Selectivity in Methanol Oxidation as Studied on Model Systems Involving Vanadium Oxides

The experiments have been performed in a number of different ultrahigh vacuum chambers. The setups are described in the literature. The V₂O₅(001) single crystal surfaces were prepared by cleavage in UHV using adhesive tape [21].

The preparation of vanadia clusters on silica follows the recipes described in ref. [8‐11] and the synthesis of the thin vanadia films has been detailed in Ref. [16, 19]. The V₂O₃ films have been either used as prepared or they have been reduced by electron bombardment before exposure to the alcohol. Methanol has been dosed from a capillary doser typically at low temperature (~80 K) before annealing to given temperature.

Calculations are based on spin-DFT and employ a plane-wave basis set as implemented in the Vienna Ab initio Simulation Package (VASP). We use gradient corrected exchange-correlation (GGA) functionals (PW91 or PBE) [22, 23]. The electron–core interaction is described by the projector augmented wave (PAW) method [24]. The clean and defective V₂O₅(001) surfaces are modeled using a fully relaxed two-layer slab and an energy cutoff of 800 eV. We have considered a wide range of defect concentrations (1/6 ≤ Θ ≤ 1). In order to rationalize the scanning tunneling spectra we also perform GGA + U calculations with U = 3 eV [25]. Furthermore, we employ the Monte Carlo method with pair interaction energies between reduced sites fitted to the DFT results to simulate the reduced V₂O₅(001) surface at a given temperature and defect concentration. The network consists of 50 double rows of 2 × 150 sites, which corresponds to an area of 57.75 × 53.7 nm². Ten trajectories, starting from random states, are simulated using the Metropolis algorithm [26] for a varying fraction of reduced sites (0–100%, every 2.5%).

For the study of the oxidation of methanol to formaldehyde on V₂O₃(0001) surfaces we employed a four-layer slab with a (2 × 2) periodicity and a 400 eV cutoff; the (V–O₃–V) bottom layer is kept fixed at its position in the fully relaxed slab representing the clean surface.

To analyze the surface properties of vanadia based catalysts we have studied single crystalline systems as in Ref. [21]. Figure 1a shows a scanning tunneling microscopy image of a V₂O₅(001) single crystal surface revealing the double rows of vanadyl–oxygen atoms protruding out of the surface of the V₂O₅(001) sheets which are held together by van der Waals forces along the (001) direction as expected from the bulk structure schematically represented in Fig. 1b. This surface undergoes a transformation via V₆O₁₃ towards V₂O₃(0001) keeping its single crystallinity over large areas of the macroscopic surface. The final result of a V₂O₃(0001) surface shows in Fig. 1c the typical hexagonal lattice of protrusions of a depolarized corundum type (0001) surface, which is expected to be terminated by half a metal ion layer with exactly the exposed surface unit cell, if the surface structure is directly derived from a cut through the bulk structure in between the puckered vanadium ion layers [16]. This situation is depicted in Fig. 1d. It will become clear further below in this paper that the surface is, in fact, not terminated in this way but rather by a layer of V=O (vanadyl) bonds with the same registry [16].

Before we discuss the evidence for this we present details for the first steps of the processes involved in the transformation from V₂O₅(001) towards V₆O₁₃ and V₂O₃ respectively. Figure 2a and b shows color coded gap maps of a V₂O₅ surface after heating to 350 and 400 K, respectively. To construct these maps we measured scanning tunneling spectra for every pixel, i.e. the derivative of the I–V-curve as a function of the tip voltage. The blue, red and yellow areas correspond to regions of different gap width. The corresponding scanning tunneling spectra are revealed in Fig. 2c. The blue trace has been taken on a fully oxidized V₂O₅(001) surface at 300 K exhibiting a 1.8 eV gap and well ordered structure, while the yellow and red pixels stem from disordered and already reduced parts in which we observe a smaller gap (yellow) and a closed gap (red). Those changes have been modeled using DFT + U and Monte Carlo simulations. The DFT + U calculated density of states (DOS) of the clean V₂O₅(001) surface yields a band gap of ~2.3 eV (Fig. 2f), which is similar to the measured value (~1.8 eV). The formation of isolated vanadyl oxygen defects results in the reduction of the band gap down to ~1.1 eV, whereas the formation of defect pairs along the [010] direction leads to a band gap of ~0.5 eV, and a trench-like structure with missing rows of vanadyl–oxygen atoms exhibits a metallic character. Earlier plain DFT-PW91 calculations of reduced V₂O₅(001) surfaces predicted the facile reduction along the direction of the vanadyl rows [27]. Therefore, we ascribe the red and yellow areas in Fig. 2a and b to randomly distributed isolated vanadyl oxygen defects and to the missing-row phase, respectively. Monte Carlo simulations at 300 K and UHV conditions (10⁻¹⁰ mbar of O₂) for 3% (Fig. 2d) and 7% (Fig. 2e) of reduced sites combined with the DFT + U band gaps result in simulated band gap maps which exhibit a disordered pattern with the same amount of red and yellow areas as the experimental ones. Hence, the anisotropic growth of vanadyl oxygen vacancies is concomitant with an insulator-to-metal transition at the surface.

We now return to the V₂O₃(0001) surface as imaged in Fig. 1c. We have shown earlier that this surface is terminated by a layer of vanadyl groups, increasing the formal oxidation state from V³⁺ in the bulk to V⁵⁺ at the surface [14, 15]. Synchrotron based photoemission studies were employed to prove this. Such a surface is completely unreactive with respect to methanol. Only physisorbed species have been observed and characterized by TDS and spectroscopic techniques. However, if the surface is reduced the situation changes considerably. We note that, different from Cr₂O₃(0001) [28], where we encountered a similar structural situation, the vanadyl groups cannot be reduced thermally without destroying the surface structure [16, 19]. In fact, it is necessary to electron bombard the surface to selectively remove the vanadyl oxygen atoms as shown in Fig. 3 and discussed in a previous paper [16]. Ultimately, all vanadyl oxygen atoms may be removed by this procedure and we are left with vanadium terminated surfaces just as it was indicated in Fig. 1d. This surface is very reactive with respect to a variety of different molecules, including CO, CO₂, H₂O and alcohols [29]. There is no specific selectivity observed under those circumstances. The situation changes when we consider intermediate vanadyl coverages, i.e. situations where vanadyl groups as well as surface vanadium ion sites are exposed simultaneously. The results presented in the following have been obtained on thin V₂O₃ films grown on Au(111). Figure 4a shows an area at 5 K where a few vanadyl oxygen atoms have been removed (dark spots) [30]. Figure 4b shows scanning tunneling spectra taken within an area where full vanadyl coverage is observed as opposed to a position where apparently an individual vanadyl oxygen atom is missing. Obviously, this very small variation in vanadyl oxygen coverage leads to a change in the local electronic structure, i.e. the surface band gap starts to close! Note that the band gap referred to here is the band gap in the surface region which may be different from the bulk band gap.

There are indications that even defects within the material influence the band gap at the surface, and that this band gap varies across the surface [30]. The band gap is believed to be important in oxidation reactions where redox processes involve electron transfer and intermediate storage [4]. It is therefore a working hypothesis to assume that a combination of the ensemble and the local band gap governs the reactivity [4‐6].

In Fig. 5a we plot the reactivity of a V₂O₃(0001) surface at two different degrees of reduction with respect to the formation of various reaction products as measured via TDS after exposing the surface to methanol at ~90 K. The black and the red traces correspond to mass 31 (methanol) and mass 29 (formaldehyde plus a methanol fragment which results from cracking in the mass-spectrometer), respectively. In addition we show water (mass 18), hydrogen (mass 2) and methane (mass 16). Below 200 K the desorption signals are basically determined by desorption of the methanol multilayer. A prominent peak at ~275 K is assigned to the desorption of the methanol monolayer [29], essentially up to 500 K the desorption traces of 29 amu (formaldehyde) and 18 amu (water) follow the methanol signal (31 amu). The water signal at ~250 K which does not follow methanol desorption is probably to be assigned to the reaction of hydroxyl groups resulting from the fission of methanol into methoxy and hydrogen [29]. An interesting point to note is the dependence of the selectivity for formaldehyde formation on the presence of vanadyl groups on the surface: their presence leads to an increased amount of produced formaldehyde and decreases the amount of formed methane indicating that the surface vanadyl groups are an important ingredient for the control of the selectivity of the V₂O₃(0001) surface.

Above 500 K small traces of H₂O are observed and a considerable signal at mass 29 which largely stems from formed formaldehyde as judged by the small intensity of the signal at mass 31. After the onset of the formaldehyde signal we see desorption of methane as well as traces of H₂O and hydrogen. All of the signals above 500 K are absent for a fully vanadyl covered surface (not shown here).

Figure 5b shows a series of traces for mass 29 as a function of the degree of reduction of the V₂O₃(0001) surface. There is a clear maximum of formaldehyde production at a moderate state of reduction corresponding to a 4 mC dose of electrons reducing the surface (full reduction needs about 80 mC). Figure 6a shows STM images of this intermediate reduction stage before and after exposure to methanol and heating to 400 K. The XPS data in Fig. 6b as well as infrared data (Fig. 7 to be discussed below) indicate the presence of methoxy groups on this surface. Figure 6b contains the C1s XPS data for methanol dosed at 100 K and consecutively heated to three different temperatures. The lowest trace is for methanol exposure at 100 K revealing spectra characteristic of methanol. At 550 K the states characteristic of methanol have entirely disappeared and a species with a C1s binding energy of 286.5 eV remains. Indeed, the XPS binding energy of 286.5 eV is typical for methoxy. Figure 7 shows temperature dependent IR spectra after exposure of the partially reduced surface (second trace from bottom) to methanol at 88 K and subsequent stepwise heating to 660 K. In comparison, the spectrum of the fully vanadyl terminated surface is shown. The vanadyl groups had been partially labeled with ¹⁸O isotopes giving rise to the doublet at 998 and 1032 cm⁻¹ respectively. It is interesting to note that this doublet shifts by more than 10 cm⁻¹ to lower energies (982 and 1021 cm⁻¹, see the right panel) upon reduction indicating the influence of the modified electronic structure discussed above in connection with the low temperature STM data. Upon exposure to methanol even at the lowest temperature the vanadyl vibrations disappear and methoxy bands appear in the spectra. This indicates that methanol first loses the hydrogen atom from the hydroxyl group and a V–OH species must be formed.

The vanadyl groups are actively involved in the formaldehyde production as the formaldehyde produced contains both ¹⁶O and ¹⁸O if CH₃ ¹⁶OH is dosed onto the surface that contains vanadyl groups formed from both isotopes as shown in Fig. 8.

It is clear that the mechanism involves both vanadyl and open vanadium surface ion sites, as schematically depicted in Scheme 3, see also Fig. 9. Methanol binds to the vanadium ion sites transferring the hydroxyl hydrogen to a vanadyl group with low activation energy. The methyl hydrogen is then transferred to another vanadyl group or directly to a V–OH group leading to the formation of water. Alternatively, the methyl group could take up a hydrogen atom to form methane.

It is the hydrogen attachment to the vanadyl group that goes hand in hand with a change in the electron count and could be influenced by the ability of the vanadium oxide to accept and release electrons, which in turn is controlled by the vanadium oxide surface band gap.

The mechanism suggested by Scheme 3 is supported by DFT calculations using the PBE functional. The energy for molecular adsorption of methanol on the fully oxidized V₂O₃(0001) surface is only −16 kJ/mol. This step differs substantially from adsorption on silica supported vanadia shown in Scheme 1, for V₂O₃ forms a closely packed lattice and insertion of a methoxy group in a V–O bond is not possible. On the reduced surface with a vanadyl defect concentration of 1/4 we find stronger adsorption with energies of −122 and −193 kJ/mol for non-dissociative (Fig. 9a) and dissociative adsorption (Fig. 9b). The latter involves hydrogen transfer to a neighboring vanadyl group. Hydrogen transfer to a three-fold coordinated surface oxygen atom is much less favorable and results in an adsorption energy of only −103 kJ/mol. The large energy difference underlines the importance of having both reduced and oxidized surface ion sites for dissociative methanol adsorption creating a hydroxyl and a methoxy site, respectively. In a second step, the methoxy group formed is oxidized to formaldehyde by hydrogen transfer to a neighboring vanadyl group (Fig. 9c). This redox step is similar to silica supported vanadia (Scheme 2). For this reaction we find a reaction energy of +141 kJ/mol with respect to the dissociative adsorption state. This reaction is only possible if two adjacent vanadyl sites are available on the surface, one for the initial adsorption and a second one for the oxidation step. For higher defect concentrations (1/2), dissociative adsorption is 20–30 kJ/mol less exothermic. However, the main difference is that after adsorption vanadyl groups are no longer available. Hence, in the redox step, hydrogen has to be transferred to a three-fold coordinated surface oxygen atom, which is far more endothermic (244 kJ/mol with respect to the dissociative adsorption state). Hence, the effect of the defect concentration on these reactions energies is sizable. This result is supported by the TDS data presented in Fig. 5b which show a formaldehyde desorption peak at 550 K which grows at the expense of the state at ~515 K when the degree of surface reduction is increased. In the light of the results of the calculation presented above, the peak at 515 K could be attributed to formaldehyde formed using vanadyl oxygen while the peak at higher temperature might be assigned to formaldehyde formed involving three-fold coordinated surface oxygen atoms.

So far we have discussed the reactivity of a well ordered surface. Realistically the reactions happen on systems where the vanadium oxide is supported on another oxide. We have studied in the past vanadium oxide (VO _x ) deposits on alumina and on silica in order to clarify spectroscopic assignments previously proposed for powder samples using Raman spectroscopy [11]. The typical sample shows an STM image as shown in Fig. 10 where the vanadia is prepared by physical vapor deposition of vanadium in an oxygen ambient onto the surface of a silica film. The morphology of the film is granular indicating three-dimensional growth of the vanadium oxide particles, which have been proven to be mainly V₂O₃ particles exposing vanadyl groups at their surface [8‐11]. In this sense they are similar to the single crystal V₂O₃(0001) surface. The morphology of the vanadium oxide deposits can only be modified within a very narrow range. Attempts have been made to deposit the vanadium within an oxygen ambient into a pre-deposited water (ice) layer [31] on the thin silica film surface. While this procedure does not lead to hydroxylation of the silica, the vanadium oxide particles expose OH and partially V=O groups at 300 K. Upon heating to 500 K, V₂O₃ particles are formed which mainly expose V sites [12]. Note that water assisted preparation slightly smoothes the vanadium oxide layer. When oxidized at 500 K in 10⁻⁶ mbar of O₂, the vibrational properties are very similar to that of the fully oxidized V₂O₃(0001) surface. If such a system, fully vanadyl covered, is exposed to methanol, it is as unreactive as a fully V=O covered V₂O₃(0001) surface. For fully reduced vanadia particles (i.e. before oxidation treatment) we find some amount of formaldehyde in the desorption spectra. However, if the reduced particles prepared using water assisted preparation are exposed to oxygen at 300 K which leads to the partially reduced (or partially oxidized) surface (note that quantifying the state of reduction is even more difficult here than in the case of single crystal surface), the amount of formaldehyde produced is much greater than for fully reduced or fully oxidized particles. Figure 11 shows the TDS traces equivalent to the ones shown for the single crystal in the relevant temperature range. It is clear that the deposited particles behave qualitatively similar, and it is tempting to assume the same mechanistic steps as more clearly underpinned by the single crystal data.