Development of Parallel Car-Parrinello Code and Application to Simulations of Liquid Silicon and Aluminum

Hannes Jonsson得到了化学部理论和计算化学项目、材料研究部门的材料理论项目和高级科学计算部门的新技术项目的资助，以开发可扩展的、可移植的计算算法，用于在MIMD计算机上执行CAR-Parrinello(CP)模拟。这笔赠款是材料研究部的真实材料计算方法倡议和高性能计算和通信倡议的一部分。Jonnson将优化用于大规模并行计算机的CP代码，并使用它来进行硅和铝散体液体以及固液界面的短程有序的模拟研究。为了与这些第一原理模拟进行比较，Jonnson还将使用经验势函数进行蒙特卡罗和分子动力学模拟，特别是对硅的Stillinger-Weber势和对铝的嵌入原子方法势。%分子水平上对原子和分子相互作用的理解在新材料的合理设计中至关重要。通过使用大规模并行计算机，这个问题的理论和计算方法将得到显著改进。CAR-Parrinello方法显示出作为理解这种相互作用的第一原理方法的巨大前景，但它的应用目前受到向量处理器速度的限制。Jonnson建议将CAR-Parrinello算法应用于大规模并行计算机，这将显著增强其对更复杂材料系统的适用性。

Hannes Jonsson is supported by a grant from the Theoretical and Computational Chemistry Program in the Chemistry Division, the Materials Theory Program in the Division of Materials Research, and the New Technologies Program in the Division of Advanced Scientific Computing to develop scalable, portable computational algorithms for performing Car-Parrinello (CP) simulations on MIMD computers. This grant is part of the Computational Approaches to Real Materials initiative in Division of Materials Research and the High Performance Computing and Communications Initiative. Jonnson will optimize the CP code for massively parallel computers, and use it to perform simulation studies of short range order in Silicon and Aluminum bulk liquids and solid-liquid interfaces. For comparison with these first principles simulations, Jonnson will also carry out Monte Carlo and Molecular Dynamics Simulations using empirical potential functions, in particular the Stillinger- Weber potential for Silicon and and Embedded Atom Method potential for Aluminum. %%% A molecular level understanding of the atomic and molecular interactions is of key importance in the rational design of new materials. Theoretical and computational approaches to this problem will be significantly enhanced through the use of massively parallel computers. The Car-Parrinello methodology shows great promise as a first principles approach to understanding such interactions, but its applications are currently limited by the speed of vector processors. Jonnson proposes to adapt the Car-Parrinello algorithms for massively parallel computers which should significantly enhance their applicability to more complex materials systems.