多层次-跨尺度物理中并行DVM-DAC算法

A Study of Parallel DVM-DAC Algorithm in Multiscale Physics

多尺度现象及相关理论方法是复杂物质系统研究中重要的科学问题.单一的量子力学或分子动力学方法无法解释多尺度体系中存在的现象.第一原理离散变分线性标度(DVM-DAC)算法是一种有效的大尺度体系计算方法.它采用分而治之方案,获得了O(n)的计算复杂性.但由于需要求解大量的特征方程,实现中存在严重的计算瓶颈.发展了一种并行DVM-DAC算法并付诸实现,有效地解决了原有算法的计算瓶颈问题.测试结果表明,并行DVM-DAC算法具有很好的可扩展性,并成功完成104碳纳米管原子体系的计算,为多尺度体系研究提供重要工具.

The phenomenon of multiscale and the corresponding theoretical model is an important scientific problem in the complex system.It is difficult for the traditional theory to describe the nature of spanning scale and multilevel problem.Being an efficient model for large scale system,the DVM-DAC algorithm has the O（n） linear scaling character originated from using the divide-and-conquer theory.However,the solving of eigenvalue equations brings serious computing bottleneck.In this paper,a parallel DVM-DAC method is presented to solve this problem.The performed results demonstrate that the parallel DVM-DAC algorithm which has succeeded in processing the carbon nanotube system with 104 atoms,and has good scalability.It can be predicted that the parallel DVM-DAC method is a important tool for the study of multiscale physics.