DFT+U modelling of a Mo-containing oxide catalyst

Trabajo presentado en Chemistry and Molecular Sciences and Technologies COST Action CM1104, celebrado en Viena (Austria) del 18 al 19 de abril de 2013.Complex mixed oxide catalysts are frequently used today for selective hydrocarbon oxidation processes. A system receiving attention lately is that constituted by the Mo-V-Nb-Te oxide system, in which the so-called M1 phase has shown interesting performances for the selective ammoxidation of propane and dehydrogenation of ethane [1]. The layered structure of this phase [2] is formed by a system of 5-, 6- and 7-member rings of MO6 octahedra (M=Mo,V) containing (-Te-O-)n and (-Nb-O-)n chains in tunnels formed by the 6- and 5- member rings respectively. The unit cell (neglecting the disorder at the M sites) has an overall stoichiometry similar to (MO3)36(TeO)4(NbO)4, i.e. 160 atoms in total. Furthermore all M sites have an occupancy by V of at most ~0.5, suggesting that all V atoms alternate with Mo along the M=O…M chains perpendicular to the layers, which would need double cells with 320 atoms for a correct modelling. Careful topological analysis has allowed to identify in this lab a halfsize model (Fig. 1) which is equivalent enough to the M1 phase structure in all relevant aspects and makes more tractable its DFT modelling. The latter has been thus undertaken. A typical Mo30V6Te4Nb4O116 stoichiometry (160 atoms) is adopted for modelling the twolayer cell. Assuming the redox states V4+, Te4+ and Nb5+ (experimentally verified), 8 excess electrons must be distributed over 30 Mo atoms, giving the average redox state Mo+5.733. The system is indeed akin to a Mo bronze. The question arises, whether these electrons are localized. It is known that spurious self-interaction in DFT may lead to unphysical delocalization. This problem has been much studied for systems containing 3d or 4f partially filled shells, but much less for partially filled 4d shells. On the other hand recent work [3] claims that the experimental value of the energetics of the reduction of MoO3 to MoO2 is reproduced with a DFT+U treatment, not with just DFT. Thus a first principles determination of U for Mo has been undertaken here, following Cococcioni’s method [4] (as implemented in program Quantum-Espresso) and using a simple ideal NaMo2O6 structure model derived from the high temperature tetragonal WO3 lattice. The results indicate a value of U≈4 eV (ca. half that given in [2]). Its suitability is tested on other Mo bronzes of experimentally known structure and average Mo redox state between 5.0 and 6.0 which are known to display semiconducting or metallic behaviour and charge density wave effects at low temperature. With this Mo U value a modelling of the (Mo,V,Nb,Te)Ox catalyst using the said 160-atom two-layer cell has been then started using program VASP. It shows that most, but not necessarily all, of the excess electrons are localized on specific Mo ions, but only if the DFT+U approach is used. A study on the preferred location of V in the several M sites (a relevant issue [5]) is under way; its progress will be reportedPeer Reviewe