Light alkenes are of paramount importance in chemical industries as they are raw materials for numerous bulk and fine chemical products. Light alkenes are usually obtained directly from high-temperature pyrolysis of the corresponding alkanes. As these processes require massive energy and the catalysts used are easy to coke, their applications are typically strictly determined by the economic profits between the alkenes and highly purified alkanes, which are in turn affected by the oil price in the global market. In comparison with these conventional nonoxidative routes, the so-called oxidative dehydrogenation (ODH) of alkanes to olefins is a promising alternative approach because this reaction is not confined by the chemical equilibrium in pyrolysis and the oxygen used here can prevent coke formation [1-3]. However, one important problem for this process is the strong oxidation of alkenes, which decreases the productivity of the desired olefin product. Therefore, it is required to design catalysts that increase the selectivity toward dehydrogenation products instead of in particular oxygen-containing molecules or carbon oxides.

Supported vanadium oxides are known to be one of the best catalysts for the ODH reactions of alkanes as they can exhibit a high thermal stability and excellent selectivity [4-12]. As a carrier of the catalyst, reducible CeO₂ has been found to be much better than irreducible SiO₂ and Al₂O₃ [13-18]. It is widely believed that the ODH reaction of alkanes on ceria-supported vanadium catalysts proceeds via the Mars-van Krevelen mechanism and the rate-determining step should be the first C-H bond cleavage [19-22]. Moreover, the presence of highly dispersed vanadium oxides is also crucial to achieve better activity and selectivity [5, 23-27]. Therefore, it is very desirable to understand the structure and corresponding catalytic activity of such ceria-supported vanadia catalysts. Indeed, tremendous efforts have been devoted to resolving their structure-activity relationship. By using high-resolution scanning tunneling microscopy (STM), some studies found that vanadium oxide species were highly dispersed as monomers on the ceria support under very low vanadium loadings, indicating that a strong interaction existed between the vanadia and ceria surfaces [28]. Moreover, subsequent studies highlighted that the ODH activity of VO_x/CeO₂ was affected by the loading of vanadium, with the VO_x monomers yielding much better performances than other larger oligomers, whereas vanadia supported on irreducible oxides, such as SiO₂, exhibited the opposite result [17, 18].

In an early work, Shapovalov et al. [29] studied the atomic structures of supported monomeric VO_x. They performed density functional theory (DFT) calculations to examine the properties of isolated VO_x (x = 1-4) clusters on rutile TiO₂(110) and CeO₂(111) surfaces. They proposed that the redox reactions occurring over the VO_x/CeO₂ catalysts should follow the Mars-van Krevelen mechanism because the oxidized substances take oxygen from VO_x, which is then oxidized by the gas-phase oxygen. Recently, Liu et al. [18] and Wu et al. [30, 31] conducted DFT calculations to illustrate the catalytic activities of supported vanadia/ceria catalysts for the ODH reactions of alcohols. They found that the so-called characteristic new empty localized states (NELS) were crucial for the oxidation of methanol to formaldehyde. In particular, it was proposed that the presence of the NELS in the supported monomeric vanadia/ceria catalysts such as VO₃/CeO₂(111) could drastically increase the hydrogen adsorption energy. Moreover, they are determined to be the most active species for the ODH reactions.

Though the ODH reactions of alcohols over supported vanadia/ceria catalysts have been studied in detail, the mechanism for the ODH of alkanes to olefins is still unclear, particularly regarding the detailed role of the catalytically active VO_x species in the ceria-supported system and its unique performance for this specific catalytic reaction. Therefore, with the aim to gain a deeper insight into the alkane ODH mechanism, in this study, we investigate the conversion of propane to propene mainly on the characteristic VO₃ cluster supported on CeO₂(111) surface. Its catalytic activity and particularly the electronic effect of the NELS on the reaction selectivity are carefully examined and discussed.

Spin-polarized DFT + U calculations were performed with the GGA-PW91 functional by using the Vienna Ab Initio Simulation Package (VASP) [32, 33]. The project-augmented wave (PAW) [34] method was used to describe the electron-core interaction, with Ce (5s, 5p, 6s, 5d, 4f), C and O (2s, 2p), and H (1s) electrons being treated as valence electrons. The on-site Coulomb interaction correction was necessary for the appropriate description of the localized 4f orbitals/electrons and related electronic and geometric properties of ceria. As suggested by Nolan and coworkers [35, 36], in this work, the U value was set as 5 eV. The wavefunctions were expanded in the planewave basis set with a kinetic energy cut-off of 400 eV. The calculated lattice parameter of bulk ceria using the DFT + U method (5.445 Å) was in good agreement with the experimental value (5.411 Å) [37].

The CeO₂(111) surface was modeled using a 12-atomic- layer slab, which represents a p(3 × 3) lateral cell, with the bottom three layers being fixed to the bulk parameters and other layers being allowed to fully relax. To avoid interactions between neighboring slabs, they were separated by a vacuum gap larger than 10 Å. The Brillouin-zone integration was performed with a 2 × 2 × 1 Monkhorst-Pack grid for each surface. Transition states of the surface reactions were located by using the constrained optimization scheme [38]. The adsorption energy (E_ads) was calculated as follows:

where E_total is the calculated total energy of the adsorption system, E_substrate is the energy of the clean substrate, and E_{gas-phase adsorbate} is the energy of the gas-phase molecule.

To understand the mechanism of the ODH reactions of propane, it is important to know and build the structures of the VO_x/CeO₂(111) catalysts. In the recent years, the atomic structures of supported VO_x species have been clarified with the assistance of DFT calculations. The identified most stable configurations of VO_x for given compositions (x = 2-4) have been well illustrated by different groups [29-31]. The calculated phase diagrams at an oxygen pressure of 0.21 atm show that VO₄ is the most stable species below 520 K, whereas VO₂ is the most stable one within a higher temperature range of 520-950 K. At ~520 K, these three different species are in close competition in terms of the stability. Particularly, supported VO₃ has a rather high H adsorption energy [31], indicating the unique role of VO₃ in the dehydrogenation processes. It is also interesting to note the existence of several different types of oxygen atoms on the VO₃/CeO₂(111) catalyst, which may have different chemical activities.

The detailed structure of the supported VO₃/CeO₂(111) catalyst is illustrated in Fig. 1. We can see that there are mainly three types of active oxygen atoms in this configuration, i.e., (1) singly coordinated terminal oxygen O_V=O, which is a vandal oxygen, (2) two-fold coordinated oxygen O_V-O-Ce, which bridges the V and Ce cations, and (3) three-fold coordinated oxygen O_Ce-O-Ce, which is simply a top-surface lattice O besides the adsorbed VO₃. To identify the type of oxygen atom responsible for the propane ODH reaction, we first modeled propane adsorption at these different sites on the VO₃/CeO₂(111) surface.

The molecular adsorption of propane on the supported VO_x in different catalysts has been well investigated in previous studies [39-43]. It is widely believed that similar to methane, propane can only be weakly adsorbed on various catalysts. Therefore, the ODH reaction of propane to propylene may prefer to follow an Eley-Rideal type mechanism and breaking of the first C-H bond of the propane molecule can be considered as the rate-determining step.

In principle, the breaking of the first C-H bond can occur in the methyl (-CH₃) or methylene (-CH₂) group of the propane molecule, so that the corresponding adsorbed products are n-propoxide or i-propoxide species, respectively. It can be learned from the thermodynamic data that the C-H bond energy in the methyl group is higher than that in the methylene group (420 vs. 401 kJ/mol) [42]. In an early theoretical study, Gilardoni et al. [44] by performing DFT calculations also showed that the formation of i-propoxide was both thermodynamically and kinetically favored. Thus, in this work we only focused on the activation of the C-H bond, which leads to the formation i-propoxide.

In Fig. 2, we illustrate the calculated structures of dissociatively adsorbed propane in two possible configurations. As one can see, in the calculated structure shown in Fig. 2(a), i-propoxide is present on the surface O_Ce-O-Ce and a H binds to the nearby O_V-O-Ce of the supported VO₃. In comparison, in the calculated structure shown in Fig. 2(b), the surface sites of these two species are interchanged. The adsorption energies corresponding to these two dissociative adsorption configurations were determined to be 2.78 and 3.08 eV, respectively. The calculated adsorption energies indicate that the dissociative adsorption structure with the C₃H₇ moiety located at O_V-O-Ce and H at a nearby surface O_Ce-O-Ce is energetically favorable to form. From the side views of the two adsorption structures it is inferred that more favorable H bonding between H-O_Ce-O-Ce and O_V-O-Ce in the structure depicted in Fig. 2(b) compared with that between H-O_V-O-Ce and O_Ce-O-Ce in the structure presented in Fig. 2(a) may account for the difference in their stabilities. Moreover, we understand from the work of Wu et al. [30] that H adsorption at O_Ce-O-Ce is also slightly stronger than that at O_V-O-Ce.

In Fig. 3, we illustrate the calculated energy profiles for the formation of the two dissociative adsorption structures discussed above. It can be seen that these two dissociative adsorption processes are quite similar. They are both initiated by a weak propane adsorption (IS, E_a = 0.18 eV), following which the C-H bond breaking occurs in the methylene group (TS1 and TS2), resulting in an adsorbed H and a C₃H₇ radical in the gas phase (IM1 and IM2). The C₃H₇ radical then binds with different O atoms in the final state (FS1 and FS2). Significantly, it can be easily noticed that the calculated barrier for such C-H bond breaking is 0.26 eV in the first process (Fig. 3(a)), whereas it is as low as 0.12 eV in the second process (Fig. 3(b)).

The above results show that in agreement with the Bronsted-Evans-Polanyi relationship, both the dissociative adsorption energy and activation energy for the evolution of the propane molecule in path B are better than those in path A. For the breaking of the first C-H bond of the propane molecule at the interface of VO₃ and CeO₂(111), the H is preferably left at the surface oxygen of the CeO₂(111) support close to VO₃ and i-propoxide is formed at the interfacial oxygen of the supported VO₃. In addition, we also believe that the high catalytic activity of the VO₃/CeO₂(111) catalyst toward the first C-H breaking process in the propane ODH reaction is mainly owing to its unique electronic effect, which strongly favors the capture of a H atom, as we have discussed in a recent work [30]. In particular, the NELS being largely constituted of O 2p orbitals of the supported VO₃ are strongly involved in this first C-H bond breaking process similar to the manner in the oxidation of methanol to formaldehyde [30].

Similar to the complete propane ODH reaction, the second C-H bond is also needed to be broken in the adsorbed C₃H₇ to obtain the target product of propylene. In Fig. 4, we present the calculated reaction profiles for the further C-H bond activation of the dissociatively adsorbed propane. As one can see from the calculated structures, to break the second C-H bond, the i-propoxide at the supported VO₃ leaves one H atom of a methyl group to a nearby free O_V-O-Ce. From the calculated results, one can clearly find that the activation energy for the breaking this second C-H bond (1.81 eV) is much higher than that for the first one (0.12 eV). As a direct product, a C₃H₆ species is formed on the supported VO₃ (IM3). We then calculated and compared the energetics of the two possible pathways for the further evolution of this intermediate state. As one can see from Fig. 4, C₃H₆ desorption as a propene molecule is much more favorable than the further breaking of the third C-H bond on the surface.

In Table 1, we list the calculated barriers for the first and second C-H bond breaking processes discussed above. We also list the calculated results of the same processes for propane and ethane on V₂O₅(001) reported in previous studies by Fan and co-workers [45, 46]. As it can be seen from the table, the calculated reaction barriers for both the first and second C-H bond breaking processes on the clean V₂O₅(001) are quite high. More significantly, the first C-H bond typically appears to be more difficult to break than the second one for both propane and ethane; this is in agreement with the regular order of the bond strengths. Most interestingly, from our calculations, we can find that on the supported VO₃/CeO₂(111) system, the first C-H bond breaking step is actually very easy to occur, which has an activation barrier of only 0.12 eV. In contrast, the second C-H bond breaking step is relatively significantly difficult with a barrier of 1.81 eV. To better understand the reason for the reversed activities toward C-H bond activation in alkanes, we conducted further electronic analyses for this interesting catalyst.

The partial electronic density of states (PDOS) for the five oxygen atoms constituting the NELS at VO₃/CeO₂(111) were calculated (Fig. 5). It can be clearly seen from Fig. 5(a) that for the system with weakly adsorbed C₃H₈ on the clean VO₃/CeO₂(111) surface, there is a high sharp peak above the Fermi level. As we have explained in a recent study [30], this corresponds to the empty states constituted mainly by the (empty) 2p orbitals of the O ions of the supported VO₃ as they fail to take enough electrons from the vanadium atom, which has five valence electrons only. Such unique NELS are located below the unoccupied 4f states of the ceria support in energy, which are then obviously more powerful in capturing electrons compared with the localized 4f states of the surface Ce cations. This is the simple reason for the very low activation barrier for the first C-H bond breaking process in the calculated propane ODH reaction. In comparison, such unique NELS are crucial for the H abstraction from propane, and in addition, we find that the NELS disappear immediately after the first C-H bond breaks on the VO₃/CeO₂(111) surface. As it can be obviously seen from Fig. 5(b) and (c), the NELS practically vanish when the first H atom adsorbs on the O atom of CeO₂ and C₃H₇ radical remains in the gas phase, whereas there are nearly no NELS when the C₃H₇ radical adsorbs on an O of VO₃. Therefore, these results clearly indicate that without the assistance of NELS, the breaking of the second C-H or any additional C-H bonds would be very difficult.

In fact, we also calculated the spin-polarized charge density (Fig. 6) of the important structures for the VO₃/CeO₂(111) catalyst during the propane ODH reaction. It is well recognized that the unique NELS practically vanish when the first H atom adsorbs on the O atom of CeO₂. Consistent with the radical nature of the C₃H₇ species in IM2, we determined the existence of a spin electron at this species (Fig. 6(a)). Interestingly, we then found that after the radical species binds to O_V-O-Ce, this spin electron transfers to the surface and localizes at one Ce cation. Thus, owing to the disappearance of the NELS, the 4f states capture the spin electron, converting the Ce cation to Ce³⁺. Furthermore, for the system with co-adsorbed C₃H₆ and two H, which is a result of the second C-H bond breaking process (FS2, Fig. 4), we can observe that two spin electrons are localized at two Ce cations and one spin electron remains in the double bond of C₃H₆ (Fig. 6(c)). These results clearly indicate that the Ce 4f orbitals would be the only empty states to accommodate the transferred electrons from H and other moieties during the further dissociation of C₃H₇, and the obvious repulsions between the crowded localized 4f electrons may strongly inhibit further dissociation of other adsorbents like C₃H₆.

Finally, it needs to be mentioned that as another important VO_x species on CeO₂(111), the VO₄ cluster can form via O₂ adsorption on VO₂, which forms after the VO₃ species results in water formation and desorption. Moreover, the VO₄/CeO₂(111) catalyst contains more NELS than VO₃, which can capture altogether three extra electrons. Therefore, such NELS of VO₄ may be effective in the two successive C-H bond breaking processes of propane, and in particular, they should be able to significantly reduce the barrier for the breaking of the C-H bond in the adsorbed C₃H₇. Nevertheless, the further dissociation of C₃H₆ would still be inhibited without their assistance.

In summary, we present a comprehensive theoretical study of the mechanism for the oxidative dehydrogenation reaction of propane over VO₃ supported on CeO₂(111). We propose that the extraordinary catalytic performance of VO₃/CeO₂(111) for breaking the first C-H bond of propane is mainly owing to the unique new empty localized states, which can capture one electron easily. After the disappearance of the NELS, it is difficult for the VO₃/CeO₂(111) catalyst to break the second C-H bond of propane because the less favorable empty 4f states of Ce cations become a reservoir for the other electron. Therefore, this work illustrates that the varied abilities of the involved NELS and Ce-4f in the different C-H bond breaking processes contribute to both the specific activity and selectivity of supported VO_x/CeO₂ catalysts for the ODH reactions of alkanes.