Enabling Petascale Applications in the Chemical Sciences

Mark S. Gordon、Monica H. Lamm、Masha Sosonkina和Theresa L. Windus通过PetaApps招标得到了网络基础设施办公室的支持，以开发可有效扩展到千万亿级计算机架构的软件。本提案的共同经费由数学和物理科学理事会化学司和多学科活动办公室以及计算机和信息科学与工程理事会提供。本研究的重点是使高质量的电子结构理论和多尺度计算方法在千万亿级计算机系统上有效地运行。化学科学中的许多重要问题对计算的要求如此之高，以至于目前解决这些问题的唯一选择是采用可靠性值得怀疑的低级方法。电子结构界已经建立了很好的协议，通过采用越来越复杂的方法来评估应用于中小型系统的方法的可靠性，这些方法通常被认为是可靠的，并研究结果的收敛性。这样的方案还不能用于非常大的系统，在本建议的三个例子中，水，气溶胶和树突状聚合物-配体结合。研究的主要动机是确保可靠水平的电子结构理论（例如，多体摄动理论和耦合簇理论）可以实际应用于解决具有挑战性的国家利益问题。结合新颖的多尺度方法和计算机和计算科学的新发展，本研究正在扩展化学系统的尺寸，可以用精确的理论方法来解决。代码的开发为计算化学社区提供了一种利用大规模并行计算机体系结构来使用高度精确的方法解决重要问题的方法。在这个社区中，推动千万亿次计算需求的是一系列巨大的问题，从具有无可置疑的准确性的液体模拟到凝聚态反应机制的研究，再到聚合物、生物分子和其他扩展系统的基态和激发态研究。两个主要的电子结构代码正在被修改以在千万亿次架构上运行，它们是GAMESS和NWChem。这两个代码实现了广泛可扩展的电子结构算法，并免费提供给所有用户。这项工作正在产生更广泛的影响，使这项研究所支持的所有开发都可以免费提供给一般社区，并可供其他代码开发人员使用。新的课程正在设计中，现有的课程正在修改中，以教育学生（本科生和研究生）和教师关于将代码从少量处理器扩展到数千个处理器或更多处理器时所产生的关键注意事项。

Mark S. Gordon, Monica H. Lamm, Masha Sosonkina, and Theresa L. Windus are supported by the Office of Cyberinfrastructure through the PetaApps solicitation to develop software that will scale efficiently to petascale computer architectures. Co-funding for this proposal is provided by the Division of Chemistry and the Office of Multidisciplinary Activities in the Mathematical and Physical Sciences Directorate, and by the Computer and Information Science and Engineering Directorate. The focus of this research is on enabling high quality electronic structure theory and multi-scale computational methods to run efficiently on petascale-class computer systems. Many important problems in the chemical sciences are so computationally demanding that the only current option for addressing such problems is to employ low-level methods whose reliability is suspect. The electronic structure community has well established protocols for assessing the reliability of methods applied to small and medium systems by employing increasingly sophisticated methods that are generally accepted to be reliable, and studying convergence of results. Such protocols have not been available for the very large systems that are typified by the three examples in the present proposal, water, aerosols, and dendrimer-ligand binding. A primary motivation for the research is to ensure that reliable levels of electronic structure theory (e.g., many body perturbation theory and coupled cluster theory) can realistically be employed to solve challenging problems of national interest. Combined with novel multi-scale methods and new developments in computer and computational science, this research is extending the sizes of chemical systems that can be addressed with accurate theoretical methods. The code developments are providing the computational chemistry community with a way forward to make use of massively parallel computer architectures to solve important problems using highly accurate methods. Driving the need for petascale computing in this community are a vast array of problems, ranging from simulations of liquids with unquestionable accuracy to studies of reaction mechanisms in the condensed phase to ground and excited state studies of polymers, biomolecules, and other extended systems. The two primary electronic structure codes that are being modified to run on petascale architectures are GAMESS and NWChem. These two codes implement broadly scalable electronic structure algorithms, and are made available to all users at no cost. This work is having a broader impact in making all developments enabled by this research freely available to the general community and accessible to other code developers. New courses are being designed and existing courses are being modified to educate students (undergraduate and graduate) and faculty regarding the crucial considerations that arise when scaling codes from a small number of processors to thousands of processors or more.