You are on page 1of 16

Journal of Catalysis 374 (2019) 183–198

Contents lists available at ScienceDirect

Journal of Catalysis
journal homepage: www.elsevier.com/locate/jcat

Microkinetic analysis of acetone hydrogenation over Pt/SiO2


Xin Gao a, Andreas Heyden b, Omar A. Abdelrahman c,⇑, Jesse Q. Bond a,⇑
a
Department of Biomedical and Chemical Engineering, Syracuse University, Syracuse, NY 13244, USA
b
Department of Chemical Engineering, University of South Carolina, Columbia, SC 29208, USA
c
Department of Chemical Engineering, University of Massachusetts Amherst, Amherst, MA 01003, USA

a r t i c l e i n f o a b s t r a c t

Article history: The kinetics of vapor-phase acetone hydrogenation were investigated over a Pt/SiO2 catalyst through a
Received 2 December 2018 range of temperatures (303–435 K) and reactant partial pressures (pACE = 0.3–140 Torr, pH2 = 100–
Revised 2 April 2019 930 Torr). We compile macroscopically observable reaction orders, apparent barriers, kinetic isotope
Accepted 21 April 2019
effects, and reaction rates; and we present a microkinetic analysis that considers multiple rate control
Available online 10 May 2019
scenarios in two Horuiti-Polanyi-type mechanisms. Overall, our data are best described by a two-site
model wherein atomic hydrogen adsorbs at a distinct set of active sites from organic species. We further
Keywords:
propose that two surface reactions each exert partial rate control, and we illustrate that degree of rate
Acetone hydrogenation
Ketone hydrogenation
control is sensitive to reaction conditions. Shifts in rate control are attributed primarily to changing sur-
Platinum face coverage regimes, which highlights the significance of both free energies of activation and chemical-
Silica potential driving forces in dictating the kinetic significance of elementary steps. Despite our conclusion
Kinetics that surface reactions involving H-X bond formation are likely rate controlling, we observe that H2/D2
Microkinetic modelling switching minimally perturbs acetone hydrogenation rates. We interpret this by considering that
Horiuti-Polanyi observed kinetic isotope effects reflect a convolution of kinetic and thermodynamic impacts, which
can obscure the elementary effects of changes in isotopic mass.
Ó 2019 Elsevier Inc. All rights reserved.

1. Introduction sugars (e.g., glucose or fructose) occurs through carbonyl hydro-


genation [9–13]. Commonly, supported metals are used for the
Converting biomass into petrochemical commodities generally aforementioned, and Ru, Pt, Pd, Ni, and Cu have been used for both
involves oxygen removal, and a number of thermal, biological, gas- and liquid-phase processes [7,8,14–20].
and catalytic processes are available for doing so. From this land- Studies of carbonyl hydrogenation have highlighted many
scape of reductive technologies, we narrow our focus to the hydro- interesting macroscale phenomena. Observed rates and selectivi-
genation of carbonyl groups using molecular H2 to form their ties are sensitive to the nature, structure, and provenance of the
corresponding alcohols. This chemistry is an important component metal catalyst [21–23]; the nature and structure of the support
of many catalytic ‘‘hydrodeoxygenation” strategies; as such, car- [24]; the identity of the carbonyl compound subjected to hydro-
bonyl hydrogenation appears frequently in the preparation of genation [25,26]; and the nature and composition of the bulk
bio-based fuels and chemicals. For example, c-valerolactone is syn- gas- or liquid-phase [27–31]. Our group is interested in under-
thesized by hydrogenation of the ketone moiety in levulinic acid standing the fundamental origins of these phenomena. To this
[1]; synthesis of long-chain linear alkanes from furfural or 5- end, we recently reported a microkinetic analysis of gas-phase
hydroxymethylfurfural condensation products requires ketone ketone hydrogenation on supported Ru catalysts [25]. Consistent
hydrogenation [2,3]; furfuryl alcohol is prepared by reducing the with prior reports, this study revealed that the most likely rate
aldehyde group in furfural [4–8]; and production of sugar alcohols controlling step during ketone hydrogenation on Ru is the addition
(e.g., sorbitol or mannitol) from their corresponding aldo- or keto- of a hydrogen adatom to a partially hydrogenated surface interme-
diate [28,30], which is generally thought to be an alkoxy [32–37].
We used data reconciliation and parameter estimation to quantify
⇑ Corresponding authors at: Department of Biomedical and Chemical Engineer- elementary reaction and activation free energies, which enabled
ing, 329 Link Hall, Syracuse University, Syracuse, NY 13244, USA (J.Q. Bond).
prediction of turnover frequencies and reactor performance at
Department of Chemical Engineering, 112F Goessmann Laboratory, University of
Massachusetts Amherst, Amherst, MA 01003, USA (O.A. Abdelrahman).
any temperature, pressure, and system composition. Our prior
E-mail addresses: abdel@umass.edu (O.A. Abdelrahman), jqbond@syr.edu (J.Q. work therefore established baseline expectations for the relatively
Bond). facile and thermodynamically ideal case of gas-phase ketone

https://doi.org/10.1016/j.jcat.2019.04.033
0021-9517/Ó 2019 Elsevier Inc. All rights reserved.
184 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

hydrogenation over Ru [25]. Relative to this point, we are able to has been no definitive resolution within the field, we consider both
interpret impacts of the many possible perturbations to the car- mechanisms here. In particular, we develop microkinetic models
bonyl hydrogenation system. One interesting observation is that that describe acetone hydrogenation using single-site and two-
Ru is sensitive to solvent effects [27,28] and is particularly well- site mechanisms. In order to identify the most likely mechanism
suited to aqueous-phase carbonyl hydrogenation [31], whereas Pt and rate controlling step(s), we attempt to reconcile each mecha-
is weakly influenced by solvents [38] and reportedly offers better nism with experimentally measured rates of acetone hydrogena-
performance in the gas phase [30,39]. Toward the aim of elucidat- tion, as well as observed barriers, reaction orders, and kinetic
ing the fundamental underpinnings of observed differences isotope effects during H2/D2 switching experiments.
between Pt and Ru, we focus this manuscript on a rigorous
microkinetic analysis for acetone hydrogenation in the gas phase 2. Experimental
over Pt/SiO2. This is a foundational study that is intended to facil-
itate a direct comparison with our prior microkinetic analysis of 2.1. Materials and methods
gas-phase ketone hydrogenation over Ru/SiO2. This will require
an in-depth, concerted analysis of the two systems that extends Hexachloroplatinic (IV) acid hydrate (37.5% Pt, Acros Organics)
beyond the scope of the present work and will comprise a separate and amorphous SiO2 (481 m2/g, Sigma Aldrich) were used for cat-
manuscript. alyst preparation. Acetone (99+%, Acros Organics) and isopropyl
On metal surfaces, reduction of ketones using molecular hydro- alcohol (reagent grade, Acros Organics) were used as reactor feeds
gen is thought to proceed through a Horiuti-Polanyi-type mecha- and for instrument calibration. H2 (99.999%, Airgas), He (99.999%,
nism [40]. Specifically, metal surfaces bind the ketone, dissociate Airgas), N2 (99.999%, Airgas), and CO (99.99% Praxair) were
molecular hydrogen, and facilitate sequential addition of adsorbed employed for kinetic studies, catalyst pretreatment, and catalyst
hydrogen atoms to adsorbed ketones to form the surface-bound characterization. H2 and D2 blends (5% H2 or D2, 1% Ar, 94% He, Air-
alcohol. Though there is consensus on this conceptual view, speci- gas) were used in probing kinetic isotope effects. Water used in
fic site requirements and rate controlling phenomena remain preparation of catalysts was purified in house by reverse osmosis,
unclear [25,41–43]. Prior analyses of ketone hydrogenation and UV oxidation, and ion exchange to a resistivity 18.2 MX cm.
alcohol dehydrogenation over Pt have considered two distinct
mechanisms [41–43], which are illustrated in Fig. 1. The first 2.2. Catalyst synthesis
involves a single type of surface site (Scheme A) and the second
involves two distinct surface sites (Scheme B). A Pt/SiO2 catalyst was prepared by incipient wetness impregna-
In both schemes, we represent gas-phase acetone, molecular tion of amorphous SiO2 with aqueous hexachloroplatinic (IV) acid
hydrogen, and isopropanol as ACE, H2, and IPOH, while H repre- hydrate. We selected SiO2 because it has no strong acid, base, or
sents atomic hydrogen. Acetone is assumed to bind molecularly redox functionality; and reference experiments confirmed it to
(Step 1), while H2 adsorbs dissociatively (Step 2). Hydrogen atoms be inert to all gas-phase species under our reaction conditions
then add sequentially to surface acetone and the partially hydro- [25]. Our synthesis used an incipient volume of 1.6 ml of precursor
genated intermediate (Step 3 and Step 4). This forms surface iso- solution per gram of support. Impregnated catalysts were dried
propanol, which desorbs to form the gas phase product (Step 5). overnight in air at 393 K and then reduced in flowing H2
Results from both surface science and computational analysis sug- (100 ml min1, 673 K, 3 K min1). Prior to removal from reduction
gest that the partially hydrogenated intermediate on Pt surfaces is cells, samples were passivated at 298 K in a stream of 1% O2 in He.
an isopropoxy fragment, which is bound atop a Pt atom via its
oxygen [37,44]. We label it accordingly (IPOX). Distinct vacant sites 2.3. Catalyst characterization
are represented as ‘‘*” and ‘‘s,” and subscripts on surface-bound
species designate the type of site where they adsorb. The sole Catalyst surface area and average pore diameter were deter-
difference between the two mechanisms is that the single-site mined by N2 physisorption at 77 K (Micromeritics ASAP 2020).
model considers that atomic hydrogen and all surface hydrocar- Before N2 dosing, samples were outgassed under vacuum (6 h,
bons compete for adsorption at a single type of active site [42], 523 K). Total surface areas and pore size distributions were
while the two-site model allows hydrocarbons and hydrogen obtained through BET and BJH analyses of the N2 adsorption/des-
atoms to bind at different surface sites [41,43]. Previously, we orption isotherm [45,46]. Pt surface site densities were quantified
showed good agreement between a two-site mechanism and by CO chemisorption at 308 K (Micromeritics ASAP 2020) [47].
measured ketone hydrogenation rates over Ru, and we rationalized Prior to CO dosing, samples were reduced in flowing H2 (3 h,
a two-site mechanism based on the idea that atomic hydrogen will 673 K, 3 K min1), evacuated at 673 K for 1 h to remove chemi-
generally bind in three-fold hollow sites on Ru, whereas larger sorbed hydrogen, and cooled to 308 K under vacuum. The analysis
adsorbates are frequently observed in atop or bridged configura- was performed at 308 K by collecting a CO adsorption isotherm,
tions [25]. The same rationale stands for Pt, and, because there evacuating the sample for 1 h to remove physisorbed CO, and

Fig. 1. Elementary steps comprising single-site and two-site mechanisms for ketone hydrogenation.
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 185

collecting a second isotherm. Irreversible CO uptake was calculated ical equilibrium that species production rates in all experiments
as the difference in CO uptake between the first and second iso- are minimally influenced by the reverse reaction. Finally, reference
therms. Irreversibly bound CO was taken to be equivalent to the experiments confirmed that, in our experimental range of acetone
Pt surface site density, which assumes a CO:Pt adsorption stoi- conversions, there is no strong product inhibition during acetone
chiometry of 1:1 [47]. Replicate chemisorption experiments on a hydrogenation over Pt (See Online Supporting Information). Based
single Pt/SiO2 catalyst indicate that this method is generally able on the above arguments, we conclude that the differential reactor
to quantify CO uptake within ± 5%. approximation is reasonable, and, for the entirety of our analysis,
we assume that hydrogenation rates are constant throughout the
2.4. Catalytic activity testing bed at feed conditions. In the interest of assessing intrinsic reactiv-
ity, hydrogenation rates are estimated on an active site basis as iso-
We performed acetone hydrogenation experiments in a down- propanol site time yields (STY, Eq. (2)) [49]. FIPOH is the molar
flow, packed bed reactor. The bed was comprised of an active phase flowrate of isopropanol out of the packed bed (±10%), and SPt is
(2.1 wt% Pt/SiO2, 45–90 lm) diluted 10:1 in amorphous SiO2 (45– the total molar quantity of Pt surface sites in the catalyst bed as
90 lm). The mixture was loaded into a 6.35 mm OD 316 stainless estimated by CO chemisorption (±5%). Propagation of uncertainty
steel tube, and the bed was held in place by quartz wool plugs. The suggests that site time yields are precise to within ± 15% in this
void volume below the catalyst bed was packed with 850–2000 mm system.
quartz granules. Reactor temperature was controlled at the outer
F IPOH
wall of the packed bed using a Type K thermocouple and a PID tem- STY ¼ ð2Þ
perature controller (LOVE 16A 3010). Kinetic data are reported at SPt
the bed temperature, which was measured by an in-line, Type K During reactor startup, effluent isopropanol flowrates decrease
thermocouple positioned in the void space just above the catalyst as a function of time on stream (see Online Supporting Informa-
bed. Prior to activity testing, the catalyst bed was reduced in situ tion). This decline in activity occurs over time scales (1–2 h) that
under H2 flow (100 ml min1) at 673 K for 4 h with a ramp rate are significantly longer than the transient periods that we observe
of 5 K min1. The bed was then cooled to the desired reaction tem- after making a perturbation in operating conditions at steady state
perature under H2. Gas-phase reactor feeds (He, H2, D2, various (<10 min). We thus attribute this initial decline in activity to cata-
blends, etc.) were regulated using mass flow controllers (Brooks lyst deactivation. Because we have employed an ex-situ method for
5850S). Liquid feeds (acetone and isopropanol) were introduced surface site quantification (CO chemisorption), it is important to
using a syringe pump (Cole-Parmer series 100) and fed through a account for deactivation in specifying reaction rates. Active site
130 mm PEEK capillary into a heated vaporization chamber where densities estimated by CO chemisorption are representative of
they were contacted with pre-heated gas feeds. The process stream the catalyst as it exists before exposure to reactants; therefore,
then passed through a heat-traced static mixer, where it was all reported site-time-yields reflect an extrapolation of hydrogena-
brought to reaction temperature. During reactor startup, the feed tion rates to zero time-on stream. We obtained these estimates by
stream was diverted through a bypass until reaching steady state fitting temporal deactivation profiles measured over the initial
composition, at which point the feed stream was introduced into  2 h on stream to a first order deactivation model with an
the reactor. The time of valve switching was taken as zero time approach to a non-zero steady state rate. This approach is permis-
on stream. sible because packed bed reactors were operated differentially over
Reaction products were analyzed using an in-line Agilent 7890A the entire deactivation profile and there is no strong product inhi-
GC equipped with a 6-port gas sampling valve, an HP-INNOWAX bition in our experimental conversion ranges [50]. The methodol-
column, and a flame ionization detector. This configuration is suf- ogy is further detailed in the online supporting information.
ficient to resolve and quantify all of the species considered in this
study, and carbon balances closed to within ± 5%. Under conditions
2.5. Kinetic isotope effects
reported here, acetone hydrogenation over Pt/SiO2 is 100% selec-
tive to isopropanol, and 10 replicate experiments at a single ref-
We examined the kinetic impact of H2/D2 switching during ace-
erence condition indicate that our precision in quantifying
tone hydrogenation over Pt using the apparatus described in the
isopropanol formation rates is ± 10%. Weight-hourly-space veloci-
preceding section. The system was permitted to reach steady state
ties were adjusted to maintain reactors at or below 10% acetone
under a flow of 5% H2 in He; subsequently, the reducing gas was
conversion, with 94% of acetone hydrogenation rates measured
switched to 5% D2 in He, and the system was again permitted to
below 5% acetone conversion. Acetone hydrogenation is exother-
reach steady state. Finally, the reducing gas was switched back to
mic (DH° = 55.2 kJ mol1), but it incurs a substantial entropy loss
5% H2 in He to quantify any perturbations to the baseline activity
(DS° = 116.3 J mol1 K1); accordingly, it is important to consider
of the catalyst (e.g., deactivation or regeneration) induced by the
experimental conversions in the context of the equilibrium limit
switch. Under all conditions, the catalyst returned to its reference
under a given set of reaction conditions. This is particularly true
state activity within  10 min of the exchange, and we conclude
at high reaction temperatures or low hydrogen partial pressures.
that there was no appreciable deactivation or induction under D2
The majority of our data (95%) were collected at or below fractional
flow. Impacts of H2/D2 switching on acetone hydrogenation kinet-
conversions and/or feed compositions where the reversibility, z
ics are inferred from site time yields of isopropanol under H2 and
[48], of the overall hydrogenation reaction (Eq. (1)) is less than
D2, which we use to establish an apparent kinetic isotope effect
0.01.
(KIEA) as defined by Eq. (3).
aIPOH
z¼ ð1Þ STY H2
K  aACE  aH2 KIEA ¼ ð3Þ
STY D2
Increased proximity to equilibrium occurred at high reaction
temperatures, low H2 pressures, and in the presence of isopropanol
co-feeds. Specifically, during IPA co-feeding experiments, H2/D2 2.6. Electronic structure calculations
switching experiments, and H2 order studies at 420–440 K, we
occasionally observed reversibilities above 0.01 and up to 0.05. Where necessary, gas phase thermodynamic data were
Regardless, these values remain sufficiently displaced from chem- supplemented with results from simulations. Specifically, we used
186 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

Gaussian’09 software to predict the standard state entropy and nation (R2, Eq. (6)) and the Akaike Information Criterion (AIC, Eq.
heat capacity of the gas-phase isopropoxy radical [51]. Geometry (7)) for each least squares solution [53]. Both are useful indicators
optimization and frequency calculations were performed using of fit quality. Importantly, the AIC penalizes a regression for
the Hartree-Fock method with a 6-311++g (3df, 3pd) basis set. increasing variable parameter numbers, which helps to identify
Benchmarking simulations revealed that this level of theory is suf- cases of over-fitting and to determine a statistically preferred
ficient to predict reference state entropies and heat capacities for model.
gas-phase acetone within  2% of tabulated values. !2
X
n
  1 Xn
 
SST ¼ log10 STY EXPi   log10 STY EXPi ð5Þ
n i¼1
2.7. Data reconciliation and parameter estimation i¼1

One of our primary methods of kinetic analysis is to determine SSE


R2 ¼ 1  ð6Þ
whether a given kinetic model corresponding to a specified rate SST
control scenario is able to reconcile with experimental observa-
AIC ¼ 2  nk þ n  lnðSSEÞ ð7Þ
tions with a physically realistic parameter set. Quantitatively, we
assess this for each proposed kinetic model by solving the least In Eqs. (5)–(7), n is the number of experimental measurements
squares problem, i.e., by determining the parameter values that and nk is the number of parameters regressed in a given rate con-
minimize the objective function given in Eq. (4). trol scenario.

X
n
    2
SSE ¼ log10 STY EXPi  log10 STY MODi ð4Þ 3. Results & discussion
i¼1
3.1. Catalyst synthesis & characterization
Here, the square error for any experiment, i, is defined by the
difference between measured isopropanol site time yields at each
Data were obtained using a Pt/SiO2 catalyst that had a total
experimental condition and the corresponding model prediction
metal loading of 2.1 wt%. Physicochemical properties of this cata-
at the same experimental condition. Our experiments span large
lyst and its parent SiO2 support are summarized in Table 1.
ranges of temperature and species partial pressures, and site time
yields vary over several orders of magnitude. To provide a uniform
weighting of all measurements and avoid bias toward preferential 3.2. Demonstrating kinetic control
minimization of errors in large site time yields, all data were
rescaled by taking their common logarithm [52]. This ensured that One can assess the extent of diffusion control using experimen-
all of our measurements and model predictions range from approx- tal tests, such as those described by Madon and Boudart [54], or by
imately 2 to 1. The objective function (Eq. (4)) was minimized by comparing rates of reaction to rates of diffusion as formalized in
varying parameter values using a trust-region reflective algorithm the Weisz-Prater criterion (Eq. (8)) [55]. Here, robs is the observed
via the lsqnonlin routine in MATLAB. Analogous to the above reaction rate per unit volume of catalyst, Rp is the radius of the cat-
described case of disparate scales in our measured hydrogenation alyst particle, Dj is the effective diffusivity of the reacting species in
rates, parameter estimates were passed to the optimization routine the catalyst pores, and Cj is the concentration of species j at the
on a logarithmic scale, ensuring that all parameter values ranged external surface of the catalyst.
from approximately 3 to 3. robs  R2p
All of our kinetic models either have analytical solutions or /¼ ð8Þ
Dj  C j
require numerical solution of a small set of nonlinear algebraic
equations; further, all of our data are collected under differential Previously, we studied the gas-phase hydrogenation of acetone
conditions. This avoids the need for iterative solution of stiff DAE over a 1.5 wt% Ru/SiO2 catalyst (45–90 lm) with a surface site den-
systems, making our optimization routines computationally facile. sity of 62 lmol g1, and we probed the influence of mass transfer
We leverage this to increase the likelihood of locating global min- using a Madon-Boudart test [25,54]. Ultimately, we demonstrated
ima. Specifically, we performed an initial regression to identify an that kinetic control is maintained over Ru/SiO2 for acetone hydro-
optimal parameter set and quantify a baseline minimum residual genation rates up to 2850 lmol g1 min1 (pA = 4.8 Torr,
sum of squares. We then introduced random perturbations pH2 = 910 Torr, T = 341 K). Assuming a bulk catalyst density of
between 0 and ± 25% to each of the optimal parameter values, 1.2 g mL1, a particle diameter of 90 lm, an effective diffusivity
and we used the perturbed optimum set as the starting state for of 1.0  106 m2 sec1, and that acetone diffusion is more likely
a second optimization. This process was continued until no further limiting than H2 diffusion, we estimate the value of the Weisz-
improvement in the value of the objective function was observed. Prater criterion to be  0.51 at these conditions. Threshold values
We repeated this strategy for multiple, randomized starting for the Weisz Prater criterion are dependent on reaction orders,
parameter sets and found the optima to be robust and insensitive with the likelihood of diffusion control increasing for higher order
to the starting state of the model. Optimum parameter estimates reactions [56]. For Ru/SiO2, macroscopically observed reaction
are reported with 95% confidence intervals. These were initially orders in hydrogen and acetone are between 0 and 1 [25]. This
calculated for logarithmically-scaled parameters using the nlparci implies that the kinetic control threshold of the Weisz-Prater crite-
routine in MATLAB, which uses the set of residual errors and the rion for acetone hydrogenation is between 1 (first order reaction)
Jacobian Matrix computed by lsqnonlin at the optimum parameter and 6 (zero order reaction). Our estimated value of 0.51 can there-
set. They were converted to the linear scale uncertainties reported fore be taken to indicate a kinetically-controlled hydrogenation
in the manuscript using propagation of error analysis. A sample rate; however, considering the uncertainty in estimating the
optimization routine, which is easily generalizable for kinetic anal- Weisz-Prater criterion, this number is sufficiently close to the
ysis of other systems, is available for download in the online sup- threshold range to raise concern. Here we emphasize two points.
porting information. First, kinetic control at this condition was rigorously demonstrated
The residual sum of squares (SSE) was used alongside the total using an experimental Madon-Boudart test, which is generally a
sum of squares (SST, Eq. (5)) to compute the coefficient of determi- more reliable assessment of kinetic control than estimated values
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 187

Table 1
Physicochemical characterization of SiO2 and Pt/SiO2 used in this study.

Pt wt%a SA (m2 g1) Pore Diameter (Å)b Irreversible CO uptake (lmol g1) COirr/Pt
0 481 56.5 – –
2.1 471 49.9 52 0.48
a
Based upon concentrations of impregnating solutions, incipient volume, and support mass.
b
Average pore diameter determined by BJH analysis of N2 adsorption/desorption data.

of the Weisz-Prater criterion. Second, these represent the most 3.4. The temperature dependence of acetone hydrogenation on Pt/SiO2
demanding conditions with respect to mass transfer limitations
in the prior study, i.e., those that had the largest ratio of reaction Fig. 2 illustrates acetone hydrogenation rates over Pt/SiO2 mea-
rate to acetone concentration. All hydrogenation rates measured sured between 322 K and 421 K at various acetone partial pres-
over Ru/SiO2 were obtained at or below this threshold, allowing sures. In general, one observes nonlinearity over the entire
us to infer kinetic control for the entire data set. temperature range. This indicates that apparent barriers are a
In the current study, we consider gas-phase acetone hydrogena- strong function of temperature; accordingly, it is not possible to
tion over a 2.1 wt% Pt/SiO2 catalyst (45–90 lm) that has a surface quantify a single barrier that describes acetone hydrogenation on
site density of 52 lmol g1. This catalyst used an identical SiO2 Pt. That said, subsets of the data are linear, and we can discuss
support and sieve fraction to the previously discussed Ru/SiO2 cat- trends in apparent barriers over different temperature and partial
alyst; accordingly, surface area, porosity, and (bulk) particle sizes pressure ranges. The most notable feature is that apparent barriers
are comparable between the two materials. The most demanding decrease as a function of temperature; further, the onset tempera-
condition for kinetic control in the current study was a measured ture and magnitude of this decrease are functions of acetone par-
rate of acetone hydrogenation of 313 lmol g1 min1 at an acetone tial pressure. For example, acetone hydrogenation rates
partial pressure of 1.23 Torr (pH2 = 926 Torr, T = 416 K). We esti- measured between 322 K and 381 K at 4.0 Torr acetone suggest
mate the value of the Weisz-Prater criterion to be roughly 0.26 an apparent barrier of 33 ± 6 kJ mol1 whereas between 381 K
at this condition, which is the maximum value for the current data and 421 K at the same acetone pressure, we observe that the
set. This is roughly half of the value of 0.51 estimated at the kinetic apparent barrier decreases to 11 ± 8 kJ mol1. These transitions
control threshold over Ru/SiO2. We are thus confident in conclud- become less pronounced at high acetone partial pressures and
ing that all hydrogenation rates reported here over Pt/SiO2 are more pronounced at low acetone partial pressures. At 21 Torr ace-
kinetically controlled. As a final note, reported rates and reactant tone, the entire trend shows reasonable linearity and can be well-
concentrations in this manuscript often result in Weisz-Prater described by an apparent barrier of approximately 34 kJ mol1. In
numbers that exceed the aforementioned kinetic control thresh- contrast, apparent barriers become near-zero (1.4 Torr acetone)
olds. We emphasize that reported rates represent zero-time correc- or even slightly negative (0.3 Torr) as temperatures exceed 381 K.
tions to measured rates. The latter were obtained over deactivated These observations are most likely attributable to differences in
catalysts under clear kinetically-controlled regimes (WP < 0.26) surface coverage under various operating conditions. In general,
and subsequently corrected to zero time on stream to provide a one expects high surface coverages at low temperatures and ele-
common basis for comparison. vated partial pressures. Catalysts studied at these conditions tend

3.3. Experimentally measured rates of acetone hydrogenation

Our kinetic data set is comprised of 185 isopropanol site time


yields measured over a range of reaction temperatures and feed
partial pressures. It is available for download as a CSV file in the
supplementary online materials. Our intent is to provide a trans-
parent, reliable, and comprehensive experimental kinetic bench-
mark for acetone hydrogenation on supported Pt catalysts, and
we hope this data will be useful to others that are interested in fun-
damental aspects of ketone hydrogenation. Our final conclusions in
this manuscript are drawn from reconciliation of kinetic models
with experimental data, and we must acknowledge that this can
provide only partial resolution of observed phenomena. Comple-
mentary insights from future efforts in surface science, in situ or
operando spectroscopy, data science, and/or computational cataly-
sis will help to broaden our understanding of ketone hydrogena-
tion. We hope that by making our data accessible, we can help to
bridge otherwise independent efforts in these areas. This data set
excludes H2/D2 switching and isopropanol co-feeding experiments
as these measurements were obtained under steady state condi-
tions without recording the preceding deactivation profile. We
are therefore not confident in our ability to precisely estimate
zero-time reaction rates for these experiments. For this reason,
we only discuss perturbations to steady state reaction rates upon Fig. 2. Arrhenius plot for Acetone hydrogenation from 322 to 421 K at acetone
H2/D2 switching (apparent kinetic isotope effects) or introducing partial pressures of (e) 0.3 Torr, (h) 1.4 Torr, (s) 4.0 Torr and (D) 21 Torr. Hydrogen
an isopropanol co-feed (apparent reaction orders in isopropanol). partial pressures were fixed at 900 Torr.
188 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

to display saturation kinetics, where heats of chemisorption largely 3.6. Analysis of apparent reaction orders
cancel from observed rate constants, and apparent barriers reflect
the elementary barrier for the rate determining step. At lower ace- Over the range of experimental conditions summarized in
tone partial pressures and/or higher reaction temperatures, sur- Table 2, acetone orders vary between 0 and 0.5, hydrogen orders
faces become sparsely covered, and reaction rates are often vary between 0.5 and 1, and isopropanol orders are near zero.
positive order in the reactants. In these regimes, apparent barriers We next consider whether various Langmuir-Hinshelwood-type
reflect a weighted sum of the endothermic barrier for the rate rate expressions can reconcile with our observations. Specifically,
determining step and exothermic heats of adsorption, and they we develop overall rate expressions as a function of acetone,
are often substantially diminished relative to elementary barriers. hydrogen, and isopropanol partial pressures in the limit where
Similar observations and analyses have been applied for acetone there is a single rate determining step (RDS) relative to which all
hydrogenation over Ru/SiO2 [25], levulinic acid hydrogenation over others are equilibrated. We assume that each elementary step in
Ru [14], and isopropanol dehydrogenation over Pt-Cu [57]. Further, Fig. 1 may be rate determining, which results in ten rate expres-
apparent barriers in the range of 40–70 kJ mol1 have been previ- sions. These are fully derived in the supporting information (Sec-
ously reported for acetone hydrogenation over Pt. Relative to our tion S6). The SI also presents a limit analysis for each rate
conditions, prior studies have considered acetone hydrogenation expression, which establishes permissible reaction orders in ace-
at low temperatures (300–370 K) and high acetone partial pres- tone, hydrogen, and isopropanol for each rate control scenario.
sures (50–300 Torr) [41–43]. These conditions roughly align with Results are summarized in Table 3.
our posited ‘‘saturation” conditions (e.g., 21 Torr Acetone), where A macroscopic reaction rate in an adsorption-limited regime
we observe comparable barriers in the range of 30–40 kJ mol1. will always be first order in the (rate-limiting) adsorbate. Because
we observe zero- or fractional orders in both acetone and H2 under
3.5. Apparent reaction orders all reaction conditions, one can exclude dissociative hydrogen
adsorption (Step 1) and molecular adsorption of acetone (Step 2)
Apparent reaction orders for acetone, hydrogen, and iso- in both the single-site and two-site mechanisms as rate controlling
propanol were determined at multiple temperatures by linear steps during acetone hydrogenation on Pt. Otherwise, depending
regression of hydrogenation rates as a function of species partial on the dominant coverage regime, any of the remaining steps in
pressures on logarithmic scales (Figs. S2 and S6). We observe lin- either the single- or the two-site mechanism can be consistent
earity over the measured partial pressure ranges at each reaction with our observed reaction orders. A definitive resolution is not
temperature, and apparent reaction orders are summarized in possible without invoking assumptions about dominant surface
Table 2. At 322 K, acetone hydrogenation has an apparent species, and we next consider observed kinetic isotope effects.
0.06 ± 0.14 order in acetone, which increases to 0.33 ± 0.09 at
381 K and 0.45 ± 0.14 at 421 K. With respect to hydrogen, we 3.7. Kinetic isotope effects induced by H2/D2 switching
observe a reaction order of 0.42 ± 0.02 at 322 K, which increases
to 0.60 ± 0.16 at 381 K and 0.81 ± 0.02 at 421 K. Increases in appar- To further probe rate controlling phenomena, we performed H2/
ent reaction orders are likely associated with a transition from sat- D2 switching experiments during acetone hydrogenation over Pt/
urated surfaces to sparsely covered surfaces with increasing SiO2. Impacts are presented as function of time on stream in
reaction temperature. This is consistent with our observed Fig. 3; the corresponding apparent kinetic isotope effects (Eq. (3))
decreases in apparent barrier as temperatures increase and partial are summarized in Table 4.
pressures decrease. To probe the extent of product inhibition, we At 322 K, the observed KIE is minor (1.17 ± 0.04), and it dimin-
introduced isopropanol as a co-feed alongside acetone and H2. ishes as reaction temperature increases to 378 K (1.04 ± 0.01) and
For these experiments, isopropanol partial pressures ranged from 416 K (1.05 ± 0.03). These observations are within range of those
0.1 to 1.1 Torr (pACE = 4.4 Torr, pH2 = 830 Torr, pIPOH:pACE = 0.02– previously reported by Baiker (KIEA  1.3) during liquid-phase tri-
0.25). These conditions simulate the impact of increasing fluoroacetophenone hydrogenation over a supported Pt/Al2O3 cat-
isopropanol-to-acetone ratios, i.e., conditions where one is most alyst at 298 K, 0.2–1.0 M trifluoroacetophenone in toluene, and
likely to observe competitive inhibition by isopropanol. In the pH2 = 5 bar [58]. Kinetic isotope effects of 1.0–1.2 are small relative
complete kinetic data set, our maximum measured isopropanol: to the 2- to 6-fold changes in rate constant that are expected when
acetone partial pressure ratio at the reactor exit was  0.1, which replacing hydrogen with deuterium in an elementary reaction that
is within the span of this experiment. Independent of reaction tem- involves either scission or formation of an H-X bond (i.e., a primary
perature, we find that acetone hydrogenation is roughly zero order kinetic isotope effect). Instead, our observations fall into the range
in isopropanol. Specifically, we observe apparent reaction orders in expected for secondary isotope effects. This suggests that the for-
IPA of 0.14 ± 0.26 at 322 K, 0.13 ± 0.23 at 381 K, and mation or scission of an H-X bond cannot be rate controlling, leav-
0.06 ± 0.21 at 422 K. The lack of temperature sensitivity in these ing only acetone adsorption and isopropanol desorption as possible
apparent reaction orders suggests a sparse IPA coverage at all reac- rate determining steps. In the preceding section, we excluded ace-
tion temperatures. tone adsorption as a rate controlling step because our maximum
observed acetone order is  0.5, whereas a system limited by ace-
tone adsorption should be first order in acetone. This leads to the
Table 2 tentative conclusion that acetone hydrogenation is limited by iso-
Measured reaction orders for acetone and hydrogen at various temperatures. propanol desorption under our experimental conditions; however,
Temperature (K) Reaction order we temper this by noting that the aforementioned magnitudes of
Acetonea Hydrogenb Isopropanolc primary and secondary kinetic isotope effects apply to true kinetic
isotope effects, which reflect perturbations to elementary rate con-
322 0.06 ± 0.14 0.42 ± 0.02 0.14 ± 0.26
381 0.33 ± 0.09 0.60 ± 0.16 0.13 ± 0.20
stants induced by changes in isotopic mass. Here, we report appar-
421 0.45 ± 0.14 0.81 ± 0.02 0.06 ± 0.21y ent kinetic isotope effects, which reflect perturbations in the
a
overall hydrogenation rate induced by a change in isotopic mass.
PACE = 0.3–21 Torr, PH2 = 900 Torr, PIPA = 0 Torr.
b
PACE = 4.0 Torr, PH2 = 100–900 Torr, PIPA = 0 Torr.
Whereas a true kinetic isotope effect represents a single elemen-
c
PACE = 4.4 Torr, PH2 = 830 Torr, PIPA = 0.1–1.1 Torr. tary kinetic impact, apparent kinetic isotope effects may reflect a
y
Isopropanol reaction order measured at 422 K. convolution of multiple elementary kinetic and thermodynamic
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 189

Table 3
Observable reaction orders for acetone hydrogenation via single- or two-site mechanisms in the limit of a single rate controlling step.

Mechanism RDS Rate expression equation Possible orders


ACE H2 IPOH
Single-site 1 S12 2 to 1 1 2 to 0
2 S21 1 1 to 1/2 1 to 0
3 S30 1 to 1 1/2 to 3/2 2 to 0
4 S39 1 to 1 0 to 1 2 to 0
5 S48 0 to 1 0 to 1 0
Two-site 1 S75 0 to 1 1 1 to 0
2 S85 1 0 to 1 1 to 0
3 S95 0 to 1 0 to 1 1 to 0
4 S105 0 to 1 0 to 1 1 to 0
5 S115 0 to 1 0 to 1 0

effects upon switching the reducing gas from H2 to D2. Next, we


take a quantitative, statistical approach by attempting to reconcile
experimental observations with predictions from microkinetic
models developed for single-site and two-site mechanisms
(Fig. 1). At the outset, each model is fully parameterized using pub-
lished data and correlations, which enables base-case estimation of
rate constants and equilibrium constants at any temperature. Sub-
sequently, we apply limiting assumptions that facilitate prediction
of macroscopic rates of acetone hydrogenation as a function of
temperature, pressure, and system composition within a specific
rate control scenario. We then regress the full set of potentially sen-
sitive parameters in each rate control scenario by minimizing the
objective function given by Eq. (4). This generates an optimal
parameter set as well as a minimum residual sum of squares
(SSE), which is used to compute the corresponding coefficient of
determination (Eq. (6)) and Akaike Information Criterion value
(Eq. (7)). Our rationale for regressing all potentially sensitive
parameters instead of the initially sensitive parameter set is two-
fold. First, we are always working with small sets of potentially
sensitive parameters (nk  6) relative to the number of measure-
ments (n = 185), which increases the likelihood that we can regress
each parameter with confidence. Second, the initial state of the
Fig. 3. Impact of a switch from H2 to D2 on observed isopropanol site time yields
model is generally far from an optimum, and parameters that are
during acetone hydrogenation at 322 K (d), 378 K (j), and 416 K (▲). Rates
obtained under H2 are shown with filled symbols (d). Rates obtained under D2 are initially insensitive often become sensitive as the parameter set
shown with unfilled symbols (s). PACE = 4.2 Torr, PH2 or PD2 = 42 Torr, is allowed to change during optimization. To increase confidence
PHe = 826 Torr. that least squares solutions identify global optima, we regress
the full set of parameters that are potentially sensitive in each rate
control scenario.
Table 4
Observed kinetic isotope effects.y 3.8.1. Thermodynamic consistency
A microkinetic model must satisfy two thermodynamic con-
Temperature (K) KIEA
straints [56]. First, as expressed in Eq. (9), any thermodynamic
322 1.17 ± 0.04
state function, C, that describes the formation of isopropanol from
378 1.04 ± 0.01
416 1.05 ± 0.04 acetone and molecular hydrogen in the gas phase must equal the
value of that state function computed by the sum of changes in
y
PACE = 4.2 Torr; PH2 or PD2 = 42 Torr.
state function for each elementary step (DCi) weighted by their
stoichiometric numbers (ri).
  X
impacts, and their observable value may depend on surface cover- 0 ¼ CIPOH;g  CACE;g  CH2 ;g  ri  DCi ð9Þ
ages under reaction conditions [59–63]. A rigorous interpretation i
of the apparent kinetic isotope effect is therefore non-trivial, and
Second, rate laws for each elementary step, i, must satisfy
we are cautious in excluding surface reactions as kinetically signif-
microscopic reversibility. This constraint is easily implemented
icant based on apparent KIEs alone. We will defer further consider-
by developing the rate expression for step i as a function of the for-
ation until after a more detailed and quantitative kinetic analysis of
ward rate constant for that step (ki), the equilibrium constant for
our data set.
that step (Ki), the thermodynamic activities of species participating
in that step (aj), and the stoichiometric coefficients of species par-
3.8. Microkinetic analysis ticipating in that step (mj).
Y jmi;j j ki Y jmi;j j
To this point, our discussion has been qualitative and based r i ¼ ki  aj   aj ð10Þ
reactants
K i products
upon permissible reaction orders and anticipated kinetic isotope
190 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

Rate and equilibrium constants are defined in Eqs. (11)–(13) Hj;x ðT Þ ¼ Hj ðT Þ  BEj;x ð17Þ
using standard conventions. Euler’s number appears in the pre-
Initial binding energy estimates are taken from the surface
exponential factor, Ai, because we use an Arrhenius formulation
science literature; however, these generally reflect bound species
in Eq. (11), and we distinguish between the activation barrier (EA,
à à on well-defined model surfaces at low coverage, and this may be
i) and the enthalpy of activation (DHi ). Specifically, EA,i = DHi +
a poor approximation for adsorption enthalpies on technical cata-
RT [64].
lysts under working conditions. Accordingly, we view species bind-
 
EA;i ing energies as uncertain and potentially variable parameters that
ki ¼ Ai  exp ð11Þ
RT may be estimated via regression. One’s ability to regress a binding
energy for a gas-phase species relies on data being collected over a
!
ekB T DSzi large range of partial pressures, coverage regimes, and/or reaction
Ai ¼ exp ð12Þ orders for the adsorbing species. We satisfy these criteria for H2
h R
and acetone, but our reactors always operate at low conversion
    and thus low isopropanol pressures. Further, our analyses always
DSi DHi apply the differential reactor approximation, i.e., we assume that
K i ¼ exp  exp ð13Þ
R RT reaction rates are constant and defined by the reactor inlet condi-
Using the above parameterization, solution of the microkinetic tions, where isopropanol partial pressures are zero. We are thus
model requires a complete definition of entropic and enthalpic not confident in our ability to regress a binding energy for iso-
changes of reaction (DSi, DHi) and activation (DSài , EA,i) for each ele- propanol, and we define the isopropanol binding energy as an
mentary step. Our approach for doing so is specified in the follow- explicit function of the acetone binding energy based on trends
ing sections. We note that initial estimates for single-site and two- observed during surface science experiments. Specifically, on a Pt
site mechanisms are identical because the nature of each elemen- (1 1 1) surface, Vannice has reported binding energies of 48.1
tary step and the species involved therein do not change between [65] and 55.2 [66] kJ mol1 for acetone and isopropanol, respec-
the two mechanisms; they only differ as to whether or not hydro- tively. Similarly, Sexton reported binding energies of 46.2 [67]
gen atoms and organic species compete for adsorption at a single and 52.7 [68] kJ mol1 for acetone and isopropanol. Finally, Avery
type of active site or if they bind at distinct types of active sites. reported the binding energy of acetone on Pt(1 1 1) to be
48.5 kJ mol1 [69]. Based on these observations, our initial esti-
3.8.2. Thermodynamics of elementary reactions mate of the acetone binding energy is 49 kJ mol1, and our
We rigorously apply the constraint specified in Eq. (9) by defin- microkinetic model constrains the isopropanol binding energy to
ing thermodynamic state functions for all surface species relative be 110% of the acetone binding energy. We estimate the binding
to their gas phase values. Relevant gas-phase thermodynamic energy of an isopropoxy species (190 kJ mol1) from that of meth-
properties include enthalpies, entropies, and heat capacities; they oxy on Pt (1 1 1) [70]. We account for the additional Van der Waals
are summarized for all reactants, intermediates, and products as stabilization provided by the longer carbon chain of isopropoxy
pure, gas-phase species at 1 bar and 298 K (Table 5). Data for H2, using the correlation given by Sexton (5.4 kJ mol1 per carbon in
ACE, IPOH, and atomic hydrogen (H) are taken from existing data- longest side chain) [68]. Initial Binding Energy estimates, enthal-
bases. The enthalpy of formation for a gas-phase isopropoxy radi- pies of adsorbed species at 298 K, and reference sources are sum-
cal was computed from the enthalpy of formation of isopropanol, marized in Table 6.
the enthalpy of formation for a hydrogen atom, and the homolytic Next, we consider estimation of entropies for surface species. In
bond dissociation energy (BDE) of O–H, which is 443 kJ mol1. prior microkinetic analysis, we assumed that bound species lose all
translational entropy upon adsorption [52]. Although this
HIPOX ¼ HIPOH þ BDEOH  HH ð14Þ approach can provide a reasonable data fit [72], recent studies by
The entropy and heat capacity for gas-phase isopropoxy at Árnadóttir [73], Schneider [74], and Iglesia [75] indicate that sur-
298 K were estimated using Gaussian 09 software [51]. face species retain a significantly larger fraction of their gas-
Enthalpies and entropies for each gas-phase species are phase entropies than allowed by our previous assumptions. Camp-
adjusted to reaction temperature using Eqs. (15) and (16). We bell reported similar observations from calorimetry data, and he
assume that heat capacities are temperature-independent in our developed a useful correlation for estimating surface entropies
experimental range (303–435 K). for molecular adsorbates (e.g., hexane, ethanol, etc.) as a function
  
of their gas phase entropies [76]. Unfortunately, it is unclear
H j ð T Þ ¼ H j þ C pj T  T ð15Þ whether these correlations extend to molecular fragments, such
as the alkoxy and atomic hydrogen intermediates that form on
 
 T the surface during acetone hydrogenation, and we are hesitant to
Sj ðT Þ ¼ Sj þ CPj  ln  ð16Þ apply them universally in our analysis. A specific concern is that
T
radical species bind strongly relative to molecular analogs [77],
Enthalpies for surface-bound species are calculated by subtract- and one might expect their mobility to scale accordingly, i.e.,
ing the binding energy for that species from its gas-phase enthalpy strongly-bound species should be less likely to diffuse across a
at a given temperature. We assume that the binding energy (BE) for surface, whereas a weakly-bound molecular adsorbate might have
a given species (j) at a given surface site (x) is not a function of tem- relatively unconstrained translational mobility in two dimensions.
perature or coverage.

Table 5 Table 6
Gas phase thermodynamic data (298 K). Binding energies and surface enthalpies for acetone, isopropoxy, atomic hydrogen,
Species H°j (kJ mol1) S°j (J mol1 K1) CPj (J mol1 K1) and isopropanol.

Acetone 217.1 295.5 75.02 Species BEj,x (kJ mol1) Hj,x (kJ mol1, 298 K) Reference
H2 0 130.7 28.84 Acetone 49 266 [65]
IPOH 272.7 309.2 89.32 IPOX 190 238 [68,70]
H 218.0 114.7 20.79 H 240 22 [71]
IPOX 47.70 296.7 77.78 IPOH 54 326 [66,68]
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 191

Eq. (18)–(24) summarize our approach for building this rationale An important aspect of this approach is that it provides a frame-
into the calculation of surface entropies for species formed during work for computing surface entropies as an explicit function of
acetone hydrogenation. First, we assume that surface species lose gas-phase entropies and species binding energies. This allows
all of their gas-phase translational mobility and thus their transla- one to estimate differences in surface mobility between strongly
tional entropy; however, they retain their gas-phase ‘‘local” and weakly bound adsorbates without introducing additional vari-
entropy, which is comprised of all gas-phase rotational and vibra- able parameters into the microkinetic model. We summarize the
tional modes [72,78]. quantities used to evaluate the above equations alongside surface
entropies for bound species at 298 K in Table 7. Though this
Sj; LOC ¼ Sj  S3D
j;TRAN  Sj;ELEC ð18Þ method invokes many approximations, it results in a surface
For gas phase species, the total (3D) translational entropy is entropy for atomic hydrogen that is consistent with prior kinetic
computed as a function of atomic mass (mj) at the reaction temper- analyses on Pt, Ru, Cu, and Fe [57,81–87]. Further, it yields surface
ature (T) and 1 bar reference pressure (P0). entropies for acetone, isopropoxy, and isopropanol that are compa-
 3 ! ! rable to those estimated using Campbell’s empirical correlation
2pmj kB T 2 kB T 5 [76]. The exception (relative to Campbell’s method) is that, by vir-
S3D
j;TRAN ¼ R  ln  þ ð19Þ
h
2 P0 2 tue of isopropoxy’s large binding energy compared to acetone and
isopropanol, it has a lower surface entropy than either of the
Electronic entropy is a function of the degeneracy of the elec- molecular adsorbates (see Table 7).
tronic ground state (gj). Acetone, isopropanol, and dihydrogen have This approach uses the harmonic oscillator approximation to
a degeneracy of 1, while atomic hydrogen and isopropoxy have an estimate the entropy associated with hindered translations in
unpaired electron and a degeneracy of 2. Electronic entropy con- two dimensions, and the accuracy of that assumption is a matter
tributes little to the total entropy of gas phase species in this sys- of current debate. For example, multiple studies suggest that the
tem (Eq. (20)), and its inclusion in the local entropy calculation harmonic oscillator approximation will, in certain cases, under-
does not substantially change our results. Regardless, we consider predict entropic contributions from translational motion of surface
it in our calculations to explicitly acknowledge differences in species [73–75]. Árnadóttir has considered the topic in detail, ulti-
entropy contributions between gas-phase species with unpaired mately suggesting that one can estimate the threshold of validity
electrons and their surface counterparts. for the harmonic oscillator approximation by comparing the diffu-
  sion barrier to the temperature where adsorption is considered. If
Sj;ELEC ¼ R  ln g j ð20Þ the diffusion barrier for an adsorbate significantly exceeds kT, then
the translational mobility of that adsorbate can be represented as
The three translational degrees of freedom lost upon adsorption low frequency vibrational modes within the harmonic oscillator
are treated as new vibrational modes [79]. One is normal to the approximation. Taking surface science binding energies (Table 6)
metal surface, and it reflects the vibrational frequency of the Pt- as a starting point; using the estimate that diffusion barriers are
adsorbate bond. We assume that this frequency is relatively high approximately 12% of the binding energy [80]; and considering
such that it contributes little entropy to the bound adsorbate [75], that our maximum reaction temperature in this study was approx-
and we exclude it from consideration. The remaining two modes imately 430 K, we estimate that the diffusion barrier for the most
comprise hindered translations and are treated as surface diffusions weakly bound species in this system—acetone—will exceed kT by
in the xy plane. For large and/or weakly bound species, these should a factor of 1.6 at our highest reaction temperature. This suggests
be very low frequency modes that contribute significantly to the the harmonic oscillator approximation is adequate for quantifying
total entropy of the adsorbate. To quantify these vibrational fre- entropic contributions from hindered translations for acetone,
quencies, we employ a correlation reported by Mavrikakis and even at our highest reaction temperature. Because acetone binding
Dumesic, which defines them as a function of the diffusion barrier at 430 K represents the smallest multiple of kT in our system, we
as shown in Eq. (21) [79]. Per Mavrikakis, we estimate the diffusion conclude that a universal application of this method is reasonable
barrier for species j to be 12% of its binding energy [80]. in our analysis.
The thermodynamic approximations detailed above allow com-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0:12  BEj plete quantification of enthalpies and entropies for all gas-phase
mj;DIFF ¼ 2
ð21Þ and surface species at any reaction temperature. This is sufficient
2l mj
to calculate enthalpies of reaction, entropies of reaction, free ener-
gies of reaction, and equilibrium constants for each elementary
To give a frequency in Hz, the binding energy for each species step summarized in Fig. 1, i.e., it comprises a full initial thermody-
(BEj) reflects that of a single molecule and is expressed in Joules, namic parameterization of the microkinetic model. Importantly,
the unit cell edge length (l) is given in m, and the mass of each spe- using the correlations and conventions described above, the vari-
cies (mj) is that of a single molecule expressed in kg. We assume able set of thermodynamic parameters are restricted to three bind-
the unit cell edge length for Pt is 3.0  1010 m. We then use this ing energies—acetone, the hydrogen atom, and the isopropoxy
frequency to compute the vibrational entropy associated with each radical.
diffusion mode.
  
Xj    3.8.3. Kinetics of elementary reactions
Sj;DIFF;1D ¼ R   ln 1  exp X j ð22Þ Rate constants for each elementary step are defined in terms of
expðX j Þ  1
two elementary parameters: the entropy of activation and the acti-
hmj;DIFF vation barrier (Eq. (11) and Eq. (12)). Molecular adsorption steps (1
Xj ¼ ð23Þ and 5) are assumed to be non-activated. Barriers for bond-breaking
kB T
and bond-forming steps are estimated using UBI-QEP correlations
Once the diffusional entropy is estimated for an adsorbate in a [77], which define the activation barrier for an elementary step
single dimension, the total entropy of the bound species is given by as a function of its reaction enthalpy and the binding energies of
Eq. (24). species participating in that step. This allows one to estimate ele-
Sj;x ¼ Sj;LOC þ 2  Sj;DIFF;1D ð24Þ mentary activation barriers without adding variable parameters
to the microkinetic model. In Table 8, UBI-QEP expressions are
192 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

Table 7 3.8.4. Data reconciliation


Approximation of diffusion modes and total surface entropies at 298 K. The above sections describe the initial kinetic and thermody-
Species mj,Diff (Hz)
m 1
j;Diff (cm )
Sj,Diff,1D Sj,x namic parameterization of microkinetic models that reflect ace-
tone hydrogenation in both single- and two-site mechanisms. We
Acetone 8.30  1011 25 25.9 187.8
H 1.26  1013 420 3.73 7.46
next apply limiting assumptions to develop 16 models that
IPOX 1.47  1012 49 20.3 171.9 describe macroscopic isopropanol site time yields as a function of
IPOH 8.81  1011 26 25.6 200.6 temperature, pressure, and species composition in various rate
control scenarios. A detailed explanation of the assumptions that
underlie each scenario is presented alongside the derivation of
Table 8 overall rate expressions in the online supporting information. Each
UBI-QEP estimates for elementary activation barriers for acetone hydrogenation on
model is developed in terms of a maximum of eight uncertain and
Pt.
potentially variable parameters: binding energies for acetone,
Step UBI-QEP correlation EA,i (298 K) atomic hydrogen, and the isopropoxy radical (isopropanol binding
!
1 1 ðBEH Þ2 38 energies are developed as an explicit function of acetone binding
EA;1 ¼  DH 1 þ
2 2  BEH energies), as well as activation entropies for each of the five ele-
2 – 0 mentary steps. In general, we regress all three binding energies
 
3 1 BEA  BEH 46 and the activation entropy for any rate constant appearing in the
EA;3 ¼ DH3 þ  DH3 þ
2 BEA þ BEH overall rate expression. Where steps are assumed to be equili-
 
4 1 BEIPOX  BEH 20
EA;4 ¼ DH4 þ  DH4 þ brated, the model is insensitive to their activation entropies; thus,
2 BEIPOX þ BEH
5 – 0 they cannot be regressed with any precision, and they are excluded
from parameter estimation. Results of regression analysis are sum-
marized in Table 10 and illustrated in Fig. 4.
summarized for relevant steps alongside their estimated barriers, Considering the single-site model, we observe relatively poor
which are based on initial binding energies; all are generally rea- fits (R2 < 0.90) for solutions that attribute rate control to H2 disso-
sonable in light of prior DFT and/or microkinetic analyses ciation (Case 1), acetone adsorption (Case 2), addition of atomic
[37,88]. For example, Alcalá has reported DFT barriers of hydrogen to surface acetone (Case 3), and addition of atomic
61 kJ mol1 and 18 kJ mol1 for Steps 3 and 4 (respectively) during hydrogen to surface isopropoxy (Case 4). For the single-site model,
acetone hydrogenation on a Pt(1 1 1) surface [37]. Bearing in mind the only single RDS scenario that provides a reasonable fit to data
the anticipated minimum precision of DFT barriers (0.1 ev), these (R2 > 0.9) is the one limited by isopropanol desorption (Case 5).
numbers are in excellent agreement with those derived here using Case 6 applies the pseudo-steady state approximation to surface
surface science binding energies and UBI-QEP correlations. isopropoxy and further assumes that all adsorption steps are equi-
Eq. (25) illustrates the method used for estimating activation librated. This relaxes the constraint that a single step is rate con-
entropies for gas adsorption steps (1, 2, and 5). We equated the trolling and allows the possibility that both surface reactions are
pre-exponential factor given by collision theory with that derived kinetically significant. We therefore regress activation entropies
from transition state theory to solve for the activation entropy for both of these steps, increasing the total number of variable
[56]. The calculation was performed at 298 K and a reference pres- parameters to five. Despite this, the fit remains poor (R2 = 0.858),
sure (P0) of 1 bar. We have assumed the cross sectional area of a and we observe only a modest improvement in residual error rel-
surface site (Ac) is 1.5  1019 m2. We additionally assume a stick- ative to Cases 3 and 4, both of which correspond to full rate control
ing coefficient (Sc) of 1.0 for molecular adsorption of acetone and by a single surface reaction. Case 7 applies the pseudo-steady state
isopropanol and a sticking coefficient of 0.08 for dissociative approximation to surface isopropanol and assumes that steps 1, 2,
adsorption of hydrogen [89]. These assumptions result in activa- and 3 are equilibrated. This allows the possibility that the second
tion entropies for gas adsorption that are on the order of surface reaction and isopropanol desorption are both kinetically
100 J mol1 K1 and adsorption pre-exponential factors of significant. For this case, we regressed activation entropies for steps
roughly 108 sec1. Both align with expectations [56]. 4 and 5 (5 total parameters); however, the optimal parameter set
!
h A c P 0 Sc predicts that the rate of hydrogenation is fully controlled by iso-
DSzi;ADS ¼ R  ln  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð25Þ propanol desorption, and we observe no improvement over the
e  kB T 2pmj kB T
desorption-limited solution (Case 5). Finally, Case 8 applies the
We initially estimate that activation entropies for the two sur- pseudo-steady state approximation to both surface isopropoxy
face reactions (Steps 3 and 4) are 50 J mol1 K1. This results in and surface isopropanol, and it assumes that acetone and hydrogen
pre-exponential factors of roughly 1010 s1, which are reasonable adsorption (Steps 1 and 2) are equilibrated. This allows that both
for bimolecular surface reactions [90]. We are unaware of any cor- surface reactions and isopropanol desorption can be kinetically sig-
relations that allow one to couple activation entropies to binding nificant (6 parameters). Despite this relatively flexible scenario, we
energies, so we take activation entropies as potentially variable were unable to find an optimum that improved upon the minimum
parameters that may be estimated through regression analysis. Ini- residual relative to Cases 5 and 7. For the single-site mechanism,
tial activation entropies, alongside their corresponding pre- Case 5, which assumes that isopropanol desorption fully controls
exponential factors are summarized in Table 9. the rate of acetone hydrogenation, achieves the minimum AIC value
and represents the statistically preferred rate control scenario.
Considering the two-site mechanism, we observe poor fits for
Table 9 models that ascribe rate control to H2 dissociation (Case 9) and
Initial Estimates for elementary activation entropies and pre-exponential factors.
acetone adsorption (Case 10). Otherwise, addition of atomic hydro-
Step DSài (J mol1 K1) Ai (s1) gen to surface acetone (Case 11), addition of atomic hydrogen to
1 98.0 1.29  108 surface isopropoxy (Case 12), and desorption of isopropanol (Case
2 90.9 3.00  108 13) all capture trends in experimental observations reasonably
3 50.0 4.13  1010 well (R2 > 0.9). Case 14 applies the pseudo-steady state approxima-
4 50.0 4.13  1010
tion to surface isopropoxy, which allows that both surface
5 91.1 2.95  108
reactions are potentially kinetically significant and adds an
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 193

Table 10
Results from Least Squares Regression for all rate control scenarios.

Mechanism Case Variable parameters SSE AIC R2 Details


Single-site 1 4 5.48 323 0.850 H2 adsorption is RDS
2 4 59.9 765 0.637 Acetone adsorption is RDS
3 4 9.78 430 0.733 Surface Reaction (1) is RDS
4 4 7.44 379 0.797 Surface Reaction (2) is RDS
5 4 2.39 169 0.935 IPOH Desorption is RDS
6 5 5.18 314 0.858 PSSH on IPOX
7 5 2.40 172 0.935 PSSH on IPOH
8 6 5.76 336 0.843 PSSH on IPOX and IPOH
Two-site 9 4 5.48 323 0.850 H2 adsorption is RDS
10 4 59.9 765 0.637 Acetone adsorption is RDS
11 4 1.66 101 0.955 Surface Reaction (1) is RDS
12 4 1.73 109 0.953 Surface Reaction (2) is RDS
13 4 2.39 169 0.935 IPOH Desorption is RDS
14 5 1.52 87.6 0.958 PSSH on IPOX
15 5 1.73 111 0.953 PSSH on IPOH
16 6 1.52 89.6 0.958 PSSH on IPOX and IPOH

PSSH = Pseudo-Steady State Hypothesis.

Fig. 4. (a) Minimum Residual Sum of Squares and (b) Akaike Information Criterion for all proposed rate control scenarios during acetone hydrogenation on single-site
(pattern fill) and two-site (solid fill) mechanisms.

additional variable parameter (5 parameters). For this case, we 12, indicating no statistical improvement and that the additional
observe a significant decrease in the minimum residual error parameter is unwarranted. Finally, Case 16 applies the pseudo-
(SSE = 1.52) relative to Case 11 and Case 12, which each assign rate steady state approximation to both the surface isopropoxy and sur-
control to a single surface reaction (SSE11 = 1.66; SSE12 = 1.73). This face isopropanol, allowing that that both surface reactions and iso-
decrease in residual error is sufficient to offset the penalty for an propanol desorption are potentially kinetically significant (6 total
increased number of free parameters such that there is a commen- parameters). Despite the increase in free parameters, this model
surate decrease in the AIC value (87.6) relative to the single- achieves the same minimum residual as observed in Case 14; thus,
reaction rate control scenarios in Case 11 and 12 (AIC11 = 101, there is no improvement in the model’s ability to describe experi-
AIC12 = 109). Case 15 applies the pseudo-steady state approxima- mental trends, and the AIC value increases. On the whole, our data
tion to surface isopropanol, allowing that hydrogen addition to appears to be better described by a two-site mechanism than a
the isopropoxy and isopropanol desorption can both be kinetically single-site mechanism. From a statistical perspective, the optimal
significant. Similar to the previous PSSH case, this introduces an model is the one that achieves a minimum value of AIC, which is
additional variable parameter (5 total). Despite this, we observe Case 14. This solution is based on a two-site mechanism, applies
that the minimum residual error (SSE15 = 1.73) is the same as that the pseudo-steady state approximation to surface isopropoxy,
observed in Case 12. Effectively, the optimal PSSH solution here is and assumes all adsorption/desorption processes are equilibrated.
driven to the limiting solution where the second surface reaction is The optimal parameter set for this model is summarized in
fully rate determining. Since there is no decrease in minimum Table 11. The parity plot generated using the Case 14 model and
residual error, there is an increase in the AIC value relative to Case the optimized parameter set is illustrated in Fig. 5.
194 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

Table 11 that pre-exponential factors/activation entropies are perhaps the


Optimal parameter set obtained by least squares solution of case 14. most difficult kinetic parameters to accurately regress from kinetic
Parameter Initial value Regressed value Units data, and we highlight areas of uncertainty that may impact their
BEACE 49 67 ± 2.5 kJ mol1 values. The absolute magnitude of pre-exponential factors is sensi-
BEH 240 237 ± 6.7 kJ mol1 tive to the absolute magnitude of the turnover frequency; thus, any
BEIPOX 190 227 ± 25 kJ mol1 inaccuracy in experimental site-time-yield estimation will trans-
DSà3 0.05 0.143 ± 0.029 kJ mol1 K1 late to inaccuracies in regressed pre-exponential factors. While
DSà4 0.05 0.051 ± 0.034 kJ mol1 K1
our determination of reaction rates (±10%) and CO uptake (±5%)
are relatively precise, one may raise questions about the validity
of using CO chemisorption as a basis for defining site-specific reac-
tion rates. Equating total CO uptake with available Pt surface area
is common practice; however, CO is significantly smaller than ace-
tone, isopropanol, and isopropoxy. This may result in fewer steric
restrictions and/or weaker lateral interactions for CO relative to
the larger hydrocarbons at a given surface coverage, which may
cause CO chemisorption to over-estimate the Pt surface area that
is available for acetone hydrogenation. In addition, our microki-
netic analysis tacitly invokes the mean field approximation, which
assumes a random organization of adsorbates on catalytic surfaces.
Surface science experiments have shown that, for certain metal
catalyzed reactions, adsorbates may instead segregate into ‘‘is-
lands,” where one might expect the majority of hydrogenation to
occur at the perimeter [64,91–93]. Again, CO chemisorption would
likely over-estimate the Pt surface area available for reaction in
this case. Both of the above situations are realistic, and either
would cause site-time-yields based on ex-situ CO chemisorption
to under-estimate the absolute magnitude of the experimental
turnover frequency. During regression, this would manifest as
unfavorable activation entropies and pre-exponential factors that
are low relative to expectations.
Aforementioned concerns over the absolute magnitude of the
Fig. 5. Parity plot illustrating level of agreement between model predictions and
activation entropy aside, the model presented in Case 14, along
experimental measurements during acetone hydrogenation over Pt. with the regressed parameter set, is adequate for capturing exper-
imental trends in acetone hydrogenation; accordingly, we examine
its predictions for degree of rate control over a large range of
For a more tangible set of kinetic and thermodynamic parame- experimental conditions (well beyond those where have collected
ters, Table 12 compiles the corresponding pre-exponential factors; data). We define the Degree of Rate Control, XRC, for elementary
activation barriers; enthalpies, entropies, and free energies of reac- step, i, per Campbell as shown in Eq. (26) [94].
tion; rate constants; and equilibrium constants for each elemen-    
ki dr
tary step in the proposed mechanism. All were computed at XRC i ¼  ð26Þ
298 K using the methods summarized in section 3.8 and the opti- r dki
mal parameter set summarized in Table 11. For comparison, the Degree of rate control was evaluated for each step by introduc-
overall thermodynamics of acetone hydrogenation are also given, ing a 1% perturbation to the rate constant for that step (ki) while
and one observes that the sum of each step weighted by its stoi- holding constant the equilibrium constant for that step (Ki) and
chiometric number returns the value of the overall reaction the rate constant for all other steps (kj – i). Results are summarized
(r1 = r2 = r3 = r4 = 1; r5 = 1). in Fig. 6. Interestingly, our statistically optimal model predicts that,
With the exception of the activation entropy for step 3, in general, both surface reactions are kinetically significant and the
regressed and calculated parameters appear reasonable based on degree of rate control from each is sensitive to operating condi-
preliminary estimates, general expectations for acetone hydro- tions. Indeed, one can only identify regimes where a single surface
genation on Pt, and prior computational analyses for ketone hydro- reaction exerts full rate control at extremely high or extremely low
genation on metals. Our view is that the optimum activation H2 pressures, which suggests that the assumption of a single, uni-
entropy for step 3 is excessively unfavorable versal rate determining step may be invalid during acetone hydro-
(0.143 ± 0.029 kJ mol1 K1). This results in a pre-exponential genation on Pt.
factor on the order of 106 s1, which falls outside the range gener- By far, the most dramatic shifts in rate control occur with mod-
ally expected for a surface reaction [64]. It is worth mentioning ulation of the hydrogen partial pressure. We attribute these

Table 12
Kinetic and Thermodynamic quantities calculated at 298 K for each elementary step in optimal model solution.

Step A,f (s1) EA,f (kJ mol1) DHrxn (kJ mol1) DSrxn (kJ mol1 K1) DGrxn (kJ mol1) kf (s1) K
8
1 1.29  10 40.0 38.7 0.116 4.17 12.6 5.4
2 3.00  108 0 66.8 0.110 33.4 3.0  108 8.8  105
3 5.72  105 40.4 28.6 0.022 35.2 4.8  102 6.6  107
4 3.66  1010 31.9 52.3 0.020 58.2 9.6  104 1.6  1010
5 2.95  108 0 73.5 0.111 40.3 3.0  108 1.2  107
Overall – – 55.6 0.117 20.8 – 4.3  103
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 195

Fig. 6. Degree of Rate Control for step 3 (solid lines) and step 4 (dashed lines) for a two-site acetone hydrogenation mechanism. Results were generated using the optimal
parameter set in Case 14, which applies the pseudo-steady state approximation to the surface isopropoxy intermediate. Case 14 also assumes that adsorption-desorption
processes are equilibrated such that steps 1, 2, and 5 have a zero degree of rate control. (a) XRC as a function of Temperature; PACE = 4 Torr, pH2 = 750 Torr, (b) XRC as a
function of Acetone partial pressure; T = 323 K, pH2 = 750 Torr, (c) XRC as a function of hydrogen partial pressure; pA = 4 Torr, T = 323 K, (d) XRC as a function of Temperature;
pACE = 4 Torr, pH2 = 37.5 Torr.

impacts to significant changes in surface coverage under different control lies primarily with Step 4: addition of a hydrogen atom
operating conditions (Fig. 7). We observe that acetone is always to surface isopropoxy. In contrast, in regimes where hydrogen cov-
the dominant species on * sites, while the coverage of isopropoxy erages approach 1, rate control shifts toward Step 3: addition of a
is always low on * sites. This seems counterintuitive because iso- hydrogen atom to surface acetone. Hydrogenation of acetone on a
propoxy has a large binding energy (227 kJ mol1) relative to ace- Pt surface (Step 3) has a relatively large free energy of activation
tone (67 kJ mol1); however, its large binding energy is offset by and is therefore kinetically challenging, whereas hydrogenation
the fact that the gas-phase isopropoxy radical has a substantially of isopropoxy on a Pt surface (Step 4) has a relatively small free
higher enthalpy (47.7 kJ mol1) than gas-phase acetone energy of activation and is kinetically facile. Accordingly, at our
(217 kJ mol1). Consequently, bound acetone and bound iso- typical reaction temperatures, the rate constant for isopropoxy
propoxy have comparable surface enthalpies, and the addition of hydrogenation (Step 4) exceeds that of acetone hydrogenation
a hydrogen atom to a surface acetone (Step 3) is enthalpically (Step 3) by 6–7 orders of magnitude (Table 12). Under conditions
and entropically unfavorable (DH3 = 28.6 kJ mol1, DS3 = where hydrogen atoms and/or isopropoxy species are abundant
22 J mol1 K1). Ultimately, the free energy penalty of hydro- on the catalyst surface, rate control is dictated by free energies of
genating surface acetone severely restricts the maximum iso- activation, and the more kinetically demanding reaction (Step 3)
propoxy coverage under our typical reaction conditions. In controls the overall reaction rate. Conversely, under conditions
regimes where hydrogen coverages are also low (e.g., at very low where isopropoxy and/or hydrogen coverages are sparse (but ace-
hydrogen partial pressures or at high reaction temperatures), rate tone coverages remain high), the kinetically facile step (Step 4)
196 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

Fig. 7. Fractional surface coverages of acetone (solid lines), atomic hydrogen (dotted lines), and isopropoxy (dashed lines) for a two-site acetone hydrogenation mechanism.
Results were generated using the optimal parameter set in Case 14, which applies the pseudo-steady state approximation to the surface isopropoxy intermediate. Case 14 also
assumes that adsorption-desorption processes are equilibrated such that steps 1, 2, and 5 have zero degree of rate control. (a) Fractional Coverages as a function of
Temperature; PACE = 4 Torr, pH2 = 750 Torr, (b) Fractional Coverages as a function of Acetone partial pressure; T = 323 K, pH2 = 750 Torr, (c) Fractional Coverages as a function
of hydrogen partial pressure; pA = 4 Torr, T = 323 K, (d) Fractional Coverages as a function of Temperature; pACE = 4 Torr, pH2 = 37.5 Torr.

becomes rate controlling because there is a small driving force or isopropanol desorption, which are the only two steps that should
(chemical potential) for reaction. This highlights the importance display no primary kinetic isotope effect when switching the reduc-
of a holistic consideration of activation parameters, surface cover- ing gas from H2 to D2. We attempt to resolve this below.
ages, and/or species partial pressures in assessing rate control, and The statistically optimal kinetic model was developed by apply-
it provides an interesting illustration of how rate controlling steps, ing the pseudo-steady state approximation to the isopropoxy
much like apparent barriers and reaction orders, may change intermediate while assuming that all adsorption-desorption pro-
depending on regimes where data are collected. cesses were equilibrated. The overall rate of hydrogenation antici-
The above analysis suggests that acetone hydrogenation can be pated in this scenario is given by Eqs. (27)–(34), and we observe
effectively modelled using a two-site Horiuti-Polanyi mechanism that it has a complex dependence on the equilibrium constants
by assuming that all adsorption processes are equilibrated and that for Steps 1, 2, and 5 as well as forward and reverse rate constants
each of the two surface reactions partially controls the rate of reac- for steps 3 and 4. Further, since we always define reverse rate con-
tion. While these assumptions capture trends in data, the conclu- stants as a function of forward rate constants and equilibrium con-
sion does not immediately reconcile with our observation that an stants, the overall hydrogenation rate has an implicit dependence
H2/D2 switch has a minor impact on observed rates of acetone on the equilibrium constants for steps 3 and 4.
hydrogenation (Section 3.7). Each surface reaction involves the for- 1 1

mation of an H-X bond; thus, the lack of a prominent kinetic iso- hHs ¼ K 21 a2H2  hs ð27Þ
tope effect argues against either of these steps being rate
controlling. That said, regression analyses show that, within the hACE ¼ K 2  aACE  h
ð28Þ
set of constraints outlined in section 3.8, we are unable to satisfac-
torily reconcile our data with kinetic control by acetone adsorption hIPOH ¼ K 5  aIPOH  h
ð29Þ
X. Gao et al. / Journal of Catalysis 374 (2019) 183–198 197

1 these observations, * sites must be acetone saturated and s sites


hs ¼ 1 1
ð30Þ
1 þ K 21  a2H2 must be mostly vacant, which generally agrees with our simulated
coverage regimes under reaction conditions (e.g., see Fig. 7 at near-
ambient temperatures and H2 pressures). In this limit, the rate
hIPOX ¼ A  h
ð31Þ
expression reduces to Eq. (36).
 
k3f  K 2  aACE  hHs þ k4r  K 5  aIPOH  hs 1 1
A¼ ð32Þ r ¼ k3f  K 21  a2H2 ð36Þ
k3r  hs þ k4f  hHs

1 For a fixed pressure of H2/D2, the experimentally observed KIE


h
¼ ð33Þ would be given by Eq. (37), which depends on the rate constant
1 þ K 2  aACE þ A þ K 5  aIPOH
for step 3 as well as the equilibrium constant for step 1 (H2 or D2
r ¼ k4f  hIPOX  hHs  k4r  hIPOH  hs ð34Þ dissociation).

One expects that replacing H with D will decrease forward    1


k3f ;H K 1;H 2
and reverse rate constants for both surface hydrogenations KIEA ¼  ð37Þ
k3f ;D K 1;D
(k3 and k4) as each involves the formation of an H-X bond. The
exact magnitude of this decrease will depend upon mass differ-
One generally expects that a change from H to D will decrease
ences in the isotopologues involved in the reaction; differences
the magnitude of the elementary rate constants for Step 3, i.e., this
in zero-point vibrational energies of the reactants, transition
step involves formation of an H-X bond and so should display a pri-
states, and products; and the position of the transition state
mary kinetic isotope effect. However, it is also realistic that replac-
along the reaction coordinate (i.e., early vs. late transition state).
ing H with D can increase the magnitude of the equilibrium
Owing to this complexity, there is no reason to expect that each
constant for Step 1, which may offset the anticipated kinetic iso-
surface reaction should exhibit an identical primary kinetic iso-
tope effect. For example, for a fixed H/D binding energy, Pt-D bonds
tope effect, and it is hard to predict the exact magnitude without
should have lower frequency vibrational modes than Pt-H bonds,
a full vibrational mode analysis for all surface species. That said,
which could result in D2 dissociation being more thermodynami-
one typically expects that a switch from H to D will decrease the
cally favorable than H2 dissociation (K1,H < K1,D). In the case where
rate constant for a reaction directly involving an H-X bond by a
replacing H with D causes a factor-of-two decrease in the rate con-
factor of 2–6 [95]; however, one may not observe the full impact
stant for step 3 and a factor-of-two increase in the equilibrium
in experimentally measurable hydrogenation rates [59–63]. A
constant for step 1, one would observe an apparent KIE of  1.4,
switch from H to D may also perturb equilibrium constants for
which is significantly diminished from the true KIE and in line with
any reactions where reactants and products are deuterated to
our experimental observation of a low temperature KIE of  1.2.
different extents. For example, we estimate that the equilibrium
This example is, of course, speculative. It is only intended to high-
constant for the gas-phase hydrogenation of acetone is roughly
light the fact that observed kinetic impacts of isotope exchange are
1–3 times larger than the analogous gas-phase deuteration of
often non-elementary and may represent a convolution of many
acetone (See Online supporting information) at our typical reac-
fundamental impacts of changes in isotope mass. Ultimately, a full
tion temperatures. In our proposed mechansim, one would
vibrational mode analysis of adsorbed intermediates and transition
expect such Thermodynamic Isotope Effects to arise in Step 1
states is necessary to determine whether rate control by surface
(H2 dissociation); Step 3 (hydrogen addition to surface acetone);
reactions can reconcile with the lack of an apparent kinetic isotope
and Step 4 (hydrogen addition to surface isopropoxy). In con-
effect upon switching the reducing gas from H2 to D2 during ace-
trast, the reaction thermochemistry for Step 2 (Acetone adsorp-
tone hydrogenation on Pt/SiO2.
tion) and Step 5 (Isopropanol Adsorption) should not be
strongly impacted by switching the reducing gas between H2
and D2. Equilibrium constants for steps 1, 3, and 4 appear 4. Conclusion
throughout our PSSH solution; thus, it is reasonable to expect
that thermodynamic isotope effects will contribute to observed Microkinetic analysis of acetone hydrogenation over Pt/SiO2
changes in hydrogenation rates upon H2/D2 switching. Depend- suggests that observed kinetic trends are consistent with a
ing on the magnitude of both kinetic and thermodynamic isotope Horuiti-Polanyi type mechanism involving two distinct active sites.
effects as well as the surface coverage under reaction conditions, Rate control in this system appears to lie primarily with surface
the impact of isotope substitution may largely cancel from reactions, namely the addition of hydrogen atoms to bound ace-
experimental observations. Unfortunately, the PSSH solution tone and isopropoxy fragments. Both steps appear kinetically sig-
described in Eqs. (27)–(34) does not lend itself to a concise rate nificant, and the degree of rate control exerted by each shifts
expression, which makes it difficult to visualize the tension with operating conditions, particularly the hydrogen partial pres-
between kinetic and thermodynamic isotope effects. As a simple sure. This result highlights the importance of considering both free
illustration, we consider the limiting case where the rate of ace- energies of activation (i.e., rate constants) and potentials of react-
tone hydrogenation is entirely controlled by addition of a hydro- ing species (e.g., surface coverages and partial pressures) in identi-
gen atom to a surface bound acetone (Step 3). In the zero- fying a rate controlling step. At first glance, a lack of observed
conversion limit, the isopropanol partial pressure is zero, and kinetic isotope effect upon switching between H2 and D2 stands
the macroscopic rate expression reduces to Eq. (35). in apparent contrast with the conclusion that surface reactions
1 1 are rate controlling; however, observed reaction rates have a com-
k3f  K 21 K 2  aACE  a2H2 plex dependence on multiple elementary rate and equilibrium con-
r¼  1 1
 ð35Þ
ð1 þ K 2  aACE Þ  1 þ K 21  a2H2 stants, any of which may be impacted by a change in isotopic mass.
It is thus plausible that opposing kinetic and thermodynamic
At low temperatures, we observe that acetone hydrogenation is impacts of isotope exchange prevent the observation of a full, pri-
zero order in acetone and half order in hydrogen. To reproduce mary kinetic isotope effect in this system.
198 X. Gao et al. / Journal of Catalysis 374 (2019) 183–198

Acknowledgements [42] B. Sen, M.A. Vannice, J. Catal. 113 (1988) 52–71.


[43] N.V. Pavlenko, A.I. Tripol’skii, G.I. Golodets, Theor. Exp. Chem. 22 (1987) 667–
675.
The authors acknowledge financial support for this work from [44] B.A. Sexton, K.D. Rendulic, A.E. Huges, Surf. Sci. 121 (1982) 181–198.
the National Science Foundation of the United States of America [45] S. Brunauer, P.H. Emmett, E. Teller, J. Am. Chem. Soc. 60 (1938) 309–319.
[46] E.P. Barrett, L.G. Joyner, P.P. Halenda, J. Am. Chem. Soc. 73 (1951) 373–380.
(Award Number CBET-1159739 and CBET-1159863).
[47] J. Freel, J. Catal. 25 (1972) 149–160.
[48] J.A. Dumesic, J. Catal. 185 (1999) 496–505.
Appendix A. Supplementary material [49] M. Boudart, Chem. Rev. 95 (1995) 661–666.
[50] M. Boudart, Chem. Eng. Sci. 22 (1967) 1387.
[51] M.J. Frisch, G.W. Trucks, H.B. Schlegel, G.E. Scuseria, M.A. Robb, J.R. Cheeseman,
Supplementary data to this article can be found online at G. Scalmani, V. Barone, B. Mennucci, G.A. Petersson, H. Nakatsuji, M. Caricato,
https://doi.org/10.1016/j.jcat.2019.04.033. X. Li, H.P. Hratchian, A.F. Izmaylov, J. Bloino, G. Zheng, J.L. Sonnenberg, M.
Hada, M. Ehara, K. Toyota, R. Fukuda, J. Hasegawa, M. Ishida, T. Nakajima, Y.
Honda, O. Kitao, H. Nakai, T. Vreven, J.A. Montgomery, J.E. Peralta, F. Ogliaro, M.
References Bearpark, J.J. Heyd, E. Brothers, K.N. Kudin, V.N. Staroverov, R. Kobayashi, J.
Normand, K. Raghavachari, A. Rendell, J.C. Burant, S.S. Iyengar, J. Tomasi, M.
[1] V.V. Ordomsky, J.C. Schouten, J. van der Schaaf, T.A. Nijhuis, Appl. Catal., A 451 Cossi, N. Rega, J.M. Millam, M. Klene, J.E. Knox, J.B. Cross, V. Bakken, C. Adamo,
(2013) 6–13. J. Jaramillo, R. Gomperts, R.E. Stratmann, O. Yazyev, A.J. Austin, R. Cammi, C.
[2] G.W. Huber, J.N. Chheda, C.J. Barrett, J.A. Dumesic, Science 308 (2005) 1446– Pomelli, J.W. Ochterski, R.L. Martin, K. Morokuma, V.G. Zakrzewski, G.A. Voth,
1450. P. Salvador, J.J. Dannenberg, S. Dapprich, A.D. Daniels, Farkas, J.B. Foresman, J.V.
[3] J.Q. Bond, A.A. Upadhye, H. Olcay, G.A. Tompsett, J. Jae, R. Xing, D.M. Alonso, D. Ortiz, J. Cioslowski, D.J. Fox, Gaussian 09, Revision A.02, in, Wallingford CT,
Wang, T.Y. Zhang, R. Kumar, A. Foster, S.M. Sen, C.T. Maravelias, R. Malina, S.R. 2009.
H. Barrett, R. Lobo, C.E. Wyman, J.A. Dumesic, G.W. Huber, Energy Environ. Sci. [52] J.Q. Bond, C.S. Jungong, A. Chatzidimitriou, J. Catal. 344 (2016) 640–656.
7 (2014) 1500–1523. [53] H. Akaike, IEEE Trans. Autom. Control 19 (1974) 716–723.
[4] H.B. Zhang, C. Canlas, A.J. Kropf, J.W. Elam, J.A. Dumesic, C.L. Marshall, J. Catal. [54] R.J. Madon, M. Boudart, Ind. Eng. Chem. Fundam. 21 (1982) 438–447.
326 (2015) 172–181. [55] M.A. Vannice, Kinetics of Catalytic Reactions, Springer, New York, 2005.
[5] Y. Nakagawa, M. Tamura, K. Tomishige, ACS Catal. 3 (2013) 2655–2668. [56] J.A. Dumesic, D.F. Rudd, L.M. Aparicio, J.E. Rekoske, A.A. Trevino, The
[6] S. Sitthisa, T. Sooknoi, Y.G. Ma, P.B. Balbuena, D.E. Resasco, J. Catal. 277 (2011) Microkinetics of Heterogeneous Catalysis, American Chemical Society,
1–13. Washington, D.C., 1993.
[7] R. Rao, A. Dandekar, R.T.K. Baker, M.A. Vannice, J. Catal. 171 (1997) 406–419. [57] R.M. Rioux, M.A. Vannice, J. Catal. 233 (2005) 147–165.
[8] R.S. Rao, R.T.K. Baker, M.A. Vannice, Catal. Lett. 60 (1999) 51–57. [58] F. Meemken, A. Baiker, J. Dupré, K. Hungerbühler, ACS Catal. 4 (2014) 344–354.
[9] H. Kobayashi, Y. Ito, T. Komanoya, Y. Hosaka, P.L. Dhepe, K. Kasai, K. Hara, A. [59] Y. Bai, B.W.J. Chen, G. Peng, M. Mavrikakis, Catal. Sci. Technol. 8 (2018) 3321–
Fukuoka, Green Chem. 13 (2011) 326–333. 3335.
[10] P. Gallezot, P.J. Cerino, B. Blanc, G. Flèche, P. Fuertes, J. Catal. 146 (1994) 93– [60] H.B. Zhang, M.Y.S. Ibrahim, D.W. Flaherty, J. Catal. 361 (2018) 290–302.
102. [61] S.M. Davis, W.D. Gillespie, G.A. Somorjai, J. Catal. 83 (1983) 131–140.
[11] P. Gallezot, N. Nicolaus, G. Flèche, P. Fuertes, A. Perrard, J. Catal. 180 (1998) 51– [62] E.M. Simmons, J.F. Hartwig, Angew. Chem. Int. Ed. 51 (2012) 3066–3072.
55. [63] M. Ojeda, A. Li, R. Nabar, A.U. Nilekar, M. Mavrikakis, E. Iglesia, J. Phys. Chem. C
[12] R.F. Perez, O. Soares, A.M.D. de Farias, M.F.R. Pereira, M.A. Fraga, Appl. Catal. B – 114 (2010) 19761–19770.
Environ. 232 (2018) 101–107. [64] I. Chorkendorff, J.W. Niemantsverdriet, Concepts of Modern Catalysis and
[13] C. Montassier, J.C. Ménézo, L.C. Hoang, C. Renaud, J. Barbier, J. Mol. Catal. 70 Kinetics, John Wiley & Sons, 2006.
(1991) 99–110. [65] M.A. Vannice, W. Erley, H. Ibach, Surf. Sci. 254 (1991) 1–11.
[14] O.A. Abdelrahman, A. Heyden, J.Q. Bond, ACS Catal. 4 (2014) 1171–1181. [66] M.A. Vannice, W. Erley, H. Ibach, Surf. Sci. 254 (1991) 12–20.
[15] E.I. Gürbüz, E.L. Kunkes, J.A. Dumesic, Green Chem. 12 (2010) 223–227. [67] B.A. Sexton, A.E. Hughes, Surf. Sci. 140 (1984) 227–248.
[16] Y. Nakagawa, S.B. Liu, M. Tamura, K. Tomishige, ChemSusChem 8 (2015) 1114– [68] K.D. Rendulic, B.A. Sexton, J. Catal. 78 (1982) 126–135.
1132. [69] N.R. Avery, Surf. Sci. 125 (1983) 771–786.
[17] R.M. West, Z.Y. Liu, M. Peter, J.A. Dumesic, ChemSusChem 1 (2008) 417–424. [70] E.M. Karp, T.L. Silbaugh, M.C. Crowe, C.T. Campbell, J. Am. Chem. Soc. 134
[18] J.N. Chheda, J.A. Dumesic, Catal. Today 123 (2007) 59–70. (2012) 20388–20395.
[19] P. Maki-Arvela, J. Hajek, T. Salmi, D.Y. Murzin, Appl. Catal. A – General 292 [71] K. Christmann, G. Ertl, T. Pignet, Surf. Sci. 54 (1976) 365–392.
(2005) 1–49. [72] S. Kandoi, J. Greeley, M.A. Sanchez-Castillo, S.T. Evans, A.A. Gokhale, J.A.
[20] J.E. De Vrieze, J.W. Thybaut, M. Saeys, ACS Catal. 8 (2018) 7539–7548. Dumesic, M. Mavrikakis, Top. Catal. 37 (2006) 17–28.
[21] S. Cao, J.R. Monnier, C.T. Williams, W.J. Diao, J.R. Regalbuto, J. Catal. 326 (2015) [73] L.H. Sprowl, C.T. Campbell, L. Árnadóttir, J. Phys. Chem. C 120 (2016) 9719–
69–81. 9731.
[22] W.H. Luo, M. Sankar, A.M. Beale, Q. He, C.J. Kiely, P.C.A. Bruijnincx, B.M. [74] A. Bajpai, P. Mehta, K. Frey, A.M. Lehmer, W.F. Schneider, ACS Catal. 8 (2018)
Weckhuysen, Nat. Commun. 6 (2015). 1945–1954.
[23] A.S. Piskun, J. Ftouni, Z. Tang, B.M. Weckhuysen, P.C.A. Bruijnincx, H.J. Heeres, [75] M. García-Diéguez, D.D. Hibbitts, E. Iglesia, J. Phys. Chem. C (2019).
Appl. Catal. A – General 549 (2018) 197–206. [76] C.T. Campbell, J.R.V. Sellers, J. Am. Chem. Soc. 134 (2012) 18109–18115.
[24] W.H. Luo, U. Deka, A.M. Beale, E.R.H. van Eck, P.C.A. Bruijnincx, B.M. [77] E. Shustorovich, H. Sellers, Surf. Sci. Rep. 31 (1998) 1–119.
Weckhuysen, J. Catal. 301 (2013) 175–186. [78] C.A. Gaertner, J.C. Serrano-Ruiz, D.J. Braden, J.A. Dumesic, J. Catal. 266 (2009)
[25] O.A. Abdelrahman, A. Heyden, J.Q. Bond, J. Catal. 348 (2017) 59–74. 71–78.
[26] G.M.R. van Druten, V. Ponec, Appl. Catal. A 191 (2000) 153–162. [79] A.A. Gokhale, S. Kandoi, J.P. Greeley, M. Mavrikakis, J.A. Dumesic, Chem. Eng.
[27] H. Wan, A. Vitter, R.V. Chaudhari, B. Subramaniam, J. Catal. 309 (2014) 174– Sci. 59 (2004) 4679–4691.
184. [80] A.U. Nilekar, J. Greeley, M. Mavrikakis, Angew. Chem. Int. Ed. 45 (2006) 7046–
[28] B.S. Akpa, C. D’Agostino, L.F. Gladden, K. Hindle, H. Manyar, J. McGregor, R. Li, 7049.
M. Neurock, N. Sinha, E.H. Stitt, D. Weber, J.A. Zeitler, D.W. Rooney, J. Catal. 289 [81] A.M. Karim, V. Prasad, G. Mpourmpakis, W.W. Lonergan, A.I. Frenkel, J.G. Chen,
(2012) 30–41. D.G. Vlachos, J. Am. Chem. Soc. 131 (2009) 12230–12239.
[29] I. McManus, H. Daly, J.M. Thompson, E. Connor, C. Hardacre, S.K. Wilkinson, N. [82] O. Hinrichsen, F. Rosowski, M. Muhler, G. Ertl, Chem. Eng. Sci. 51 (1996) 1683–
S. Bonab, J. ten Dam, M.J.H. Simmons, E.H. Stitt, C. D’Agostino, J. McGregor, L.F. 1690.
Gladden, J.J. Delgado, J. Catal. 330 (2015) 344–353. [83] M. Salciccioli, Y. Chen, D.G. Vlachos, Ind. Eng. Chem. Res. 50 (2011) 28–40.
[30] C. Michel, J. Zaffran, A.M. Ruppert, J. Matras-Michalska, M. Jedrzejczyk, J. [84] W. Rachmady, M.A. Vannice, J. Catal. 192 (2000) 322–334.
Grams, P. Sautet, Chem. Commun. 50 (2014) 12450–12453. [85] W. Rachmady, M.A. Vannice, J. Catal. 207 (2002) 317–330.
[31] C. Michel, P. Gallezot, ACS Catal. 5 (2015) 4130–4132. [86] W. Rachmady, M.A. Vannice, J. Catal. 208 (2002) 158–169.
[32] M. Mavrikakis, M.A. Barteau, J. Mol. Catal. A: Chem. 131 (1998) 135–147. [87] R.M. Rioux, M.A. Vannice, J. Catal. 216 (2003) 362–376.
[33] R.B. Barros, A.R. Garcia, L.M. Ilharco, J. Phys. Chem. B 108 (2004) 4831–4839. [88] S. Wang, B. Temel, J. Shen, G. Jones, L.C. Grabow, F. Studt, T. Bligaard, F. Abild-
[34] R.B. Barros, A.R. Garcia, L.M. Ilharco, Surf. Sci. 572 (2004) 277–282. Pedersen, C.H. Christensen, J.K. Nørskov, Catal. Lett. 141 (2011) 370–373.
[35] J. Hrbek, R.A. DePaola, F.M. Hoffmann, J. Chem. Phys. 81 (1984) 2818–2827. [89] R.J. Madix, G. Ertl, K. Christmann, Chem. Phys. Lett. 62 (1979) 38–41.
[36] J.M. Sturm, C.J. Lee, F. Bijkerk, Surf. Sci. 612 (2013) 42–47. [90] C.T. Campbell, Y.K. Sun, W.H. Weinberg, Chem. Phys. Lett. 179 (1991) 53–57.
[37] R. Alcalá, J. Greeley, M. Mavrikakis, J.A. Dumesic, J. Chem. Phys. 116 (2002) [91] K. Lenz, B. Poelsema, S.L. Bernasek, G. Comsa, Surf. Sci. 189–190 (1987) 431–
8973–8980. 437.
[38] S.K. Wilkinson, I. McManus, H. Daly, J.M. Thompson, C. Hardacre, N.S. Bonab, J. [92] S.L. Bernasek, K. Lenz, B. Poelsema, G. Comsa, Surf. Sci. Lett. 183 (1987) L319–
ten Dam, M.J.H. Simmons, C. D’Agostino, J. McGregor, L.F. Gladden, E.H. Stitt, J. L324.
Catal. 330 (2015) 362–373. [93] C. Sachs, M. Hildebrand, S. Völkening, J. Wintterlin, G. Ertl, J. Chem. Phys. 116
[39] N.N. Rimar, G.N. Pirogova, Russ. Chem. Bull. 47 (1998) 398–401. (2002) 5759–5773.
[40] I. Horiuti, M. Polanyi, Trans. Faraday Soc. 30 (1934) 1164–1172. [94] C.T. Campbell, Top. Catal. 1 (1994) 353–366.
[41] F. Rositani, S. Galvagno, Z. Poltarzewski, P. Staiti, P.L. Antonucci, J. Chem. [95] K.J. Laidler, Chemical Kinetics, McGraw-Hill, New York, 1965.
Technol. Biotechnol. Chem. Technol. 35 (1985) 234–240.

You might also like