Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Catalysis Today
journal homepage: www.elsevier.com/locate/cattod
a r t i c l e i n f o a b s t r a c t
Article history: A detailed kinetic model of the Fischer–Tropsch synthesis (FTS) product distribution based
Received 29 May 2013 on the CO-insertion mechanism has been derived. The model was developed using the
Received in revised form 13 August 2013 Langmuir–Hinshelwood–Hougen–Watson approach. The intrinsic kinetic parameters were estimated
Accepted 14 August 2013
using a set of data obtained in a stirred tank slurry reactor with a rhenium promoted cobalt cat-
Available online 8 September 2013
alyst over a range of operating conditions (T = 478, 493, 503 K; P = 1.5, 2.5 MPa; H2 /CO = 1.4, 2.1;
WHSV = 1.0–22.5 NL/gcat /h). Physical meaningfulness of the model and its parameters was verified. Con-
Keywords:
sistent with reported measurements, model predicts that adsorbed CO is the most abundant surface
Fischer–Tropsch synthesis
Kinetic modeling
species. The observed increase in the chain growth probability factor and decrease in olefin-to-paraffin
CO-insertion ratio with increase in carbon number is explained utilizing the chain length dependent desorption of
Slurry reactor 1-olefins concept.
Cobalt catalyst © 2013 Elsevier B.V. All rights reserved.
Rhenium promoter
0920-5861/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.cattod.2013.08.008
B. Todic et al. / Catalysis Today 228 (2014) 32–39 33
Fig. 1. Examples of CO activation pathways: (a) direct CO dissociation (carbide mechanism); and (b) H-assisted CO dissociation (carbide mechanism); (c) CO hydrogenation
(CO-insertion mechanism).
the carbide (most often CH2 ) and adsorbed CO for the CO-insertion desorption [29]. This dependency is caused by the weak Van der
mechanism. Waal’s (VdW) interactions of the 1-olefin precursor, a -complex,
Storsæter et al. [14] compared versions of the two mechanisms with the catalyst surface. Nguyen et al. [33] modeled olefin adsorp-
– carbide (including direct and H-assisted CO dissociation) and tion on the surface of a zeolite. Coupling density function theory
CO-insertion – using the UBI-QEP (unity bond index – quadratic (DFT) and statistical thermodynamic calculations, the authors
exponential potential) method [31] and micro-kinetic modeling of showed that inclusion of weak Van der Waal’s forces leads to
C1 and C2 species formation. Their results showed that the cho- linearly increasing chemisorption energy with chain length. Cheng
sen CO-insertion pathway had a lower activation barrier compared et al. [34,35] also followed the same general idea and a theoretical
to both direct and H-assisted CO dissociation mechanisms that approach (based on an ab-initio DFT study) and showed that
were utilized. Based on these findings, they suggested that the adding VdW interactions increases the chemisorption energy of
CO-insertion mechanism is likely the main mechanism of FTS. Our 1-olefin and, in turn, leads to the explanation of non-ASF product
kinetic model is based on the mechanism used by Storsæter et al. distribution and decrease of OPR with carbon number.
[14], with a few modifications (e.g. grouping of elementary steps In our recent study [23] we showed that the linear dependency
of the same type) that are made in order to reduce the number of of the heats of chemisorption of 1-olefins with carbon number
model parameters and facilitate calculation of products higher than results in a linear dependency of desorption activation energy:
C2 . n 0
Ed,o = Ed,o + E × n (1)
The elementary steps of the FTS mechanism and associated rate
and equilibrium constants used in the model derivations are shown This is then used to directly derive the 1-olefin desorption rate
in Table 1. Steps 1 and 2 are the adsorption equilibrium steps for CO constant, which has an exponential dependency with carbon num-
and H2 , respectively. Activation of adsorbed CO by hydrogenation to ber:
CHO S and insertion of CO S into the growing chain (Cn H2n+1 S)
are grouped into Step 3. This combined step is assumed to be rate kd,n = kd,0 × ec×n (2)
determining and its kinetic rate constant is independent of chain
Constant c in Eq. (2) is related to the weak VdW interactions by the
length. It is followed by a series of fast hydrogenation steps in which
following relation:
the inserted CO is hydrogenated to CH2 , forming an extended chain
Cn H2n+1 CH2 S (Steps 4 and 5). Water formation and its removal E
c=− (3)
from the surface (Step 6) are considered to be fast and at pseudo- R×T
equilibrium. Normal paraffins are formed by hydrogenation of the where E is the contribution of VdW interactions of the chain with
adsorbed alkyl chain in Step 7. Formation of 1-olefin is by dehy- the surface for every C-atom (or CH2 -group).
drogenation (-hydrogen elimination) of Cn H2n+1 S followed by The formation of 1-olefin molecule consists of two steps:
desorption (Step 8). These two steps are combined into a one-step -hydrogen elimination from the growing chain (Cn H2n+1 S, -
desorption process [32], which is dependent on chain length as complex) and desorption of the -complex, whereby the latter is
described below. considered to be rate limiting. Our mechanism groups these two
steps into a hypothetical one-step desorption process (Fig. 2) [32].
3.2. Chain-length-dependent 1-olefin desorption As the chain length increases, the VdW attractive forces cause an
increase in the activation energy for the one-step desorption of 1-
n in Fig. 2), which is followed by a decreased probability in
olefin (Ed,o
One of the key assumptions of our model is that the 1-olefin
desorption rate constant depends exponentially on carbon num- 1-olefin formation. Therefore, we will see a higher residence time
ber, due to linearly increasing activation energy of the 1-olefin of the adsorbed alkyl chain (-complex) on the surface, which is in
B. Todic et al. / Catalysis Today 228 (2014) 32–39 35
Table 1
Elementary steps of the CO-insertion mechanism used in kinetic model derivation.
(1) CO + S ↔ CO–S K1
(2) H2 + 2S ↔ 2H–S K2
equilibrium with the -complex, and this in turn results in a higher considered to be negligible. The final equations of the model are
probability for chain growth and hydrogenation to n-paraffin with summarized below.
increasing chain length. The chain growth probabilities are dependent on carbon number
Potential secondary reactions of initially formed 1-olefins are and can be calculated as:
hydrogenation, isomerization and readsorption. These reactions
k3 K1 PCO
are not considered in the present model. ˛1 = (4)
k3 K1 PCO + k7M K2 PH2
3.3. Model equations
k3 K1 PCO [S]
˛2 = (5)
The model derivation is based on Langmuir–Hinshelwood– k3 K1 PCO [S] + k7 K2 PH2 [S] + k8,E ec·2
Hougen–Watson (LHHW) methodology. Detailed descriptions on
how this approach is used to derive kinetic models of FTS can be k3 K1 PCO [S]
˛n = n≥3 (6)
found in the literature [23,26,28]. The model is derived using the k3 K1 PCO [S] + k7 K2 PH2 [S] + k8,0 ec·n
CO-insertion mechanism shown in Table 1. Assumptions made in
the derivation are: (1) only one type of FTS active sites is present where [S] is the fraction of vacant sites and is calculated by solving
on the Co catalyst surface and the number of these sites is con- the site balance:
stant; (2) steady state concentrations of surface intermediates and
vacant sites; (3) the rate determining steps (RDS) are 3, 7 and 8, 1 PH2 O 1 PH2 O
while all others are considered to be at quasi-equilibrium; (4) the
[S] = 1/ 1 + K1 PCO + K2 PH2 + +
K6 K2 PH2 K2 K4 K5 PH2
rate constants of chain propagation and chain hydrogenation (n-
⎛ ⎞⎫
paraffin formation) step are independent of carbon number, while ⎬
1
n
i
K2 PH2 ⎝˛1 + ˛1 ˛2 + ˛1 ˛2 ˛j ⎠
the rate constant of 1-olefin formation is dependent on chain length
+ +
(as described above); (5) methane and ethylene are assigned sep- K5 K2 PH2 ⎭
i=3 j=3
arate formation rate constants; (6) coverage of species that are
(7)
formed and disappear in fast equilibrium steps (Steps 4–6) are
n
RCn H2n+2 = k7 K20.5 PH
0.5
˛1 ˛2 ˛i · [S]2 n≥2 (9)
2
i=3
n
RCn H2n = k8,0 ec·n ˛1 ˛2 ˛i · [S] n≥3 (11)
i=3
Fig. 2. Potential energy diagram for the 1-olefin desorption step.
36 B. Todic et al. / Catalysis Today 228 (2014) 32–39
3.4. Parameter estimation and model validation mation compared to other n-paraffins explains the higher yield of
this species. Additional reasons that could also explain high selec-
The parameters of this model are activation energies Ea,i , tivity toward methane, e.g. existence of pure methanation sites [39]
enthalpies Hi , preexponential factors Ai and a weak interaction or separate methane formation pathways [40], were not included
contribution to 1-olefin desorption energy E. They are calculated in this model. DFT studies [34,41] showed that ethene is more
by minimizing a multi-response objective function [36]: strongly adsorbed to the surface than the higher 1-olefins, which
exp 2 would result in its higher desorption activation energy. This is the
Ri,j
Nresp Nexp cal
− Ri,j reason why our model, consistent with the experimental results,
Fobj = exp (12) predicts a low rate of ethene formation as compared to other 1-
Ri,j
i=1 j=1 olefins. It should be noted that low ethene formation has often been
In total, 696 responses (points) at different temperatures (478, ascribed to its high reactivity in secondary reactions as evidenced
493 and 503 K) were used simultaneously in this optimization. The from ethene cofeeding studies [16,39]. The hypothesis that ethene
genetic algorithm and Levenberg–Marquardt method are used as adsorbs to the surface more strongly can also provide an explana-
global and local optimization tools [37]. Detailed algorithm of the tion for the high reactivity of externally added ethene molecules.
parameter estimation can be found elsewhere [23]. In order to The weak physical interactions with the catalyst surface cause
speed up convergence, the parameter estimation was confined to a the activation energy for the 1-olefin desorption step to increase
wide range of realistic values. Initial guesses for some of the model from 100 kJ/mol by 1.1 kJ/mol for very C-atom in the molecule. It
parameters were taken from Storsaeter et al. [14], who utilized the is important to note that the value of approximately 1 kJ/mol/CH2
UBI-QEP method. for the contribution of weak Van der Waal’s forces is consistent
The mean absolute relative residual (MARR) is used to evalu- with our understanding of these interactions. From this value, we
ate the quality of model predictions compared to the experimental can calculate value of constant c, appearing in the term ec×n , as
values: approximately −0.27, which is in a very good agreement with the
exp literature values for cobalt FTS catalysts [16,23].
Ri,j cal
Nresp Nexp
− Ri,j 1 Fit quality of the model for C1 –C15 n-paraffins and C2 –C15 1-
MARR = × N · N × 100% (13) olefins is shown in Fig. 3 for selected conditions at the three
Ri,j exp
resp exp
i=1 j=1 temperatures tested. Overall, the MARR value for the 29 species
considered is 23.5%. Compared to our previous study [23], which
The model was validated using physicochemical and statistical
was based on the Lox and Froment [26] form of the carbide mech-
tests [23,36].
anism and had MARR = 26.6%, there is a slight improvement in the
fit quality. Results of the F-test show that the model fit is sta-
4. Results and discussion tistically meaningful. Consistent with previous results, the model
reliably predicts changing growth probability with carbon number
4.1. Estimated parameters and quality of model predictions (0.8–0.95 in C3 –C15 hydrocarbon range). Deviations of C1 and C2 in
the total hydrocarbon distribution are well predicted as shown in
Table 2 shows the estimated values of model parameters. The Fig. 3. The exponential drop in OPR with carbon number in C3 –C15
results of the t-test show that all of the parameters are statis- range is also accounted for. It might be possible to further improve
tically different from zero. The physicochemical tests show that the model performance by including the olefin secondary reactions
the parameters are consistent with physicochemical laws [36] – to the kinetic scheme; however, that would lead to an increase in
i.e. activation energies have positive values and heats of reactant model complexity and number of parameters.
adsorption are negative. In order to further analyze the meaningful- Rates of reactant disappearance can be calculated from the
ness of these values, we compare the relevant activation energies product formation rates utilizing the reaction stoichiometry. Due
and enthalpies to literature values. to the fact that the CO insertion mechanism has approximately
The activation energy of the chain propagation step (Step 3 zeroth order dependence with respect to CO and half order depend-
in Table 1) is also the overall energy barrier of FTS. It was esti- ence with respect hydrogen, a reasonable agreement is obtained
mated to be 93 kJ/mol, which is within the reported range of between the model predicted and experimentally measured reac-
80–120 kJ/mol [38]. The activation energies for n-paraffin and 1- tant rates. MARR values are 18.0% and 17.8% for CO and H2 ,
olefin, as well as methane and ethene formation, are very similar to respectively.
values obtained using the carbide mechanism [23]. The estimated
heats of reactant adsorption and enthalpies of elementary reac- 4.2. FTS reaction mechanism, surface species and implications to
tions are low. This may be related to high surface coverage of CO kinetic modeling
[15], as discussed below. Lower activation energy of methane for-
Even though CO-insertion has long been recognized as a plau-
Table 2 sible main pathway for FTS, the proposed kinetic models have
Estimated values of model parameters and statistical results. typically not utilized this concept. Majority of FTS kinetic models
Parameter Value Unit Parameter Value Unit utilizing the LHHW approach are simple CO disappearance mod-
els [42,43], where selection of carbide mechanism as the basis for
A1 6.59 × 10−5 MPa−1 H1 −48.9 kJ/mol
A2 1.64 × 10−4 MPa−1 H2 −9.4 kJ/mol derivation has a distinct advantage; all CO disappearance is due
A3 4.14 × 108 mol/(gcat × h) E3 92.8 kJ/mol to C1 species formation, which in turn only involves three to five
A4 3.59 × 105 – H4 16.2 kJ/mol elementary steps. The model derivation is more complex if CO-
A5 9.81 × 10−2 – H5 11.9 kJ/mol insertion mechanism is used because CO is consumed not only
A6 1.59 × 106 MPa H6 14.5 kJ/mol
A7 4.53 × 107 mol/(gcat × h) E7 75.5 kJ/mol
to initiate chain growth (C1 species), but also during propagation,
A8 4.11 × 108 mol/(gcat × h) E8 100.4 kJ/mol where CO is inserted into the growing chain. Therefore, it becomes
A7M 7.35 × 107 mol/(gcat × h) E7M 65.4 kJ/mol necessary to define the surface concentration of the growing chains,
A8E 4.60 × 107 mol/(gcat × h) E8E 103.2 kJ/mol which is in turn related to the chain termination steps. Thus, a more
E 1.1 kJ/mol/CH2 MARR 23.5 %
comprehensive approach is required to take into account all of the
Statistical results: Fc = 90.1 > F0.05 (11, 47) = 1.70, lowest tc = 72.59 > t0.05 (21) = 1.72. necessary elementary steps of the FTS reaction mechanism.
B. Todic et al. / Catalysis Today 228 (2014) 32–39 37
Fig. 3. Experimental and calculated product distributions: (a) n-paraffin and 1-olefin formation rates; (b) total hydrocarbon formation, i.e. ASF plot; (c) 1-olefin-to-n-paraffin
ratio (process conditions: fist row – T = 478 K, P = 1.5 MPa, H2 /CO = 2.1, WHSV = 3.7 NL/gcat /h, XCO = 37%; second row – T = 493 K, P = 1.5 MPa, H2 /CO = 1.4, WHSV = 5.6 NL/gcat /h,
XCO = 34%; third row – T = 503 K, P = 1.5 MPa, H2 /CO = 2.1, WHSV = 11.3 NL/gcat /h, XCO = 42%).
One of the advantages of using a structured LHHW approach covering 40–80% of total active sites. SSITKA (steady state isotopic
compared to more empirical selectivity models is the ability to transient kinetic analysis) experiments of van Dijk et al. [44,45]
obtain information about the surface coverage of various species show that about 65% of the surface is covered with adsorbed CO.
and their changes with varying process conditions. Surface cov- High coverage of CO is a prerequisite for the CO-insertion mecha-
erages of adsorbed CO and H can be calculated from the model nism to be favored over the carbide FTS mechanism [10]. Significant
as: coverage of growing chain intermediates and vacant sites is also
predicted by the model with average values of 17% and 14%, respec-
[CO–S] = K1 PCO [S] (14)
tively. SSITKA studies usually report chain coverage lower than
[H–S] = K2 PH2 [S] (15) 10% [44,46]. Contrary to these the coverage of atomic hydrogen
is usually disregarded in the experimental transient kinetic stud-
where [S] is the fraction of vacant sites defined by Eq. (7). The con- ies, mainly due to difficulties related to kinetic isotope effects.
centration of growing chains on the surface is determined by chain Therefore our knowledge of it mostly comes from micro-kinetic
growth probabilities and is given by: models (e.g. [14]), which are in turn most often determined under
⎛ ⎞ methanation conditions (high H2 /CO ratio and low pressure), giv-
n
i
ing an atomic H coverage usually above 10%. A recent SSITKA study
[R–S] = K2 PH2 ⎝˛1 + ˛1 ˛2 + ˛1 ˛2 ˛j ⎠ [S] (16)
by den Breejen et al. [47] showed that even at H2 /CO = 10 and
i=3 j=3 P = 1.85 bar, hydrogen covered only about 10% of the surface. If the
tests were conducted under realistic FTS conditions (i.e. H2 /CO ≤ 2
Fig. 4 shows the predicted surface coverage for a selected (base-
and P ≥ 10 bar) one would expect a decrease in atomic H surface
line) condition. Under all conditions used in this study, the model
coverage. Similar conclusion can be made from DFT studies [9]
predicts that adsorbed CO is the most abundant surface species,
which show that increasing CO coverage reduces H2 adsorption
enthalpy, and would then result in a decrease of H coverage. Our
model predicts the H coverage of around 1%. This is mainly due to
the fact that the employed mechanism assumes that most of the
reactions in which hydrogen is consumed are fast.
Rprop k3 [CO–S]
˛n = = n≥3
Fig. 4. Calculated fractions of surface intermediates at T = 493 K, P = 1.5 MPa, Rprop + Rn-par + R1-ole k3 [CO–S] + k7 [H–S] + k8,0 ec·n
H2 /CO = 2.1, WHSV = 8 NL/gcat /h, XCO = 45%. (17)
38 B. Todic et al. / Catalysis Today 228 (2014) 32–39
5. Conclusions
Acknowledgments
References
[1] F.G. Botes, J.W. Niemantsverdriet, J. van de Loosdrecht, Catalysis Today 215
(2013) 112.
[2] B.H. Davis, Fuel Process. Technol. 71 (2001) 157–166.
[3] J.T. Kummer, T.W. DeWitt, P.H. Emmett, J. Am. Chem. Soc. 70 (1948) 3632–3643.
[4] R.C. Brady III, R. Pettit, J. Am. Chem. Soc. 103 (1981) 1287–1289.
[5] P.M. Maitlis, R. Quyoum, H.C. Long, M.L. Turner, Appl. Catal. A 186 (1999)
Fig. 5. Simulation of FTS product distribution behavior with carbon number (at
363–374.
T = 493 K, P = 1.5 MPa, H2 /CO = 2.1, WHSV = 8 NL/gcat /h, XCO = 45%): (a) 1-olefin/n- [6] H. Pichler, H. Schulz, Chem. Ing. Tech. 42 (1970) 1162–1174.
paraffin ratio (OPR); (b) chain growth probability (˛n ); (c) total hydrocarbon [7] M. Corral Valero, P. Raybaud, Catal. Lett. 143 (2013) 1–17.
distribution (ASF) plot. [8] M. Zhuo, K.F. Tan, A. Borgna, M. Saeys, J. Phys. Chem. C 113 (2009) 8357–8365.
[9] M. Ojeda, R. Nabar, A.U. Nilekar, A. Ishikawa, M. Mavrikakis, E. Iglesia, J. Catal.
272 (2010) 287–297.
[10] R.A. van Santen, I.M. Ciobîcă, E. van Steen, M.M. Ghouri, in: C.G. Bruce, K. Helmut
Having the ec×n term (where c is a constant approximately equal (Eds.), Advances in Catalysis, vol. 54, Academic Press, Amsterdam, 2011, pp.
to −0.3), means that as the carbon number increases, the term 127–187.
[11] F. Fischer, H. Tropsch, Brennstoff Chem. 7 (1926) 97.
ec×n will decrease, causing ˛n to increase. Therefore, chain growth [12] M. Claeys, E. van Steen, in: A.P. Steynberg, M.E. Dry (Eds.), Studies in Surface
probability (˛n ) will follow the trend plotted in Fig. 5b. Having a Science and Catalysis, vol. 152, Elsevier B.V., Amsterdam, 2004, pp. 601–680.
continuously increasing ˛n with carbon number means that the [13] J. Schweicher, A. Bundhoo, N. Kruse, J. Am. Chem. Soc. 134 (2012) 16135–16138.
[14] S. Storsæter, D. Chen, A. Holmen, Surf. Sci. 600 (2006) 2051–2063.
positive bend in total hydrocarbons is predicted, as is expected [15] M. Zhuo, A. Borgna, M. Saeys, J. Catal. 297 (2013) 217–226.
from the experimental data (Fig. 5c). It is also interesting to point [16] G.P. Van der Laan, A.A.C.M. Beenackers, Catal. Rev. Sci. Eng. 41 (1999) 255–318.
out that after a certain carbon number is reached (above C15 ), the [17] A.T. Bell, Stud. Surf. Sci. Catal. 92 (1995) 63–72.
B. Todic et al. / Catalysis Today 228 (2014) 32–39 39
[18] J. Anfray, M. Bremaud, P. Fongarland, A. Khodakov, S. Jallais, D. Schweich, Chem. [32] J. Cheng, P. Hu, P. Ellis, S. French, G. Kelly, C.M. Lok, J. Phys. Chem. C 112 (2008)
Eng. Sci. 62 (2007) 5353–5356. 1308–1311.
[19] C.G. Visconti, E. Tronconi, L. Lietti, R. Zennaro, P. Forzatti, Chem. Eng. Sci. 62 [33] C.M. Nguyen, B.A. De Moor, M.-F. Reyniers, G.B. Marin, J. Phys. Chem. C 115
(2007) 5338–5343. (2011) 23831–23847.
[20] C.G. Visconti, E. Tronconi, L. Lietti, P. Forzatti, S. Rossini, R. Zennaro, Top. Catal. [34] J. Cheng, T. Song, P. Hu, C.M. Lok, P. Ellis, S. French, J. Catal. 255 (2008) 20–28.
54 (2011) 786–800. [35] J. Cheng, P. Hu, P. Ellis, S. French, G. Kelly, C.M. Lok, J. Catal. 257 (2008) 221–228.
[21] S.-H. Kwack, J.W. Bae, M.-J. Park, S.-M. Kim, K.-S. Ha, K.-W. Jun, Fuel 90 (2011) [36] G.F. Froment, K.B. Bischoff, J. De Wilde, Chemical Reactor Analysis and Design,
1383–1394. John Wiley & Sons, Hoboken, NJ, 2011.
[22] S.-H. Kwack, M.-J. Park, J.W. Bae, S.-J. Park, K.-S. Ha, K.-W. Jun, Fuel Process. [37] T.-Y. Park, G.F. Froment, Comput. Chem. Eng. 22 (Suppl. 1) (1998) S103–S110.
Technol. 92 (2011) 2264–2271. [38] F.H. Ribeiro, A.E.S.V. Wittenau, C.H. Bartholomew, G.A. Somorjai, Catal. Rev. Sci.
[23] B. Todic, T. Bhatelia, G.F. Froment, W. Ma, G. Jacobs, B.H. Davis, D.B. Bukur, Ind. Eng. 39 (1997) 49–76.
Eng. Chem. Res. 52 (2013) 669–679. [39] H. Schulz, Top. Catal. 26 (2003) 73–85.
[24] W. Zimmerman, D. Bukur, S. Ledakowicz, Chem. Eng. Sci. 47 (1992) 2707–2712. [40] W.H. Lee, C.H. Bartholomew, J. Catal. 120 (1989) 256–271.
[25] H. Schulz, M. Claeys, Appl. Catal. A 186 (1999) 91–107. [41] A.M. Goda, M. Neurock, M.A. Barteau, J.G. Chen, Surf. Sci. 602 (2008) 2513–2523.
[26] E.S. Lox, G.F. Froment, Ind. Eng. Chem. Res. 32 (1993) 71–82. [42] I.C. Yates, C.N. Satterfield, Energy Fuels 5 (1991) 168–173.
[27] Y.-N. Wang, W.-P. Ma, Y.-J. Lu, J. Yang, Y.-Y. Xu, H.-W. Xiang, Y.-W. Li, Y.-L. Zhao, [43] B. Sarup, B.W. Wojciechowski, Can. J. Chem. Eng. 67 (1989) 62–74.
B.-J. Zhang, Fuel 82 (2003) 195–213. [44] H.A.J. van Dijk, J.H.B.J. Hoebink, J.C. Schouten, Top. Catal. 26 (2003) 111–119.
[28] J. Yang, Y. Liu, J. Chang, Y.-N. Wang, L. Bai, Y.-Y. Xu, H.-W. Xiang, Y.-W. Li, B. [45] H.A.J. van Dijk, J.H.B.J. Hoebink, J.C. Schouten, Top. Catal. 26 (2003) 163–171.
Zhong, Ind. Eng. Chem. Res. 42 (2003) 5066–5090. [46] T. Komaya, A.T. Bell, J. Catal. 146 (1994) 237–248.
[29] F.G. Botes, Energy Fuels 21 (2007) 1379–1389. [47] J.P. den Breejen, P.B. Radstake, G.L. Bezemer, J.H. Bitter, V. Frøseth, A. Holmen,
[30] C.M. Masuku, W.D. Shafer, W. Ma, M.K. Gnanamani, G. Jacobs, D. Hildebrandt, K.P.d. Jong, J. Am. Chem. Soc. 131 (2009) 7197–7203.
D. Glasser, B.H. Davis, J. Catal. 287 (2012) 93–101. [48] T.J. Donnelly, I.C. Yates, C.N. Satterfield, Energy Fuels 2 (1988) 734–739.
[31] E. Shustorovich, H. Sellers, Surf. Sci. Rep. 31 (1998) 1–119.