The understanding of H diffusion in materials is pivotal to designing suitable processes. Though a nudged elastic band (NEB)+molecular dynamics (MD)/quantum mechanics (QM) algorithm has been developed to simulate H diffusion in materials by
IntroudctionAlumina exhibits a remarkable series of metastable polymorphs other than the most stable alpha-alumina. Alumina in particular gamma-alumina has been used in a variety of industrial applications, where H-diffusion in nanostructured alumina plays important roles. Gamma-alumina is believed to have a spinel structure. The spinel structure possesses a face-centered cubic sublattice of O atoms. The Al 2 O 3 stoichiometry necessitates vacant cation positions,locations of which has been a subject of intensive research. The H atoms are expected to diffuse through such vacant positions. Johnsson et al. proposed the nudged elastic band (NEB) method [1] to calculate such a diffusion path in an inhomogeneous material. The NEB method is an efficient method for finding the minimum energy path (MEP) between two given states (i.e., before and after the reaction). Relating to the long MEP, a large number of atoms need to be involved in the NEB calculations. However, since accurate evaluation of the energy profile through MEP usually requires an electronic-structure calculation method such as the densityfunctional theory (DFT), treating all the involved atoms with such a compute-intensive method is impractical. Our group then has combined the NEB method with a hybrid MD/QM method recently (called NEB+MD/QM method below) [2,3,4]. In this method, accurate but computationintensive quantum mechanical (QM) simulations are embedded within a classical molecular dynamics (MD) simulation only when and where high fidelity is required.Like other similar methods, the NEB+MD/QM method must take account of a large number of quantum states in the QM computations. It is of course computationally intensive. Though we have parallelized the program and achieved impressive performance on clusters, it cannot yet give results in reasonable time for large scale and complex models in our simulations. We thus decided to harness the power of large numbers of heterogeneous, distributed CPUs provided by a grid infrastructure. The NEB+MD/QM algorithm then has been gridified based on an integrated frame-