Mechanism of Hydrodeoxygenation of Acrolein on a Cluster Model of

Jul 28, 2010 - We have explored the potential energy surface for reactions of acrolein on a Mo3O9 cluster model of the MoO3 surface to investigate the...
0 downloads 4 Views 4MB Size
13782

J. Phys. Chem. C 2010, 114, 13782–13795

Mechanism of Hydrodeoxygenation of Acrolein on a Cluster Model of MoO3 Daniel R. Moberg,†,‡,§ Timothy J. Thibodeau,†,‡,§ Franc¸ois G. Amar,† and Brian G. Frederick*,†,‡,§ Department of Chemistry, Laboratory for Surface Science and Technology, and Forest Bioproducts Research Institute, UniVersity of Maine, Orono, Maine 04469-5708 ReceiVed: May 14, 2010; ReVised Manuscript ReceiVed: July 3, 2010

We have explored the potential energy surface for reactions of acrolein on a Mo3O9 cluster model of the MoO3 surface to investigate the thermodynamics and kinetics of hydrogenation and selective hydrodeoxygenation. In the presence of hydrogen, conversions of acrolein to allyl alcohol, 1-propanol, and propene are all thermodynamically favorable, but the selectivity is controlled kinetically to form the least favorable product, allyl alcohol. We propose a mechanism in which coordinatively unsaturated Mo sites (i.e., oxygen vacancies) selectively chemisorb acrolein. On the basis of experimental and theoretical evidence, the active phase of the catalyst is a reduced hydrogen bronze, HxMoO3-y, and surface hydroxyl sites are occupied when x is in the range 1.1-1.2. Surface hydroxyls are important for both oxygen vacancy formation and as Brønsted acids. The reaction rate is essentially controlled by protonation of the C-1 carbon of chemisorbed acrolein. Additional reaction barriers for proton donation to the C-2 or C-3 sites are similar in magnitude (104-134 kJ/mol), limiting the rates of formation of propene and 1-propanol, respectively. In contrast, the selectivity toward allyl alcohol is determined by a smaller O-H bond formation barrier (33 kJ/mol). The estimated reaction rate is comparable to the rate of oxygen vacancy creation, so that operation in a continuous flow process appears to be feasible. The calculated reaction barrier for C-O scission is 104 kJ/mol; we discuss the advantages of other oxides, particularly WO3, that have stronger metal oxygen bonds and stronger Brønsted acidity of surface hydroxyls. 1. Introduction The production of transportation fuels from lignocellulosic biomass has the potential to reduce dependence on nonrenewable fossil fuel sources.1,2 Fast pyrolysis of biomass3-6 produces a bio-oil composed of a complex mixture of oxygenated organic compounds, including carboxylic acids, aldehydes, ketones, alcohols, and phenols.4,7-11 These bio-oils are highly acidic, viscous, and have low energy density due to the oxygencontaining compounds. To make a fungible fuel with high energy density and stability, we are focusing on development of thermal12 and catalytic hydrotreating processes to remove oxygen.13 Because bio-oils contain a large fraction of unsaturated compounds, catalysts that could selectively break carbon-oxygen bonds, without hydrogenating unsaturated carbon bonds, would be desirable economically because they minimize hydrogen consumption. Despite significant efforts in the 1970s and 1980s to develop hydrotreating catalysts, many fundamental questions remain and there is a strong interest and continuing effort to develop commercially viable catalysts. Development of hydrodeoxygenation catalysts for upgrading biofuels has been reviewed recently.5 Three classes of catalysts have been investigated: noble metal hydrogenation catalysts,14 zeolites,15,16 and sulfides of molybdenum, cobalt, and nickel.17 Each class has limitations. Hydrodeoxygenation using metal catalysts is not selective toward carbon-oxygen bonds. Zeolites tend to produce polycyclic aromatic hydrocarbons and a * Corresponding author: Brian G. Frederick, 153 ESRB/Barrows Hall, LASST, University of Maine. E-mail: [email protected]. Tel: +001 (207) 581-2268. Fax: +001 (207) 581-2255. † Department of Chemistry. ‡ Laboratory for Surface Science and Technology. § Forest Bioproducts Research Institute.

relatively large yield of CO2.18-23 The sulfide catalysts, based on hydrodesulfurization, are not yet commercially economical, due to poor stability in the presence of water, and require a sulfur source to prevent conversion to the oxide. Furthermore, the addition of sulfur to biofuel feedstocks that are inherently low in sulfur content, seems counterproductive to meeting future low sulfur fuel standards.24 Acrolein is the simplest unsaturated aldehyde and so serves as a good model compound to explore the competition between CdO and CdC hydrogenation, leading to the compounds illustrated in Figure 1. Hydrogenation of acrolein to propanal is the least desirable product; reduction of the carbonyl group to allyl alcohol and 1-propanol is better. Loss of oxygen while preserving the CdC double bond would lead to the most desirable product, propene. In this work, we focused on reducible metal oxides as a new class of hydrodeoxygenation catalysts. An extensive literature exists regarding the selective oxidation of olefins to form aldehydes and ketones, using reducible metal oxides such as the molybdates, vanadates, antimonates, and tungstates.25,26 Selective oxidation is essentially the reverse of hydrodeoxygenation. The goal of this theoretical study is to explore the mechanism of catalytic hydrodeoxygenation of a model compound, acrolein, on MoO3 based catalysts and to compare these results with available experimental data for both MoO3 and the closely related WO3 in order to understand the important characteristics of metal oxide catalysts for hydrodeoxygenation. Fundamental considerations of metal oxide properties, elaborated by Grasselli25 for selective oxidation reactions, guide our thinking. Grasselli et al.25,27-31 have shown that bismuth molybdate catalysts can be used to oxidize propene to acrolein. This

10.1021/jp104421a  2010 American Chemical Society Published on Web 07/28/2010

Hydrodeoxygenation of Acrolein on MoO3

Figure 1. Hydrotreating of acrolein may produce propanal (least desirable), 1-propanol, and allyl alcohol through hydrogenation (HYD) reactions, and propene (most desirable) via hydrodeoxygenation (HDO).

Figure 2. Mars-van Krevelen mechanism26 for selective oxidation of propylene to acrolein25 on a bismuth molybdate catalyst.

process, also known as the Mars-van Krevelen mechanism,26 is outlined in Figure 2. Propene adsorbs on the surface of bismuth oxide, where hydrogen abstraction forms a π-allyl radical. This species then migrates to the molybdenum oxide phase, forming a C-O bond with lattice oxygen. Abstraction of a second hydrogen from the intermediate forms acrolein, which desorbs and leaves the catalyst with a surface hydroxyl and an oxygen vacancy. The catalyst is regenerated by dissociation of oxygen (from the feedstream) on the bismuth oxide phase, followed by solid-state diffusion to the molybdenum oxide phase, where the catalyst is regenerated. A review of the literature on hydrotreating reactions of acrolein reveals mostly hydrogenation over noble metal catalysts that have high activity but a strong selectivity toward propanal.32-39 By contrast, Hoang-Van and Zegaoui40 demonstrated that acrolein reacts at 323 K to form 1-propanol over hydrogen-reduced MoO3 and allyl alcohol over hydrogenreduced WO3. Furthermore, Matsuda et al.41,42 have shown that 2-propanol can be converted at 398 K to propene over MoO3

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13783 that has been treated in hydrogen at 623 K. While both groups41-43 suggest that a reduced metal oxide bronze forms, the precise stoichiometry of the active phase was not determined. We have recently carried out a systematic study of this reduction “pretreatment” process for WO3, using gravimetric analysis and mass spectrometry and find that the stoichiometry, following reduction in hydrogen at 623 K for 10 h, corresponds to a tungsten oxide bronze with a range of compositions, H0.9-1.3WO2.9-2.7.13 We found that this catalyst converts allyl alcohol to 1-propanol at low temperatures (323-423 K). It is active for the hydrodeoxygenation of allyl alcohol to propene and 1,5-hexadiene at higher temperatures (523-623 K).13 These results collectively suggest that oxygen vacancies and hydrogen bronze formation may be important factors in designing metal oxide based catalysts for hydrodeoxygenation and that there is experimental evidence for the concept of reversing the catalytic cycle depicted in Figure 2. Pudar et al.44 have used a Mo3O9 cluster model of the MoO3 (010) surface to successfully describe the basic features of selective oxidation of propene to acrolein. The overall thermodynamics for oxidation of π-allyl to acrolein is illustrated in Figure 3, in which we plot ∆G593 obtained from Pudar et al.,44 starting from their π-allyl and Mo3O9, 8, forming an allyl alkoxide intermediate 10, and ending with acrolein and a defective cluster 12, which is highly unfavorable (numbering as in Pudar et al.44). Pudar et al.44 within the constraints of the Mo3O9 cluster, argued that, in the presence of oxygen, the selective oxidation reaction becomes favorable if molecular oxygen adsorbs on the vacancy, 14salthough the catalyst is not strictly returned to the initial state, 8. This figure suggests that the reverse process, hydrodeoxygenation of acrolein over a defective MoO3 catalyst, 12, to the allyl radical, 8, and on to propene, may be thermodynamically favorable. We first present calculations to show that, in a hydrogen feedstream, oxygen vacancy formation on the Mo3O9 cluster model most likely occurs by hydrogen adsorption and surface hydroxyl formation at terminal sites, followed by desorption of water. As Pudar et al.44 have shown (Figure 3), adsorption of acrolein at an oxygen vacancy to form the allyl alkoxide, 10, is favorable, but the activation energy barrier for hydrogen insertion and simultaneous C-O bond scission to form propene is about 100 kJ/mol (not shown; see below). We have extended the work of Pudar et al.44 to explore the transition states involved in two additional hydrogenation pathways, starting from the allyl alkoxide, 10, leading to allyl alcohol and 1-propanol. The activation energy barrier we found for formation of the allyl alcohol O-H bond was substantially smaller than the barriers in the propene and 1-propanol pathways, suggesting that kinetic control can produce the least thermodynamically favorable product, allyl alcohol, on this cluster. We discuss the results in the context of several important fundamental aspects of metal oxide chemistry, including: the role of M-O bond strength in C-O bond breaking and oxygen vacancy formation; the structural capability of MoO3 (and WO3) to form hydrogen bronzes; and the corollary role of O-H bond strength in H transport and hydrogen insertion. 2. Theoretical Methods Density functional calculations were performed using the Gaussian 03 program45 with the unrestricted B3LYP functional and the LANL2DZ effective core potential (ECP) and basis set for Mo and the 6-31G(d,p) basis set for all other atoms. This level of theory was chosen for compatibility with the Mo15O56H22 cluster results of Tokarz-Sobieraj et al.46 and the

13784

J. Phys. Chem. C, Vol. 114, No. 32, 2010

Moberg et al.

Figure 3. Free energy calculations at 593 K for conversion of π-allyl(g) + Mo3O9, 8, via an allyl alkoxide, 10, to acrolein (g) + Mo3O8H, 12, calculated by Pudar et al.,44 suggesting that the thermodynamics of the reverse process, hydrodeoxygenation of acrolein to propene over a defective molybdenum oxide catalyst, may be favorable. Intermediate 10 is taken as the starting point in this work for two alternative pathways, leading to allyl alcohol and 1-propanol.

Mo3O9 cluster results of Pudar et al.44 Geometry optimization of their structures produced similar total electronic energies. Using structures optimized at the 6-31G(d,p) level, we performed single point calculations with the 6-311+G(d,p) basis set for all atoms besides Mo. Free energies were calculated using thermodynamic scaled47,48 frequency calculations at 323.15 K and 1 atm, which are typical of the experimental conditions utilized in hydrodeoxygenation. Transition state searching was performed with the TS or QST3 algorithms49-51 followed by frequency calculations to verify first-order saddle points and intrinsic reaction coordinate (IRC) calculations to confirm that the transition states led toward the correct reactant and product conformations. Rate constants52 for individual steps were calculated from classical energy barriers at the 6-311+G(d,p) level of theory. Zero point energy corrections and partition functions were calculated from geometries optimized at the 6-31G(d,p) level. 3. Results 3.1. Oxygen Vacancy Creation. The first step in the hydrodeoxygenation process is the creation of an oxygen vacancy in the MoO3 lattice. The MoO3 (010) surface has three different types of oxygen sites: terminal (singly coordinated), asymmetric bridging (doubly coordinated), and symmetric bridging (triply coordinated).46 In the Mo3O9 cluster,44 however, there are only two types: terminal and symmetric bridging, as shown in Figure 4a. Although oxygen vacancy creation has been studied on larger clusters,46 hydrogen dissociation and oxygen vacancy creation energies were not reported by Pudar et al.44 for the Mo3O9 cluster. We calculated the energy change of these processes at the B3LYP/6-31G(d,p)/LANL2DZ level for comparison with the values of Tokarz-Sobieraj et al.,46 who performed DFT calculations at a comparable level on a Mo15O56H22 cluster. This allowed us to assess the accuracy of defect energies and evaluate computational methods. We considered three ways to create an oxygen vacancy, either in the terminal or bridging position for

Figure 4. (a) The Mo3O9 cluster used in the calculations. (b) The Mo3O9H3 cluster which has three hydroxyl groups on the underside. (c) The Mo3O9H cluster with terminal surface hydroxyl hydrogen, Ht. (d) The Mo3O9H cluster with bridging surface hydroxyl hydrogen, Hb.

clusters with (a) enough hydrogen to result in a defective Mo3O8 cluster or (b) the same clusters with the terminal oxygen atoms on the underside (nonreacting side) of the cluster converted to hydroxyl groups, shown in Figure 4b. First, a vacancy can be formed by removing atomic oxygen

Mo3O9 f Mo3O8 + O(g)

∆E(O)

Mo3O9H3 f Mo3O8H3 + O(g)

∆E(O)

(1a) (1b)

where ∆E(O) is the energy required to remove an oxygen from the terminal or bridging site. Second, if the cluster is partially reduced with one surface hydroxyl at either a terminal oxygen or a bridging oxygen, as illustrated in part

Hydrodeoxygenation of Acrolein on MoO3

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13785

Figure 5. Comparison of oxygen vacancy defect creation energies on terminal and bridging sites via production of atomic oxygen (O), hydroxyl radical (OH), or water (H2O), compared to the Mo15O56H22 cluster calculations of Tokarz-Sobieraj et al.46 The black bar at the center of each group is the result of ref 46; on either side of this are the results for optimized final geometries for nonterminated (left) and H-terminated (right) starting structures (Opt). At the outside of each group are the results corresponding to the unrelaxed (SP) final structures.

c or d in Figure 4, a vacancy could be created by removing the hydroxyl

Mo3O9H f Mo3O8 + OH(g) Mo3O9H4 f Mo3O8H3 + OH(g)

∆E(OH) ∆E(OH)

(2a) (2b)

and forming a hydroxyl radical in the gas phase, where ∆E(OH) is the energy to remove the hydroxyl from the cluster. A third possibility is the loss of water, either from chemisorbed molecular water or via dehydroxylation, i.e., through the reaction between two hydroxyl groups followed by the desorption of water

Mo3O9H2 f Mo3O8 + H2O(g) Mo3O9H5 f Mo3O8H3 + H2O(g)

∆E(H2O) ∆E(H2O)

(3a) (3b)

where ∆E(H2O) is the energy to create the oxygen vacancy and gas phase water. For each of the six reactions above, we calculated the energy required to remove either a terminal or bridging oxygen and further investigated the effect of fully optimizing the product cluster (“Opt”) as opposed to freezing the remaining atoms of the final cluster at the initial cluster’s geometry (“SP”). This last comparison may be important for small cluster models of surfaces as these clusters are not as rigidly constrained as larger models or slabs.53 We compare the results of these 24 calculations with the results of Tokarz-Sobieraj46 for oxygen vacancy creation at the terminal and symmetric bridging sites of the Mo15O56H22 cluster in Figure 5. The pathways which were compared included formation of O, OH, and H2O. The overall trend is similar: the most favorable pathway is the removal of water from the cluster, whether from terminal or bridging sites and irrespective of the degree of hydrogenation of the underside of the cluster. However, it is clear that, in agreement with the results on the

larger cluster,46 removal of water is easier to achieve (by 100 kJ/mol or more) from the bridging sites in both Mo3O9H2 or Mo3O9H5 clusters. Production of atomic ground state oxygen is least likely, with energies of 568 and 621 kJ/mol for terminal and bridging sites, although formation of 1/2O2 would lower the energies by 180 kJ/mol to about 388 and 441 kJ/mol, respectively; this means that the thermodynamic formation energy for production of 1/2O2 is comparable to that for producing OH. Of the pathways considered, oxygen vacancy creation with production of water via dehydroxylation is the only energetically viable route. On average, the optimized small cluster defect energies are closer to the results for the larger cluster with the notable exception of the SP energy for OH removal from a terminal oxygen site. Further comparison of the optimized energies for defect creation on the small clusters with and without hydrogen termination (a series vs b series) shows no clear pattern: in some cases, the H-terminated cluster gives higher defect energies than the unterminated cluster while in other cases the reverse is true. As might be expected, the largest relaxation effects are observed for the removal of bridging oxygens with the largest such effect (∼200 kJ/mol) being observed for the removal of OH from the hydrogen terminated cluster (reaction 2b). Indeed there is significant distortion in the relaxed structure of this product, with formation of a weak Mo-Mo bond and some hydrogen bonding between the hydroxyls on the underside of the cluster. Production of water and creation of a vacancy on a bridging site for the H-terminated cluster also show a large relaxation effect with the cluster experiencing a great degree of distortion, including the formation of an eight-membered ring structure that is closed by a hydrogen bond between a hydroxyl hydrogen and a hydroxyl oxygen. The distortion and hydrogen bonding observed in these hydrogen terminated clusters caused us to exclude these bridging-defect structures from further consideration in the energetics of the full acrolein reduction pathway. However, the relatively good agreement of the optimized defect energies with those of the larger cluster prompts us to retain the use of fully optimized geometries in the reaction pathway study of section 3.3. 3.2. H2 Adsorption and Hydroxyl Formation. In order to form the oxygen defects at which acrolein can bind, the most likely path is by way of water desorption from the lattice. For that to happen, it is necessary for hydrogen to adsorb to the surface oxygens forming hydroxyl groups and then for these surface hydroxyls to react, forming water, which can finally desorb. We therefore studied the energy changes for H2 adsorption onto the fully oxidized cluster at both terminal and bridging sites. For these calculations, geometry optimization was carried out at the double-ζ level and the reported energies calculated with the 6-311+G(d,p) basis set for light atoms. In the Mo3O9 cluster, there are only three possible combinations of sites at which an H2 molecule can adsorb: two terminal positions, two bridging positions, or one terminal and one bridging position. The energy changes for these reactions are shown in Figure 6. Note that we are treating the three oxygens on the underside of the cluster as unavailable since they would normally be shared with other Mo atoms in the bulk. We also calculated the energy for the reaction

Mo3O9 + 1/2H2 f Mo3O9H

(4)

adding a single H atom to either a bridging or a terminal site. As Figure 6 shows, the energy of single atom addition to the

13786

J. Phys. Chem. C, Vol. 114, No. 32, 2010

Moberg et al.

Figure 6. Adsorption energies for the reaction (n/2)H2 + Mo3O9 f Mo3O9Hn for different final hydroxyl site combinations.

TABLE 1: Overall Energetics of Hydrogen Adsorption and Oxygen Vacancy Formation for the Two-Step Reactions (eq 5)a n 0 0 1 1 2 2

H-site

product (O-site)

∆E/(kJ/mol)

t b t,t b,b

O (t) O (b) OH (t) OH (b) H2O (t) H2O (b)

568.145282 620.781232 351.866607 404.502557 73.7337113 126.369661

C3H6O (allyl alcohol) + H2 Mo3O8H + C3H4O + 2H2 f Mo3O8H + C3H6 (propene) + H2O C3H8O (1-propanol)

(6)

a The letters “t” and “b” refer to H-atom adsorption to terminal and bridging oxygens, respectively, and to the site of the final oxygen vacancy. All energies refer to B3LYP/6-311+G(d,p) single point calculations for structures optimized at the B3LYP/6-31G(d,p) level.

unterminated cluster is found to be favorable by about 42 kJ/ mol for the terminal (hydroxyl) site and unfavorable by roughly 34 kJ/mol for the bridging site. The energetics for the adsorption of H2 are qualitatively additive for these structures, though clearly there are more complicated cluster energetic effects in play. In section 3.1 we found that oxygen defects are energetically favored to form at bridging sitessespecially for the dehydroxylation reaction with production of water. However, the hydroxyls required to form those defects prefer the terminal sites. The net 0 K energetics for the two-step processes

n Mo3O9 + H2 f Mo3O9Hn f Mo3O8 + OHn 2

the adsorption of H2 to the oxygen vacancy site) with no barrier. The H-H bond is lengthened from 0.74 Å for gas phase H2 to 1.51 Å for H2 adsorbed at the Mo(IV) site. Transfer of the two hydrogen atoms to the terminal oxygen sites is downhill by another 7 kJ/mol. The issue of the detailed mechanism of hydrogen regeneration of the catalyst is discussed further in section 4.3, but a more definitive understanding of this aspect of the catalytic cycle will require further calculations on realistic slab models of the MoO3 surface. 3.3. Acrolein Reactions on MoO3. We next report the results of theoretical calculations of the energetics of hydrogenation of acrolein on reduced molybdenum oxide to explore the suitability of this material as a selective hydrogenation or hydrodeoxygenation catalyst. Starting with acrolein, the defective cluster Mo3O8H (labeled 12 in Figure 3), and gas phase hydrogen, we constructed the PES for acrolein reacting to yield allyl alcohol, 1-propanol, and propene as indicated in the reaction scheme (all species considered in the gas phase)

(5)

n ) 0, 1, 2 is summarized in Table 1. The overall trends for the energetics of defect formation remain unchanged from the results of Figure 5: O atoms are the hardest and H2O molecules the easiest to remove from the cluster. It is significant, however, that in contrast to Figure 5, the terminal positions are more favorable sites for defect creation than the bridging positions by about 50 kJ/mol for all of the defect formation processes studied here when the thermodynamics of hydrogen adsorption is taken into consideration. Thus in the reactions of acrolein with the Mo3O8H cluster, we focus on oxygen vacancies only at the terminal oxygen positions. Before leaving the topic of H2 adsorption, we consider the process of H-atom production on the MoO3 surface at coordinatively unsaturated sites. We calculate the energy of H2 + Mo3O8 f Mo3O8\H2 to be -51.3 kJ/mol (Mo3O8\H2 denotes

where we note that care has been taken to restore the catalyst to its initial state for each reaction. Figure 7 graphs the total optimized energy at 0 K of these reaction pathways relative to the 12 + acrolein + 2H2 starting point. The inset of Figure 7 at the upper right, shows the theoretical values of ∆rG for each of the three products at 0 and 323 K as well as the experimental value of ∆rG at 323 K.54 In Figure 7, we have used the numbering scheme implemented in Pudar et al.44 and have added higher numbers (>14 or >TS4) for the structures and transition states created for this work. There are generally slight differences in structure and energy of species that are in common with Pudar et al.44 because of slight methodological differences. The geometry of each structure is shown in Figure 8, and the total energies and free energies are given in Table 2. More details are available in the Supporting Information. Starting at 12, the adsorption of acrolein at the site of the terminal oxygen vacancy is favorable, forming a reduced molybdenum oxide cluster with adsorbed acrolein and a surface hydroxyl (11). Next, a proton migration occurs from the surface hydroxyl over TS4 to acrolein’s C-1 forming an allyl alkoxide intermediate (10). The activation energy barrier is 104 kJ/mol. Structure 10 is a branch point in the mechanism leading to three potential products, propene, allyl alcohol, and 1-propanol. To add a second hydrogen atom to the allyl alkoxide intermediate, it must first be added to the cluster. This was done through the addition of 1/2H2 to form either the structure 10\Ht or the structure 10\Hb where the notation indicates that an H atom is bonded to structure 10 at either a terminal or bridging oxygen site (see Figure 4c,d). Note that the thermodynamics of addition of hydrogen to the cluster complex is in accordance with the results of section 3.2 (Figure 6). The most desirable product from the perspective of hydrodeoxygenation catalysis is propene. This involves breaking the carbon-oxygen bond of the allyl alkoxide without hydrogenating the carbon-carbon double bond. In Figure 7, this pathway is shown by a solid line. The transition state for the formation of propene, TS2′, involves the proton transfer from the surface hydroxyl on 10\Ht to C-3 of the allyl alkoxide forming propene coordinated to the surface with an extended Mo-O bond (5). The prime notation indicates that our TS2′ is somewhat different

Hydrodeoxygenation of Acrolein on MoO3

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13787

Figure 7. Calculation of total optimized energy change at 0 K for reaction of acrolein and hydrogen on a Mo3O9 cluster, leading to either propene (solid line), allyl alchohol (dotted line), or 1-propanol (dashed line). See scheme of eq 6 for details. The notation 10\Ht means a hydrogen atom is covalently bonded to a terminal oxygen of structure 10 while 12+propene means two gas phase species. Structure 10 is a common branch point on the PES. The inset gives overall ∆rG for the three reactions, with (a) reactants at 0 K, (b) products at 0 K, (c) products at 323 K, and (d) experimental values of the products at 323 K.54 At 0 K, 1-propanol is thermodynamically favored, while allyl alcohol is least favorable. At 323 K, allyl alcohol remains less favorable thermodynamically than either propene or 1-propanol, but propene becomes very slightly more favorable than 1-propanol due to entropic effects (see eq 6).

in structure and properties from TS2 reported by Pudar et al.44 The activation energy barrier is 104.2 kJ/mol. The desorption energy for propene, essentially equal to the desorption barrier, is 13.9 kJ/mol. The energy required to regenerate the catalyst (structure 12) by adding hydrogen and removing H2O is 25.7 kJ/mol. Overall the ∆E for this branch of the reaction scheme is -185.8 kJ/mol. The value of ∆rG323 for this process is -116.5 kJ/mol (see insert of Figure 7c). The next most desirable product is allyl alcohol, since the carbon-oxygen bond is hydrogenated while leaving the carbon-carbon double bond unaffected. In order to produce allyl alcohol, a proton has to be added to the oxygen of the allyl alkoxide. The proton on a terminal site is too far away to make this migration, so the proton must come from a bridging site.55 From the bridging position the proton can be transferred to the oxygen of the allyl alkoxide (15) through TS6 with an activation energy of 43 kJ/mol. The large energy (127 kJ/mol) required to desorb the loosely coordinated allyl alcohol from the cluster (structure 15) to form Mo3O8 (16) is not so problematic in free energy terms as the temperature is raised: the free energy of desorption is 66 kJ/mol at 323 K. The last step of this pathway regenerates the original structure 12 by adding 1/2H2 to a terminal site. Overall the ∆E for the production of allyl alcohol in this scheme is -76.6 kJ/mol. The value of ∆rG323 for this process is -9.9 kJ/mol. The least desirable product, 1-propanol, involves the hydrogenation of both double bonds. From 10\Ht, the next step in

forming 1-propanol is the transfer of a proton to the C-3 position of the allyl alkoxide (17) similar to the first step in the formation of propene. We note that the hybridization is still largely sp2 and the C-O bond length is rather extended. The activation energy barrier for this process through TS7 is 107.1 kJ/mol. Since this barrier is very similar to the barrier for the production of propene (TS2′), the relative selectivity of propene and 1-propanol cannot be determined by these barriers alone. The addition of 1/2H2 to form a surface hydroxyl is favorable, forming 17\Ht. The proton can then be transferred to the C-2 position forming surface propoxide (18) via TS8. This activation energy barrier is 128 kJ/mol, which is significantly larger than TS7 because C-2 must stretch to span the distance to the surface hydroxyl. In 18, the C-O bond has re-formed and the hybridization within the intermediate has converted to sp3. Another 1/2H2 must be added to the cluster at a bridging oxygen forming the structure, 18\Hb.56 From this point, the energetics of alcohol formation parallels the allyl alcohol pathway. The proton migrates from a bridging oxygen site to the oxygen of the propoxide forming chemisorbed 1-propanol (19). The activation energy barrier (TS10) for the formation of this species is 38 kJ/mol. Desorption of 1-propanol requires 120 kJ/mol and leaves behind the same structure, 16, generated at the end of the allyl alcohol pathway. Overall, ∆E for 1-propanol production is -225.3 kJ/mol. The value of ∆rG323 for this part of the scheme is -93.4 kJ/mol.

13788

J. Phys. Chem. C, Vol. 114, No. 32, 2010

Moberg et al. TABLE 2: Energy and Free Energy of Species in PES of Figure 7 Relative to Structure 12+acrolein+2H2 species

∆rE/(kJ/mol)b

∆rG323/(kJ/mol)b

12+acrolein (+2H2) 11 TS4 10 10\Ht TS2′ 5 4+propene 12+propene TS5 10\Hb TS6 15 16 + allyl alcohol 12 + allyl alcohol TS7 17 17\Ht TS8 18 18\Ht TS9 18\Hb TS10 19 16 + 1-propanol 12 + 1-propanol

0a -119 -17 -164 -184 -80 -225 -211 -186 22 -138 -95 -149 -23 -71 -77 -225 -276 -141 -317 -340 -131 -288 -250 -297 -177 -226

0 -58 37 -97 -85 8 -133 -172 -117 111 -38 -5 -48 18 -5 20 -133 -155 -27 -186 -177 22 -125 -93 -121 -70 -93

a Total energy at this level of theory is -999.7564229 hartree (material balance is respected as in eq 6). Additional energies and free energies in hartrees: acrolein, -191.97412464; 12, -805.4231562; H2, -1.179571018. b All energies and free energies refer to B3LYP/6-311+G(d,p) single point calculations for structures optimized at the B3LYP/6-31G(d,p) level; LANL2DZ ECP and basis used for Mo atoms. c Additional free energies in hartrees: G323 values: acrolein, -191.943328; 12, -805.437659; H2, -1.182464; (12+acrolein +2H2), -999.745915.

4. Discussion

Figure 8. Structures associated with the acrolein reaction pathways shown in the potential energy surface of Figure 7.

Based on the barriers found along the three pathways in the scheme of Figure 7, our results suggest that the cluster would be selective toward allyl alcohol. The formation of propene is reaction limited with ∆Gq323 (TS2′) ) 93 kJ/mol, since the desorption barrier for propene is ∆desG323 ) 14 kJ/mol. For allyl alcohol, the formation barrier and desorption barriers are comparable: ∆Gq323 (TS6) ) 33 kJ/mol and ∆desG323 ) 34 kJ/ mol, respectively. Finally for 1-propanol, while the desorption barrier is somewhat larger than that for allyl alcohol (∆desG323(16 + 1-propanol) ) 51 kJ/mol) the reaction barrier is quite a bit larger: ∆Gq323 (TS8) ) 128 kJ/mol. Table 2 provides a summary of the energies of the stable species and transition states depicted in Figures 7 and 8. Also given, as appropriate, are the ∆G323 values of these species, both calculated at the UB3LYP/6-311+G(d,p)/LANL2DZ level using structures optimized at the UB3LYP/6-31G(d,p)/LANL2DZ level.

For selective oxidation catalysis, Grasselli has laid out a series of principles that are crucial to understanding and designing effective catalytic materials: role of “lattice oxygen, metal-oxygen bond strength, host structure, redox, multifunctionality of active sites, site isolation, and phase cooperation.”25 As hydrodeoxygenation is essentially the reverse process, we frame our discussion in terms of a condensed set of principles, namely, (1) the importance of lattice oxygen defects and the relative strength of metal-oxygen vs carbon-oxygen bonds; (2) the structural stability of metal oxides undergoing significant reduction and hydrogen bronze formation; (3) the relative rate of substrate reactivity and catalyst regeneration; and (4) the importance of acidity in catalyst selectivity. 4.1. Lattice Oxygen and Relative Strength of Metal-oxygen vs Carbon-oxygen Bonds. For HDO, a key feature of the reducible metal oxide is the ease with which an oxygen vacancy can be created, influencing the number of active sites where the substrate can adsorb and, therefore, the activity of the catalyst. The coordinatively unsaturated surface metal ion is therefore the reducing agent. The number and nature of the coordinatively unsaturated sites can be controlled through both choice of metal oxide and the hydrogen pressure. The thermodynamic relationships between oxygen pressure and oxygen vacancy creation energy are well-known.57,58 For WO3, which is an n-type semiconducting oxide, the vacancy concentration varies as PO2-1/6.59-61 We expect MoO3 to behave similarly due

Hydrodeoxygenation of Acrolein on MoO3

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13789

to the large charge on the Mo6+ ion that forms polarons in these materials. This behavior derives from the reaction

Oxo H V••o + 2e′ + 1/2O2(g)

(7)

in Kro¨ger-Vink notation,57 and the assumption that the oxygen vacancy, Vo••, is doubly ionized, such that [e′] ) 2[Vo••], so that the fractional vacancy concentration,

θV ) [V••o ]/[Oxo] ) (K7 /4)1/3pO2-1/6

(8)

where K7 ) exp(-∆G7/RT) is the equilibrium constant for reaction 7. Using our estimate for surface terminal oxygen vacancy creation of ∆Gterm ) 568 kJ/mol, K7 ) 2 × 10-48 at 623 K, or θV ) 8.4 × 10-17pO2-1/6, which is too small to provide sufficient surface oxygen vacancies for the proposed mechanism. However, combining reaction 7 with the equilibrium, 1/2O2 + H2 H H2O, gives the reaction

Oxo + H2(g) H V••o + 2e′ + H2O(g)

(9)

The oxygen vacancy concentration is then

θV )

( ) K9pH2 4pH2O

1/3

(10)

Using our estimate for the surface terminal oxygen vacancy creation energy of 74 kJ/mol for reaction 9, K9 ) 6.2 × 10-7 at 623 K, or θV ) 5 × 10-3(pH2/pH2O)1/3, which is qualitatively consistent with observation of a substantial coverage of surface vacancies. For example, creation of surface oxygen vacancies during reduction of WO3(100) has been studied at the atomic scale by scanning tunneling microscopy,62 and the morphological changes during reduction of MoO3(010) in H2/N2 mixtures on the mesoscale have been elucidated by atomic force microscopy.63 We emphasize that through a process of oxygen vacancy diffusion, the site of vacancy creation may differ from the site of oxygen abstraction from the oxygenate. Our cluster calculations only consider the thermodynamic energy costs of oxygen vacancy creation on the Mo site, but the reaction of hydrogen to produce water (∆E )74 kJ/mol) lowers the cost dramatically compared to the surface metal-oxygen bond strength (∆E ) 568 kJ/mol), making surface oxygen vacancy creation viable. Furthermore, the pH2 provides an independent means of controlling the surface oxygen vacancy concentration. For HDO, the stronger the surface metal-oxygen bond, the harder it will be to create surface vacancies for adsorption of the substrate, while if the metal-oxygen bond is too weak, the material will not be able to abstract oxygen from the molecule. We estimate the ModO and MosO bond strengths using the reaction scheme of eq 5 to be 568 and 352 kJ/mol from the first and third entries of Table 1 which report the energy to produce ground state atomic oxygen and OH radical, respectively. Our number for the strength of the double bond is reasonably consistent with that reported by Dixon and coworkers,64 who report 574.88 kJ/mol for ModO and 405.85 kJ/mol for MosO, based on higher level calculations of several cluster models. That our single bond energy is lower is consistent with the fact that our value is based on the desorption of OH rather than O(3P).

As the average CdO and CsO bond enthalpies (at 298 K) are 745 and 358 kJ/mol, respectively,65,66 if the CdO bond of acrolein can be reduced to CsO while the molecule is adsorbed at an oxygen vacancy on the surface, then the energy cost of breaking the CsO bond (358 kJ/mol) becomes less than the energy gain of forming a ModO bond (574.88 kJ/mol). Mechanistically, this suggests that hydrogenation of the CdO bond needs to occur first, so that the CsO bond will be weaker than the ModO bond, thus favoring CsO bond cleavage (forming propene) over MosO cleavage (forming the alcohol). Theoretical calculations by Fu et al.67 and Li et al.64 determined the strength of a variety of MsO bonds. The bond energies of metal oxides of W, Mo, Cr, and V were calculated from the atomization energy of each metal oxide cluster. From their work the metal oxide bond strengths can be ranked in order of decreasing bond energy: W > Mo > V > Cr. From these results we expect the oxides of W or Mo to be better for hydrodeoxygenation as the V-O and Cr-O bonds are too weak to perform oxygen abstraction. This agrees with the experimental work that shows that WO3 and MoO3 can be used as selective reduction catalysts.13,40 4.2. Structural Aspects of Vacancy Formation and Hydrogen Storage. For coordinatively unsaturated metal sites to abstract oxygen, the metal oxide must support a high degree of oxygen vacancy formation without suffering collapse, since for the surface oxygen vacancy reaction site to be separated from the site of oxygen vacancy creation, vacancy diffusion is required. Likewise, for selective hydrogenation to occur according to the mechanism proposed, dissociation of hydrogen should occur at different sites than the Brønsted acid sites where proton donation occurs. For similar reasons, the capability to absorb a significant quantity of hydrogen atoms that can diffuse easily through the structure appears to be advantageous. Although our calculations on the Mo3O9 cluster do not allow for explicit calculation of the surface vs bulk populations of oxygen vacancies and hydrogen atoms, we consider the ability of MoO3 and WO3 to support oxygen vacancies and form bronzes, in part to justify the thermodynamic (as opposed to kinetic) arguments utilized in creating surface hydroxyls and oxygen vacancies within the mechanism shown in Figure 7. The family of reducible metal oxides, including the molybdates, tungstates, and vanadates, is a group of materials that can accommodate oxygen vacancies without structural collapse.25 The compositions of these nonstoichiometric materials have been characterized as a function of the partial pressure of oxygen and of temperature. Higher degrees of reduction occur through structural defect formation, known as crystallographic shear planes, leading to the so-called Magneli phases, as in the MoO3 system.68 The mechanism of shear plane formation has been studied extensively for WO3 and MoO3 through detailed transmission electron microscopy and X-ray diffraction studies.69,70 Electrostatic interactions between the shear planes have been proposed to account for the regular periodic arrangement of the planes.71 The formation of hydrogen bronzes, HxMoO3, with compositions, x, ranging from 0 to 2, is well-known. There are four distinct crystallographic phases of the HxMoO3 bronzes: type I (blue orthorhombic, 0.21 < x < 0.40), type II (blue monoclinic, 0.85 < x < 1.04), type III (red monoclinic, 1.55 < x < 1.72), and type IV (green monoclinic, x ) 2.0).72 Under the conditions that hydrogen-treated MoO3 is active for acrolein hydrogenation,43 Hoang-Van and Zegaoui provide evidence for bronze formation: although oxygen substoichiometry was suggested, the stoichiometry was not determined. From the composition

13790

J. Phys. Chem. C, Vol. 114, No. 32, 2010

Figure 9. Sequential absorption energy as a function of hydrogen composition, x, in HxMoO3, from Sha, et al.73 compared to surface terminal (solid line) and symmetric bridging (dotted line) sites calculated by the same group,74 showing that surface hydroxyl sites become populated when the bronze composition reaches the range x ) 1.1-1.2. Energies of surface terminal (dash-dot) and symmetric bridging (dash) hydroxyls on the Mo3O9 cluster are shown for comparison.

dependence of the (600) diffraction line, Matsuda et al.42 imply that the composition of the active phase for 2-propanol dehydration is intermediate between H0.93MoO3 and H2MoO3. For WO3, we determined13 that the active phase for hydrodeoxygenation of allyl alcohol to propene and 1,5-hexadiene at 523-623 K, following hydrogen pretreatment at 623 K, is a reduced tungsten oxide bronze of composition H1.3WO2.8. The evidence shows that the hydrogen composition of these bronzes is in the range x g 1.0. In the mechanism proposed for HDO in the PES of Figure 7, we have utilized hydroxyls on either bridging or terminal sites as needed. We first consider the thermodynamic population of protons on surface vs bulk sites at low hydrogen composition. Comparing the absorption energies of the four H binding sites within the HxMoO3 lattice, as reported by Sha,73 we calculate that, based on one proton per unit cell (x ) 0.25) at 323 K, the Boltzmann distribution results in an occupancy of 0.75 on the terminal oxygens (BE ) -61.8 kJ/mol, relative to 1/2H2), an occupancy of 0.25 on the asymmetric bridging oxygen (BE ) -58.9 kJ/mol), and negligible occupancy in the weaker asymmetric and symmetric bridging sites. Comparison of the bulk absorption energies with surface adsorption energies on the MoO3(010) bilayer slab, calculated by the same group,74 suggests that the surface terminal site (BE ) -18.4 kJ/mol relative to 1/2H2) has an occupancy 10-7 that of the most strongly bound “inward” asymmetric bridge site (BE ) -62.8 kJ/mol). For comparison, the binding energy of hydrogen on the terminal oxygen, Ht, in our Mo3O9 cluster calculations is -42 kJ/mol, which improves population of surface sites but is detrimental to their acidity for hydrogen insertion. Clearly, at low hydrogen composition, the depletion of surface hydroxyls would severely limit the hydrogenation functionality of the catalyst. However, as we have shown above, there is strong evidence that metal oxides which are active for acrolein reduction have high hydrogen content. As the hydrogen content increases from x ) 0.25 to x ) 1, the sequential absorption energy (at 0 K) in the bulk increases as shown in Figure 9 (relative to 1/2H2),73 becoming comparable energetically with the surface terminal sites when x ) 1.1 and

Moberg et al. similar to surface symmetric bridging sites when x ) 1.2. Our symmetric bridging and terminal hydroxyl adsorption energies are also shown in Figure 9 for comparison. This demonstrates that the population of hydrogen in surface sites is a function of the bulk hydrogen composition, which depends on the hydrogen pressure. Furthermore, to the extent that bridging sites differ in energy from terminal sites, the relative population of these species may be controlled by hydrogen pressure. 4.3. Relative Rates of Acrolein Reaction and Catalyst Regeneration. In this section, we turn from thermodynamics to kinetics. For hydrodeoxygenation, the rate of oxygen vacancy creation and diffusion must be faster than the rate of oxygen abstraction from the substrate to preserve the active phase of the catalyst. Likewise, hydrogen dissociation and transport to the surface must exceed hydrogenation rates. Furthermore, because dissociation of hydrogen and oxygen vacancy creation may take place distant from the hydrogenation and oxygen abstraction sites, electron transport must be fast to maintain redox activity. We first examine the factors involved in solidstate kinetics and then consider the acrolein reactions. Solid State Kinetics. Maintaining oxygen vacancy and surface hydroxyl concentrations requires oxygen vacancy creation and diffusion, hydrogen dissociation and diffusion, and electron transport. Oxygen vacancy diffusion in reducible metal oxides is facile. Isotope tracer measurements of the oxygen diffusion rate in these materials prove that transport is extremely fast75 and that, in fact, surface reactions control the rate of oxide reduction. We expect the activation energy for vacancy creation on MoO3 to be smaller than the activation energy that we measured13 during hydrogen reduction of WO3. We obtained apparent activation energies of 140-170 kJ/mol, depending on specific surface area, with negligible rates of reduction at temperatures below 473 K. Our calculated desorption energy for H2O chemisorbed at a Mo3O8 vacancy is 102 kJ/mol. Assuming a standard preexponential factor of 1013 s-1, we estimate the rate constant for H2O desorption as 3 × 104 s-1 at 623 K, which is a typical hydrogen pretreatment temperature and a temperature at which Matsuda et al.42 observe dehydration of 2-propanol to propene. At 323 K, where Hoang-Van and Zegaoui40 observe only hydrogenation of acrolein to allyl alcohol, the first-order rate constant for desorption is orders of magnitude smaller: 3 × 10-4 s-1. Hydrogen transport within the bulk is also facile. Ritter et al.76 have measured the activation energy of hydrogen diffusion through molybdenum oxide bronzes, which is 10-30 kJ/mol depending on hydrogen concentration. Bulk MoO3 calculations by Sha et al.73 show that the barrier to diffusion of hydrogen in the bulk is only 10 kJ/mol. Chen et al.74 calculate barriers of 50-60 kJ/mol for migration of hydrogen on the surface of a bilayer slab, which is significantly smaller than the barrier we calculate for migration between the terminal and bridging sites (∼200 kJ/mol) on the Mo3O9 cluster. The calculations of Chen et al.74 employ the PW91 functional77 implemented in VASP.78,79 There is evidence that the PW91 functional systematically underestimates desorption barriers of H2 from slab and cluster models of the reconstructed Si(100) surface by about 40 kJ/ mol when compared to the B3LYP functional with the same basis set.53 At the same time, a systematic increase in size of the Si cluster models appears to reduce the desorption barrier by about 40 kJ/mol as the cluster approaches a slab.53 Thus, these factors probably compensate each other, suggesting that the surface diffusion barriers calculated by Chen et al.74 are good estimates. We conclude that our barriers for H atom hopping

Hydrodeoxygenation of Acrolein on MoO3 from terminal to bridging sites on the Mo3O9 cluster are likely to be too large by a factor of 4. Experimental evidence suggests that on oxides such as sapphire, the surface diffusion barriers for hydroxyls and water are comparable to the desorption barrier.80 Monte Carlo simulations indicate that on WO3(001) surfaces, diffusion of alcohols is more facile but dehydroxylation leads to a substantial degree of surface heterogeneity,81 indicating that surface diffusion of hydroxyls (i.e., protons) is a potential limitation and that bulk diffusion may be a much faster transport process in the bronzes. Our result for molecular hydrogen adsorption on the terminal oxygen vacancy site, i.e., Lewis acid sites, (section 3.2) gives an adsorption energy of -42 kJ/mol with a negligible barrier to adsorption. We are not aware of other calculations or experiments to confirm this, but to fully explore the hydrogen dissociation and diffusion through the bulk, vis-a`-vis the surface, will require slab calculations. The pressure dependence of hydrogen adsorption, suggested by eq 9 has also not been explored, to our knowledge. Hydrogen spillover has been investigated to facilitate catalyst reduction and oxygen vacancy creation at lower temperatures. Hoang-Van and Zegaoui showed that impregnating WO3 and MoO3 with 0.1% platinum increases the rate of acrolein hydrogenation from 0.07 mmol h-1 g-1 to 0.55 mmol h-1 g-1.82 Chen et al.74 have recently calculated the barrier for migration of H atoms from a six-atom Pt cluster to the MoO3 (010) (slab) surface to be ∼35 kJ/mol, and as reviewed above, their calculated barriers for H atom diffusion from surface to bulk sites are of the order of 50 kJ/mol while diffusion barriers within the bulk are of the order of 13 kJ/mol or less.73,74 Thus it is possible that at high temperatures, a catalyst like MoO3 (or its bronze) may display the necessary bifunctionality on its own while at low temperature, the addition of a second material such as clusters or nanoparticles of Pt will provide the necessary functionality to maintain the catalyst activity. We assume that hydrogen dissociation occurs at sites separated from proton donation sites and also that rapid electron transport is required to prevent surface charging. The nonstoichiometric oxides in the molybdate and tungstate families are n-type, narrow band gap semiconductors deriving from electron transport in the conduction band.83 The mobilities of the electrons are rather low compared to most semiconductors84 due to strong interactions with the M6+ ions, but conductivities of 10-6 to 10-4 Ω-1 cm-1 have been reported for WO3 and MoO3 films.85,86 At low temperatures, transport involves formation of polarons, which have been studied extensively with ESR.87 The conductivity increases with temperature and degree of nonstoichiometry.59 Electron transport is also facilitated by formation of bronzes in the tungstates and molybdates. Band structure calculations of the HxMoO3 bronzes explain the well-known metallic character.73 As x increases from 0.25 to 1.5, the Fermi level rises into the free-electron-like conduction band. Thus, electronic transport requirements are easily met in the reduced bronzes Acrolein Reaction Rates. We begin by noting that Li et al.88 have calculated the energy barriers for hydrogenation of the CdO (316 kJ/mol) and CdC (306 kJ/mol) bonds in the gas phase. The barriers on our cluster are much smaller. The similarity of the gas phase barriers will not allow for selectivity between the functional groups, but from gas phase thermodynamics, propene is the most favorable product. Selectivity toward allyl alcohol or 1-propanol requires catalytic lowering of the CdO barrier and kinetic control.

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13791 We make an order of magnitude estimate of the rate of conversion of acrolein on the Mo3O9 cluster. On the basis of the activation energy barriers for the surface reactions, we expect that the rate of the reaction will be controlled by the barrier TS4, while the selectivity between propene, 1-propanol, and allyl alcohol will be determined by the relative rates associated with TS2′, TS6, and TS8. The transition state TS4 leading to 10 has a classical activation energy barrier of 104 kJ/mol, which is essentially identical to the classical activation energy barrier leading to propene (TS2′). By contrast, the classical energy barrier to allyl alcohol (TS6) is only 33 kJ/mol, while the ratelimiting step in the 1-propanol pathway is 135 kJ/mol. A transition state theory calculation52 of the rate constant for TS4 is kTS4 ) 8.2 × 1011 s-1 exp{(-89 kJ/mol)/RT}, which is 2.8 × 10-3 at 323 K and 160 s-1 at 473 K. The rate constant for hydrogen transfer from the surface hydroxyl to the C3 carbon of the allyl alkoxide intermediate 10\Ht through TS2′ gives kTS2 ) 1 × 1012 s-1 exp{(- 93 kJ/mol)/RT}, which is 1 × 10-3 s-1 at 323 K and 58 s-1 at 473 K. The rate constant for the formation of allyl alcohol through TS6 was calculated to be kTS6 ) 8 × 1012 s-1 exp{(- 33 kJ/mol)/RT}, which is 3 × 107 s-1 at 323 K and 1.6 × 109 s-1 at 473 K. The rate-limiting step for the formation of 1-propanol is the addition of a proton to C-2 to form propoxide (18) through TS8, with a rate of kTS8 ) 2 × 1010 s-1 exp{(- 126 kJ/mol)/RT}, which is 8 × 10-11 s-1 at 323 K and 2.3 × 104 s-1 at 473 K. Note that these are all calculated as first-order rate constants, assuming a proton exists in the bridging or terminal surface hydroxyl sites, as required. As discussed in section 4.2, the occupancy of these sites is controlled by the bronze composition, and therefore, at lower hydrogen composition (i.e., lower pH2), the production of alcohols could be diminished. For higher hydrogen compositions, in which the terminal and bridging surface hydroxyls are fully occupied (θHt ) θHb ≈ 1), the rates of allyl alcohol, 1-propanol, and propene become pseudo-first-order in the concentrations of species 10, 17, and 10, respectively. To estimate the rate of the acrolein reaction, we observe that the cluster is highly selective toward allyl alcohol and the rates of the elementary steps from 10 to gas phase allyl alcohol are fast compared to the rate of crossing the barrier TS4. Therefore, we apply the steady-state approximation for intermediate 11, including the reversible adsorption and desorption of acrolein onto surface terminal oxygen vacancies and the rate of crossing TS4. At steady state

pAcθV d[Ac] ) - [Ac](kdes + kTS4) ) 0 dt √2πmAckT

(11)

where pAc is the partial pressure of acrolein, mAc is the molecular mass of acrolein, [Ac]is the coverage of intermediate 11, and kdes is the rate constant for desorption of acrolein. The steadystate coverage of 11 is then

[Ac] )

pAcθV

1 (k + √2πmAckT des kTS4)

(12)

and the rate of production of 10, which controls the rate of allyl alcohol production, is

r10 ) kTS4[Ac]

(13)

13792

J. Phys. Chem. C, Vol. 114, No. 32, 2010

Moberg et al.

For the typical conditions of pAc ) 2 kPa, a fractional terminal vacancy coverage of 5 × 10-3 (eq 10 with pH2 ) pH2O), kdes ) 2 × 105 s-1, we find that the coverage of species 11 is 9 × 1013 cm-2 leading to a reaction rate of 2.6 × 1011 cm-2 s-1. If the reaction is selective for allyl alcohol, then little hydrodeoxygenation occurs and the vacancy creation rate required to regenerate the catalyst is negligible. However, if the alcohol pathways were blocked and hydrodeoxygenation to propene became dominant, then we would need to estimate the rate of water desorption for comparison. At 323 K, the rate constant for water desorption from the oxide would be 3 × 10-4 s-1, so the rate of oxygen vacancy creation

rv ) kvθterm

(14)

is roughly 2 × 1011 cm-2 s-1, where θterm ) 6.8 × 1014 cm-2 (1 - θV) is the coverage of terminal surface hydroxyls. The above mechanism predicts that the fractional vacancy concentration increases with the ratio pH2/pH2O, which increases the acrolein reaction rate, potentially becoming greater than the vacancy creation rate, limiting the rate of the acrolein adsorption. Thus, our estimates indicate that the rate of acrolein conversion, based on activation energy barriers identified on the Mo3O9 cluster, are similar in magnitude to the oxygen vacancy creation rate. 4.4. Role of Acidity in Catalyst Selectivity. In hydrodeoxygenation, the following processes must occur: chemisorption, proton donation, and desorption. Adsorption and desorption energies are controlled by the Lewis acidity of oxygen vacancy sites, while proton donation is dependent on the Brønsted acidity of the surface hydroxyls. Lewis Acidity. Chemisorption of aldehydes, ketones and alcohols onto the surface oxygen vacancies are favorable due to the Lewis acid-base interaction between the coordinatively unsaturated metal ion and the oxygen lone pair. The trends in ammonia thermogravimetric analysis89 and IR spectroscopy of pyridine adsorption90 show that the unsaturated metal ions in MoO3 are relatively strong Lewis acids. The surface concentration of Lewis acid sites upon reduction varies for different oxides, decreasing in the order Cr2O3, WO3, Nb2O5, Ta2O5, V2O5 ≈ MoO3.89 The strengths of the Lewis acid sites increase as Cr2O3 < MoO3 < WO3, based on cluster calculations by Li and Dixon.91 Comparison between the adsorption energies of propene (14 kJ/mol) and that of acrolein (119 kJ/mol) and the alcohols (120 kJ/mol) shows that the surface defects provide a substantial degree of selectivity toward oxygen-containing carbonyl and alcohol functionalities over unsaturated carbon double bonds. The substantial binding energy for acrolein is accompanied by a lengthening of the CdO bond from 1.216 to 1.262 Å in the process of forming the Mo-O bond. Desorption barriers of products play an important role in the selectivity of the catalyst only if the reaction is desorption limited. Propene is loosely coordinated to the surface of the cluster, which is a factor that favors hydrodeoxygenation. The C-O bond distance extended going through TS2′ from 1.451 Å in the allyl alkoxide (10\Ht) to 3.549 Å in 5. The Mo-O bonds shortened from 1.866 to 1.702 Å, which corresponds essentially to a ModO bond. The stronger interaction of allyl alcohol with the oxygen vacancy site (121 kJ/mol) can also be seen in changes in bond lengths. When allyl alcohol formed on the surface, the Mo-O distance increased by 0.254 Å from 1.701 to 2.120 Å as the C-O bond increased slightly from 1.423 Å in the gas phase to

1.467 Å. Likewise, the O-H bond length of 0.981 Å is longer than in gas phase allyl alcohol (0.965 Å). Chemisorbed 1-propanol displays similar interactions with Mo sites and a nearly identical desorption energy of 120 kJ/ mol. When 1-propanol is formed from propoxide, the Mo-O distance extends from 1.867 to 2.134 Å. At the same time the C-O bond increases to 1.479 Å and an O-H bond forms with a bond length of 0.966 Å. After 1-propanol desorbs, the C-O shrinks to 1.424 Å but the O-H bond distance remains relatively constant (0.966-0.965 Å). Brønsted Acidity. An HDO catalyst must have the capability to donate protons; i.e., it must have strong Brønsted acid surface hydroxyls. Matsuda et al.41,42 have shown that H2 pretreatment of MoO3 is critical to dehydration of 2-propanol to form propene and attribute this behavior to an increase in the concentration of acidic sites, as confirmed by ammonia temperature programmed desorption. On the basis of infrared spectroscopy of amines, Busca92 reports the order of Brønsted acidity as WO3 > MoO3 > V2O5 > Nb2O5. Murrell et al.,93 determined the Brønsted acidity of a series of oxides based on their ability to crack light gas oil. Of the oxides they studied, WO3, Nb2O5, MoO3, and V2O5 were the most acidic, listed in order of decreasing acidity. The details of the acid strength series may depend on the oxide pretreatment conditions as well as the type of measurement. Brønsted acidity has also been investigated theoretically. Bernholc et al.94 determined the Brønsted acidity of the metal oxide clusters TiO4H4, WO4H2, and NbO4H3 by calculating the energy cost of removing a proton from the clusters. The structure with the lowest energy cost was the most acidic: the Brønsted acidity decreases in the order WOx > NbOx . TiOx. In a recent paper, Fu et al.67 calculate the Brønsted basicity of the Mo3O9 cluster used in this work, its chromium and tungsten analogues, and the V3O6Cl3 analogue from the difference of bond energies, BE(CH) - BE(OH), for reactions of the clusters with alkanes (C1 to C4) to form surface OH (in both terminal and bridging varieties). This energy difference increases from 33 kJ/mol for H abstraction from isobutane at the Cr3O9 terminal oxygen position to a high of 332 kJ/mol for H abstraction from methane at the W3O9 cluster bridging oxygen position. For propane on Mo3O9, they report abstraction barriers of 138 and 212 kJ/mol at the terminal and bridging positions, respectively. In our study, the reverse activation energy barrier (from 5 to TS2′) corresponds to abstraction of a proton from propene. The barrier is ∆E(TS2′ - 5) ) 146 kJ/mol while the forward barrier is 104 kJ/mol, which is similar in magnitude. Similar numbers can be deduced on the 1-propanol pathway, where the reverse barrier, ∆E(TS7 - 17) ) 149 kJ/mol and ∆E(TS8 - 18) ) 175 kJ/ mol. The good agreement between our results indicates that, to the extent that the Brønsted acidity (i.e., forward barriers) largely controls the rate of reaction, a more acidic oxide, e.g., WO3, would be advantageous. In exploring the configuration space for the adsorbed allyl alkoxide intermediate 10/Ht for transition states leading to hydrogenation of the C-1, C-2, and C-3 carbons, the proximity of hydroxyl groups on the Mo3O9 cluster imposed significant geometric constraints that affect the size of the proton donation barrier. We note that the distance between the terminal oxygens varies between 3.6 Å in the geometry optimized cluster 11 of Pudar et al.44 and 5.45 Å in 5. In the MoO3(110) surface, the oxygen distance is about 3.5 Å, while in the geometry-optimized Mo3O9 cluster it is about 5 Å. Decreasing the oxygen distance to 3.5 Å increases the cluster energy by about 30 kJ/mol, which may artificially increase the proton donation barriers on the

Hydrodeoxygenation of Acrolein on MoO3 Mo3O9 cluster. Several low-energy pathways via rotation about the two dihedral angles were explored in an unsuccessful effort to bring the C-2 hydrogen near a surface hydroxyl, while the C-3 carbon could easily interact with either of the terminal surface OH groups. It is interesting to note that in TS8, the C-O bond must stretch in order to get close enough to the hydroxyl for hydrogen transfer to occur. This appears to contribute to the higher barrier to hydrogenation to form 1-propanol at C-2 than at C-3. Clearly, the atomic arrangement of surface hydroxyls is dependent on the crystallographic structure of a particular facet of a particular phase of oxide. 4.5. Assessment of Cluster Model and Theoretical Method Accuracy. The inconsistency between the adsorption energies and activation energy barriers for hydrogen on the Mo3O9 cluster and the results of Tokarz-Sobieraj et al.46 and Chen and co-workers73,74 raises the question of the accuracy of the theoretical method and the cluster model itself. We can evaluate the accuracy of the B3LYP functional and 6-311+G(d,p) basis set from comparison of the Gibbs free energy of the gas phase reaction to experimental Gibbs free energies. Experimental data were obtained from the Gibbs free energy of formation as a function of temperature as reported in Yaws’ handbook.54 The calculated ∆rG323 values are plotted at the far right of the insert of Figure 7. At 323 K the free energies calculated by DFT are in the correct order, with propene being the thermodynamically favored product. All of the ∆rG323 values calculated from DFT are within 10 kJ/mol of the experimental values. To gauge the accuracy of the core potentials for Mo, we compare the vibrational frequencies of the Mo3O9 cluster to experimental results. Experimental infrared spectra of the (MoO3)3 polymer have been recorded in a solid neon matrix.95 They report bands at 977.6 and 840 cm-1 which can be compared to the calculated vibrational frequencies of 976 and 833 cm-1, respectively, at the B3LYP/LANL2DZ/6-311+G(d,p) level, and using the scaling factors for harmonic frequencies given by Andersson.48 This level of agreement provides good support for the accuracy of the basis set for oxygen and the effective core potentials we are using for Mo. We have chosen to use the Mo3O9 cluster to model HDO because it has been used by Pudar et al.44 to model the oxidation of propene to acrolein. A direct comparison with experimental activation energy barriers is provided by their calculation, at the double-ζ level, of the barrier for CH activation on both the Bi4O7 and Mo3O9 clusters. They find that CH activation is feasible at BiV sites with a barrier of 46 kJ/mol while the barrier on Mo3O9 is 136 kJ/mol. Their activation energy barrier on Bi4O7 is 12 kJ/mol lower than the experimentally determined barrier,96 which they attribute to the small size of the cluster. The barrier on Mo3O9 is substantially higher, which agrees qualitatively with the fact that propene activation is not observed on MoO3. In summary, the level of theory can be expected to provide estimates of energy differences within 10 kJ/mol and scaled vibrational frequencies within 1%. As discussed in section 4.3, Steckel et al.53 have shown that the cluster size effects are particularly important for hydrogen adsorption and diffusion. Coupled with the much lower barriers computed by Chen for a one layer slab, it is quite likely that the rather large hydrogen migration barriers on the Mo3O9 cluster are due to the small cluster size. However, for the reactions other than hydrogen adsorption and diffusion, the cluster model is likely to give accuracies of order 10 kJ/mol.

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13793 5. Conclusions and Perspectives on Further Work We have extended the potential energy surface for reactions of propene to acrolein by Goddard’s group44,97 on a Mo3O9 cluster model of the MoO3 surface to investigate the thermodynamics and kinetics of hydrogenation and selective hydrodeoxygenation of acrolein. We find that in the presence of hydrogen, conversion of acrolein to allyl alcohol, 1-propanol, and propene is thermodynamically favorable, but the selectivity can be controlled through kinetic barriers to form the least favorable product, allyl alcohol. We propose a mechanism in which oxygen vacancies are formed through dehydroxylation of surface terminal sites to form water. The oxygen vacancy fractional coverage was estimated to be in the range of θV ) 0.005-0.05 at 623 K, depending on the pH2/pH2O ratio. The coordinatively unsaturated Mo sites (i.e., oxygen vacancies) serve as selective sites for chemisorption of acrolein. Experimental evidence13,41-43 indicates that under typical hydrogen reduction conditions, the active phase of the catalyst is a reduced hydrogen bronze, HxMoO3-y. Comparison with the work of Sha et al.73 and Chen et al.74 indicates that surface hydroxyl sites are occupied when y is in the range 1.1-1.2, which appears to be consistent with experimental data of Matsuda et al.41,42 Surface hydroxyls are important for both oxygen vacancy formation and as Brønsted acids. The reaction rate is essentially controlled by protonation of the C-1 carbon of chemisorbed acrolein. Additional reaction barriers for proton donation to the C-2 or C-3 sites are similar in magnitude (100-135 kJ/mol), limiting the rates to 1-propanol and propene, while the barrier to O-H bond formation is much smaller (33 kJ/mol) leading to allyl alcohol. The calculated reaction barrier for C-O scission leading to the hydrodeoxygenation product is 104 kJ/mol. On the basis of our proposed mechanism, we expect that the barrier may be lower for metal oxides such as WO3 that have stronger metal-oxygen bonds. The stronger Brønsted acidity of surface hydroxyls on WO3 may also lower the barriers to proton donation. However, the ability to form surface oxygen vacancies may decrease, potentially lowering the activity of the catalyst. Evidence from our own work has shown that hydrogen reduction of WO3 produces a bronze, HxWO3-y, that is, in fact, active for hydrodeoxygenation of allyl alcohol to propane and propene and 1,5-hexadiene.13 Although the calculations suggest that the Mo3O9 cluster is highly selective toward allyl alcohol, the flexibility of the ring structure allows the O-O distance to expand to values (∼5 Å) much larger than would be expected, for example, on the MoO3(010) surface (3.6 Å) and the binding and activation energies for proton transfer may be overestimated due to the small cluster size. For these reasons, as well as to be able to explore the role of hydrogen bronze on protonation, we plan to extend this work using slab models and may find qualitative changes in the selectivity toward propene. The estimated reaction rate is comparable to the rate of oxygen vacancy creation, so that operation in a continuous flow process appears to be feasible. Acknowledgment. We thank the Department of Energy, under Grant No. DE-FG02-07ER46373 for financial support of this work and the UMaine High Performance Computing Facility for computing time. We are grateful to Rachel Austin, Ping Liu, and Clay Wheeler for valuable discussions and critical reading of the manuscript. Supporting Information Available: Computational details including Cartesian coordinates, total energies at absolute zero,

13794

J. Phys. Chem. C, Vol. 114, No. 32, 2010

and free energies at 323 K of individual species at the 6-31Gd,p and 6-311+G(d,p) levels. This material is available free of charge via the Internet at http://pubs.acs.org. References and Notes (1) Breaking the Chemical and Engineering Barriers to Lignocellulosic Biofuels: Next Generation Hydrocarbon Biorefineries; Huber, G. W., Ed.; National Science Foundation. Chemical, Bioengineering, Environmental, and Transport Systems Division: Washington, D.C., 2008. (2) Huber, G. W.; Iborra, S.; Corma, A. Chem. ReV. 2006, 4044. (3) Churin, E.; Maggi, R.; Grange, P.; Delmon, B. Characterization and upgrading of a bio-oil produced by pyrolysis of biomass. In Research in Thermochemical Biomass ConVersion; Bridgwater, A. V., Kuester, J. L., Eds.; Elsevier Applied Science: New York, 1998. (4) Ingram, L.; Mohan, D.; Bricka, M.; Steele, P.; Strobel, D.; Crocker, D.; Mitchell, B.; Mohammad, J.; Cantrell, K.; Pittman, C. U. Energy Fuels 2008, 22, 614. (5) Elliott, D. C. Energy Fuels 2007, 21, 1792. (6) DeSisto, W. J.; Hill, N.; Beis, S. H.; Mukkamala, S.; Joseph, J.; Baker, C.; Ong, T.-H.; Stemmler, E. A.; Wheeler, M. C.; Frederick, B. G.; van Heiningen, A. Energy Fuels 2010, 24, 2642. (7) Garcia-Perez, M.; Chaala, A.; Pakdel, H.; Kretschmer, D.; Roy, C. Biomass Bioenergy 2007, 31, 222. (8) Mullen, C. A.; Strahan, G. D.; Boateng, A. A. Energy Fuels 2009, 23, 2707. (9) Oasmaa, A.; Meier, D. Characterisation, Analysis, Norms & Standards. In Fast Pyrolysis of Biomass: A Handbook; Bridgwater, A. V., Ed.; CPL Press: Newbury, U.K., 2005; Vol. 3, p 19. (10) Joseph, J.; Baker, C.; Jensen, B. L.; Wheeler, M. C.; DeSisto, W. J.; Frederick, B. G. Energy Fuels 2010, in press. (11) Piskorz, J.; Scott, D. S.; Radlein, D. Composition of Oils Obtained by Fast Pyrolysis of Different Woods. In Pyrolysis Oils from Biomass: Producing, Analyzing and Upgrading; Soltes, E. J., Milne, T. A., Eds.; ACS Symposium Series 376; American Chemical Society: Washington, DC, 1988; p 167. (12) Schwartz, T. J.; van Heiningen, A. R. P.; Wheeler, M. C. Green Chem. DOI: 10.1039/c005067a. (13) Thibodeau, T. J.; Canney, A. C.; DeSisto, W. J.; Wheeler, M. C.; Amar, F. G.; Frederick, B. G. Appl. Catal., A, submitted for publication. (14) Gutierrez, A.; Kaila, R. K.; Honkela, M. L.; Slioor, R.; Krause, A. O. I. Catal. Today 2009, 147, 239. (15) Gayubo, A. G.; Aguayo, A. T.; Atutxa, A.; Aguado, R.; Olazar, M.; Bilbao, J. Ind. Eng. Chem. Res. 2004, 43, 2619. (16) Churin, E.; Delmon, B.“What can we do with pyrolysis oils?”; Pyroysis and Gasification, 1989, Luxembourg. (17) Laurent, E.; Delmon, B. Appl. Catal., A 1994, 109, 77. (18) Williams, P. T.; Nugranad, N. Energy 2000, 25, 493. (19) Williams, P. T.; Horne, P. A. Biomass Bioenergy 1994, 7, 223. (20) Bridgwater, A. V.; Cottam, M.-L. Energy Fuels 1992, 6, 113. (21) Lu, Q.; Zhu, X.; Li, W. Z.; Zhang, Y.; Chen, D. Y. Chin. Sci. Bull. 2009, 54, 1941. (22) Yang, X.; Chatterjee, S.; Zhang, Z.; Zhu, X.; C. U.; Pittman, J. Ind. Eng. Chem. Res. 2010, 49, 2003. (23) Zhu, X.; Lobban, L. L.; Mallinson, R. G.; Resasco, D. E. J. Catal. 2010, 271, 88. (24) Elliott, D. C.; Baker, E. G. Hydrodeoxygenation of wood-derived liquids to produce hydrocarbon fuels. 20th Intersociety Energy ConVersion Engineering Conference, American Institute of Chemical Engineers: New York, 1985. (25) Grasselli, R. K. Top. Catal. 2002, 21, 79. (26) Mars, P.; van Krevelen, D. W. Chem Eng. Sci., Spec. Suppl. 1954, 3, 41. (27) Burrington, J. D.; Kartisek, C. T.; Grasselli, R. K. J. Catal. 1980, 63, 235. (28) Grasselli, R. K.; Burrington, J. D. Selective Oxidation and Ammoxidation of Propylene by Heterogeneous Catalysis. AdV. Catal. 1981, 30, 133. (29) Grasselli, R. K.; Burrington, J. D.; Buttrey, D. J.; DeSanto, P.; Lugmair, C. G.; Volpe, A. F.; Weingand, T. Top. Catal. 2003, 23, 5. (30) Grasselli, R. K. Top. Catal. 2001, 15, 93. (31) Callahan, J. L.; Graselli, R. K.; Milberger, E. C.; Strecker, H. A. Ind. Eng. Chem. Prod. Res. DeV. 1970, 9, 134. (32) Claus, P.; Bruckner, A.; Mohr, C.; Hofmeister, H. J. Am. Chem. Soc. 2000, 122, 11430. (33) Claus, P.; Hofmeister, H.; Mohr, C. Gold Bull. 2004, 37, 181. (34) Coq, B.; Figueras, F.; Geneste, P.; Moreau, C.; Moreau, P.; Warawdekar, M. J. Mol. Catal. 1993, 78, 211. (35) Grunert, W.; Bruckner, A.; Hofmeister, H.; Claus, P. J. Phys. Chem. B 2004, 108, 5709. (36) Marinelli, T.; Nabuurs, S.; Ponec, V. J. Catal. 1995, 151, 431. (37) Mohr, C.; Hofmeister, H.; Claus, P. J. Catal. 2003, 213, 86.

Moberg et al. (38) Mohr, C.; Hofmeister, N.; Lucas, M.; Claus, P. Chem. Eng. Technol. 2000, 23, 324. (39) Nagase, Y.; Hattori, H.; Tanabe, K. Chem. Lett. 1983, 1615. (40) Hoang-Van, C.; Zegaoui, O. Appl. Catal., A 1997, 164, 91. (41) Matsuda, T.; Hirata, Y.; Sakagami, H.; Takahashi, N. Chem. Lett. 1997, 1997, 1261. (42) Matsuda, T.; Hirata, Y.; Suga, S.; Sakagami, H.; Takahashi, N. Appl. Catal., A 2000, 193, 185. (43) Hoang-Van, C.; Zegaoui, O. Appl. Catal., A 1995, 130, 89. (44) Pudar, S.; Oxgaard, J.; Chenoweth, K.; van Duin, A. C. T.; Goddard, W. A., III J. Phys. Chem. C 2007, 111, 16405. (45) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Montgomery, J. A., Jr.; Vreven, T.; Kudin, K. N.; Burant, J. C.; Millam, J. M.; Iyengar, S. S.; Tomasi, J.; Barone, V.; Mennucci, B.; Cossi, M.; Scalmani, G.; Rega, N.; Petersson, G. A.; Nakatsuji, H.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Klene, M.; Li, X.; Knox, J. E.; Hratchian, H. P.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.; Ochterski, J. W.; Ayala, P. Y.; Morokuma, K.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Zakrzewski, V. G.; Dapprich, S.; Daniels, A. D.; Strain, M. C.; Farkas, O.; Malick, D. K.; Rabuck, A. D.; Raghavachari, K.; Foresman, J. B.; Ortiz, J. V.; Cui, Q.; Baboul, A. G.; Clifford, S.; Cioslowski, J.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, P.; Komaromi, I.; Martin, R. L.; Fox, D. J.; Keith, T.; Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Challacombe, M.; Gill, P. M. W.; Johnson, B.; Chen, W.; Wong, M. W.; Gonzalez, C.; Pople, J. A. Gaussian 03, ReVision E.01; Gaussian, Inc.: Wallingford, CT, 2004. (46) Tokarz-Sobieraj, R.; Witko, M.; Grybos, R. Catal. Today 2005, 99, 241. (47) Irikura, K. K.; Johnson, R. D.; Kacker, R. N.; Kessel., R. J. Chem. Phys. 2009, 130, 114102. (48) Andersson, M. P.; Uvdal, P. J. Phys. Chem. A 2005, 109, 2937. (49) Peng, C.; Schlegel, H. B. Isr. J. Chem. 1993, 33, 449. (50) Peng, C.; Ayala, P. Y.; Schlegel, H. B. J. Comput. Chem. 1996, 17, 49. (51) Ayala, P. Y.; Schlegel, H. B. J. Chem. Phys. 1997, 107, 375. (52) Eyring, H. J. Chem. Phys. 1935, 3, 107. (53) Steckel, J. A.; Phung, T.; Jordan, K. D.; Nachtigall, P. J. Phys. Chem. B 2001, 105, 4031. (54) Yaws, C. L. Chemical Properties Handbook: Physical, Thermodynamic, EnVironmental, Transport, Safety, and Health Related Properties for Organic and Inorganic Chemicals; McGraw-Hill: New York, 1999. (55) As shown in section 3.2, it is energetically more favorable to adsorb a hydrogen on a terminal site than on a bridging site. However, if we first add hydrogen to the terminal site and then transfer it to the bridging site, (10\Ht) f (10\Hb), this process has a substantial energy barrier, 206.3 kJ/ mol, represented by TS5. In the discussion, we argue that H addition to a bridging site is energetically accessible if the composition of the bronze is about H1.2MoO3. Also the barriers for H transport on the (010) surface have been shown to be much smaller than these on the cluster.74 (56) In order to move the hydrogen close enough to the oxygen in the propoxide to form 1-propanol, the proton transfer must occur from a bridging site, similar to TS5 and TS6 in the allyl alcohol pathway. The transfer from the terminal (18\Ht) to a bridging site (18\Hb) proceeds through TS9 and has an activation energy of 209 kJ/mol, comparable to TS5’s barrier. (57) Kroger, F.; Vink, V. Solid State Phys. 1956, 3, 307–n435. (58) Schmalzried, H. Solid State Reactions, 2nd ed.; Verlag Chemie: Weinheim, 1981. (59) Berak, J. M.; Sienko, M. J. J. Solid State Chem. 1970, 2, 109. (60) Ryu, K. H.; Oh, E. J.; Kim, K. H.; Yo, C. H. J. Korean Chem. Soc. 1995, 39, 157. (61) Gillet, M.; Lemire, C.; Gillet, E.; Aguir, K. Surf. Sci. 2003, 532535, 519. (62) Tanner, R. E.; Altman, E. J. J. Vac. Sci. Technol., A 2001, 19, 1502. (63) Smith, R. L.; Rohrer, G. S. J. Catal. 1996, 163, 12. (64) Li, S.; Hennigan, J. M.; Dixon, D. A.; Peterson, K. A. J. Phys. Chem. A 2009, 113, 7861. (65) Pauling, L. The Nature of the Chemical Bond; Cornell University Press: Ithaca, NY, 1960. (66) Atkins, P.; de Paula, J. Physical Chemistry, 8th ed.; W.H. Freeman and Company: New York, 2006. (67) Fu, G.; Chen, Z.-N.; Xu, X.; Wan, H. J. Phys. Chem. A 2008, 112, 717. (68) Magneli, A.; Blomberg-Hansson, B.; Kihlborg, L.; Sundkvist, G. Acta Chem. Scand. 1955, 9, 1382. (69) Booth, J.; Ekstrom, T.; Iguchi, E.; Tilley, R. J. D. J. Solid State Chem. 1982, 41, 293. (70) Sundberg, M.; Tilley, R. J. D. J. Solid State Chem. 1974, 11, 150. (71) Stoneham, A. M.; Durham, P. J. J. Phys. Chem. Solids 1973, 34, 2127.

Hydrodeoxygenation of Acrolein on MoO3 (72) Eda, K.; Sotani, N.; Kunitomo, M.; Kaburagi, M. J. Solid State Chem. 1998, 141, 255. (73) Sha, X.; Chen, L.; Cooper, A. C.; Pez, G. P.; Cheng, H. J. Phys. Chem. C 2009, 113, 11399. (74) Chen, L.; Cooper, A. C.; Pez, G. P.; Cheng, H. J. Phys. Chem. C 2008, 112, 1755. (75) Winter, E. R. S. J. Chem. Soc. A 1968, 2889. (76) Ritter, C.; Mu¨ller-Warmuth, W.; Scho¨llhorn, R. J. Chem. Phys. 1985, 83, 6130. (77) Perdew, J. P.; Chevary, J. A.; Vosko, S. H.; Jackson, K. A.; Pederson, M. R.; Singh, D. J.; Fiolhais, C. Phys. ReV. B 1992, 46, 6671. (78) Kresse, G.; Furthmu¨ller, J. Comput. Mater. Sci. 1996, 6, 15. (79) Kresse, G.; Hafner, J. Phys. ReV. B 1993, 48, 13115. (80) Nelson, C. E.; Elam, J. W.; Cameron, M. A.; Tolbert, M. A.; George, S. M. Surf. Sci. 1998, 416, 341. (81) Ma, S.; Amar, F. G.; Frederick, B. G. J. Phys. Chem. A 2003, 107, 1413. (82) These rates differ from those reported in ref 40 in that they are renormalized to mass of catalyst rather than mass of Pt in the 0.1 wt % Pt/WO3 catalyst. (83) Cora, F.; Stachiotti, M. G.; Catlow, C. R. A.; Rodriguez, C. O. J. Phys. Chem. B 1997, 101, 3945. (84) Moulzolf, S. C.; Frankel, D. J.; Lad, R. J. ReV. Sci. Instrum. 2002, 73, 2325. (85) Moulzolf, S. C.; Ding, S. A.; Lad, R. J. Sens. Actuators 2001, 77, 375.

J. Phys. Chem. C, Vol. 114, No. 32, 2010 13795 (86) Nirupama, V.; Gunasekhar, K. R.; Sreedhar, B.; Uthanna, S. Curr. Appl. Phys. 2010, 10, 272. (87) Salje, E. K. H. Polarons and bipolarons in WO3-x and YBa2Cu3O7. In Polarons and Bipolarons in High-Tc Superconductors and Related Materials; Salje, E. K. H., Alexandrov, A. S., Liang, W. V., Eds.; Cambridge University Press: Cambridge, 1995. (88) Li, Z.; Chen, Z.-X.; Kang, G.; He, X. J. Mol. Struct.: THEOCHEM 2008, 870, 61. (89) Auroux, A.; Gervasini, A. J. Phys. Chem. 1990, 94, 6371. (90) Kataoka, T.; Dumesic, J. A. J. Catal. 1988, 112, 66. (91) Li, S.; Dixon, D. A. J.Phys. Chem. A. 2006, 110, 6231. (92) Busca, G. The Surface Acidity and Basicity of Solid Oxides and Zeolites. In Metal Oxides: chemistry and applications; Fierro, J. L. G., Ed.; CRC Taylor & Francis: Boca Raton, FL, 2006; pp 247. (93) Murrell, L. L.; Kim, C. J.; Grenoble, D. C. Acid catalyzed hydrocarbon conversion proceses utilizing a catalyst comprising a Group IVB, VB or VIB metal oxide on an inorganic refractory oxide support. Exxon Research & Engineering Co., 1980, patent 4,233,139. (94) Bernholc, J.; Horsley, J. A.; Murrell, L. L.; Sherman, L. G.; Soled, S. J. Phys. Chem. 1987, 91, 1526. (95) Hewett, W.D. J.; Newton, J. H.; Weltner, W. J. J. Phys. Chem. 1975, 79, 2640. (96) Martir, W.; Lunsford, J. H. J. Am. Chem. Soc. 1981, 103, 3728. (97) Jang, Y. H.; Goddard, W. A., III Top. Catal. 2001, 15, 273. (98) Chen, K. D.; Iglesia, E.; Bell, A. T. J. Phys. Chem. B 2001, 105.

JP104421A