH2 = 0.15 bar,
throughput computational screening of catalyst materials that is PCH4 = PH2O = 0.40 bar).11 We used the Rh(211) surface as the
based on the use of DRCs.11 It starts with a reference material
(typically the best known catalyst) for which a full mechanism reference catalyst to get the Xi values used in eq 15. We found it
and microkinetic model is developed. These are then used to to be slightly more accurate (relative to a full DFT-based
determine the DRCs of the species in the mechanism (i.e., all calculation of the rate for each metal) than the NB approach
adsorbed intermediates and transition states) at the reaction when the metals are similar to the reference metal (<0.5 eV
conditions of interest. The few species with the highest DRCs different on the NB descriptor plot where the axes are the bond
for this reference material are chosen, and their energies on energies to C and O adatoms), clearly attributable to the errors
different materials are used as the key descriptors needed to in the linear scaling approximations intrinsic to the NB
estimate their rates and thus screen materials for their activity. approach. In private communication, Jens Nørskov has argued
A derivation of the needed equations is provided in detail that those linear scaling relations may provide more accurate
elsewhere.11 Here we just give a summary of the main steps. estimates of the species’ energies than the raw DFT energies.
Rearranging eq 7 gives When the metals are very different from the reference (>1 eV
away on the NB-type descriptor plot), eq 15 is much less
⎛ −G ⎞ accurate than the NB approach. This is due to the fact that the
Xi*d⎜ i ⎟ = d(ln r ) DRCs change too much (from those of the reference metal
⎝ RT ⎠ (12)
Rh), so that the approximation in eq 15 of DRCs equal to those
where we have dropped the superscript “0” on G0i to simplify for the reference material breaks down. Different species
typing, but we still recognize that Gi refers to the standard-state become rate-controlling for these distant metals.11
free energy of species i. We can now integrate this over some To assess the validity of this “constant DRC approximation”
change in Gi, from its value Gi,o for some reference catalyst, used for eq 15, the DRCs for adsorbed species and transition
denoted by a small o subscript, to its value Gi,n for some new states were calculated by the NB approach across descriptor
catalyst, n. Assuming that Xi remains constant over the range of space, for this same methane steam reforming reaction and
interest, at constant temperature this gives same conditions as above.11 The results are summarized in
Figure 3, where the colors correspond to the species with the
⎛ −Gi , n − ( −Gi , o) ⎞ rn rn
Xi*⎜
⎝ RT
⎟=

∫r d(ln r ) = ln
ro
o (13)

Equation 13 shows the change in the total rate to the product


of interest, in moving from catalyst o to catalyst n, due only to
the change in the standard-state free energy of species i, Gi, and
its degree of rate control Xi. This is only true if all other Xi = 0
in the reaction. If there are N different adsorbates and transition
states in the mechanism, then the total change in ln(rate) is the
sum of terms of identical form to the left-hand side of eq 13:
N ⎛ −Gi , n − ( −Gi , o) ⎞ r
∑ Xi*⎜ ⎟ = ln n
i=1 ⎝ RT ⎠ ro (14) Figure 3. (a) Coverages (or −DRCs) of adsorbed species and (b) the
DRCs of transition states for methane steam reforming, computed
Taking the exponent of both sides of this equation and using the NB approach, plotted as a function of the two NB
assuming that the standard-state entropy of the same adsorbed descriptors (the adsorption energies of O and C atoms). Reaction
species or transition state is similar on different surfaces, so that conditions are the same as in Table 1. The colors in the legend labels
changes in standard-state free energies approximately equal on the bottom left represent values of absolute magnitude 1.0, with
changes in the electronic potential energy (i.e., the internal lighter coloring being smaller coverages down to 0.0 (white). The
energy at 0 K, Ei,n), gives coverages are equal to the absolute magnitude of the degree of rate
control for adsorbed species, and the color map for coverages is
rn N −Ei , n − (−Ei , o) created by letting the red, green, and blue channels be weighted by the
= e∑i=1 Xi * ( RT
)
coverage of O*, C*, and CO* respectively. The color map for
ro (15) transition states is created similarly, with the color corresponding to
each transition-state given to the bottom right corresponding to that
One may use this equation to estimate rn/ro for new (but DRC = 1.0. From ref 11.
similar) materials to the reference material. Since Xi is near zero
for most species, one need only to execute this sum over those
few species with non-negligible Xi, so one only needs to dominant DRC. It is clear that the DRCs change drastically
computationally estimate the internal energies (on both throughout this (EC*, EO*) two-descriptor plane. However, there
materials) for those few intermediates and transition states. are large areas where the DRCs are constant, and these cover the
In our first application of eq 15,11 DFT was used to calculate vast majority of descriptor space. Within each of these regions, the
these internal energies. In principle, one could estimate these approximations in eq 15 are valid. Thus, if a new class of catalyst
by any other electronic structure method, by experiments, or by is discovered for some reaction, and it has not yet been
semiempirical methods. optimized, it is most likely to fall in such an area with constant
2777 DOI: 10.1021/acscatal.7b00115
ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

DRCs where eq 15 is valid. It thus could prove excellent for One then changes each energy slightly to numerically
computationally optimizing such new catalyst materials. determine its DRC using only a tiny time step beyond that,
The fact that Rh falls on the border of several such areas for making sure (numerically) that any changes in coverages are
methane steam reforming is due to the fact that Rh was nearly tiny on a relative (i.e., %) basis and that a true differential
the optimum metal for these reaction conditions already. response in relative rate to that energy change is approached.
Optimizing a catalyst material generally means moving on
descriptor space to such a border area where there is no longer
a single adsorbate and single transition state that dominate the
■ AUTHOR INFORMATION
Corresponding Author
DRCs. The fact that eq 15 is still more accurate than the NB *E-mail: charliec@uw.edu.
method within 0.5 eV of Rh on this plot indicates that this
ORCID
DRC method is reasonably robust even when the DRCs are
changing near the reference material. To bypass the issue of Charles T. Campbell: 0000-0002-5024-8210
changing DRCs for the key species, one could adapt this DRC Notes
method by moving only small distances in descriptor space The author declares no competing financial interest.
using constant DRC values, and then, when a better material is
found, recalculate the DRC values by using full DFT for that
new reference material, tiling out descriptor space with small
■ ACKNOWLEDGMENTS
The authors would like to acknowledge the Department of
steps in that way.11 Energy, Office of Basic Energy Sciences, Chemical Sciences
Note that one could improve the accuracy of the NB method Division grant number DE-FG02-96ER14630, for support of
for catalyst screening by using full DFT instead of linear scaling this work, and the reviewers for helpful suggestions.
to calculate the energies for all species with large DRCs but
retain most of the speed by using linear scaling for all species
with small DRCs.
■ REFERENCES
(1) Campbell, C. T. Top. Catal. 1994, 1, 353−366.
This DRC-based approach to catalyst screening in eq 15 is (2) Stegelmann, C.; Andreasen, A.; Campbell, C. T. J. Am. Chem. Soc.
computationally faster than the NB method when screening 2009, 131, 8077−8082.
less than 100 new catalysts, thus adding a valuable approach (3) Laidler, K. J. J. Chem. Educ. 1988, 65, 250−254.
that is complementary to the Nørskov−Bligaard method. It can (4) Campbell, C. T.; Sprowl, L. H.; Arnadottir, L. J. Phys. Chem. C
be implemented without a microkinetic model if the DRCs of 2016, 120, 10283−10297.
the dominant species are known approximately (e.g., from (5) Campbell, C. T. J. Catal. 2001, 204, 520−524.
experiments). (6) Kozuch, S.; Shaik, S. J. Am. Chem. Soc. 2006, 128, 3355−3365.
(7) Stegelmann, C.; Andreasen, A.; Campbell, C. T. J. Am. Chem. Soc.
In computational catalyst screening, one is often more 2009, 131, 13563−13563.
interested to optimize selectivity than activity. A simple (8) Kozuch, S.; Shaik, S. Acc. Chem. Res. 2011, 44, 101−110.
derivation11 shows that the selectivity ratio equivalent of the (9) Kozuch, S.; Shaik, S. J. Phys. Chem. A 2008, 112, 6032−6041.
activity ratio in eq 15 above (i.e., the selectivity to the desired (10) Uhe, A.; Kozuch, S.; Shaik, S. J. Comput. Chem. 2011, 32, 978−
product P from reactant R for a new material n relative to 985.
reference material o) can be estimated by (11) Wolcott, C. A.; Medford, A. J.; Studt, F.; Campbell, C. T. J.
Catal. 2015, 330, 197−207.
Sn −Ei , n − (−Ei , o)
N (12) Dumesic, J. A.; Huber, G. W.; Boudart, M. In Handbook of
= e∑i=1(XP,i − XR,i) * ( RT
)
Heterogeneous Catalysis, 2nd ed.; Ertl, G., Knözinger, H., Schüth, F.,
So (16)
Weitkamp, J., Eds.; Wiley-VCH Verlag Gmbh: Weinheim, Germany,
Note the similarity of this to the right-hand side of eq 15: Xi 2008; pp 1445−1462.
is simply replaced with XP,i − XR,I, which is the difference in (13) Agarwal, N.; Sanchez-Castillo, M. A.; Cortright, R. D.; Madon,
R. J.; Dumesic, J. A. Ind. Eng. Chem. Res. 2002, 41, 4016−4027.
DRCs for species i in the rate of making the desired product P (14) Sanchez-Castillo, M. A.; Agarwal, N.; Miller, C.; Cortright, R.
and the rate of consuming the most valuable reactant R. Thus, D.; Madon, R. J.; Dumesic, J. A. J. Catal. 2002, 205, 67−85.
eq 16 is almost as easy to apply as eq 15 except that it uses two (15) Baranski, A. Solid State Ionics 1999, 117, 123−8.
DRCs instead of one. (16) Linic, S.; Barteau, M. A. J. Catal. 2003, 214, 200−212.
These DRC-based methods for catalyst improvement are (17) Grabow, L.; Xu, Y.; Mavrikakis, M. Phys. Chem. Chem. Phys.
somewhat related to a mathematical optimization technique 2006, 8, 3369−3374.
known as the “steepest descent” or “gradient descent” method. (18) Grabow, L. C.; Gokhale, A. A.; Evans, S. T.; Dumesic, J. A.;
Mavrikakis, M. J. Phys. Chem. C 2008, 112, 4608−4617.
12. DEGREES OF RATE CONTROL UNDER TRANSIENT (19) Gokhale, A. A.; Kandoi, S.; Greeley, J. P.; Mavrikakis, M.;
Dumesic, J. A. Chem. Eng. Sci. 2004, 59, 4679−4691.
REACTION CONDITIONS (20) Gokhale, A. A.; Dumesic, J. A.; Mavrikakis, M. J. Am. Chem. Soc.
The analysis of DRC values had mainly been applied to kinetics 2008, 130, 1402−1414.
where a steady-state or quasi-steady-state rate has been (21) Bhan, A.; Delgass, W. N. Catal. Rev.: Sci. Eng. 2008, 50, 19−151.
established, as, for example, in a flow reactor, or for low (22) Stoltze, P. Prog. Surf. Sci. 2000, 65, 65−150.
conversion rates in batch reactors. However, it can be applied (23) Stegelmann, C.; Schiødt, N. C.; Campbell, C. T.; Stoltze, P. J.
to transient kinetics within some microkinetic models. To do Catal. 2004, 221, 630−649.
(24) Bligaard, T.; Norskov, J. K.; Dahl, S.; Matthiesen, J.;
this, one must follow a careful procedure that is a bit tricky, as Christensen, C. H.; Sehested, J. J. Catal. 2004, 224, 206−217.
described previously.5 Basically, one must recognize that the (25) Campbell, C. T.; Koel, B. E. J. Catal. 1985, 92, 272−283.
DRCs vary with time (just as they do with reactant and product (26) Campbell, C. T. J. Catal. 1986, 99, 28−38.
concentrations, even at steady state). Thus, one must propagate (27) Stegelmann, C.; Stoltze, P. J. Catal. 2004, 226, 129−137.
the transient kinetics up to the time of interest using the (28) Norskov, J. K.; Abild-Pedersen, F.; Studt, F.; Bligaard, T. Proc.
unperturbed energies of the microkinetic model being analyzed. Natl. Acad. Sci. U. S. A. 2011, 108, 937−943.

2778 DOI: 10.1021/acscatal.7b00115


ACS Catal. 2017, 7, 2770−2779
ACS Catalysis Viewpoint

(29) Norskov, J. K.; Bligaard, T.; Rossmeisl, J.; Christensen, C. H. (61) Avanesian, T.; Gusmao, G. S.; Christopher, P. J. Catal. 2016,
Nat. Chem. 2009, 1, 37−46. 343, 86−96.
(30) Mavrikakis, M. Nat. Mater. 2006, 5, 847−848. (62) Dix, S. T.; Scott, J. K.; Getman, R. B.; Campbell, C. T. Faraday
(31) Greeley, J.; Mavrikakis, M. J. Am. Chem. Soc. 2002, 124, 7193− Discuss. 2016, 188, 21−38.
7201. (63) Tran, H. V.; Doan, H. A.; Chandler, B. D.; Grabow, L. C. Curr.
(32) Greeley, J.; Mavrikakis, M. J. Am. Chem. Soc. 2004, 126, 3910− Opin. Chem. Eng. 2016, 13, 100−108.
3919. (64) Motagamwala, A. H.; Dumesic, J. A. Proc. Natl. Acad. Sci. U. S. A.
(33) Greeley, J.; Mavrikakis, M. Nat. Mater. 2004, 3, 810−815. 2016, 113, E2879−E2888.
(34) Greeley, J.; Jaramillo, T. F.; Bonde, J.; Chorkendorff, I. B.; (65) Andersson, M. P.; Bligaard, T.; Kustov, A.; Larsen, K. E.;
Norskov, J. K. Nat. Mater. 2006, 5, 909−913. Greeley, J.; Johannessen, T.; Christensen, C. H.; Norskov, J. K. J.
(35) Greeley, J.; Norskov, J. K.; Kibler, L. A.; El-Aziz, A. M.; Kolb, D. Catal. 2006, 239, 501−506.
M. ChemPhysChem 2006, 7, 1032−1035. (66) Alayoglu, S.; Nilekar, A. U.; Mavrikakis, M.; Eichhorn, B. Nat.
(36) Greeley, J.; Norskov, J. K. Surf. Sci. 2007, 601, 1590−1598. Mater. 2008, 7, 333−338.
(37) Greeley, J.; Stephens, I. E. L.; Bondarenko, A. S.; Johansson, T. (67) Studt, F.; Sharafutdinov, I.; Abild-Pedersen, F.; Elkjaer, C. F.;
P.; Hansen, H. A.; Jaramillo, T. F.; Rossmeisl, J.; Chorkendorff, I.; Hummelshoj, J. S.; Dahl, S.; Chorkendorff, I.; Norskov, J. K. Nat.
Norskov, J. K. Nat. Chem. 2009, 1, 552−556. Chem. 2014, 6, 320−324.
(38) Reuter, K.; Scheffler, M. Phys. Rev. Lett. 2003, 90, Article No.
046103.
(39) Reuter, K.; Frenkel, D.; Scheffler, M. Phys. Rev. Lett. 2004, 93,
Article No. 116105.
(40) Reuter, K.; Scheffler, M. Appl. Phys. A: Mater. Sci. Process. 2004,
78, 793−798.
(41) Reuter, K. In Nanocatalysis: Principles, Methods, Case Studies;
Heiz, U., Hakkinen, H., Landman, U., Eds.; Springer: Berlin, 2005.
(42) Reuter, K.; Scheffler, M. Phys. Rev. B: Condens. Matter Mater.
Phys. 2006, 73, 045433.
(43) Reuter, K. Catal. Lett. 2016, 146, 541−563.
(44) Wellendorff, J.; Silbaugh, T. L.; Garcia-Pintos, D.; Norskov, J.
K.; Bligaard, T.; Studt, F.; Campbell, C. T. Surf. Sci. 2015, 640, 36−44.
(45) Choksi, T.; Greeley, J. ACS Catal. 2016, 6, 7260−7277.
(46) Zhao, Z.; Li, Z.; Cui, Y.; Zhu, H.; Schneider, W. F.; Delgass, W.
N.; Ribeiro, F.; Greeley, J. J. Catal. 2017, 345, 157−169.
(47) Sprowl, L. H.; Campbell, C. T.; Arnadottir, L. J. Phys. Chem. C
2016, 120, 9719−9731 (An Erratum to this paper will soon appear to
point out that the x-axis was mislabelled on Figures 6, 7 and the
Abstract graphic, whereby the natural logarithm was incorrectly
written as a logarithm to the base 10.).
(48) Medford, A. J.; Shi, C.; Hoffmann, M. J.; Lausche, A. C.;
Fitzgibbon, S. R.; Bligaard, T.; Norskov, J. K. Catal. Lett. 2015, 145,
794−807.
(49) Gardiner, W. C. Rates and Mechanisms of Chemical Reactions; W.
A. Benjamin, Inc.: Menlo Park, CA, 1972.
(50) Laidler, K. J. Chemical Kinetics, 3rd ed.; Harper Collins: New
York, 1987.
(51) Abild-Pedersen, F.; Greeley, J.; Studt, F.; Rossmeisl, J.; Munter,
T. R.; Moses, P. G.; Skulason, E.; Bligaard, T.; Norskov, J. K. Phys. Rev.
Lett. 2007, 99 (4), 016105.
(52) Studt, F.; Abild-Pedersen, F.; Bligaard, T.; Sorensen, R. Z.;
Christensen, C. H.; Norskov, J. K. Science 2008, 320, 1320−1322.
(53) Norskov, J. K.; Bligaard, T.; Logadottir, A.; Bahn, S.; Hansen, L.
B.; Bollinger, M.; Bengaard, H.; Hammer, B.; Sljivancanin, Z.;
Mavrikakis, M.; Xu, Y.; Dahl, S.; Jacobsen, C. J. H. J. Catal. 2002,
209, 275−278.
(54) Medford, A. J.; Lausche, A. C.; Abild-Pedersen, F.; Temel, B.;
Schjodt, N. C.; Norskov, J. K.; Studt, F. Top. Catal. 2014, 57, 135−
142.
(55) Norskov, J. K.; Bligaard, T.; Kleis, J. Science 2009, 324, 1655−
1656.
(56) Bligaard, T.; Bullock, R. M.; Campbell, C. T.; Chen, J. G. G.;
Gates, B. C.; Gorte, R. J.; Jones, C. W.; Jones, W. D.; Kitchin, J. R.;
Scott, S. L. ACS Catal. 2016, 6, 2590−2602.
(57) Peterson, A. A.; Norskov, J. K. J. Phys. Chem. Lett. 2012, 3, 251−
258.
(58) Vojvodic, A.; Norskov, J. K. National Science Review 2015, 2,
140−143.
(59) Medford, A. J.; Vojvodic, A.; Hummelshoj, J. S.; Voss, J.; Abild-
Pedersen, F.; Studt, F.; Bligaard, T.; Nilsson, A.; Norskov, J. K. J. Catal.
2015, 328, 36−42.
(60) Avanesian, T.; Christopher, P. ACS Catal. 2016, 6, 5268−5272.

2779 DOI: 10.1021/acscatal.7b00115


ACS Catal. 2017, 7, 2770−2779

You might also like