Modeling aqueous-phase hydrodeoxygenation of sorbitol over Pt/SiO 2-Al2O3

Article abstract of DOI:10.1039/c3ra45179h

In this paper, we investigated the effects of temperature, hydrogen partial pressure, and sorbitol concentration on the aqueous-phase hydrodeoxygenation (APHDO) of sorbitol over a bifunctional 4 wt% Pt/SiO₂-Al ₂O₃ catalyst in a trickle bed reactor. APHDO involves four fundamental reactions: (1) hydrogenation; (2) dehydration; (3) C-C bond cleavage by dehydrogenation and decarbonylation; and (4) C-C bond cleavage by dehydrogenation and retro-aldol condensation. The main deoxygenation routes are decarbonylation and alcohol dehydration. Retro-aldol condensation plays a critical role in reducing the carbon number of the products. The key products in this system are C1-C6 n-alkanes, primary and secondary alcohols, and carbon dioxide. As shown in this paper, the reaction conditions can dramatically change the product selectivity for APHDO of biomass-derived feedstocks (e.g., sorbitol). A sorbitol hydrodeoxygenation reaction network was generated that predicts all of the 43 experimentally measured species. The reaction network consists of 4804 reactions and produces a total of 1178 distinct chemical species. The associated material balance equations were solved numerically to model the experimentally observed species as a function of temperature, concentration, and pressure. The model concentrations fit well the experimentally measured values, demonstrating that the model was accurately able to model the reaction families and capture the salient features of the experimental observations. The trend observed in this paper can be used for the optimization of reactors and new catalysts to selectively make targeted products by hydrodeoxygenation of biomass-derived feedstocks.

Full text of DOI:10.1039/c3ra45179h

RSC Advances
View Article Online
View Journal | View Issue
PAPER
Modeling aqueous-phase hydrodeoxygenation of
sorbitol over Pt/SiO2–Al2O3†
Cite this: RSC Adv., 2013, 3, 23769
a
b
b
b
*
Brian M. Moreno, Ning Li,‡ Jechan Lee,§ George W. Huber§
a
*
and Michael T. Klein
In this paper, we investigated the effects of temperature, hydrogen partial pressure, and sorbitol
concentration on the aqueous-phase hydrodeoxygenation (APHDO) of sorbitol over a bifunctional 4 wt
% Pt/SiO2–Al2O3 catalyst in a trickle bed reactor. APHDO involves four fundamental reactions: (1)
hydrogenation; (2) dehydration; (3) C–C bond cleavage by dehydrogenation and decarbonylation; and
(4) C–C bond cleavage by dehydrogenation and retro-aldol condensation. The main deoxygenation
routes are decarbonylation and alcohol dehydration. Retro-aldol condensation plays a critical role in
reducing the carbon number of the products. The key products in this system are C1–C6 n-alkanes,
primary and secondary alcohols, and carbon dioxide. As shown in this paper, the reaction conditions can
dramatically change the product selectivity for APHDO of biomass-derived feedstocks (e.g., sorbitol). A
sorbitol hydrodeoxygenation reaction network was generated that predicts all of the 43 experimentally
measured species. The reaction network consists of 4804 reactions and produces a total of 1178 distinct
chemical species. The associated material balance equations were solved numerically to model the
experimentally observed species as a function of temperature, concentration, and pressure. The model
concentrations fit well the experimentally measured values, demonstrating that the model was
accurately able to model the reaction families and capture the salient features of the experimental
observations. The trend observed in this paper can be used for the optimization of reactors and new
catalysts to selectively make targeted products by hydrodeoxygenation of biomass-derived feedstocks.
Received 17th September 2013
Accepted 25th September 2013
DOI: 10.1039/c3ra45179h
www.rsc.org/advances
During sorbitol conversion, oxygen is removed by two
different pathways: (1) alcohol dehydration (removal of
1. Introduction
The sustainable production of liquid fuels and chemicals from
biomass has attracted much attention as an alternative to fossil
fuels because of the availability and low cost of lignocellulosic
biomass.1–6 Biomass feedstocks have a high number of highly
reactive organic functional groups. Biomass conversion involves
selective removal of oxygen from the biomass.6–8 The biomass
feedstock reacts with hydrogen via aqueous-phase hydro-
deoxygenation (APHDO) to produce alkanes, alcohols, and
polyols.9–22 Bifunctional catalyst systems are typically used for
APHDO reactions and involve metal23,24 and acid sites.13–25
Sorbitol has previously been used as a model compound to
study the HDO of biomass.14,16,26
hydroxyl groups as water primarily on acid catalytic sites) and
(2) decarbonylation (removal of aldehyde groups as carbon
monoxide primarily on metal catalytic sites).27,28 Hydrogen
then reacts with the C]C and C]O bonds generated during
alcohol dehydration to produce a saturated compound. The
dehydration route is advantageous compared to decarbon-
ylation because no carbon from the biomass is lost in this
process. However, the dehydration route does consume large
amounts of hydrogen, and produces very large quantities of
water.29
A schematic network of the reaction families for HDO of
sorbitol is presented in Fig. 1. The reactions that occur during
this process include alcohol dehydration, cyclization, dehydro-
genation, decarbonylation, hydrogenolysis, and retro-aldol
condensation. Measured products are produced by a stepwise
combination of these reaction pathways, as indicated in the
gure. Furan and pyran species are produced via cyclization of
C4–C6 alcohols, followed by numerous dehydration and
cracking reactions. Several of the smaller species are produced
via retro-aldol condensation. The carbon monoxide produced
from decarbonylation is converted to carbon dioxide via the
water–gas shi reaction.
aDepartment of Chemical and Bimolecular Engineering, University of Delaware, 150
Academy Street, Newark, DE 19716, USA. E-mail: mtk@udel.edu
bDepartment of Chemical Engineering, University of Massachusetts Amherst, 686 North
Pleasant Street, Amherst, MA 01003, USA
† Electronic supplementary information (ESI) available. See DOI:
10.1039/c3ra45179h
‡ Current address: Dalian Institute of Chemical Physics, Chinese Academy of
Sciences, Dalian 116023, China.
§ Current address: Department of Chemical and Biological Engineering, University
of Wisconsin-Madison, 1415 Engineering Drive, Madison, WI 53706, USA.
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23769

View Article Online
RSC Advances
Paper
Fig. 1 Major reaction pathways for aqueous-phase hydrodeoxygenation (APHDO) of sorbitol.
While there has been a tremendous interest in using HDO loaded on different supports by incipient wetness impregna-
reactions to convert biomass into fuels and chemicals, there has tion with tetra-amine platinum nitrate (Strem Chemicals)
been no kinetic study done for HDO. The objective of this paper aqueous solution according to literature.9,10 The platinum
is to develop a kinetic model for HDO of sorbitol that can content in the catalyst was 4 wt%. The mixture was then dried
describe the complicated products that are produced in this in an oven overnight at 373 K and calcined at 533 K for 3 h in
process. Molecular-level kinetic models preserve the identities air. The BET surface area of the Pt/SiO2–Al2O3 as prepared was
of each species within a chemical system. Since the species measured as 430 m2 gꢀ1 by nitrogen adsorption at 77 K using a
produced during biomass conversion are not as well docu- Quantchrome Autosorb Automated Gas Sorption System. From
mented as those produced from traditional petroleum fuels, the hydrogen chemisorption experiment carried out at 303 K
molecular-level models are very helpful to identify reaction on the same instrument, the catalysts had a hydrogen uptake
intermediates. With these specic molecular identities, it is of 74.25 mmol H2 gcatalystꢀ1, which corresponds to an H to Pt
possible to predict the physical properties of the novel ratio of 0.724.
oxygenate species in biomass pyrolysis oils and deoxygenated
products. Furthermore, the kinetic rates of each deoxygenation
pathway could direct research for the design of improved cata-
lysts as well as help to identify the reaction conditions that will
make biofuels a competitive energy source.
2.2. Catalyst test
The hydrodeoxygenation experiments were carried out in a
stainless steel tubular ow reactor heated by a Lindberg (type
54032) furnace. In order to ensure a uniform temperature
prole along the catalyst bed, an aluminum tube was split
lengthwise and was inserted in the void between the heating
tube of the furnace and tubular reactor. Subsequently, it was
2. Experimental methods
2.1. Catalyst preparation
The SiO2–Al2O3 (SIAL3125, SiO2–Al2O3 molar ratio about 4.0) found that the temperature difference between the middle and
support was supplied by Grace Davison. The platinum was top of the tubular reactor was less than 15 K.
23770 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
Prior to the reaction, the Pt/SiO2–Al2O3 was reduced in the
reactor with hydrogen owing from the bottom at approxi-
mately 200 mL minꢀ1. The reduction temperature of the Pt/
SiO2–Al2O3 catalyst increased from room temperature to 723 K
at 50 K hꢀ1 and was then held at 723 K for 2 h. The reactor was
then cooled down to the reaction temperature and the pressure
was slowly increased to the desired value.
The liquid feed (5 wt% sorbitol aqueous solution) was co-fed
with hydrogen (45 mL minꢀ1) through the catalyst from the
bottom of reactor with the help of a JASCO PU980 HPLC pump.
A gas–liquid separator was utilized downstream of the the
reactor tube. The gaseous products from the reactor owed
through a backpressure regulator, used to maintain the pres-
sure of the reaction system.
The gaseous products were further analyzed by two online
gas chromatographs (HP 5890 series II). Carbon dioxide in the
gaseous product was analyzed by a Thermal Conductivity
Detector (TCD) in conjunction with an Alltech HAYESEP DB
100/120 packed column (Part no. 2836PC) that was heated at a
constant temperature (348 K). The TCD and the injection port
were held at 433 K and at 393 K, respectively. The column ow
rate was 41 mL minꢀ1 with helium as the carrier gas.
RSC Advances
3. Modeling methods
3.1. Generation of reaction network via INGen
The Klein research group has enabled automated network
generation for biomass pyrolysis and subsequent hydro-
deoxygenation through the development of INGen, the Inter-
active Network Generator.30 User-selected seed molecules and
allowable reaction types determine the size and complexity of
the reaction network.
INGen combines relevant mechanistically informed reaction
pathways to completely describe the reactivity of a given
chemical system. The intuitive interface allows a user to select
reaction types and seed molecules to initialize the reaction
network. INGen then automatically generates an exhaustive
reaction network by searching for reactable moieties within
each molecule and writing all possible reactions. Performing
this kind of search on a computer requires the computational
denition of a chemical species as well as a way to transform
those reactant species into distinct products.
Chemical species are computationally dened in INGen
using bond electron matrices, which are intuitive representa-
tions of the atoms and their corresponding bond connections
within molecules. Each element of a bond electron matrix
indicates the formal bond order between any given pair of
atoms in a molecule. A large library of preexisting species is
included with INGen, and new species may be added by simply
sketching the molecule in ChemDraw and converting to the
appropriate le type.
Given the bond electron matrix representation of a reactant
molecule, it becomes apparent that the addition and subtrac-
tion of bonds within that matrix represents the breaking and
forming of bonds in a chemical reaction. INGen writes chemical
reactions by applying reaction matrices for each reaction family
to all of the reactable moieties within each chemical species. A
reactant matrix is converted into a different product matrix
during each chemical reaction.
For each reaction family in this model, the reaction matrix is
given along with a sample reactant and product matrix pair in
Fig. 2. The reactive atomic sites are listed in a particular order to
standardize the reaction operator for that reaction family. A
reaction is dened by three properties: the site neighborhood,
the reaction site, and the reaction operator.
The site neighborhood is the connectivity of atoms that is
required to participate in a given reaction type, although all of
the atoms within the site neighborhood do not necessarily
change connectivity during the reaction. Site neighborhoods are
most closely analogous to chemical moieties, such as the
hydroxyl group in the reactant depicted in Fig. 2b. The reaction
site itself denes which atoms will participate in the reaction,
and the order in which they will be acted upon. These are the
atoms whose connectivities will change during the reaction.
The reaction operator (the reaction matrix specic to the reac-
tion family) denes how to affect the bonds of the site atoms.
Generally, only a small number of atoms and bonds are modi-
ed in a given chemical reaction.
Alkanes in the gaseous product were analyzed with a ame
ion detector (FID) with an Alltech AT-Q capillary column (30 m,
0.32 mm I.D., Part no. 13950). Helium was used as the carrier
gas with a column ow rate of 1 mL minꢀ1. Both the injection
port and the detector were held at 473 K. The split ratio was 100.
The GC oven temperature was held at 313 K for 6 min, ramped
to 453 K at 5 K minꢀ1, and held at 453 K for 25 min.
Liquid product that accumulated in the gas–liquid separator
was drained periodically and analyzed by GC-MS and HPLC.
The GC-MS is Shimadzu GC-2010 with an Rtx-VMS capillary
column (30 m, 0.25 mm I.D., lm thickness 1.4 mm). Helium
was used as the carrier gas with a column ow rate of 1.57 mL
minꢀ1. Both the injection port and the detector were held at 513
K. For each analysis, a 1 mL liquid sample was injected with a
split ratio of 200. The column was held at 308 K for 5 min,
ramped to 323 K at 5 K minꢀ1, then to 513 K at 20 K minꢀ1, and
kept at 513 K for 7.5 min. A Shimadzu HPLC with UV-Vis (SPD-
20AV) and RID (RID-10A) detectors was also used to analyze the
liquid products. A 1 mL liquid sample was injected for each
sample. The products were separated by a BIO-RAD Aminex
HPX-87H column (Catalog no. 125-0140) that was maintained
at 303 K with 0.005 M H2SO4 as the mobile phase owing at a
rate of 0.6 mL minꢀ1
.
Finally, a total organic carbon (TOC) analyzer (Shimadzu
TOC-5000A) was used to check the carbon balance. The liquid
sample was taken at least 12 h aer the reaction conditions were
changed when the gas phase results were constant. From the GC
analysis results, reaction equilibrium was reached within 8 h.
To exclude the effect of deactivation of the catalyst, we
checked the stability of the catalyst under each reaction
condition. No evident change of activity or selectivity of the
catalyst was observed aer a continuous run of 24 h, indicating
that the catalyst is stable under the conditions used in this
paper. There was no coking on the catalyst under the reaction
conditions we used in this process.
To exemplify each of the site selection aspects of automated
network generation, we may continue to focus on the alcohol
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23771

View Article Online
RSC Advances
Paper
Fig. 2 (Contd.)
23772 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
RSC Advances
Fig. 2 (Contd.)
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23773

View Article Online
RSC Advances
Paper
Fig. 2 Chemical reactions as matrix operations. In most cases, R groups on carbon atoms are implicit if they do not directly participate in the reaction.
23774 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
RSC Advances
cyclization reaction depicted in Fig. 2b. We must rst locate a extent to which individual reactions within a given reaction
specic sequence of atoms: H1–O2–C3–C4–C5–C6–O7–H8. The family depend upon their unique energy barrier (ꢀ1 # ai # 1).
reaction occurs when the reaction matrix is added to the reac-
E*j ¼ E*0,i + aiDHR,j
(2)
tant matrix, which breaks and forms chemical bonds and gives
the product species. A single bond electron matrix temporarily
represents the two product species. This process of adding
reaction matrices to appropriately selected reactant matrices
leads to the generation of a complete reaction network for the
given chemical system.
Pinacol rearrangement (Fig. 2n) is a particularly interesting
reaction family in this model. The overall reaction in a pinacol
rearrangement produces water and a carbonyl (aldehyde or
ketone).31,32 This reaction family specically affects pairs of
hydroxyl groups on adjacent carbon atoms in which traditional
alcohol dehydration would have resulted in an unstable enol
(carbon–carbon double bond with an adjacent hydroxyl group).
Sorbitol and its products contain many adjacent hydroxyl
groups that undergo pinacol rearrangements.
An example of this relationship is the hydrogenation of hexa-
1,3-diene to hex-1-ene (Reaction 2402) and hex-3-ene (Reaction
2403), in which the hydrogenation of the secondary alkene is
more energetically favorable than the hydrogenation of the
primary alkene (DHR,2402 < DHR,2403). Therefore, although both
of these reactions are members of the same reaction family, the
rate constant for the hydrogenation of the secondary alkene is
larger (k2402 > k2403).
The enthalpy change for each of the 4804 reactions is avail-
able as ESI (Table 9).† To simplify parameter tuning and to
account for the identical reaction temperature of tuned data
sets, E*0,i was set to zero for each reaction family. Thus, the only
tuned parameters in this kinetic model are log k0,i and ai for
each of the 20 reaction families, resulting in a total of 40 tuned
kinetic parameters.
All reactions were written irreversibly to simplify the model
and shorten the solution time. When applicable, reverse reac-
tions were written separately.
4. Results
4.1. Aqueous-phase hydrodeoxygenation of sorbitol
3.2. Tuning of kinetic parameters via KME
4.1.1. Effects of reaction temperature. We investigated the
effect of reaction temperature on the catalytic performance of
the 4 wt% Pt/SiO2–Al2O3 at weight hourly space velocities
(WHSV) of 0.73 hꢀ1 and 2.91 hꢀ1 as shown in Table 1. The gas
products were composed of CO2 and C1–C6 straight-chain
alkanes. We did not detect any CO in the gas phase. The liquid
phase products included alcohols, diols, polyols, ketones, cyclic
ethers, 1,4-sorbitan, isosorbide, and unconverted sorbitol. We
were able to quantify between 76 and 100 mol% of the products
in Table 1. The carbon balances were occasionally low as a result
of some unidentied species that were not reported.
The gas phase yield increases with reaction temperature. A
lower CO2 gas phase selectivity (C–C bond cleavage) and higher
alkane selectivity (C–O bond cleavage) is observed at low WHSV
and/or high temperature. The specic gas phase selectivities of
C1–C6 alkanes were very similar.
This chemical system was modeled as a PFR with 3.3 g of Pt/
SiO2–Al2O3 catalyst at a loading density of 1.26 g cmꢀ3, resulting
in reactor dimensions of 0.246 cm in diameter and 54.9 cm in
length. Molar ow rates for positional isomers were divided
according to the percentages in Table 8 (ESI†).
From the completed reaction network, all of the mass
balances are automatically generated and subsequently solved
numerically by KME (the Kinetic Modeling Editor).33 Kinetic
rate parameters were tuned using the data sets that had 100–105
mol% carbon identied with identical sorbitol feed concentra-
tion of 5 wt%, temperature of 518 K, and weighted hourly space
velocity (WHSV) of 0.73 hꢀ1, where the pressure varied between
2.93 and 4.90 MPa (data sets 2, 8, 9, and 10). Kinetic parameter
tuning was performed using simulated annealing of a least
squares optimization function, F (minimization of the sum of
squares of residuals), via KME.
The C1 liquid phase selectivity increased with reaction
temperature at WHSV ¼ 2.91 hꢀ1. The C2–C3 selectivity went
through a maximum at low WHSV. This same fraction increased
with temperature at high WHSV. The C1–C3 products were
mainly composed of methanol, ethanol, propanol, propane-1,2-
diol, and glycerol.
The adjustable parameters in this kinetic model are given in
eqn (1) and (2). The Arrhenius expression for the rate constant,
kj, identies the pre-exponential factor, k0,i, and the activation
energy, E*j , for each reaction, j.
log kj ¼ log k0,i ꢀ E*j /2.303RT
(1)
The selectivity of C4 in the liquid phase decreased with
The Evans–Polanyi linear free energy relationship (LFER)34 temperature at low WHSV and increased with temperature at
given in eqn (2) allows for differences in the rate constants of high WHSV. At high space velocity (WHSV ¼ 2.91 hꢀ1), the
individual reactions (j) within reaction families (i). These selectivity of C6 compounds decreased with reaction tempera-
differences result from the enthalpy change of each reaction, ture (higher reaction rate at high temperature), while the C1–C5
which leads to faster kinetic rates for energetically favored selectivity increased.
reactions. E*j , the activation energy of reaction j, is a function of
E*0,i, the activation energy of reaction family i (set to zero), ai, the effect of system pressure for the APHDO of 5 wt% sorbitol
LFER constant of reaction family i (tuned by KME), and DHR,j solution over the Pt/SiO2–Al2O3 catalyst as shown in Table 2. As
4.1.2. Effects of system pressure. We also investigated the
,
the enthalpy change of reaction j (calculated by group contri- total pressure increased, the initial concentration of hydrogen
bution methods via CME).35 The LFER constant indicates the and sorbitol at the reactor inlet increased. The carbon
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23775

View Article Online
RSC Advances
Paper
Table 1 Total molar carbon selectivity of products in gas- and liquid-phase as a function of temperature for the APHDO of sorbitola
Data set number
WHSV (hꢀ1
Temperature (K)
Conversion (mol%)
Gas phase yield (mol%)
1
2
3
4
5
6
)
0.73
0.73
0.73
2.91
498
74.6
2.91
2.91
538
100
498
100
37.8
518
100
80.8
538
100
85.3
518
100
39.7
10.8
71.5
Gas phase carbon selectivity (mol%)
C1
Carbon dioxide
Methane
Ethane
Propane
Butane
59.9
5.2
10.1
7.5
7.1
5.3
4.9
46.6
1.5
7.3
11.9
4.1
40.4
5.2
13.8
9.7
10.4
8.9
11.6
20.9
2.1
5.7
8.8
1.0
1.9
41.0
5.9
14.8
9.9
10.7
8.6
9.1
18.9
1.2
2.6
3.9
0.7
90.1
0.6
1.2
1.2
1.2
2.2
3.6
66.7
0.3
1.4
5.9
0.5
8.0
50.6
62.6
5.4
9.0
6.6
6.5
5.7
4.2
36.3
1.1
1.2
51.5
5.0
11.7
7.3
8.2
8.0
8.3
26.9
1.4
2.9
4.9
2.5
8.3
7.0
C2
C3
C4
C5
C6
Pentane
Hexane
Liquid phase yield (mol%)
C1
C2
C3
C4
C5
C6
7.0
1.7
10.0
15.3
10.4
11.4
4.0
6.6
1.4
Liquid phase carbon selectivity (mol%)
C1
C2
C3
Methanol
Ethanol
Propanol
Acetone
3.3
15.6
16.0
0.9
0
9.9
27.5
40.3
0.9
0
6.2
13.5
17.7
1.0
0
0.4
2.1
1.1
0.2
0
2.9
3.3
7.4
2.1
0
5.1
10.7
13.1
2.8
0
Glycol
Propane-1,2-diol
1-Hydroxypropan-2-one
Glycerol
Butanol
Butanone
Tetrahydrofuran
Butanediol
Pentanol
8.5
0
0
4.5
0
0
4.3
3.9
0.7
3.1
2.1
1.5
11.1
5.0
0.8
1.2
3.3
0
1.0
0
0
4.6
0
0
0
0.8
0
5.6
2.1
0.7
0
0.7
0
1.7
0
0
1.6
2.0
0
3.6
0
4.0
0.5
0.2
0
9.8
0
0
0.9
2.3
0
1.6
4.0
2.0
2.7
3.5
5.0
10.2
8.7
8.9
1.8
3.2
0.5
18.9
0
2.5
0
0
3.8
5.3
0
C4
C5
0
0
0
4.9
1.0
5.5
2.1
2.6
5.3
10.7
1.3
1.0
4.1
0.7
17.1
0
0.3
0.2
0.3
0.5
7.8
2.9
1.1
1.1
0.6
0.3
1.4
6.2
0
3.3
3.2
5.0
3.8
3.1
12.2
10.4
3.7
1.0
7.2
0.8
2.9
0
Pentanone
Tetrahydropyran
2-Methyltetrahydrofuran
Pentane-1,2-diol
(Tetrahydrofuran-2-yl)methanol
Hexanol
C6
Hexanone
2,5-Dimethyltetrahydrofuran
2-Methyltetrahydropyran
Hexane-1,2-diol
(Tetrahydropyran-2-yl)methanol
Hexane-1,2,6-triol
Isosorbide and 1,4-sorbitan
Sorbitol
0.8
4.3
0
0.9
0
14.2
0
0
0
84.3
0
0
0
0
27.0
38.0
77.5
0
0
76.0
0
0
98.4
Carbon identied (mol%)
101.7
104.2
a
Reaction condition: 2.93 MPa with 5 wt% sorbitol solution as the feed and H2 ow rate of about 45 mL minꢀ1
.
conversion to the gas phase products went through a maximum alcohols and cyclic ethers at low initial hydrogen concentra-
at 3.48 MPa. The C5 and C6 selectivity in both the gas phase and tion. With the increase in initial hydrogen concentration, the
the liquid phase increased with total pressure. In contrast, the yield of C4–C6 increased, and the major compounds changed
C1 selectivity in both the gas phase (mainly composed of CO2) from cyclic ethers to alcohols (butanol, pentanol, and
and the liquid phase (methanol) decreased with total pressure hexanol).
(i.e., hydrogen concentration at the reactor inlet).
4.1.3. Effects of sorbitol feed concentration. Table 3 shows
The liquid products were mainly alcohols. The yield of C1– the effect of sorbitol feed concentration on the catalytic
C3 compounds decreased with increasing system pressure. performance of 4 wt% Pt/SiO2–Al2O3 catalyst. As sorbitol feed
The major C4–C6 compounds in the liquid phase were concentration increased, the carbon conversion to the gas
23776 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
RSC Advances
Table 2 Total molar carbon selectivity of products in gas- and liquid-phase as a function of pressure for the APHDO of sorbitola
Data set number
2
8
9
10
11
Pressure (MPa)
Conversion (mol%)
Gas phase yield (mol%)
2.93
100
80.8
3.48
100
90.5
4.17
100
86.9
4.9
100
74.7
5.65
100
64.3
Gas phase carbon selectivity (mol%)
C1
Carbon dioxide
Methane
Ethane
Propane
Butane
40.4
5.2
13.8
9.7
10.4
8.9
11.6
20.9
2.1
5.7
8.8
1.0
1.9
26.3
3.9
10.6
9.1
11.1
14.5
24.6
10.0
1.3
3.0
4.1
0.3
0.8
24.6
3.3
9.8
19.4
1.8
7.9
10.3
10.1
16.1
34.5
30.0
0.8
4.9
11.0
2.5
3.7
7.1
18.3
1.4
7.3
10.8
9.6
17.8
34.8
44.5
0.7
4.6
12.2
3.7
6.4
16.9
C2
C3
C4
C5
C6
9.3
11.4
14.7
26.8
15.7
1.3
4.8
6.9
0.7
1.1
Pentane
Hexane
Liquid phase yield (mol%)
C1
C2
C3
C4
C5
C6
1.4
0.5
0.9
Liquid phase carbon selectivity (mol%)
C1
C2
C3
Methanol
Ethanol
Propanol
Acetone
9.9
27.5
40.3
0.9
0
13.4
30.0
40.9
0
0
8.6
30.7
44.2
0
0
2.8
16.4
36.5
0
1.5
10.3
27.5
0
Glycol
0
0
Propane-1,2-diol
1-Hydroxypropan-2-one
Glycerol
1.0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
C4
C5
Butanol
Butanone
Tetrahydrofuran
Butanediol
Pentanol
4.6
0
0
0
0.8
0
2.9
0
0
0
2.7
0
4.2
0
0
0
3.4
0
8.5
0
0
0
6.7
0
8.2
0
0
0
7.9
0
Pentanone
Tetrahydropyran
2-Methyltetrahydrofuran
Pentane-1,2-diol
(Tetrahydrofuran-2-yl)methanol
Hexanol
5.6
2.1
0.7
0
0.7
0
0.8
4.3
0
0.9
0
5.4
0
0
0
1.4
0
0
3.3
0
0
0
3.3
0
0
0
2.4
0
0
3.2
0
0
0
3.0
0
2.4
2.1
0
2.1
28.5
3.9
3.3
1.4
0
1.1
1.4
15.4
1.3
3.0
3.8
0
0.2
0
0
C6
Hexanone
2,5-Dimethyltetrahydrofuran
2-Methyltetrahydropyran
Hexane-1,2-diol
(Tetrahydropyran-2-yl)methanol
Hexane-1,2,6-triol
Isosorbide and 1,4-sorbitan
Sorbitol
0.8
0
0
0
0
0
0
0
0
0
0
Carbon identied (mol%)
101.7
100.5
102.6
104.8
108.8
a
Reaction condition: 518 K, WHSV ¼ 0.73 hꢀ1, with 5 wt% sorbitol solution as the feed, and H2 ow rate of about 45 mL minꢀ1
.
phase decreased. The C5 and C6 gas phase selectivity went hexanol and hexanediol were maximized at a sorbitol feed
through a maximum with respect to sorbitol feed concentra- concentration of 20 wt%.
tion. As the sorbitol feed concentration increased from 10 to 50
wt%, the C1 gas phase selectivity increased and the C2–C6 gas went through a maximum around a sorbitol feed concentration
phase selectivity decreased. of 10 wt%. The major C1–C3 product at a sorbitol feed
In the liquid phase products, the yield of C1–C3 compounds
The increase in the sorbitol feed concentration increased the concentration of 10 wt% to 50 wt% was propane-1,2-diol. At the
yield of C6 carbon that was in the liquid phase. At the highest lowest sorbitol feed concentration, the C1–C3 product selec-
sorbitol feed concentration, most of this carbon was present as tivity favors primarily alcohols (selectivity sequence: propanol >
isosorbide and unconverted sorbitol. The selectivities of ethanol > methanol).
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23777

View Article Online
RSC Advances
Paper
Table 3 Total molar carbon selectivity of products in gas- and liquid-phase as a
function of sorbitol feed concentration for the APHDO of sorbitola
4.2. Reaction network statistics
The automatically generated network consists of 4804 irrevers-
ible chemical reactions in total. The reaction count for each
individual reaction family is given in Table 4. The most preva-
lent reaction families are alcohol dehydration and pinacol
rearrangement.
Data set number
2
13
14
15
Sorbitol feed concentration (wt%)
Conversion (mol%)
Gas phase yield (mol%)
5
10
100
26.6
20
100
10.9
50
79.0
4.2
100
80.8
A total of 1178 distinct chemical species are included in the
reaction network, while only three species (sorbitol, hydrogen,
and water) were used for network generation. The maximum
number of experimentally measured species (including posi-
tional isomers) in a given data set was 43. The complete reaction
network and species list are available as ESI (Tables 9 and 10,
Gas phase carbon selectivity (mol%)
C1 Carbon dioxide
Methane
C2 Ethane
C3 Propane
C4 Butane
C5 Pentane
C6 Hexane
40.4
5.2
13.8
9.7
10.4
8.9
11.6
20.9
2.1
5.7
8.8
1.0
1.9
30.2
1.8
6.4
7.5
7.0
14.0
33.2
86.8
1.2
52.4 75.8
2.6
7.5
7.6
2.0
2.4
1.8
3.8
5.7
9.5
5.3 respectively†).
14.7
8.9
Liquid phase yield (mol%)
60.2 85.2
4.3. Tuning of kinetic parameters
C1
C2
C3
C4
C5
C6
0.7
1.8
0.3
0.6
3.5
Each of the tuned kinetic rate parameters for all 20 of the
reaction families is given in Table 5. The LFER given in eqn (2)
is nontrivial only for the 17 reaction families with nonzero
ai parameters. The remaining reaction families share a
single rate constant (log k0,i) for all reactions. With all E*0,i set
to zero, there are a total of 37 kinetic parameters in this tuned
model.
The relative rates of competing reaction pathways are shown
by the values of the tuned kinetic parameters. The bifunctional
catalyst promotes both dehydration (primarily on acid catalytic
sites) and decarbonylation (primarily on metal catalytic sites),
with a slightly faster rate of decarbonylation than alcohol
dehydration (RF4 versus RF9). The rate constants for aldehyde
hydrogenation and the opposing primary alcohol dehydroge-
nation reaction (RF6 and RF16) suggest that aldehydes do not
readily convert into primary alcohols. The high rates of decar-
bonylation and water–gas shi (RF9 and RF20) lead to signi-
cant production of carbon dioxide via the consumption of
aldehydes. The relative parameter values for RF10 and RF11
indicate that dihydrofuran rings saturate more readily than
aromatic furan rings.
43.4
19.8
8.4
13.5 11.7
15.1
8.5
4.3
4.3
1.4
10.5
20.8 64.0
Liquid phase carbon selectivity (mol%)
C1 Methanol
C2 Ethanol
9.9
27.5
40.3
0.9
0
1.0
0
0
4.6
0
0
0
0.8
0
5.6
2.1
0.7
0
0.7
0
0.8
4.3
0
0.9
0
1.4
4.1
7.2
0.4
8.5
32.6
1.3
0
1.8
7.9
3.3
9.8
2.1
0.5
0.6
1.4
1.9
3.3
4.1
1.1
1.7
0.2
3.1
0
1.1
2.9
7.8
0.5
0
13.1
1.0
0
2.1
13.4
4.0
5.5
2.7
1.5
0.8
2.5
2.5
4.0
5.4
10.3
3.6
0.8
4.9
2.0
3.7
0.4
0.7
0.8
0
C3 Propanol
Acetone
Glycol
Propane-1,2-diol
1-Hydroxypropan-2-one
Glycerol
2.4
7.0
1.4
2.2
0.3
0.5
1.3
3.0
0.7
0.6
0.2
1.1
1.1
1.3
0.5
1.8
0.9
0.2
2.3
0.6
0.5
C4 Butanol
Butanone
Tetrahydrofuran
Butanediol
C5 Pentanol
Pentanone
Tetrahydropyran
2-Methyltetrahydrofuran
Pentane-1,2-diol
(Tetrahydrofuran-2-yl)methanol
C6 Hexanol
Hexanone
2,5-Dimethyltetrahydrofuran
2-Methyltetrahydropyran
Hexane-1,2-diol
(Tetrahydropyran-2-yl)methanol
Hexane-1,2,6-triol
Isosorbide and 1,4-sorbitan
Sorbitol
4.4. Parity plots
A parity plot of each data set used for parameter tuning is given
in Fig. 3. Each plot compares the model prediction for the
product composition to the experimentally measured product
composition, with a diagonal line to indicate the ideal model t.
The parity plots for the data sets with the poorest t are shown
in Fig. 3a and d. These data sets are at the high and low ends of
the pressure range used for parameter tuning. The data set with
the worst model t also deviated the most from 100 mol%
identied carbon. The products with the lowest prediction
accuracy are ethanol and propanol. Fig. 4 shows a parity plot
1.6
0.3
0
0
0
3.7 48.7
19.7
71.1 89.4
0
Carbon identied (mol%)
101.7 113.5
a
Reaction condition: 518 K, 2.93 MPa, WHSV ¼ 0.73 hꢀ1, and H2 ow
rate of about 45 mL minꢀ1
.
The yield of C4–C5 liquid phase products also went through a with species labels to show the relative accuracy for these
maximum with respect to sorbitol feed concentration. At high products. The accuracy of the model t to the experimental data
sorbitol feed concentration, the major C4–C5 products were is summarized in Table 6.
butanediol, butanone, and tetrahydrofuran compounds. No C4–
Fig. 5 shows how the reaction pressure and identied carbon
C5 product contained more than two oxygen atoms. At the lowest percentage affect the accuracy of the model predictions. The
sorbitol feed concentration, the selectivity sequence of C4–C5 models were the most accurate (i.e., had the highest coefficients
products is tetrahydropyran > butanol > 2-methyltetrahydrofuran. of determination) for the data sets at the center of the pressure
23778 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
RSC Advances
Table 4 Reaction count by reaction family
conditions, such as temperature, pressure, and feed concen-
tration, can affect the reaction products. In this paper, we have
shown how yield and/or selectivity to C5–C6 alkanes, C1–C3
alcohols and diols, and C4–C6 alcohols and diols change with
reaction conditions. As we proposed previously,16 another
promising application of APHDO of sorbitol is the production of
low-oxygen-content species or monofunctional compounds.11
Some of these compounds, including 2-methyltetrahydrofuran
and 2,5-dimethyltetrahydrofuran, can be directly used as fuels
or fuel additives.36
RF#
Reaction family
Count
RF1
RF2
RF3
RF4
Acyclic thermal cracking
Alcohol cyclization 5
Alcohol cyclization 6
Alcohol dehydration
12
68
30
958
6
487
140
56
RF5
Alcohol ring closure
RF6
RF7
RF8
RF9
Aldehyde hydrogenation
C–O hydrogenolysis 5
C–O hydrogenolysis 6
Decarbonylation
487
126
18
C4–C6 alcohols, ketones, and diols can be used to produce
liquid fuels by aldol condensation followed by aqueous-phase
hydrodeoxygenation.11 The selectivity of C4–C6 alcohols and
diols increases with increasing pressure and goes through a
RF10
RF11
RF12
RF13
RF14
RF15
RF16
RF17
RF18
RF19
RF20
Furan ring saturation 2H
Furan ring saturation 4H
Hydrogenation (n-olen)
Hydrogenation (oxygenated species)
Ketone hydrogenation
Pinacol rearrangement
Primary alcohol dehydrogenation
Pyran ring saturation
Retro-aldol condensation
Reverse water–gas shi
Water–gas shi
21
417
491
613
487
167
218
1
maximum with respect to temperature (at WHSV ¼ 2.91 hꢀ1
)
and sorbitol feed concentration. The C4–C6 products are
formed via a series of reaction pathways from sorbitol. These
C4–C6 products are then converted into C4–C6 alkanes, C1–C3
oxygenates, or C1–C3 alkanes and CO2. Increasing the sorbitol
feed concentration decreases the overall sorbitol conversion
(see Table 3), which in turn decreases the yield of C4–C6 alco-
hols and diols.
1
4804
Total
Cyclic ether intermediates (such as isosorbide) are generated
by the dehydration of sorbitol. These products have a low rate of
C–C bond cleavage and can selectively produce large amounts of
C5 and C6 alkanes.16 The C5–C6 alkanes could be isomerized
and added into gasoline. The yield of C5–C6 alkanes increases
with increasing temperature at higher WHSV. This increase
results from the conversion of C5–C6 alcohols and diols into
alkanes. The yield of C5–C6 alkanes also increases dramatically
as the pressure increases from 2.93 to 3.48 MPa. Aer a further
increase in pressure, the C5–C6 yield does not continue to
increase. The maximum in the gasoline-range (C5 and C6
normal alkanes) product yield with respect to sorbitol feed
concentration results from the decrease in water concentration
with increased sorbitol concentration, which encourages the
dehydration of sorbitol. The comparable selectivity of C1–C4
light alkanes at similar gas phase carbon conversion implies
that the reaction pathway does not signicantly change with
temperature.
Table 5 Tuned kinetic parameters of each reaction family. n h reaction order, k0
h rate constant (M1ꢀn ꢀ1), a h LFER constant
s
RF#
Reaction family
n
log k0
a
RF1
RF2
RF3
RF4
RF5
RF6
RF7
RF8
Acyclic thermal cracking
Alcohol cyclization 5
Alcohol cyclization 6
Alcohol dehydration
Alcohol ring closure
Aldehyde hydrogenation
C–O hydrogenolysis 5
C–O hydrogenolysis 6
Decarbonylation
Furan ring saturation 2H
Furan ring saturation 4H
Hydrogenation (n-olen)
Hydrogenation (oxygenated species)
Ketone hydrogenation
Pinacol rearrangement
Primary alcohol dehydrogenation
Pyran ring saturation
Retro-aldol condensation
Reverse water–gas shi
Water–gas shi
1
1
1
1
1
2
2
2
1
2
3
2
2
2
1
1
2
1
2
2
5.5
6.4
7.1
7.3
3.8
5.7
11.8
11.4
7.8
8.1
4.5
13.3
6.1
6.2
7.2
6.6
7.9
7.6
0.0
10.0
1.00
ꢀ0.12
0.01
ꢀ0.01
0.16
0.04
1.00
1.00
RF9
0.07
0.09
RF10
RF11
RF12
RF13
RF14
RF15
RF16
RF17
RF18
RF19
RF20
ꢀ0.05
0.13
0.01
ꢀ0.03
0.00
ꢀ0.06
0.22
C1–C3 alcohols and diols are valuable commodity chemicals
that have functional value beyond their fuel value. The highest
C1–C3 yield was achieved at 518 K, 2.93 MPa, and WHSV ¼ 0.73
hꢀ1 with 10 wt% of sorbitol as the feed. Combined with the
observations for C4–C6 alcohols and diols, these conditions
appear to be optimal for alcohol and diol production. The
selectivity of C1–C3 alcohols and diols increases with reaction
ꢀ0.01
0.00
0.00
range (data sets 2, 8, 9, and 10; Fig. 5a). Model t also suffers temperature.
slightly as identied carbon deviates from 100 mol%, as shown
The change in hydrogen concentration resulting from reac-
in Fig. 5b.
tion pressure has a drastic effect on the product selectivity for
the APHDO of sorbitol. The hydrogen concentration can be used
to adjust the relative rates of C–C versus C–O bond cleavage. The
decrease in C1 product selectivity with increasing total pressure
is most likely due to an increase in hydrogen partial pressure.
5. Discussion
5.1. Effects of reaction conditions on target products
Hydrodeoxygenation of biomass-derived compounds produces Increasing the system pressure increases the hydrogen
a wide variety of species and is greatly inuenced by reaction concentration and most likely results in a decrease in the rate of
conditions. It is important to understand how reaction dehydrogenation, which is the rst step in C–C bond cleavage
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23779

View Article Online
RSC Advances
Paper
Fig. 3 Parity plots: predicted molar flow versus observed molar flow (ꢁ 10ꢀ7), mol sꢀ1. x0 h sorbitol feed concentration (wt%), P h pressure (MPa), T h temperature
(K), WHSV h weight hourly space velocity (hꢀ1), y h carbon identified in measured products (mol%), F h objective function value, R2 h coefficient of determination.
by decarbonylation.16 Therefore, higher hydrogen concentration these products were primarily formed via decarbonylation
inhibits decarbonylation.
reactions.
Propane-1,2-diol and glycerol are most likely formed by retro-
The shi in gas phase selectivity from C2–C6 to C1 with
increasing sorbitol feed concentration occurs primarily aldol condensation of sorbitol. The alcohols are formed by
because more carbon is cleaved via decarbonylation to CO, hydrodeoxygenation of propane-1,2-diol and glycerol or from
followed by immediate conversion to CO2 via water–gas shi. decarbonylation of larger oxygenates. The smaller increase in
We have previously shown that the C1 gas phase selectivity gas phase yield with increasing reaction temperature at low
increases as the WHSV increases and that most of the C1 gas WHSV can be explained because carbon conversion to the gas
phase products are formed from decarbonylation reactions.16 phase at low WHSV is very high (>80 mol%). As a result, the
The low oxygen content (two oxygen atoms or fewer) of C4–C5 oxygenate concentration in the solution was lower and the effect
products at high sorbitol feed concentration indicates that of temperature became less evident. The high selectivity of
23780 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
RSC Advances
All of the experimental data were at 100 mol% sorbitol
conversion, except for data sets 4 and 15 (which were at 74.6
mol% and 79.0 mol%, respectively). Those data sets have low
identied carbon percentages that did not allow for reliable
parameter tuning. Due to a lack of low conversion data, a
kinetic study to identify the reaction intermediates proved
challenging.
Data sets 2, 8, 9, and 10, which were used for parameter
tuning, have identical sorbitol feed concentration (5 wt%),
temperature (518 K), WHSV (0.73 hꢀ1), and identied carbon
balances near 100 mol%, while they differ only in system
pressure (2.93, 3.48, 4.17, and 4.90 MPa, respectively).
Temperature dependence could not be included in the rate
parameters for this model because there were an insufficient
number of data sets with near-100 mol% identied carbon at
varied temperature (with all other conditions constant).
Therefore, we could not effectively deduce the log A0,i and E0*,i
values and instead utilized a log k0,i value. As a result of the
temperature-independent rate parameters, accurate predictions
cannot be made at varied temperatures.
A requirement for accurate kinetic modeling is complete
product identication. If an insufficient (or extraneous)
percentage of products are measured, the numerical predic-
tions of the kinetic model will overestimate (or underestimate)
the experimental results. Most of the data sets with low coeffi-
cients of determination also have identied carbon percentages
that differ signicantly from 100 mol%.
The tuned kinetic rate parameters for competing reactions
indicate which reaction pathway controls the production of a
given species. For example, the rate constants for alcohol
dehydration and decarbonylation are very similar, which indi-
cates that these two deoxygenation methods are approximately
equally dominant at the tuned conditions. Balanced data at
various reaction temperatures would enable tuning of the acti-
vation energies for each of the reaction families in this model.
Knowing the activation energies of competing reaction path-
ways can help to direct catalyst research by modulating between
these pathways to favor desired products.
Fig. 4 Parity plot detail with selected species labels. For clarity, this scale does
not show carbon dioxide or any species with observed molar flow rates less than
1 ꢁ 10ꢀ9 mol sꢀ1. Data Set 8, x0 ¼ 5 wt%, P ¼ 3.48 MPa, T ¼ 518 K, WHSV ¼
0.73 hꢀ1, y ¼ 100.5 mol%, F ¼ 0.0902, R2 ¼ 0.987.
C1–C3 alcohols and diols is in accordance with the pathway we
assumed for the retro-aldol condensation of sorbitol in our
previous work.16 Table 7 summarizes the effects that the reac-
tion conditions have on the selectivities of each species type in
this system.
In practical applications, we can choose the proper reaction
conditions depending upon which products are desired. At
higher pressure, for example, we can selectively produce
pentane and hexane, which can be used to produce gasoline
with a high octane value by hydroisomerization. Likewise, we
can also selectively produce alcohols to make C1–C3 low-
oxygen-content compounds for the production of hydrogen or
make C4–C5 low-oxygen-content for liquid, jet-fuel-range, large,
normal alkanes by dehydrogenation and aldol condensation of
alcohols followed by aqueous phase hydrodeoxygenation.
The current work mainly focuses on the effect of sorbitol feed
concentration. In the future, further detailed work on the effect
of concentration of reaction intermediates are still needed to
provide deeper insight into how to control the selectivity for
these series-parallel reaction networks.
Within the constraints of this tuned kinetic model, we can
identify reaction intermediates and compare rate constants for
competing reactions. Given additional data with approximately
100 mol% identied carbon at varied conditions, we could
obtain more detailed information about the specic effects of
reaction conditions on intermediate species and pathways.
Table 6 Summary of model fit. x0 h sorbitol feed concentration (wt%), P h
pressure (MPa), T h temperature (K), WHSV h weight hourly space velocity (hꢀ1),
y h carbon identified in measured products (mol%), F h objective function value,
R2 h coefficient of determination
5.2. Model t
The model ts are most accurate for the data sets that were used
for parameter tuning. As expected, weaknesses in the predictive
abilities of the kinetic model became more pronounced at
conditions that varied signicantly from those at which the
model was tuned. Since the data were tuned using all of the
varied pressure data sets (except for the highest pressure
because of the high identied carbon percentage), Fig. 5a is a
smooth curve with a central maximum.
Data x0
Fig. 3 set
P
T
WHSV
y
(wt%) (MPa) (K) (hꢀ1
)
(mol%)
F
R2
a
b
c
2
8
9
5
5
5
5
2.93
3.48
4.17
4.90
518 0.73
518 0.73
518 0.73
518 0.73
101.7
100.5
102.6
104.8
0.751
0.0902 0.987
0.136
0.609
0.913
0.977
0.873
d
10
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23781

View Article Online
RSC Advances
Paper
Fig. 5 Coefficient of determination as a function of reaction conditions. x0 ¼ 5 wt%, T ¼ 518 K, WHSV ¼ 0.73 hꢀ1, varied pressure.
Table 7 Summary of the effect of reaction conditions on species type selectivity. x0 h sorbitol feed concentration (wt%), P h pressure (MPa), T h temperature (K),
WHSV h weight hourly space velocity (hꢀ1). High or low values of each reaction condition maximize or minimize the selectivity of each species type. Rank refers to the
minimum number of reaction steps from sorbitol; in this case, low rank species are intermediates. Selectivity trends with temperature also depend upon weight hourly
space velocity
Sorbitol, x0
Pressure, P
Temperature, T
WHSV
Max
Species type
Max
Min
Max
Min
Max
Min
Min
Carbon dioxide
Alkane, C1–C5
Hexane
Methanol
Primary alcohol, C2–C6 Low x0
Secondary alcohol
Diol
Triol
Low x0
Low x0
Low x0
Low x0
High x0
High x0
High x0
High x0
High x0
High x0
High x0
High x0
Low P
Low P
High P Low P
Low P
High P Low P
High P Low P
N/A
N/A
High P High T
High P High T
High T
High P High T
Low T
Low T
Low T
Low T
Low T
High T
High T
High T
Low WHSV
Low WHSV
Low WHSV
Low WHSV
Low WHSV
Low WHSV
High WHSV
High WHSV
High WHSV
High WHSV
High WHSV
High WHSV
High T
Low T
Low T
Low T
Low x0
Low x0
Low x0
N/A
N/A
High WHSV Low WHSV
High WHSV Low WHSV
Ketone
20 wt% Otherwise High P Low P
High T & high WHSV Low T & high WHSV High WHSV Low WHSV
Low T & low WHSV High T & low WHSV
High T & high WHSV Low T & high WHSV High WHSV Low WHSV
Low T & low WHSV High T & low WHSV
High T & high WHSV Low T & high WHSV High WHSV Low WHSV
Low T & low WHSV High T & low WHSV
High T & high WHSV Low T & high WHSV Low WHSV
Low T & low WHSV High T & low WHSV
Furan, low rank
Furan, high rank
Pyran
High x0 Low x0
Low x0 High x0
High P Low P
High P Low P
20 wt% Otherwise High P Low P
High WHSV
alcohols, and polyols. Changes in the reaction conditions
control the product selectivity.
6. Conclusions
We investigated the effect of temperature, pressure, and
sorbitol feed concentration on the aqueous phase hydro-
deoxygenation (APHDO) of sorbitol over a 4 wt% Pt/SiO2–Al2O3
catalyst. Deoxygenation is necessary for the production of
valuable products such as fuels from biomass derivatives.
Specically, hydrodeoxygenation is desirable because the
oxygen can be removed by the addition of hydrogen while
maintaining the carbon number. APHDO of sorbitol results in a
wide variety of products including liquid alkanes, light alkanes,
We constructed a complete reaction network for sorbitol
hydrodeoxygenation that produces all experimentally measured
species. All of the kinetic balance equations were automatically
generated and parameter tuning was constrained via reaction
families for a simplied solution. The kinetic model predictions
are in very good agreement with the experimental results.
The completed kinetic model enables the prediction of
product composition for varying feed composition and experi-
mental conditions, determination of optimal conditions,
23782 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013

View Article Online
Paper
RSC Advances
identication of the major and minor pathways and interme-
diate species, as well as the potential to identify the most suitable
feedstock composition to obtain a desired product, all of which
increase energy efficiency and reduce consumption of resources.
The reaction pathway and kinetic rate information that is
captured in this model can be used to make predictions about
more realistic feedstocks including cellulose by describing the
secondary reactions following cellulose depolymerization.
Sorbitol is a hydrogenation product of glucose, which is the
monomeric unit of cellulose. Since both glucose and sorbitol
are intermediate species in the conversion of cellulose to small
molecules, the kinetic model presented in this paper describes
the chemical reactions that occur aer the initial depolymer-
ization of the cellulose macromolecule.
References
1 L. Lynd, M. Laser, D. Brandsby, B. Dale, B. Davison,
R. Hamilton, M. Himmel, M. Keller, J. McMillan, J. Sheehan
and C. Wyman, Nat. Biotechnol., 2008, 26(2), 169–172.
2 A. Carroll and C. Somerville, Annu. Rev. Plant Biol., 2009, 60,
165–182.
3 J. R. Regalbuto, Science, 2009, 325(5942), 822–824.
4 S. Brethauer and C. E. Wyman, Bioresour. Technol., 2010,
101(13), 4862–4874.
5 S. Chundawat, G. Beckham, M. Himmel and B. Dale, Annu.
Rev. Chem. Biomol. Eng., 2011, 2, 121–145.
6 G. Huber, S. Iborra and A. Corma, Chem. Rev., 2006, 106(9),
4044–4098.
Further understanding of this reaction chemistry combined
with catalyst improvements will lead to targeted product selec-
tivity from the hydrodeoxygenation of biomass-derived feed-
stocks. Results will be used to direct further experimentation
and utilization of sorbitol as a surrogate molecule for biomass-
to-fuels processing. Given more complete data, this model could
7 J. N. Chheda, G. W. Huber and J. A. Dumesic, Angew. Chem.,
Int. Ed., 2007, 46(38), 7164–7183.
8 D. M. Alonso, J. Q. Bond and J. A. Dumesic, Green Chem.,
2010, 12(9), 1493–1513.
9 G. W. Huber, R. D. Cortright and J. A. Dumesic, Angew.
Chem., Int. Ed., 2004, 43(12), 1549–1551.
direct catalyst research to modulate desired deoxygenation 10 G. W. Huber, J. N. Chheda, C. J. Barrett and J. A. Dumesic,
pathways via tuned activation energy and identify the reaction Science, 2005, 308(5727), 1446–1450.
conditions that will make biofuels a competitive energy source. 11 E. L. Kunkes, D. A. Simonetti, R. M. West, J. C. Serrano-Ruiz,
¨
C. A. Gartner and J. A. Dumesic, Science, 2008, 322(5900),
417–421.
Notation
12 T. P. Vispute and G. W. Huber, Green Chem., 2009, 11(9),
1433–1445.
13 R. M. West, M. H. Tucker, D. J. Braden and J. A. Dumesic,
Catal. Commun., 2009, 10(13), 1743–1746.
14 N. Li, G. A. Tompsett and G. W. Huber, ChemSusChem, 2010,
3(10), 1154–1157.
15 N. Li, G. A. Tompsett, T. Zhang, J. Shi, C. E. Wyman and
G. W. Huber, Green Chem., 2011, 13(1), 91–101.
16 N. Li and G. W. Huber, J. Catal., 2010, 270(1), 48–59.
17 R. Xing, A. V. Subrahmanyam, H. Olcay, W. Qi, G. P. Walsum,
H. Pendse and G. W. Huber, Green Chem., 2010, 12(11),
1933–1946.
kj
Rate constant of reaction j, M1ꢀn sꢀ1
k0,i
Ej*
E*0,i
F
n
P
Rate constant of reaction family i, M1ꢀn sꢀ1
Activation energy of reaction j, kJ molꢀ1
Activation energy of reaction family i, kJ molꢀ1
Sum of squares of residuals objective function value
Reaction order
Absolute pressure, MPa
R
Universal gas constant, cm3 MPa molꢀ1 Kꢀ1
Coefficient of determination
R2
T
Absolute temperature, K
Sorbitol feed concentration, wt%
Carbon identied in measured products, mol%
x0
y
18 R. M. West, E. L. Kunkes, D. A. Simonetti and J. A. Dumesic,
Catal. Today, 2009, 147(2), 115–125.
19 R. M. West, Z. Y. Liu, M. Peter and J. A. Dumesic,
ChemSusChem, 2008, 1(5), 417–424.
20 C. Zhao, J. He, A. A. Lemonidou, X. Li and J. A. Lercher,
J. Catal., 2011, 280(1), 8–16.
21 W. Xu, S. J. Miller, P. K. Agrawal and C. W. Jones,
ChemSusChem, 2012, 5(4), 667–675.
Greek letters
ai
LFER constant of reaction family i
Enthalpy change of reaction j, kJ molꢀ1
DHR,j
22 C. Zhao, Y. Kou, A. A. Lemonidou, X. Li and J. A. Lercher,
Angew. Chem., 2009, 121(22), 4047–4050.
23 G. W. Huber, J. W. Shabaker, S. T. Evans and J. A. Dumesic,
Appl. Catal., B, 2006, 62(3–4), 226–235.
Abbreviations
24 R. R. Davda, J. W. Shabaker, G. W. Huber, R. D. Cortright and
J. A. Dumesic, Appl. Catal., B, 2003, 43(1), 13–26.
25 J. W. Shabaker, G. W. Huber, R. R. Davda, R. D. Cortright and
J. A. Dumesic, Catal. Lett., 2003, 88(1–2), 1–8.
APHDO Aqueous-phase hydrodeoxygenation
CME
Composition Modeling Editor
INGen Interactive Network Generator
KME
KMT
LFER
RF#
Kinetic Modeling Editor
Kinetic Modeling Toolkit
Linear free energy relationship
Reaction family #
¨
¨
26 E. I. Gurbuz, E. L. Kunkes and J. A. Dumesic, Green Chem.,
2010, 12(2), 223–227.
27 M. Vasiliu, K. Guynn and D. A. Dixon, J. Phys. Chem. C, 2011,
115(31), 15686–15702.
WHSV Weight hourly space velocity, hꢀ1
This journal is ª The Royal Society of Chemistry 2013
RSC Adv., 2013, 3, 23769–23784 | 23783

View Article Online
RSC Advances
Paper
28 T. P. Vispute, H. Zhang, A. Sanna, R. Xiao and G. W. Huber, 33 W. Wei, C. A. Bennett, R. Tanaka, G. Hou, M. T. Klein, Jr
Science, 2010, 330(6008), 1222–1227.
29 Z. He and X. Wang, Catalysis for Sustainable Energy, 2012, 1,
28–52.
and M. T. Klein, Fuel Process. Technol., 2008, 89(4), 350–
363.
34 M. Boudart, Kinetics of Chemical Processes, Prentice-Hall,
Inc., Englewood Cliffs, NJ, 1968.
30 C. A. Bennett, User-controlled kinetic network generation
with INGen, PhD Dissertation, Chemical & Biochemical 35 Z. Hou, Soware tools for molecule-based kinetic
Engineering, Rutgers University, 2009.
modeling of complex systems, PhD Dissertation,
Chemical & Biochemical Engineering, Rutgers University,
2011.
´
´
´
31 I. Bucsi, A. Molnar, M. Bartok and G. A. Olah, Tetrahedron,
1994, 50(27), 8195–8202.
´
´
´
´
32 I. Bucsi, A. Molnar, M. Bartok and G. A. Olah, Tetrahedron, 36 Y. Roman-Leshkov, C. J. Barrett, Z. Y. Liu and J. A. Dumesic,
1995, 51(11), 3319–3326.
Nature, 2007, 447(7147), 982–985.
23784 | RSC Adv., 2013, 3, 23769–23784
This journal is ª The Royal Society of Chemistry 2013