Composition of tungsten oxide bronzes active for hydrodeoxygenation

Article abstract of DOI:10.1016/j.apcata.2010.08.025

We explore the composition of reduced tungsten oxide bronze materials as a new class of heterogeneous catalysts for hydrodeoxygenation of bio-fuels. We demonstrate a method using thermogravimetric analysis (TGA) and mass spectrometry to determine both the oxygen substoichiometry and hydrogen content of the H_yWO_3-z catalytic phase that is formed during hydrogen pretreatment at temperatures from 200 to 500 °C. Results for three WO₃ materials show that the composition depends on crystallinity and surface area, as well as pretreatment temperature and time. The optimal pretreatment conditions for hydrogenation of acrolein to allyl alcohol at 50 °C occur over a narrow temperature range around 350 °C. We report for the first time that at higher reaction temperatures the catalysts are active for hydrodeoxygenation of allyl alcohol to propene and 1,5-hexadiene. We have proposed a mechanism similar to the Mars-van Krevelen cycle in which an alcohol adsorbs on an oxygen vacancy, forming a tungsten-oxygen bond and breaking the carbon-oxygen bond. Our TGA and reaction kinetics data indicate that the composition of the active catalyst is between H_0.9WO_2.9 and H_1.3WO_2.7 and that the rate of hydrodeoxygenation is comparable to the oxygen vacancy creation rate.

Full text of DOI:10.1016/j.apcata.2010.08.025

Applied Catalysis A: General 388 (2010) 86–95
Contents lists available at ScienceDirect
Applied Catalysis A: General
journal homepage: www.elsevier.com/locate/apcata
Composition of tungsten oxide bronzes active for hydrodeoxygenation
Timothy J. Thibodeaua,b,c, Alexander S. Canneyb,c, William J. DeSistob,c,d
,
M. Clayton Wheelerc,d, Franc¸ ois G. Amara, Brian G. Fredericka,b,c,∗
a Department of Chemistry, University of Maine, Orono, ME 04469, USA
b Laboratory for Surface Science and Technology, University of Maine, Orono, ME 04469, USA
c Forest Bioproducts Research Institute, University of Maine, Orono, ME 04469, USA
d Department of Chemical and Biological Engineering, University of Maine, Orono, ME 04469, USA
a r t i c l e i n f o
a b s t r a c t
Article history:
We explore the composition of reduced tungsten oxide bronze materials as a new class of heterogeneous
catalysts for hydrodeoxygenation of bio-fuels. We demonstrate a method using thermogravimetric anal-
ysis (TGA) and mass spectrometry to determine both the oxygen substoichiometry and hydrogen content
of the HyWO3−z catalytic phase that is formed during hydrogen pretreatment at temperatures from 200 to
500 ◦C. Results for three WO3 materials show that the composition depends on crystallinity and surface
area, as well as pretreatment temperature and time. The optimal pretreatment conditions for hydrogena-
tion of acrolein to allyl alcohol at 50 ◦C occur over a narrow temperature range around 350 ◦C. We report
for the first time that at higher reaction temperatures the catalysts are active for hydrodeoxygenation
of allyl alcohol to propene and 1,5-hexadiene. We have proposed a mechanism similar to the Mars–van
Krevelen cycle in which an alcohol adsorbs on an oxygen vacancy, forming a tungsten–oxygen bond and
breaking the carbon–oxygen bond. Our TGA and reaction kinetics data indicate that the composition
of the active catalyst is between H0.9WO2.9 and H1.3WO2.7 and that the rate of hydrodeoxygenation is
comparable to the oxygen vacancy creation rate.
Received 19 November 2009
Received in revised form 13 August 2010
Accepted 17 August 2010
Available online 22 August 2010
Keywords:
Acrolein
Allyl alcohol
Tungsten trioxide
Bronze
Catalyst
Spill-over
Hydrogenation
Hydrodeoxygenation
Thermogravimetric analysis
© 2010 Elsevier B.V. All rights reserved.
1. Introduction
pounds in pyrolysis oil are unsaturated. For economic reasons, we
have focused on developing catalysts that can selectively reduce
Production of fuels from lignocellulosic biomass has the
potential to reduce dependence on fossil fuels and emission of
greenhouse gases. Utilization of biomass is important because
it is currently the only renewable source of liquid transporta-
tion fuels [1,2]. Pyrolysis of biomass is economically feasible at a
smaller scale and offers potential energy efficiency advantages over
Fischer–Tropsch and fermentation processes. However, pyrolysis
produces a complex mixture of oxygenated organic compounds
with low energy density due to high oxygen composition and
upgrading is necessary [1–3]. Pyrolysis oil is composed of carboxylic
acids, aldehydes, ketones, carbohydrates, alcohols and phenols
[4–6]. Early work demonstrated that several classes of materials,
including supported precious metals and cobalt/molybdenum sul-
fides, are capable of catalytically removing oxygen from the oil,
increasing its energy content and stability [7,8]. Many of the com-
or break carbon–oxygen bonds while minimizing hydrogenation
of carbon–carbon double bonds.
As an example of hydrotreating reactions for an unsaturated
aldehyde, Fig. 1 illustrates the range of activity and selectivity that
has been reported for hydrogenation of acrolein, CH2 CH-CHO,
over a variety of catalysts [9–17]. Most noble metal catalysts have
good activity but are very selective to formation of the undesirable
product, propanal. Two reducible metal oxide catalysts, as reported
by Hoang-Van and Zegaoui [13] facilitate hydrogenation of the car-
bonyl group, albeit with much lower activity. Molybdenum oxide
is selective towards 1-propanol, but both the carbon–carbon dou-
ble bond and the carbonyl group are hydrogenated. Interestingly,
over a reduced tungsten oxide catalyst, only the carbonyl group is
hydrogenated, selectively producing allyl alcohol. Hoang-Van and
Zegaoui [13] reported that pretreatment in hydrogen at tempera-
tures of 150–350 ◦C is necessary to achieve activity for both pure
MoO3 and WO3 catalysts, as well as for Pt/MoO3 and Pt/WO3 cat-
alysts, and suggest that the active phase is a hydrogen bronze,
HxWO3 (where x is non-stoichiometric) [18]. Further evidence
that bronzes are active for hydrodeoxygenation of alcohols has
been reported: Matsuda et al. have shown that 2-propanol can
∗
Corresponding author at: Laboratory for Surface Science and Technology,
University of Maine, Orono, ME 04469, USA. Tel.: +1 207 581 2268;
fax: +1 207 581 2255.
E-mail address: brian.frederick@umit.maine.edu (B.G. Frederick).
0926-860X/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.apcata.2010.08.025

T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
87
Fig. 1. Comparison of the activity (on a log scale) for hydrogenation of acrolein over a selection of catalysts (references identified in brackets). Selectivity toward propanal,
1-propanol, and allyl alcohol is indicated as the fraction of each product (on a linear scale) [11,13,15].
be converted to propene over a molybdenum oxide bronze [19]
and Herrera et al. have shown that WO3/SBA-15 is active for the
dehydration of methanol and 2-butanol [20].
bronze. The maximum weight gain measured is 0.098%, which cor-
responds to the composition of H0.3WO3, assuming no oxygen loss.
In this paper, we first describe a method using TGA and
mass spectrometry to determine the composition of a reduced
metal oxide bronze. We report the composition of bronzes made
from three different tungsten oxide materials at temperatures of
300–500 ◦C. We then show that the activity for acrolein hydrogena-
tion to allyl alcohol occurs over a narrow pretreatment temperature
range and is critically dependent on the amount of hydrogen in the
bronze. Next, we report the activity and selectivity of bronzes, pre-
treated at 350 ◦C, for hydrogenation and hydrodeoxygenation of
allyl alcohol. At 50 ◦C, propanal is the major product, with lesser
amounts of 1-propanol, differing somewhat from the results of
Hoang-Van and Zegaoui [13]. At higher temperatures, we observe
good hydrodeoxygenation activity, mostly to propene and 1,5-
hexadiene. We analyze the product distribution to place limits on
the composition of the active phase and suggest that selectivity
toward hydrodeoxygenation is limited kinetically by the C–O bond
activation barrier, while hydrogenation is facile and the products
are largely controlled by thermodynamics.
Reducible metal oxide catalysts, such as bismuth molybdate,
vanadia, and tungsten oxide, have long been known as selective
oxidation catalysts [21]. The widely accepted Mars–van Krevelen
mechanism [22] involves selective oxidation of olefins to aldehy-
des, where the oxygen in the product comes directly from the oxide
lattice. In a parallel theoretical study, we have explored the factors
controlling hydrodeoxygenation of acrolein to produce propene,
1-propanol, and allyl alchohol on reduced MoO3 surfaces [23]. The
mechanism for propene production, shown schematically in Fig. 2,
involves hydrogen adsorption, formation of oxygen vacancies via
water production, selective adsorption of acrolein at the vacancy,
hydrogenation of the carbonyl to an allyl alkoxide intermediate,
and C–O bond scission leading to propene. Thus, we proposed that
the active state of the catalyst is a reduced metal oxide bronze.
Both Hoang-Van and Zegaoui [18] and Matsuda et al. [19] suggest,
based on XRD, that the catalyst is a bronze, but do not quantify the
hydrogen content or consider oxygen substoichiometry.
Although thermogravimetric methods have been used to char-
acterize tungsten oxide bronzes formed through solid state and
electrochemical reactions [24], we are not aware of materials
characterization work that determines simultaneously the compo-
sition of hydrogen and oxygen when the bronze is formed under
prolonged hydrogen treatment of tungsten oxide at atmospheric
pressure in the 300–500 ◦C range. There is general agreement that
loss of oxygen does not occur at temperatures below 200 ◦C [25].
Under typical temperature ramp conditions, 5 ◦C/min in flowing
H2, reduction of WO3 is not observed until 600 ◦C [25]. Lackner et
al. [25] have studied the reduction of commercial tungsten oxide
at temperatures between 800 and 1000 ◦C under isothermal condi-
tions. They report that in this high temperature range the oxide is
reduced to tungsten metal, as determined by a weight loss of 20.7%.
Previous studies using thermogravimetric analysis have also
investigated hydrogen reduction with noble metals dispersed on
WO3 [26–29]. In most high temperature ramp thermogravimetric
analysis (TGA) experiments, the addition of metals causes weight
loss, corresponding to oxygen substoichiometry, to occur at lower
temperatures than on the pure WO3. Only in a study by Kanamaru
et al. [27] were low temperature isothermal TGA experiments per-
formed. They prepared 0.1 and 0.01 wt% Pt/WO3 by impregnation
with H2PtCl6, converting the WO3 to HxWO3. The catalysts were
calcined in air and cooled in argon. Then, the WO3 was treated in
10% H2/Ar at temperatures of 40–200 ◦C. At these temperatures,
they report a weight gain due to the addition of hydrogen to the
2. Experimental
2.1. Materials
Three types of non-porous tungsten oxide were examined,
a
commercially available material of 30–70 nm particle size
(Sigma–Aldrich Tungsten(VI) oxide, nanopowder, <100 nm (TEM))
that we refer to hereafter as “nano-WO3”; a material prepared at
UMaine by Lu and Tripp [30] by emulsion polymerization that we
refer to as “UM-WO3”; and a commercially available material of
20 ␮m particle size (Sigma–Aldrich) that we refer to as “20 ␮m-
WO3.”
Nitrogen adsorption isotherms were measured at 77 K using a
Micromeritics ASAP-2020 instrument. The WO3 was degassed for
4 h at 150 ◦C. The adsorption isotherms and corresponding pore size
distributions, as calculated with the BJH method [31,32], showed
no evidence of pores. Surface areas were calculated using the BET
method [33]. The nano-WO3 had a surface area of 9.1 m2/g, UM-
WO3 had a surface area of 3.0 m2/g, and 20 ␮m-WO3 had a surface
area of 1.5 m2/g.
Powder X-ray diffraction measurements were made on all three
materials using a PANalytical X’PertPro X-Ray diffractometer uti-
lizing Cu-K␣ radiation in a parallel beam optical configuration. The
diffraction patterns, shown in Fig. 3, were very similar to those
reported previously by Lu et al. [30]. The crystalline phase was

88
T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
Fig. 2. Schematic mechanism proposed for the hydrodeoxygenation activity of reduced metal oxide bronzes.
ascribed to monoclinic WO3. However, the broad background, cen-
tered at 2ꢀ = 16–17◦, indicates significant amounts of amorphous
material, which is greatest in the UM-WO3.
internal bed temperatures during the highly exothermic pretreat-
ment reaction could be as much as 200 ◦C higher than the external
temperature. The catalyst was pretreated in pure hydrogen (Parker
Ballston PEM-500, >99.999%) for periods of 3–10 h at tempera-
tures between 200 and 450 ◦C. In the TGA oven, the bronzes were
prepared, starting from fresh samples of WO3, under the same con-
ditions used in the reactor. The oxide was heated to the specified
temperature and held for 10 h in pure hydrogen at atmospheric
pressure, flow 90 sccm, then cooled to 50 ◦C.
Bronze composition was quantified in the TGA from the weight
changes during two reaction steps for each formation temperature.
The mass loss during bronze formation is the difference between
oxygen weight loss and hydrogen weight gain. To determine both
the hydrogen content and the oxygen substoichiometry, in Step 1,
the bronze was purged with nitrogen, then heated to 1000 ◦C at
a ramp rate of 20 ◦C/min. The mass lost due to reaction of hydro-
gen in the bronze with lattice oxygen produced water, which was
confirmed with mass spectrometry. The sample was then cooled
to 50 ◦C and purged with oxygen. In step 2, the reduced oxide was
heated in oxygen to 1000 ◦C at a ramp rate of 20 ◦C/min to reoxidize
the catalyst to WO3, allowing us to determine the total of the oxy-
gen substoichiometry in the bronze and the additional amount of
oxygen lost in Step 1. We elaborate on the physical evidence sup-
porting the proposed method and numerical procedures utilized in
determining the composition of the bronze, HyWO3−z, in Section 4.
The evolved gases were monitored via a 50 ␮m ID quartz
capillary sampling system (AMETEK) using a mass spectrometer
(Stanford Research Systems, RGA 300) housed in a turbo-molecular
pumped vacuum chamber. Approximately 100 mg of WO3 powder
was placed on either an alumina or platinum pan in the TGA oven.
Blank experiments, with both ceramic and platinum pans (with-
2.2. Catalyst activation and bronze composition
The WO3 catalysts were pretreated and tested in a fixed bed
reactor, described previously [34]. An amount of WO3, correspond-
ing to 0.75 m2 surface area, was supported on 0.5 g of crushed
quartz (90–180 ␮m particle size) which was packed into a 0.25 in.
OD stainless steel glass lined reactor tube (AMI). A thermocouple
inside a thermal well allowed the internal bed temperature to be
measured directly and was used for temperature control during
hydrogen pretreatment and reaction measurements. We note that
Fig. 3. X-ray diffraction patterns of three tungsten oxide materials (as pre-
pared/supplied). Patterns for 20 ␮m-WO3 and UM-WO3 displaced vertically by 2000
and 4000, respectively, for clarity.

T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
89
out WO3), showed that uptake of hydrogen was negligible. A gas
handling manifold allowed switching between hydrogen (grade
5.0, Matheson Tri-Gas) or oxygen (grade 5.0, BOC Gas) after flush-
ing the TGA system with house nitrogen (blow-off from a liquid
nitrogen dewar) for safety reasons. Control experiments confirmed
that, without hydrogen pretreatment of WO3, cycles of heating to
1000 ◦C in nitrogen and oxygen resulted in weight changes of 0.1%
or less.
2.3. Reaction characterization
After pretreatment, the catalyst was cooled to room tem-
perature in hydrogen. Then the reactor was heated to reaction
temperature (between 50 and 350 ◦C). Reactant vapor was sup-
plied by a vapor liquid equilibrator (VLE) at 10–12 ◦C using nitrogen
carrier gas (BOC, grade 5.0) and blended with hydrogen (Parker
Ballston PEM-500, >99.999%) controlled by separate mass flow con-
trollers. The flow rates of nitrogen carrier and hydrogen were both
10 sccm (cm3/min at 0 ◦C and 1 atm). The estimated vapor pressures
of acrolein (Alfa Aesar, 97%, stabilized) and allyl alcohol (Acros,
>99%) were 88 and 4 torr, respectively. Typical total flow rates were
20 sccm corresponding to space velocities of 1800–3400 h−1. Mass
hourly space velocities for acrolein were 2.0–2.4 g/h g WO3 and
for allyl alcohol were 1.6–11.7 g/h g-WO3 for the temperatures and
pressures studied.
The product distribution was analyzed using GC-MS (Thermo
Trace) for compound identification and GC-FID (Thermo Trace)
for quantification. Switching valves allowed sampling upstream
and downstream of the reactor bed to estimate conversion. The
injector and FID were heated to 250 ◦C. The column flow rate was
1.5 mL/min and the split ratio was 50. A stabilwax column (Restek,
30 m, 0.32 mm, 20 ␮m) was used to separate acrolein, allyl alco-
hol, 1-propanol and 1,5-hexadiene. The GC program consisted of
a 5 min hold at 40 ◦C, a ramp to 150 ◦C at 10 ◦C/min followed by a
2 min hold. We used a PLOT Q column (HP, 30 m, 0.32 mm, 20 ␮m)
to separate ethene, ethane, propene, propane, and 1,5-hexadiene;
however, allyl alcohol was poorly separated from 1-propanol. The
GC program consisted of an initial hold at 80 ◦C for 1 min, a ramp
to 250 ◦C at 20 ◦C/min, and a 5 min hold. To measure light gases
(CO, CO2), a J&W Gas Pro column was used (60 m, 0.32 mm) with
MS detection. To separate CO, CO2, and nitrogen the column was
cooled to −80 ◦C with liquid nitrogen and held for 3 min. Then, the
GC oven was heated to 250 ◦C at 25 ◦C/min and held for 5 min.
To close the mass balance on allyl alcohol, FID sensitivity fac-
tors were estimated based on the number of carbons less the
number of oxygens. At low reaction temperatures, adsorption of
allyl alcohol on the catalyst surface accounted for a majority of
the apparent conversion, which was confirmed by an isothermal
desorption experiment. Initially, the catalyst was exposed to allyl
alcohol for a period of 4 h at 50 ◦C. The allyl alcohol was switched
out and the downstream mixture was sampled every 15 min for 2 h
by GC-FID with only hydrogen flowing over the catalyst. The allyl
alcohol evolved accounted for 95% of the apparent conversion. A
subsequent temperature programmed reaction measurement, at a
ramp rate of 5 ◦C/min to 350 ◦C, produced additional allyl alcohol
and other reaction products.
Fig. 4. Illustration of the TGA and MS data during bronze formation, and a two-
step method to determine composition. (A) Percent weight change on a ceramic
pan loaded with the nano-WO3 (9 m2/g) at (a) 300 ◦C, (b) 350 ◦C, (c) 400 ◦C, and (d)
475 ◦C bronze formation temperatures. Illustration of (B) temperature profile and
(C) mass spectral intensities for a) hydrogen (2 amu), (b) water (18 amu), and (c)
oxygen (32 amu) vs. time for 350 ◦C bronze formation temperature.
vated temperatures, a mass loss is measured which is mostly due to
loss of oxygen as water, as indicated by the mass spec signal. Fig. 4A
shows that at lower temperatures (300–350 ◦C) the mass loss is
small and approaches steady state. The composition of the bronze
formed during pretreatment in this temperature range can be con-
trolled precisely. As the temperature is increased above 350 ◦C the
rate of mass loss is much faster, so controlling the composition is
more difficult. By 450 ◦C the mass loss is 20%, which corresponds
to complete reduction to tungsten metal.
To determine the hydrogen content, the catalyst was heated in
nitrogen (Step 1). The hydrogen reacts with lattice oxygen, pro-
ducing the water desorption peak shown in Fig. 4C(b). In Step 2,
after switching to oxygen and heating, the reduced oxide (WO3−x
)
formed at the end of Step 1 gains weight to within 0.1% of its ini-
tial value, corresponding to WO3. The weight gain corresponds to
recovery of the oxygen lost during bronze formation and the lattice
oxygen lost to water formation in Step 1.
Fig. 5 shows the mass change for (A) bronze formation, (B) Step
1, and (C) Step 2 for the three materials on a ceramic pan. In gen-
eral, as the temperature is increased during the bronze formation
step, the mass loss increases. Over the 300–400 ◦C range, the mass
loss also correlates with surface area, suggesting that the oxygen
substoichiometry depends on surface area. However, the weight
change in Step 1 at temperatures below 350 ◦C was similar for all
three materials, indicating a similar hydrogen content. At tempera-
tures of 300–350 ◦C the mass loss (due to H2O desorption) in Step 1
was similar to the mass gain (due to oxygen) in Step 2. Because these
mass changes were much larger than the mass loss during bronze
formation, the hydrogen stoichiometry, y, should be large com-
pared to the oxygen substoichiometry, z, i.e. the bronzes formed
at these temperatures were close to HyWO3−z, with y » z. At higher
3. Results
3.1. Composition of reduced bronze catalysts (HyWO3−z
)
Fig. 4 illustrates the mass changes, temperature program, and
mass spectrometer signals during bronze formation and the two
analysis steps used to determine the composition of the reduced
tungsten bronze. While flowing hydrogen over the oxides at ele-

90
T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
Fig. 6. Dependence of acrolein hydrogenation activity at 50 ◦C after pretreatment
in hydrogen for 3 h at the indicated temperature for UM-WO3 (circles) compared to
results of Hoang-Van and Zegaoui [13] (squares).
Interestingly, if the tungsten oxide is reduced at 250 ◦C for 3 h
we notice that in the succeeding reaction, the catalyst is selective
towards the formation of 1-propanol for 2 h, then allyl alcohol is
produced at a much lower rate for an additional hour before the
catalyst loses activity. This shows that initially, there is enough
hydrogen in the catalyst to completely hydrogenate acrolein to 1-
propanol but the rate of hydrogenation is faster than the rate of
hydrogen dissociation so the selectivity shifts from 1-propanol to
allyl alcohol. The amount of hydrogen consumed in the produc-
tion of allyl alcohol and 1-propanol agrees with the amount of
hydrogen in the bronze, based on the TGA experiments, to within
experimental errors.
Fig. 5. Weight loss or gain during (A) bronze formation in hydrogen, (B) heating
in nitrogen (Step 1), and (C) heating in oxygen (Step 2) as a function of the hydro-
gen treatment temperature for the 20 ␮m-WO3, UM-WO3 and nano-WO3 materials
using ceramic or platinum TGA pans, as indicated.
By extending the pretreatment duration from 3 to 10 h, we found
that the catalyst maintained steady state activity for days. This
suggests that under reaction conditions, hydrogen in the bronze
can be replenished through H2 dissociation on oxygen vacancyz
sites.
We repeated the experiment with the 20 ␮m-WO3 and nano-
WO3 samples and neither exhibited substantial activity for the
hydrogenation of acrolein. We suggest that the differences in activ-
ity may relate to the amount of amorphous phase present in each
sample, as shown in X-ray diffraction patterns (Fig. 3). As we will
show, the UM-WO3 bronze has a higher amount of hydrogen as
compared to the other materials under the same pretreatment con-
ditions.
temperatures, the weight gain in Step 2 is larger than the mass
loss in Step 1, and similar to the mass loss during bronze forma-
tion, which suggests that the bronze has a composition HyWO3−z
,
with y < z. These qualitative trends in the bronze composition are
confirmed quantitatively by our analysis given in Section 4.
Because noble metals are used as spillover catalysts in hydrogen
storage materials, we also compared the data for the UM-WO3 to
experiments on a platinum pan. In Step 1, there is a larger mass loss
above 350 ◦C on platinum compared to the ceramic pan, which cor-
responds to a higher hydrogen content. This shows that significant
hydrogen uptake occurs over macroscopic distances. At tempera-
tures up to 400 ◦C, the mass changes during bronze formation are
similar. Above 400 ◦C, when the ceramic pan is used the oxide is
reduced completely to metal while on a platinum pan it is not as
heavily reduced, suggesting that hydrogen increases the stability
of the tungsten oxide lattice.
3.3. Hydrodeoxygenation (HDO) of allyl alcohol
The reactivity of allyl alcohol was measured on all three mate-
rials at four temperatures between 50 and 350 ◦C after hydrogen
pretreatment at 350 ◦C for 10 h. A summary of the initial selectiv-
ity of each catalyst at each reaction temperature is shown in Fig. 7
and Table 1. The product distribution and activity was different on
each of these materials, although all the materials are active for iso-
merization to propanal and hydrogenation to 1-propanol at 50 ◦C.
By comparison, Hoang-Van and Zegaoui [13] report 100% selec-
tivity toward 1-propanol after pretreatment at 300 ◦C and similar
amounts of propanal and 1-propanol after 400 ◦C pretreatment. We
report, for the first time, that hydrodeoxygenation products were
formed at higher temperatures (above 150 ◦C).
As shown in Fig. 7, at lower temperatures, a large proportion
of the apparent conversion is attributed to adsorption. After allyl
alcohol reaction for 4 h at 50 ◦C, an isothermal desorption experi-
ment at 50 ◦C accounted for most of the reactant. A small amount
of reaction products desorbed during subsequent temperature pro-
grammed desorption. We note that all three catalysts hydrogenated
the C C bond to form 1-propanol, whereas only the UM-WO3 was
3.2. Hydrogenation of acrolein to allyl alcohol
Hoang-Van and Zegaoui have shown that the hydrogenation of
acrolein requires that the catalyst be pre-treated in hydrogen and
that the activity is sensitive to the pretreatment conditions [13]. We
have measured the acrolein hydrogenation activity as a function
of pretreatment temperature under similar, 3 h conditions using
the UM-WO3 material. The activities are summarized in Fig. 6, and
compared to those of Hoang-Van and Zegaoui [13]. The activity
is probably lower for our catalysts because of lower surface area
and lower partial pressure of acrolein. The peak in activity in our
results occurs 200 ◦C higher than theirs, which may be due to sur-
face area dependence or could be attributed to inaccuracies in bed
temperature measurement, as noted in the Experimental section.
The hydrogenation activity occurs over a narrower pretreatment
temperature range in our results.

T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
91
Table 1
Percent conversion and yield of products on basis of allyl alcohol molar flow rate as a function of reaction temperature for each catalyst pretreated for 10 h at 350 ◦C.
Catalyst
20 ␮m WO3
UM-WO3
50
nano-WO3
50
Reactor Temperature (◦C)
50
150
250
350
150
250
350
150
250
350
% Conversion
Ethene
Ethane
Propene
Propane
13.00
0.00
0.00
0.01
0.00
0.03
0.23
0.13
0.01
13.00
0.01
0.01
0.18
0.00
0.09
2.76
4.05
0.02
38.00
0.15
0.08
2.17
0.02
0.20
10.17
7.37
0.80
37.00
0.65
0.24
5.96
0.07
1.73
8.35
3.12
5.05
8.00
0.00
0.00
0.00
0.00
0.16
0.16
0.03
0.04
15.00
0.02
0.01
0.11
0.00
0.16
2.44
3.85
0.03
49.00
0.29
0.11
1.90
0.02
17.59
17.75
11.16
0.87
26.00
0.14
0.05
2.93
0.01
1.31
3.56
0.85
4.13
7.00
0.00
0.00
0.00
0.00
0.16
0.16
0.01
0.03
35.00
0.04
0.03
0.12
0.00
5.84
6.02
1.00
0.03
61.00
0.28
0.13
2.67
0.03
22.01
22.25
16.53
0.19
40.00
0.08
0.03
4.99
0.01
2.22
2.30
0.00
0.03
Acrolein
Propanal
1-Propanol
1,5-Hexadiene
able to hydrogenate the C O bond of acrolein (Section 3.2). This
suggests that the sites for hydrogenation of alkenes are different
from those for hydrogenation of carbonyls.
state, the HDO products predominate on the higher surface area
materials.
As reported in Fig. 7 and Table 1, we see conversion of allyl alco-
hol back to acrolein at temperatures above 150 ◦C. This is consistent
with thermodynamic control, since acrolein becomes thermody-
namically more favorable than allyl alcohol at temperatures above
150 ◦C. At most temperatures and on most catalysts, the yield of
propanal is greater than 1-propanol, despite the fact that only at
reaction temperatures above 277 ◦C is the formation of propanal
more favorable than 1-propanol. Thus, the propanal in the product
distribution appears to be kinetically controlled.
The formation of hydrodeoxygenation products (primarily
propene and 1,5-hexadiene) is observed only at high reaction
temperatures, even though the Gibbs free energy of reaction is
highly favorable over the entire temperature range due to the co-
production of water. This indicates that the activation barrier for
C–O scission must be high, limiting the production of HDO products.
In addition to the HDO products, acrolein, and propanal
observed at 350 ◦C, there was small amounts of CO and CO2. There
was little evidence for coking on the basis of carbon-nitrogen analy-
sis. Thus, we were not able to quantitatively close the mass balance
at high temperatures.
4. Discussion
4.1. Stoichiometry of the bronze
When tungsten oxide is heated in hydrogen, at temperatures
below 200 ◦C, weight increases are observed when platinum facil-
itates spill-over [25], indicating absorption of hydrogen to form a
bronze. Our data for 200 ◦C, taken on a ceramic pan also shows a
weight gain. At temperatures of 300 ◦C and higher, a net weight
loss is always reported [25,26]. The loss of oxygen during hydrogen
treatment outweighs the gain due to hydrogen absorption. Theo-
retical calculations for MoO3 suggest that the most favorable route
to loss of oxygen is through formation of water [35]. Formation of an
oxygen vacancy with production of water has a calculated reaction
energy of −0.2 eV, while formation of hydroxyls requires 4–4.5 eV
The time dependence of the selectivity of all three catalysts at
350 ◦C is shown in Fig. 8. The major hydrodeoxygenation prod-
ucts, propene and 1,5-hexadiene, decay to a steady state value. On
the 20 ␮m-WO3, propanal and 1-propanol are relatively constant
with time, whereas on the other materials they decay. At steady
Fig. 7. Comparison of conversion (y-axis, total bar height) and selectivity (propor-
tion of bar height) as a function of catalyst and reaction temperature (x-axis) for
three tungsten oxide bronzes with different surface areas. HDO products formed at
higher reaction temperatures include propene, 1,5-hexadiene, propane, ethene, and
ethane.
Fig. 8. Time dependence of product distribution on (A) 20 ␮m-WO3, (B) UM-WO3,
and (C) nano-WO3 as a function of time showing that the hydrodeoxygenation
rate (propene, 1,5-hexadiene, ethene) decays toward a steady state value but the
selectivity of HDO to isomerization products increases with surface area.

92
T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
and desorption of atomic oxygen 6.5–6.8 eV, depending on the lat-
tice site [35]. Therefore, we write a chemical reaction for the initial
hydrogen reduction step in terms of water production,
shows that the substoichiometry correlates strongly with surface
area. The hydrogen stoichiometry, y, generally peaks around 350 ◦C
pretreatment temperature, but there is no correlation with sur-
face area. The hydrogen content is highest in the UM-WO3, which
appears to have the largest proportion of amorphous material from
the XRD data. However, the correlation between hydrogen content
and crystallinity is not clear when comparing the other two mate-
rials. For all materials in the ceramic pan, and at temperatures up
to the optimal pretreatment temperature of 350 ◦C, hydrogen was
more likely to dissolve and the oxide was not heavily reduced. This
is shown by the large value of y (around 1) and the low value of z
(less than 1). At higher temperatures the oxide was more heavily
reduced and the amount of dissolved hydrogen decreased, yielding
values of y less than 1 and z greater than 2.5.
Hydrogen spillover phenomena are widely known for platinum
particles dispersed on reducible oxides [27] and are discussed
by Hoang-Van and Zegaoui [18] with respect to pretreatment of
platinum supported molybdenum and tungsten oxide catalysts.
Comparison of reduction of the UM-WO3 on ceramic vs. platinum
pans showed no differences in weight loss during H2 pretreatment
within the uncertainties of the experiment. However, the weight
loss in Step 1 between 400 and 500 ◦C shows significant differences,
implying that greater amounts of hydrogen were present in the
bronze if a platinum pan was used. By comparison with the same
material on the ceramic pan, the value of y for UM-WO3 is even
larger than at 350 ◦C. At the same high temperatures, the oxygen
substoichiometry parameter, z, is not as large on the platinum pan
as on the ceramic pan. This suggests that the thermodynamics of
the ternary phase are rather complex and that hydrogen stabilizes
the lattice oxygen.
WO3 + H2(g) → HyWO3−z + zH2O(g).
(1)
As shown in Fig. 4, water production was confirmed using mass
spectrometry. The amount of water produced during the nitrogen
step was calculated from the mass loss in TGA, which is directly
related to the amount of hydrogen dissolved in the bronze. Because
the atomic weight of oxygen, AO, is 16 times that of hydrogen, AH, an
initial estimate of the oxygen substoichiometry coefficient, z, can be
calculated from the fractional weight loss during bronze formation
in hydrogen, wH,
wHMWO
3
z0
=
,
(2)
AO
where MWO is the molecular weight of WO3. Since tungsten oxide
3
is 20% oxygen by weight, wH cannot be greater than 0.207.
To independently determine the hydrogen content during pre-
treatment, the bronze was heated in nitrogen to 1000 ◦C in Step
1. Assuming that the weight loss during the nitrogen step was
primarily due to loss of water,
HyWO3−z
→
1 yH2O(g) + WO3−x
,
(3)
2
we obtained an initial estimate of the hydrogen stoichiometric coef-
ficient, y0,
ꢀ
ꢁ
wNMWO
3
y0 ≈ 2
,
(4)
MH
O
2
from the fractional weight loss in Step 1, wN, where MH is the
O
2
The rate of oxide reduction increases with temperature and
increasing surface area of the catalysts. The TGA curves for several
temperatures are shown in Fig. 4A for the nano-WO3 on a ceramic
pan. At temperatures below 400 ◦C, the weight change is small and
slow. At higher temperatures (475 ◦C) the oxide reduces completely
to metal very quickly. The rate of reduction is generally believed to
be controlled by the rate of the surface chemical reaction, rather
than diffusion of oxygen vacancies to the surface of the particle
[36]. In previous isothermal TGA experiments [25,37,38] in the tem-
perature range from 600 to 1000 ◦C, an initial parabolic mass loss
is observed as the oxide converts from WO3 to WO2.9, followed by
three approximately linear regions corresponding to phase changes
from WO2.9 to WO2.72, WO2.72 to WO2, and WO2 to W. The dis-
continuities between linear segments in the mass loss curve agree
molecular weight of H2O. As noted, without hydrogen pretreat-
ment, heating WO3 in nitrogen to 1000 ◦C resulted in a weight loss
less than 0.1%, illustrating the necessity of dissolved hydrogen in
producing significant weight losses.
To determine the oxygen substoichiometry at the end of Step 1,
the reduced tungsten oxide was heated in oxygen in Step 2. In all
cases, the weight gain restored the WO3 to within 0.1% of its initial
weight. From the stoichiometry of the oxidation reaction,
WO3−x
+
1 xO2 → WO3,
(5)
2
we calculate, the oxygen substoichiometric coefficient, x, from the
fractional weight gain in Step 2, wO
wO · MWO
3
x =
.
(6)
reasonably well with the crystallographic phases W20O58, W18O49
,
AO
and WO2 that are known from diffraction. In these linear segments,
the overall rate of mass loss is controlled by the movement of the
hydrogen front down through the bed [37] and therefore the rate
of mass loss is limited by mass transfer processes, not the chemical
reaction rate. The activation energy of the surface chemical reac-
tion has been estimated within a shrinking core particle and moving
front model [37] from a set of isothermal TGA experiments using
plots of ln[dꢁW/dt] vs. 1/T, where dꢁW/dt is the slope of the rel-
ative total weight change. The activation energies obtained vary
between 28 and 142 kJ/mol [25].
To the extent that hydrogen bronze formation occurred (i.e.
y0 > 0), the initial estimate of the substoichiometric coefficient, z0,
is an underestimate, and from the stoichiometry of Reaction (3), we
were able to determine y and z iteratively, using
wHMWO + yi−1AH
3
zi =
(7)
(8)
AO
yi = 2(x − zi)
which converged to within 0.01% after 4–5 iterations, and the coef-
ficients x, y satisfied
The active phase of our catalysts for HDO occurs within the ini-
tial parabolic mass loss region, WO3 to WO2.9, for which activation
energies have not been estimated. From mass conservation,
MWO − M(HyWO3−z
)
3
wH
=
(9)
MWO
3
dmO
dt
to within 0.01%. Table 2 summarizes the values of x, y, and z from
the weight changes summarized in Fig. 5.
= V(−rate)
(10)
where mO is the mass of oxygen and V is the volume of the material.
If we take the rate law to be first order in oxygen mass,
4.2. Factors affecting bronze composition
ꢂ
ꢃ
dmO
dt
mO
V
Quantitative analysis of the oxygen substoichiometry, z, shown
= V −k
(11)
in Table 2, for bronze formation in the 300–400 ◦C range, indeed

T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
93
Table 2
Composition of tungsten oxide bronzes, HyWO3−z, after hydrogen treatment as a function of temperature and further reduced WO3−x after hydrogen removal as water.a
Catalyst (Surface Area)
Pan Material
Hydrogen
y
z
Bronze HyWO3−z
x
Reduced WO3−x
Pretreatment
Temperature/◦C
20 ␮m-WO3 (1.5 m2/g)
300
350
400
450
0.79
0.89
0.48
0.47
0.06
0.07
0.12
2.05
H
0.79WO2.94
0.45
0.51
0.36
2.27
WO2.55
WO2.49
WO2.64
WO0.73
Ceramic
H0.89WO2.93
H0.48WO2.88
H0.47WO0.95
UM-WO3 (3.0 m2/g)
200
300
350
400
450
475
0.93
1.11
1.29
0.77
0.21
0.29
0.06
0.08
0.23
1.13
2.84
2.80
H0.93WO2.94
H1.11WO2.92
0.52
0.63
0.88
1.51
2.95
2.95
WO2.48
WO2.37
WO2.12
WO1.49
WO0.05
WO0.05
Ceramic
H
H
1.29WO2.77
0.77WO1.87
H0.21WO0.16
H0.29WO0.20
300
350
400
500
300
325
350
400
450
475
1.02
1.31
1.87
1.48
1.22
0.95
1.28
0.80
0.25
0.09
0.07
0.27
1.22
1.95
0.12
0.22
0.25
2.40
2.59
2.89
H1.02WO2.93
0.58
0.91
2.15
2.70
0.73
0.70
0.95
2.80
2.72
2.93
WO2.42
WO2.09
WO0.85
WO0.30
WO2.27
WO2.30
WO2.05
WO0.20
WO0.28
WO0.07
Platinum
Ceramic
H
1.31WO2.73
H1.87WO1.78
H
H
1.48WO1.05
1.22WO2.88
nano-WO3 (9.1 m2/g)
H0.95WO2.78
1.28WO2.75
H
H0.80WO0.60
H0.25WO0.41
H0.09WO0.11
a
At lower temperatures, the uncertainties are y( 0.05), z( 0.01), x( 0.02); at higher temperatures they are y( 0.3), z( 0.1), x( 0.05).
and recast the equation in terms of the total mass, mt, tungsten
mass, mW, and initial oxide mass, mi,
first to form surface hydroxyls, and then dehydroxylation leads to
water desorption and surface oxygen vacancies. The initially large
values of the effective activation energy at mt/mi = 99.99% may be
due to a low population of surface hydroxyls, since the bulk sites
are much more stable at low hydrogen content. A possible explana-
tion for the decreased activation energies at mt/mi = 99.5% may be
associated with the surface hydroxyl coverage. After an extended
ꢂ
ꢃ
d(mt/mi)
dt
mt
mi
mW
mi
= −k
−
(12)
and use an Arrhenius expression for the rate constant,
k = A exp (− Ea/RT) then taking the logarithm gives
ꢅ
ꢅ
ꢅ
ꢅ
ꢄ
ꢆ
ꢅ
ꢅ
ꢅ
ꢂꢂ
ꢃ
ꢂ
ꢃꢃꢇ
d(mt/mi)
dt
Ea(z)
RT
mt
mi
mW
mi
ln
= −
+ ln
A
−
(13)
z
z
where the rate of relative mass change is evaluated for a particular
fractional mass loss, z, and the activation energy is allowed to be
a function of the substoichiomety. This shows that a plot of the
natural log of the slope vs. 1/T, for the indicated fractional mass
loss, should be linear with slope −Ea/R, and that the other factors
all appear in the intercept.
Arrhenius plots are shown in Fig. 9 for each material at fractional
masses of 99.99% (WO2.998) and 99.5% (WO2.93). The specific rate
of reduction increases with surface area of the material, which is
expected. At very small substoichiometry, the apparent activation
energy, around 100 kJ/mol, is not significantly different for the three
materials. However, the apparent activation energy decreased
with increasing surface area at mt/mi = 99.5%. For the low surface
area 20 ␮m-WO3, the apparent activation energy was 201 kJ/mol.
The activation energy for the UM-WO3 was 88 7 kJ/mol and for
the nano-WO3 the apparent activation energy was 40 13 kJ/mol.
These values span the range of activation energies reported in simi-
lar experiments performed at much higher temperature conditions
[25], suggesting that particle size effects may be important. In our
experiments, the hydrogen content may be significantly higher
than in the high temperature TGA measurements [25,37,38]; there-
fore, the reduction mechanism and activation energies may differ
due to hydrogen content as well.
Clearly, the conditions required for catalyst pretreatment are
sensitive to the WO3 surface area, reduction time, and tempera-
ture. Both mass spectrometry data and theoretical calculations for
oxygen defect creation mechanism are consistent with the reaction
of hydrogen with lattice oxygen to form water and oxygen vacan-
cies. In our work [23] and that of Tokarz-Sobieraj et al. [35], for
MoO3 surfaces, we argue that hydrogen dissociation must occur
Fig. 9. Arrhenius plots for rate of reduction during hydrogen treatment of nano-
WO3, UM-WO3, and 20 ␮m-WO3 in ceramic TGA pans as indicated at a relative
mass of (A) 99.99% (WO2.998) and 99.5% (WO2.93).

94
T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
reduction period, during which hydrogen absorbs into the bulk,
surface hydroxyl become energetically competitive with bulk sites
and dehydroxylation become facile. The surface area dependence
at mt/mi = 99.5% suggests that hydrogen content increases faster on
higher surface area materials.
For the reaction of MoO3 with H2 to form water and sur-
face oxygen vacancies, analogous to Reaction (1), we calculate
ꢁrG = 74 kJ/mol [23]. Oxygen vacancy diffusion is extremely fast
[36] in MoO3 and WO3, and the oxide reduction rate is controlled by
the surface reaction rates. From this perspective, the higher the spe-
cific surface area, the faster we would expect the rate of reduction
of the oxide to be. Under flowing hydrogen conditions, we should
not expect to achieve a true thermodynamic equilibrium, and so
the steady state composition of the bronze need not be the same
for the three materials, even if the physical state, and hence the
ꢁrG for Reaction (1) were the same.
The differential heat of absorption of hydrogen into MoO3 to
form the bronze, HaMoO3, is large and negative at low hydrogen
stoichiometry, a, decreases to zero around a = 1, and the reac-
tion becomes endothermic for a > 1 [39]. At low composition,
this reflects the electrostatic attraction between the protons dis-
solved in the bronze and lattice oxygen, implying a larger cohesive
energy of the solid. To the extent that the hydrogen content of
the bronzes varies among the three materials, we can expect the
ꢁrG for Reaction (3) to be a function of hydrogen content, y. The
apparent activation energies obtained (40–200 kJ/mol) for reduc-
tion of our reduced tungsten oxide bronze materials bracket the
value calculated for dehydroxylation of surface hydroxyls from
MoO3 (74 kJ/mol) [23,35]. While differences in metal–oxygen bond
strength certainly contribute to the magnitude of the reduction bar-
rier, the presence of hydrogen in the bronze must also be a factor.
Theoretical investigations of the effect of hydrogen in the bronze on
dehydroxylation energies would be very helpful in distinguishing
these effects.
possible that for the UM-WO3, with the highest hydrogen content,
1.29WO2.77, surface hydroxyl groups are populated and account
H
for carbonyl hydrogenation. A possible explanation for C C hydro-
genation sites may be direct transfer of hydrogen from the bronze
if allyl alcohol is adsorbed on oxygen vacancy sites. Further spec-
troscopic and theoretical work is needed to explore these potential
reaction pathways.
These reactions occur at lower temperatures (50 ◦C) where the
catalyst may or may not be able to adsorb hydrogen, but oxygen
vacancy creation is negligible [27]. Therefore, if the catalyst is oper-
ated at temperatures below the threshold for rapid oxygen vacancy
creation (which may be surface area and hydrogen partial pressure
dependent), hydrogenation may be maintained at steady state, but
hydrodeoxygenation cannot be sustained.
If the reaction temperature is increased above 150 ◦C
hydrodeoxygenation activity is observed. At these temperatures
production of propene and 1,5-hexadiene is observed and varies
with time. We have proposed that the mechanism is closely related
to the Mars–van Krevelen mechanism, as shown in Fig. 2. In one
part of the catalytic cycle, hydrogen dissociates to form surface
hydroxyls. The hydroxyls react with lattice oxygen to form water
and create oxygen vacancies. Hydrogenation steps require protons,
either from surface hydroxyls or directly from hydrogen atoms dis-
solved in the bronze. In the second part of the catalytic cycle, the
alcohol adsorbs at a coordinatively unsaturated tungsten site (i.e.
an oxygen vacancy), forming a relatively strong W–O bond and
weakening the O–C bond. At this point, the propene radicals can
desorb and couple to form 1,5-hexadiene [40–42] or they can pick
up an additional hydrogen to form propene. It is important to note
that the relative amounts of propene and 1,5-hexadiene vary with
time online. In both processes the catalyst is left in a more oxidized
state. Therefore, the HyWO3−x catalyst must be able to form oxygen
vacancies again, as well as absorb hydrogen, in order to maintain a
catalytic cycle at steady state.
From the time dependent product distribution, we can estimate
the maximum amount of oxygen taken up by the catalyst in the
hydrodeoxygenation reactions,
4.3. Implications for catalyst pretreatment and reactivity
These results provide insight into several factors affecting the
preparation and use of reduced tungsten oxide bronzes as a catalyst
for hydrogenation (at 50 ◦C) and hydrodeoxygenation (at tempera-
tures above 150 ◦C). When the catalyst is pretreated in hydrogen at
temperatures close to the 300 ◦C threshold for loss of lattice oxygen,
a steady state bronze composition can be achieved. The threshold
temperature and treatment time appear to be sensitive to the spe-
cific surface area. To the extent that hydrogen may stabilize lattice
oxygen, the hydrogen pretreatment pressure may also affect bronze
composition.
The kinetics during reduction and bronze formation provide
a guide to determining the catalyst pretreatment time. The reac-
tion is exothermic and therefore accurate temperature control in
the catalyst bed is required to reproducibly prepare the catalyst.
We should also note that if the reduced catalyst (in the bronze
state) was exposed to air, spontaneous combustion of the hydro-
gen and oxidation of the tungsten suboxide generates substantial
heat, to the extent that it glowed orange. Thus, the catalyst needs to
be maintained in an inert or reducing environment until reaction
begins.
CH2 CH–CH2–OH (allyl alcohol) + VO → CH2 CH–CH3 + OO
(14)
2CH2 CH–CH2–OH (allyl alcohol)
+2VO → CH2 CH–CH2 CH2–CH CH2 + 2OHO
(15)
where VO represents a vacancy on an oxygen site. Notice that there
is no net consumption or production of hydrogen in these reactions,
but oxygen vacancies are annihilated. Loss of vacancies, as adsorp-
tion and reaction sites, would reduce the rate of these reactions if
not regenerated. Oxygen vacancy creation occurs via the reactions
OO
+
1 H2 → OHO
(16)
(17)
2
2OHO → H2O(g) + OO + VO
In addition to the major HDO reaction products, we also observe
production of propanal, acrolein and 1-propanol, from which we
estimate the net consumption of hydrogen according to the reac-
tions,
CH2 CH–CH2–OH (allyl alcohol) → CH3–CH2–CH O (propanal)
We found that only the UM-WO3 was active for hydrogenation
of acrolein to allyl alcohol, whereas all three tungsten oxide bronzes
hydrogenated allyl alcohol to 1-propanol and catalyzed the isomer-
ization to propanal. As already mentioned, this suggests that the
sites for C C hydrogenation are different from C O hydrogenation.
Our analysis of the adsorption energies for surface terminal, sur-
face bridging, and bulk hydroxyl sites in the MoO3 bronze system
indicate that population of surface hydroxyls occurs only at fairly
high hydrogen content (HxMO3, with x ≈ 1.1–1.2). Thus it may be
(18)
CH2 CH–CH2–OH (allyl alcohol) → H2+CH2 CH–CH O (acrolein)
(19)
H2 + CH2 CH–CH2–OH (allyl alcohol)
→ CH3–CH2–CH2–OH (1-propanol)
(20)

T.J. Thibodeau et al. / Applied Catalysis A: General 388 (2010) 86–95
95
Reaction (20) decreases and Reaction (19) increases the hydrogen
content of the bronze while Reaction (18) does not change the
bronze composition.
tion of this alcohol above 150 ◦C. We have shown that the active
bronze (HyWO3−z) has the composition where 0.89 < y < 1.29 and
0.07 < z < 0.31.
When the UM-WO3 was reduced at 350 ◦C the composition of
the bronze was determined to be H1.29WO2.77. After 5 h of reac-
tion at 350 ◦C, 280 ␮mol of hydrogen was consumed by 1-propanol
production and 330 ␮mol of oxygen was taken up by the cat-
alyst. Assuming that no oxygen vacancy regeneration occurred,
the final state of the catalyst would be H1.23WO2.85. This repre-
sents a maximum change in composition of the catalyst over 5 h
of operation. Similarly, the nano-WO3 would be converted from
H1.17WO2.69 to H0.84WO2.74 and the 20 ␮m-WO3 could be con-
verted from H0.89WO2.94 to H0.85WO2.97. If we neglect catalyst
regeneration during reaction it is evident that the state of the
catalyst will approach WO3 and become inactive for hydrodeoxy-
genation. As Fig. 8 shows, the HDO reaction rates decrease, as would
be expected for a reversible A ꢀ B reaction approaching a steady
state condition, in which the HDO Reactions (14) and (15) are
competing with the oxygen vacancy Reactions (16) and (17). From
thermogravimetric analysis we have estimated an oxygen vacancy
creation barrier of order 140–200 kJ/mol. This is similar in magni-
tude to the barriers we have estimated for C–O bond scission on
a Mo3O9 cluster of 104 kJ/mol [23]. Thus, we expect that at steady
state, the bronze composition will approach an intermediate value
between the initial composition of the bronze and the maximum
values calculated above.
Acknowledgements
We gratefully acknowledge support of this work through DOE
Award #DE-FG02-07ER46373. The authors would like to thank
Rachel Pollock and I. Tyrone Ghampson for the BET and XRD anal-
ysis of the tungsten oxide catalysts; Zhixiang Lu and Carl Tripp for
synthesis of the UM-WO3; and Rachel Austin for a critical reading
of the manuscript.
References
[1] G.W. Huber, in: B. National Science Foundation. Chemical, Environmental, and
Transport Systems Division (Ed.), NSF, Washington, DC, 2008.
[2] G.W. Huber, S. Iborra, A. Corma, Chemical Reviews (2006) 4044–4098.
[3] S.H. Beis, S. Mukkamala, N. Hill, J. Joseph, C. Baker, B. Jensen, E.A. Stemmler, M.C.
Wheeler, B.G. Frederick, A. van Heiningen, A.G. Berg, W.J. DeSisto, BioResources
5 (2010) 1408–1424.
[4] L. Ingram, D. Mohan, M. Bricka, P. Steele, D. Strobel, D. Crocker, B. Mitchell, J.
Mohammad, K. Cantrell, C.U. Pittman, Energy & Fuels 22 (2008) 614–625.
[5] C.A. Mullen, G.D. Strahan, A.A. Boateng, Energy & Fuels 23 (2009) 2707–2718.
[6] J. Joseph, C. Baker, S. Mukkamala, S.H. Beis, M.C. Wheeler, W.J. DeSisto, B.L.
Jensen, B.G. Frederick, Energy & Fuels (2010), doi:10.1021/ef100504d.
[7] E. Furimsky, Catalysis Reviews, Science and Engineering 25 (1983) 421–458.
[8] D.C. Elliott, Energy & Fuels 21 (2007) 1792–1815.
[9] P. Claus, A. Bruckner, C. Mohr, H. Hofmeister, Journal of the American Chemical
Society 122 (2000) 11430–11439.
Similar arguments apply to the steady state hydrogen content
of the catalyst in its active state. The observation that hydrogena-
tion activity is relatively constant suggests that the dissociation of
hydrogen on the bronze is sufficiently fast that the steady state
hydrogen composition probably remains quite similar to the initial
values.
Comparing the activity of the three materials for acrolein hydro-
genation and allyl alcohol HDO reactions, the differences in surface
area and crystallinity offer distinct opportunities to optimize a
low temperature, first stage hydrotreatment reactor and a second
stage, higher reaction temperature process, according to the gen-
eral strategies previously employed [8]. The UM-WO3 material was
the only catalyst active for carbonyl hydrogenation, which appears
to be related to the higher hydrogen content and more amor-
phous physical state. By contrast, the low surface area 20 ␮m-WO3
showed the highest steady state selectivity toward HDO reactions.
Once surface oxygen vacancies are created, maintaining a lower
hydrogen content in the bronze may be preferable to suppress
hydrogenation reactions.
[10] P. Claus, H. Hofmeister, C. Mohr, Gold Bulletin 37 (2004) 181–186.
[11] B. Coq, F. Figueras, P. Geneste, C. Moreau, P. Moreau, M. Warawdekar, Journal
of Molecular Catalysis 78 (1993) 211–226.
[12] W. Grunert, A. Bruckner, H. Hofmeister, P. Claus, Journal of Physical Chemistry
B 108 (2004) 5709–5717.
[13] C. Hoang-Van, O. Zegaoui, Applied Catalysis A: General 164 (1997) 91–103.
[14] T. Marinelli, S. Nabuurs, V. Ponec, Journal of Catalysis 151 (1995) 431–438.
[15] C. Mohr, H. Hofmeister, P. Claus, Journal of Catalysis 213 (2003) 86–94.
[16] C. Mohr, N. Hofmeister, M. Lucas, P. Claus, Chemical Engineering & Technology
23 (2000) 324–328.
[17] Y. Nagase, H. Hattori, K. Tanabe, Chemistry Letters (1983) 1615–1618.
[18] C. Hoang-Van, O. Zegaoui, Applied Catalysis A 130 (1995) 89–103.
[19] T. Matsuda, Y. Hirata, S. Suga, H. Sakagami, N. Takahashi, Applied Catalysis A
193 (2000) 185–193.
[20] J.E. Herrera, J.H. Kwak, J.Z. Hu, Y. Wang, C.H.F. Peden, J. Macht, E. Iglesia, Journal
of Catalysis 239 (2006) 200–211.
[21] R.K. Grasselli, Topics in Catalysis 21 (2002) 79–88.
[22] P. Mars, D.W.V. Krevelen, Chemical Engineering Science 3 (Special Suppl.)
(1954) 41.
[23] D.R. Moberg, T.J. Thibodeau, F.G. Amar, B.G. Frederick, Journal of Physical Chem-
istry C 114 (2010) 13782–13795.
[24] L. Bartha, A.B. Kiss, T. Szalay, International Journal of Refractory Metals & Hard
Materials 13 (1995) 77–91.
[25] A. Lackner, T. Molinari, P. Paschen, Scandinavian Journal of Metals 25 (1996)
115–121.
[26] A.J.V. Hefedus, T. Millner, J. Neugebauer, K. Sasvári, Zeitschrift für Anorganische
und Allgemeine Chemie 281 (1955) 64–82.
5. Conclusions
[27] K. Kanamaru, H. Kuwata, T. Okino, Thermochimica Acta 267 (1995) 117–127.
[28] K.M. Sancier, Journal of Catalysis 23 (1971) 298–300.
[29] P.A. Sermon, G.C. Bond, Catalysis Review 8 (1973) 211–239.
[30] Z. Lu, S.M. Kanan, C.P. Tripp, Journal of Material Chemistry 12 (2002) 983–989.
[31] E.P. Barrett, L.G. Joyner, P.P. Halenda, Journal of American Chemical Society 73
(1951) 373–380.
[32] M. Kruk, M. Jaroniec, A. Sayari, Langmuir 13 (1997) 6267–6273.
[33] S. Brunauer, P.H. Emmett, E. Teller, Journal of American Chemical Society 60
(1938) 309–319.
[34] M. Lu, MS Thesis, Chemistry, University of Maine, 2005.
[35] R. Tokarz-Sobieraj, M. Witko, R. Grybos, Catalysis Today 99 (2005) 241–253.
[36] E.R.S. Winter, Journal of Chemical Society (A). Inorganic Physics Theory (1968)
2889–2902.
[37] J.A. Bustnes, D. Sichen, S. Seetharaman, Metallurgical Transactions B 24B (1993)
475–480.
From this study we have gained insight about the composition
of tungsten oxide catalysts during pretreatment and under reac-
tion conditions for the reduction of acrolein and allyl alcohol. Using
thermogravimetric analysis under conditions of flowing hydrogen,
nitrogen, and then oxygen, we have developed a method to deter-
mine the composition of tungsten oxide bronzes, HyWO3−z, from
the relative weight changes. The evolved gases were monitored
with a mass spectrometer. The net weight loss in the 300–500 ◦C
hydrogen treatment step includes both weight loss due to removal
of lattice oxygen and weight gain due to dissolution of hydrogen to
form a bronze. Use of a platinum TGA pan increased the hydrogen
composition, which appears to stabilize the lattice oxygen.
[38] R. Hasegawa, T. Kurosawa, T. Yagihashi, Transactions of JIM 15 (1974) 75–82.
[39] X. Sha, L. Chen, A.C. Cooper, G.P. Pez, H. Cheng, Journal of Physical Chemistry C
113 (2009) 11399–11407.
[40] B. Grzybowska, J. Haber, J. Janas, Journal of Catalysis 49 (1977) 150–163.
[41] W. Martir, J.H. Lunsford, Journal of American Chemical Society 103 (1981)
3728–3732.
Using a catalyst pretreated at 350 ◦C, we have shown that
tungsten oxide bronzes of form HyWO3−z are capable for the hydro-
genation of allyl alcohol at lower temperatures (50–150 ◦C). More
importantly, these same catalysts are capable of hydrodeoxygena-
[42] P.J. Hart, H.R. Friedli, Journal of Chemical Society (D) (1970) 621–622.