论文部分内容阅读
扩散系数计算量非常大,为提高计算效率研究采用并行计算替代单机计算:用非阻塞通信替代阻塞通信;利用元胞分解的成果创建要拷贝的外层粒子列表;用拷贝粒子和移动粒子的重叠部分来减少通信量,在空间分解(SD)算法的基础上对MD模拟的并行算法进行了改进,使并行计算的效率提高了28.64%。用改进的SD并行计算方法,以RedHat Linux 8.0位操作系统和基于Message Passing Interface(MPI)消息传递界面的四节点计算机集群系统(PC Cluster)。对水分子在宏量条件下以及受限于硅酸岩狭缝中的速度自相关函数进行了MD模拟计算。结果表明,并行算法可以使速度自相关函数及相应的自扩散系数的MD模拟时间大为缩短。
Diffusion coefficient calculation is very large, in order to improve the computational efficiency of the parallel computing alternative stand-alone calculation: non-blocking communication instead of blocking communication; use of the results of cellular decomposition to create a list of outer particles to be copied; with copy particles and the overlap of moving particles In order to reduce the traffic, the MD algorithm was improved based on the Spatial Decomposition (SD) algorithm, which improved the efficiency of parallel computing by 28.64%. With improved SD parallel computing, a four-node PC Cluster with Red Hat Linux 8.0 operating system and Message Passing Interface (MPI) messaging interface. MD simulation was performed on the autocorrelation function of water molecules in macroscopical conditions and in silicate rock slits. The results show that the parallel algorithm can make the velocity autocorrelation function and the corresponding self-diffusion coefficient of the MD simulation time greatly reduced.