Toolkit-based framework for scalable High Performance Standalone Molecular Dynamics simulations

[摘要] Computational modelling and simulation in materials science uses mathematical abstractionsof particle-particle forces to postulate, develop and understand materials that are organizedas particle systems. Real particle systems occupy macroscopic scales and can be costly tosimulate in terms of hardware and software tools and simulation time. Even the most basicsimulation can generate large amounts of intermediate data that requires innovative furtherprocessing to decipher its underlying physics or to answer fundamental questions about itsmaterial properties. These questions are increasingly being asked due the present furiousacademic and industrial interest in nanosized crystalline lattices. Of particular pertinence arequestions of whether or not the properties of the nanostructures are identical to those of theirmacrostructures. In the light of this the primary focus of this contribution is the developmentof a tool to simulate face-centered cubic (fcc) particle systems on a 'standalone hardwareplatform, and to apply it to a specified particle system. The studied particle systems are torange from nanostructures to macrostructures.This thesis is thematically divided into two main parts. In the first part, comprising thefirst five chapters, we conduct a detailed survey of the current in computation, followed by adefinition of the kinds of systems to which the study is applicable, and then we provide adetailed but not overwhelming description of the tool development, with numerous actualcodes and examples. This part culminates in a working tool, abbreviated VSV. In the secondpart, comprising the subsequent chapters, we apply the VSV and associated tools to solve actual physics problems in nanostructures thereby offering new approaches and results toanswer the current questions.In developing VSV, we discuss the pairwise-particle potential and its integration into anembedded atom model (EAM) approach that is a cost-effective way to simulate fcc metalliclattice systems as a select case that has practical and industrial relevance. To do this, Wechose the Sutton-Chen EAM as being suitable. This was followed by the application of VSVon a single computer as the 'standalone setting, and then on a small, four-computer clusterconsisting of multi-core, multi-processors to test its scaling and parallelization. The testsystem consisted of 30,261 copper atoms in an arbitrary fcc lattice. The various simulationswere then evaluated for performance enhancements in terms of execution speed and ease ofapplication. The large amounts of intermediate data made it necessary to develop smallerextensions to the VSV tool to enable output visualization. These extensions were written inVisual Basic and Matlab. We then apply VSV to simulate the systems at low energies andsuggest novel answers to various questions within the framework mentioned above.A first major, unwittingly observed result in the application of VSV is that it showedthat bond lengths between any two particles appear to develop a temporal oscillation whenperturbed by a nearby displaced atom. These oscillations are seen to propagate throughoutthe lattice and eventually form a standing wave pattern, through which temperature can bemodelled. By applying perturbations in which the bond length oscillation amplitudes areconstrained to small values by deliberately applied perturbations, we found that the useof an elastic, Hooke's Law model results in a faithful reproduction of the known elasticconstants for the copper material on which it was tested. Thus, we suggest and develop aunique impulse and oscillation method that is useful to calculate the elastic properties offcc nanostructures. A second result is the extension of this impulse-oscillation method topostulate a way to initiate wave-like energy transfer through the lattice. These waves areshown to be phonons and by constraining the energy to the first Brillouin zone we show that the temperature behavior of the lattice can be estimated. A third result is that VSV enablesthe computation of the diffusion in a nanosized lattice. Furthermore, we apply standarddiffusion models to a low temperature regime and calculate the diffusion constants of thelattice using standard models. An important result is the indication that diffusion in such acollection of atoms in not driven by Brownian motion, but by the interplay of pair-wise 12-6forces. We also show that the lattice atoms can spontaneously coalesce into a shape that isguided by the global minimum of potential energy. Finally, VSV also shows how growthinto bigger fcc structures can be simulated through atom captures.The foregoing results are compared with the literature values and the good agreementsindicate that it as a reasonable new and cost effective way to investigate many of the propertiesof fcc lattices. Aspects of the research have been published in a book chapter, journals andpresented at international conferences.Finally, we present the concluding remarks about research and suggest further directionsin terms of further development of VSV and its applications to new and novel structures.

[发布机构] University of the Free State