CDS&E: Massively Parallel Quantum Dynamics: Computing many accurate quantum states for real molecular applications

德克萨斯理工大学的莱昂内尔“比尔”普瓦里尔得到了化学部化学理论、模型和计算方法项目以及高级网络基础设施办公室的CDS E（计算和数据支持科学与工程）奖的支持。 他和他的研究小组正在开发在下一代大规模并行超级计算机上扩展精确量子动力学分子模拟的方法。分子模拟有望使计算机取代昂贵而耗时的实验室实验。 因此，这些模拟经常用于许多领域（能源，药物设计，纳米材料等）。然而，大多数模拟将组成分子的原子核视为经典粒子，忽略了核运动的量子动力学效应。量子效应可能很重要，排除它们可能会导致不可靠的结果。另一方面，将它们包括在内会带来令人生畏的数字障碍和困难的数学。为了应对这一挑战，Poirier和同事们正在开发世界上第一个大规模并行精确量子动力学代码SwitchIT，它也被设计成易于使用（即使是“非专家”）和易于访问（任何人都可以免费下载该软件包）。因此，SwitchIT可以在许多使用分子模拟的领域中显著提高分子模拟的准确性、可靠性和真实预测能力。 Poirier和他的同事计划将SwitchIT作为开源发布，并通过确保感兴趣的研究人员可以为SwitchIT代码库做出贡献，围绕该语言建立一个用户社区。这使得该项目有了更大的发展空间。高级网络基础设施办公室（Office of Advanced Cyberinfrastructure）的软件集群对此特别感兴趣，该办公室为该奖项提供了共同资助。超级计算技术的最新发展预示着科学计算的巨大飞跃，并有可能显著推进广泛的科学领域。尖端科学越来越依赖于大规模跨学科合作、共享基础设施和大规模并行高性能计算机（HPC）。与此同时，在化学动力学领域，出现了向更大和更多样化的系统的转变，包括纳米材料和生物分子。最后，在数值模拟中结合量子动力学（QD）效应以获得准确结果的需要正变得越来越被认识到，即使对于大型系统（例如，光合作用中的电子传递）。这些趋势的汇合是不可避免的。然而，由于陡峭的学习曲线和/或缺乏访问，相对较少的化学家目前受益于HPC或精确QD。Poirier的研究旨在利用HPC的力量，使精确的量子点计算能够用于比迄今为止实现的更大和更具挑战性的分子应用。这样的计算构成了一个开拓性的进步，在以前的精确量子点的努力，与大规模并行化的一个重要的CDS E功能的方法。该实现还将两个线性代数创新融合在一起：（1）使稀疏迭代线性求解器和特征值求解器能够在1000 - 10000个核之间高效并行化;（2）使用小波将基大小减小到可管理的水平，即使对于具有多达12个原子的精确QD系统也是如此。“专家”QD从业者可以在低级别上利用SwitchIT，将HPC功能引入到他们自己的QD代码中，而SwitchIT为非专家提供易于使用的前端和大量文档。 此外，Poirier和小组成员定期提供指导和培训，以及访问得克萨斯理工大学化学计算集群。Poirier的小组成员本身接受了计算科学和下一代HPC高度相关技能的密集跨学科培训，例如，美国的竞争力。 简而言之，这项NSF奖项扩展了可以用精确量子点研究的系统的复杂性，同时通过网络使更广泛的社区能够执行此类计算。

Lionel "Bill" Poirier of Texas Tech University is supported by a CDS&E (Computational and Data-Enabled Science and Engineering) award from the Chemical Theory, Models, and Computational Methods program in the Chemistry Division, and the Office of Advanced Cyberinfrastructure. He and his research group are developing methods to scale exact quantum dynamical molecular simulations across the next generation of massively parallel supercomputers. Molecular simulations promise to enable computers to replace expensive and time-consuming laboratory experiments. Thus, these simulations used routinely in many areas (energy, drug design, nanomaterials, etc). However, most simulations treat the atomic nuclei that make up the molecules as classical particles and ignore quantum dynamical effects of nuclear motion. Quantum effects can be important, and excluding them may lead to unreliable results. Including them, on the other hand, presents daunting numerical hurdles and difficult mathematics. To meet this challenge, Poirier and coworkers are developing the world's first massively parallel exact quantum dynamics code, SwitchIT, which is also designed to be easy to use (even for "non-experts") and easy to access (anyone may download the package for free). SwitchIT may thus dramatically improve the accuracy, reliability and true predictive power of molecular simulations, in the many areas where these are employed. Poirier and coworkers plan to release the SwitchIT as open source and build a user community around the language by ensuring that interested researchers can contribute to the SwitchIT codebase. This permits a wider growth of the project. This aspect is of special interest to the software cluster in the Office of Advanced Cyberinfrastructure, which has provided co-funding for this award.Recent developments in supercomputing technology promise great leaps forward in scientific computation, with the potential to significantly advance a broad range of scientific fields. Increasingly, cutting edge science has come to rely on large-scale interdisciplinary collaboration, shared infrastructure, and massively parallel high performance computers (HPC). At the same time, within the chemical dynamics field, there has been a shift towards larger and more diverse systems, including nanomaterials and biological molecules. Finally, the need to incorporate quantum dynamical (QD) effects in numerical simulations in order to obtain accurate results is becoming increasingly recognized, even for large systems (e.g., electron transport in photosynthesis). The convergence of these trends is inevitable. Yet, due to steep learning curves and/or lack of access, relatively few chemists currently benefit from either HPC or exact QD. Poirier's research seeks to leverage the power of HPC, to enable exact QD calculations for larger and more challenging molecular applications than heretofore realized. Such calculations constitute a pioneering advance over previous exact QD efforts, with massive parallelization a vital CDS&E feature of the approach taken. The implementation also fuses together two linear algebra innovations that: (1) enable efficient parallelization of sparse iterative linear solvers and eigensolvers across 1000 - 10000 cores; (2) use wavelets to reduce the basis size to a manageable level, even for exact QD systems with as many as 12 atoms. "Expert" QD practitioners can tap into SwitchIT at a low level to introduce HPC capability into their own QD codes, whereas SwitchIT provides non-experts with an easy-to-use front end and extensive documentation. In addition, Poirier and group members regularly provide guidance and training, as well as access to the Texas Tech Chemistry Computation Cluster. Poirier's group members themselves receive intensive interdisciplinary training in computational science and next-generation HPC highly relevant skills, e.g., for American competitiveness. In short, this NSF award extends the complexity of systems that can be studied with exact QD, while simultaneously cyber-enabling a much broader cross section of the community to perform such calculations.