Collaborative Research: Petascale Simulations of Quantum Systems by Stochastic Methods

该奖项支持开发、测试和优化新一代模拟代码，用于在千万亿次计算机上计算电子状态。这些代码将基于量子蒙特卡罗方法。这项工作将结合计算机科学的最新进展，如全局阵列、容错并行、图形处理单元和任务动态并行，以及最近的学科进展，可以处理相关的电子系统和研究目前遥不可及的物理效应。目标是建立生产质量的量子蒙特卡罗代码，可以利用轨道1和后续硬件实现科学突破计算。千兆级计算机的成功使用可能非常具有挑战性。将量子蒙特卡罗代码扩大两个数量级以上，将需要寻求算法组织和并行化的新途径；例如，利用节点的多核/共享内存特性。注意容错和负载平衡将导致可靠和高效的千兆级代码。将电子的量子蒙特卡罗方法与离子的经典模拟方法相结合，使几乎任何系统都能进行真实的模拟。将产生新的预测能力和见解，特别是使用电子的完整量子力学描述的动力学和有限温度现象。另一个重点是将量子蒙特卡罗应用于强相关电子系统，这是凝聚态物理中最重要的挑战之一。pi计划实现和测试新的相关波函数，例如带回流的pfaffians，以及新的量子蒙特卡罗算法，例如经典自由度和量子自由度的动态耦合。这些发展将应用于研究水，在没有任何经验或平均场输入的情况下，对所有相关自由度（如非零度温度下的电子和质子）进行完整的量子力学描述。量子蒙特卡罗计算也将用于理解过渡金属化合物，特别是过渡金属氧化物材料在高压下的磁性状态和金属-绝缘体转变。尽管经过几十年的研究，这些系统仍然没有得到充分的理解，并且需要极高的精度来揭示各种多体现象和实验观察到的现象的起源。开发的计算工具将通过开源项目QMCPACK和QWalk提供给科学界。全局数组工具包将提供基于全局地址空间编程模型的高级可伸缩编程环境。这项研究和开发工作还将支持高性能计算领域的研究生和博士后的培训。pi将组织一个暑期学校，“千万亿次量子蒙特卡罗方法”，以培训学生，博士后和研究人员开发的方法，使他们能够在千万亿次及更高的高性能计算领域开辟新的领域。该奖项支持针对最强大的计算机的新一代计算机代码的开发、测试和性能调优。这些新代码将基于量子力学，并创造一种“虚拟显微镜”，可以在原子尺度上探测材料。这些代码将提供原子间力的精确测定，通过对电子及其运动的精确量子力学描述，这些力是这些力的最终来源。由此产生的代码将用于研究水，水可能是生命中最重要的组成部分，但也是最难理解的液体和固体之一。pi代码提供的对电子的精确描述将使涉及电子的材料的最精确的计算研究成为可能，这些电子相互作用强烈，导致物质的新状态。了解这些材料可能会导致物质的新状态的发现，这可能会导致新的材料和设备技术。开发的代码将提供给更广泛的科学家计算社区，使他们能够利用最强大的计算机开辟新的领域。这项研究和开发工作还将支持高性能计算领域的研究生和博士后的培训。pi将组织一所暑期学校，“千万亿次量子蒙特卡罗方法”，以培训学生，博士后和研究人员开发的方法，使他们能够在千万亿次上开辟高性能计算的新领域。

TECHNICAL SUMMARY This award supports developing, testing, and optimizing a new generation of simulation codes for calculating electronic states on petascale computers. The codes will be based on quantum Monte Carlo methods. The effort will combine recent advances in computer science, such as Global Arrays, fault-tolerant parallelism, Graphics Processing Units and task-dynamical parallelism, with recent disciplinary advances that can treat correlated electronic systems and study physical effects currently out of reach. The goal is to establish production-quality quantum Monte Carlo codes that can exploit Track I and follow-on hardware to achieve scientific breakthrough calculations. Successful use of petascale computers will likely be very challenging. Scaling up the quantum Monte Carlo codes by more than two orders of magnitude will require pursuing new avenues for algorithm organization and parallelization; for example, utilizing the multicore/shared memory nature of the nodes. Attention to fault tolerance and load balance will lead to reliable and efficient petascale codes. A key thrust will be to couple quantum Monte Carlo methods for electrons with classical simulation methods for the ions enabling realistic simulations on virtually any system. New predictive capabilities and insights will result, particularly into dynamics and finite temperature phenomena using a full quantum mechanical description of the electrons. Another thrust involves applying quantum Monte Carlo to strongly correlated electronic systems - among the most important challenges in condensed matter physics. The PIs plan to implement and test new correlated wavefunctions such as pfaffians with backflow, and new quantum Monte Carlo algorithms, such as the dynamical coupling of classical and quantum degrees of freedom. These developments will be applied to study water with a full quantum mechanical description of all the relevant degrees of freedom such as electrons and protons at nonzero temperatures without any empirical or mean-field inputs. Quantum Monte Carlo calculations will also be used to understand transition metal compounds, in particular, the magnetic states and metal-insulator transitions at high pressures in transition metal oxide materials. Despite decades of studies these systems remain inadequately understood and require exceedingly high accuracy to reveal the origins of a variety of many-body phenomena and experimentally observed phenomena. The developed computational tools will be available to the scientific community through the open source projects QMCPACK, and QWalk. The Global Arrays toolkit will provide a high-level and scalable programming environment based on the global address space programming model. This research and development effort will also support training of graduate students and postdocs in the area of high performance computing. The PIs will organize a Summer School, "Petaflop Quantum Monte Carlo Methods," to train students, postdocs and researchers in the developed methodologies and enable them to open new frontiers with high performance computing at the petascale and beyond. NONTECHNICAL SUMMARY This award supports developing, testing, and tuning the performance of a new generation of computer codes targeted for the most powerful computers. These new codes will be based on quantum mechanics and create a "virtual microscope" that probe materials at the atomic scale. These codes will provide an accurate determination of the forces between atoms enabled by an accurate quantum mechanical description of the electrons and their motions which are the ultimate sources of these forces. The resulting code will be applied to study water, perhaps the most important component of life but also one of the most challenging liquids and solids to understand. The accurate description of electrons afforded by the PIs codes will enable the most accurate computational study of materials which involve electrons that interact strongly with each other leading to new states of matter. Understanding these materials may lead to the discovery of new states of matter that may lead to new materials and device technologies. The codes that are developed will be made available to the broader computational community of scientists and empower them to utilize the most powerful computers to open new frontiers. This research and development effort will also support training of graduate students and postdocs in the area of high performance computing. The PIs will organize a Summer School, "Petaflop Quantum Monte Carlo Methods," to train students, postdocs and researchers in the developed methodologies and enable them to open new frontiers with high performance computing at the petascale.