Daniel L. Roach

Parallel Computational Modelling of Inelastic Neutron Scattering in Multi-node and Multi-core Architectures

This paper examines the initial parallel implementation of SCATTER, a computationally intensive inelastic neutron scattering routine with polycrystalline averaging capability, for the General Utility Lattice Program (GULP). Of particular importance to structural investigation on the atomic scale, this work identifies the computational features of SCATTER relevant to a parallel implementation and presents initial results from performance tests on multi-core and multi-node environments. Our initial approach exhibits near-linear scalability up to 256 MPI processes for a significant model.

Scatter: A New Inelastic Neutron Scattering Simulation Subroutine for GULP

The General Utility Lattice Program [1] (or GULP, to its friends) is a well-established and popular simulation program combining the functionality of both lattice and molecular dynamics. GULPs functionality is broad, with intrinsic functions enabling the materials researcher to model the lattice and molecular dynamics of a large range of semi-empirical potential models as well as structure optimizations, defect calculations, and structure/potential fitting using standard or genetic algorithm-based approaches. Further functionality has been added recently, with a collaboration between the University of Salfords Institute for Materials Research (Salford, U.K.) and Curtin University of Technologys Nanochemistry Research Institute (Perth, Western Australia), in the development of a new inelastic neutron scattering subroutine called “scatter,” which calculates inelastic (both coherent and incoherent) S(Q,ω) data sets for lattice models within GULP, which can be averaged over crystal orientation (the polycrystalline average). These data sets can then be used to assist in the analysis of experimental data on a wide range of neutron inelastic instruments, especially when studying polycrystalline materials.

Inelastic and Quasi-Elastic Neutron Scattering

This chapter describes the basic principles of inelastic and quasi-elastic neutron scattering as applied to hydrogen storage systems—in particular to provide an understanding of the vibrations and diffusion of hydrogen and deuterium in the host lattice. The techniques are then illustrated using typical hydrogen storage materials. Incoherent inelastic scattering can be applied to isolated hydrogens—where the protons can be modelled as in an isolated potential well formed by the surrounding atoms. At higher concentrations, the effect of H–H interactions and the role of hydrogen vibrational density of states are described. Ab initio theory becomes important in this case. The advantages of modelling the dynamics of a deuteride by simulation of the polycrystalline coherent inelastic neutron scattering in comparison with ab initio modelling are then described. The final area of application of inelastic scattering is to the case of adsorbed H2 molecules where particular spin transitions are observed. Here the results provide important information on the geometry of the potential energy surface around the adsorbing site. Quasi-elastic neutron scattering is then described. In particular the Chudley–Elliott model is derived for a Bravais lattice and it is indicated how this approach can be extended to more general cases where there are multiple sublattices which may have differing energies of adsorption. Here the important case of intermetallic Laves phases is described.

GPU Acceleration for Hermitian Eigensystems

As a recurrent problem in numerical analysis and computational science, eigenvector and eigenvalue determination usually employs high-performance linear algebra libraries. This paper explores the implementation of high-performance routines for the solution of multiple large Hermitian eigenvector and eigenvalue systems on a Graphics Processing Unit (GPU). We report a performance increase of up to two orders of magnitude over the original \(\textsc{Eispack} {}\) routines with a NVIDIA Tesla C2050 GPU, providing an effective order of magnitude increase in unit cell size or simulated resolution for Inelastic Neutron Scattering (INS) modelling from atomistic simulations.

The interpretation of polycrystalline coherent inelastic neutron scattering from aluminium

Presented here is a method by which Q-dependent dispersive dynamics may be measured (and used to generate a lattice dynamical model) for polycrystalline samples. This method is analogous to the measurement of dispersion curves for single-crystal samples.

Computational modelling and simulation of polycrystalline coherent inelastic neutron scattering

Inelastic neutron scattering is an experimental technique widely used to investigate the vibrational characteristics of materials in condensed matter research. While coherent inelastic neutron scattering is typically restricted to single-crystal samples, analysis of the complex datasets obtained on polycrystalline samples remains challenging, even for the simplest of structures. However, given the common availability of high performance computing platforms, it is becoming feasible to apply computationally intensive calculation methods that sample millions of q-points to the simulation of polycrystalline models of increasing complexity, a technique referred to as poly-CINS analysis. This approach allows the interpretation of experimental results by comparison and fitting against theoretical models. This paper describes a new high-performance implementation of the Scatter code, a modelling package developed for the General Utility Lattice Program (Gulp) for heterogenous CPU-GPU computing architectures. It provides the ability to generate theoretical poly-CINS data sets from semi-empirical and ab-initio models. We present the computational framework behind its implementation, applying an example of a semi-empirical model for the dynamics of a large unit-cell system, namely the two (low and ambient temperature) phases of solid C60 to illustrate the methodology and its scalability characteristics.