Abstract:  High performance computing is a powerful tool to understand composition-structure-property relationships in materials. Density functional theory is mainly used to study the materials from atomistic scale to micrometer scale in the computational simulation of materials science, and is parallelized over k-points, bands and plane wave in many computing softwares of materials science with high parallel efficiency. Paralleling over k-points can be scaled well but has poor performance for big unit cell. Paralleling over bands has better efficiency for small or middle size of unit cell. Paralleling over plane wave has better performance but poor scalability for big size of unit cell, and relies on global communication over whole CPU. Rewriting or redesigning computing softwares of materials science to take full advantage of the latest hardware technology such as accelerator or many-core technology has become a recent trend.

Key words: high performance computing application, materials science, parallel computing

摘要： 高性能计算是研究材料的成分-结构-性质三者之间关系的有力工具。材料科学中的计算模拟主要使用密度泛函理论研究原子到微米尺度的材料，其并行实现方式主要分为并行k点、并行能带和并行平面波，具有较高的并行效率和大量的软件实现。并行k点方式具有较好的扩展性，但不适合于计算大晶胞体系；并行能带方式对于中小晶胞体系效率较高；并行平面波方式适合于大晶胞体系，但对全局通讯的依赖性较高，并行扩展性较差。充分利用最新的硬件技术，如加速卡、众核技术等，改写或重新设计材料科学计算软件已成为最近的发展趋势。

关键词: 高性能计算应用, 材料科学, 并行计算