面向量子力学计算的超级计算机探索性数据科学
Exploratory Data Science on Supercomputers for Quantum Mechanical Calculations
AI总结:
针对文学化编程在高性能计算环境下应用于密度泛函理论计算的障碍,研究提出PyBigDFT与remotemanager两个Python库,实现超级计算机上DFT计算的透明访问,支撑快速原型开发与数据探索。
摘要:
文学化编程——将程序代码与自然语言叙述融为一体——已成为数据科学领域的一种普遍方法。这种方法在密度泛函理论(Density Functional Theory,DFT)计算领域也颇具吸引力,尤其适用于交互式开发新方法与工作流。然而,老旧的编程范式以及使用高性能计算(High Performance Computing,HPC)资源的相关难题,阻碍了文学化编程的有效应用。本文提出两个旨在消除这些障碍的Python库。首先介绍PyBigDFT库,该库可用于构建材料或分子体系,并提供对基于小波的BigDFT代码的高层访问接口。随后介绍相关的remotemanager库,它能够对任意Python函数进行序列化,并在远程超级计算机上执行。本文展示了这两个库如何协同实现对基于HPC的DFT计算的透明访问,且可作为快速原型开发与数据探索的基础组件。
英文摘要:
Literate programming - the bringing together of program code and natural language narratives - has become a ubiquitous approach in the realm of data science. This methodology is appealing as well for the domain of Density Functional Theory (DFT) calculations, particularly for interactively developing new methodologies and workflows. However, effective use of literate programming is hampered by old programming paradigms and the difficulties associated with using High Performance Computing (HPC) resources. Here we present two Python libraries that aim to remove these hurdles. First, we describe the PyBigDFT library, which can be used to setup materials or molecular systems and provides high-level access to the wavelet based BigDFT code. We then present the related remotemanager library, which is able to serialize and execute arbitrary Python functions on remote supercomputers. We show how together these libraries enable transparent access to HPC based DFT calculations and can serve as building blocks for rapid prototyping and data exploration.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。