项目名称: 重金属材料第一原理模拟的并行算法研究
项目编号: No.61300012
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 高兴誉
作者单位: 北京应用物理与计算数学研究所
项目金额: 23万元
中文摘要: 基于密度泛函理论的第一原理方法是当今材料科学不可或缺的理论研究手段,Kohn--Sham方程特征值解法器是其中的计算核心。重金属材料第一原理模拟的计算量可达E级,对Kohn--Sham方程解法器提出了苛刻的性能要求。平面波方法是求解Kohn--Sham方程最成熟的数值方法之一。然而,一般的平面波方法难以扩展到数千处理器核,不能满足重金属材料高精度数值模拟的要求,亟需发展新方法、新技术改变现状。本项目面向数万核上的重金属材料第一原理模拟,从权衡负载平衡与通信开销的角度研究具有良好并行扩展能力的平面波算法,并结合多特征向量并行迭代发展Kohn--Sham方程的二级并行算法,研制高效并行的Kohn--Sham方程解法器。在此基础上,本项目瞄准若干典型重金属的物性研究,支撑上千个重金属原子的第一原分子动力学模拟扩展到数万核。
中文关键词: 重金属;第一性原理;Kohn-Sham方程;平面波;特征值和特征向量
英文摘要: Nowadays, the first-principles method based on density funcitional theory (DFT) is indispensable to the theoretical research in material sciences. Its computational kernel is the eigensolver for the Kohn--Sham equation . A high-performance Kohn--Sham equation solver is critically required by the first-principles simulations of high-Z metals since the whole compuational cost would probably grow to the exascale. The planewave method is most popular for solving the Kohn--Sham equation. But it is difficult for the general schemes to achieve a good scalability on thousands of cores. And these methods can hardly accommodate the high-accuracy simulations of high-Z metal meterials. Thus it is desirable to make some progress by new parallel algorithms and implementation techniques. In this program, our work is oriented to simulating high-Z metals on tens of thousands of cores. To this end, we will redesign the parallel planewave method in the perspective of the trade-off between load balancing and communication cost. Moreover, combining the distribution over planewaves with bands, we will develop a two-level parallel iteration for the Kohn--Sham equation. These algorithms are used to make a high-performance Kohn--Sham equation solver. Driven by the property study of several typical high-Z metals, we will apply the solver
英文关键词: high-Z metal;first-principles;Kohn-Sham equation;plane wave;eigenvalue and eigenvector