PolyPal: A parallel microscale virtual specimen generator

We present an open source program, PolyPal, that can generate a polycrystalline virtual specimen in the micrometer scale for atomistic calculations and visualization. Unlike regular meshes or perfect lattices, atomic positions in polycrystalline materials need to be defined before calculations, and the capability of an atom-generation code is evaluated by the maximum size of the virtual specimen it can generate as well as by the efficiency of the necessary input-output process. Present atom-generation codes are implemented in a serial fashion, and the maximum size of the virtual specimen is limited by the on-board memory. Furthermore, it is difficult to handle a single position file with billions of atoms not only because it takes a long time to read in a row but also full domain decomposition takes hours. PolyPal addresses these challenges with a fully parallelized MPI input-output scheme that supports multiple export options on a Linux cluster. It has no limit in the system size with virtually perfect scalability. Additionally by controlling the size distribution and homogeneity of grains, the program can simulate different microstructures, as typically found in the bulk system or in thin-film samples, prepared with different fabrication processes. PolyPal will harness molecular dynamics codes in the coming age of the exascale computing.

我们开发了一款开源程序PolyPal，可生成微米尺度的多晶虚拟样本，用于原子级计算与可视化。与常规网格或完美晶格不同，多晶材料的原子位置需在计算前预先定义，原子生成代码的性能可通过其可生成的虚拟样本最大尺寸，以及必要的输入输出（Input-Output）流程效率进行评估。当前的原子生成代码均采用串行实现方式，虚拟样本的最大尺寸受限于节点板载内存。此外，处理包含数十亿原子的单原子位置文件极具挑战：不仅串行读取会耗费大量时间，全域分解也需耗时数小时。PolyPal通过一套完全并行化的消息传递接口（MPI，Message Passing Interface）输入输出方案解决了上述难题，该方案支持Linux集群上的多种导出格式。该程序不受系统尺寸限制，且具备近乎完美的可扩展性。此外，通过调控晶粒的尺寸分布与均匀性，该程序可模拟不同制备工艺下块体系统或薄膜样品中常见的各类微观结构。在即将到来的百亿亿次计算时代，PolyPal将助力分子动力学（Molecular Dynamics, MD）代码的高效运行。