SPX: Collaborative Research: Asynchronous, Parallel-Adaptive Solution of Extreme Multiscale Problems in Seismology

伊利诺伊大学和田纳西大学的科学家之间的这一合作项目将开发并行软件，用于在艾级超级计算机上进行有效的地震模拟。目前，最先进的系统可以在高达1 Hz的频率下解析地震响应，但设计工程师需要高达10 Hz的分辨率。同步障碍可能会限制exascale系统的性能。该项目将开发可扩展的、无障碍的异步模拟工具和时空自适应网格，以满足艾级平台上的地震分辨率要求。它将开发改进的断层泥物理模型，并扩展异步双曲线求解器，以解决椭圆（最终抛物线）系统。这些扩展将使第一个区域，全周期地震模拟，涵盖快速地震事件和地震之间慢得多的地壳运动，以及在大多数基于PDE的科学和工程应用中使用异步艾级解算器。异步解决方案技术将支持更可靠的地震灾害地图，以及更安全、更经济的抗震建筑和基础设施的设计。鉴于其广泛的适用性，这项研究所提供的前所未有的模拟能力可能会在商业和国防领域引发许多突破。4名研究生研究助理将接受跨学科培训，本科生将通过国家超级计算应用中心参与。这是伊利诺斯大学和田纳西大学科学家的合作项目，将开发并行软件，用于在百万亿次超级计算机上进行有效的地震模拟。目前，最先进的系统可以在高达1 Hz的频率下解析地震响应，但设计工程师需要高达10 Hz的分辨率。跨子域的同步障碍和负载平衡可能会限制exascale系统的性能。该项目将用可扩展的、无障碍的异步求解器和无DDM的时空自适应网格化来取代标准的块同步并行模型和区域分解方法（DDM），以满足艾尺度平台上的地震分辨率要求。它将为断层泥物理学开发剪切过渡带模型，并使用伪时间方法来扩展异步双曲线求解器以解决椭圆（最终是抛物）系统。这些扩展将使第一个区域，全周期地震模拟，涵盖快速地震事件和地震之间慢得多的地壳运动，以及在大多数基于PDE的科学和工程应用中使用异步艾级解算器。异步解决方案技术将支持更可靠的地震灾害地图，以及更安全、更经济的抗震建筑和基础设施的设计。鉴于其广泛的适用性，这项研究所提供的前所未有的模拟能力可能会在商业和国防领域引发许多突破。4名研究生研究助理将接受跨学科培训，本科生将通过国家超级计算应用中心参与。SPIN（Students Pushing Incidence）

This collaborative project among scientists at the University of Illinois and the University of Tennessee will develop parallel software for efficient earthquake simulations on exascale supercomputers. State-of-the-art systems now resolve seismic response at frequencies up to 1 Hz, but design engineers require resolution up to 10 Hz. Synchronization barriers may limit performance on exascale systems. This project will develop scalable, barrier-free asynchronous simulation tools and space-time adaptive meshing to meet the seismic resolution requirements on exascale platforms. It will develop improved fault-gouge physics models and extend asynchronous hyperbolic solvers to address elliptic (and eventually parabolic) systems. These extensions will enable the first regional, full-cycle seismic simulations, covering fast earthquake events and much slower crustal motion between earthquakes, as well as the use of asynchronous exascale solvers in most PDE-based scientific and engineering applications. The asynchronous solution technology will support more reliable earthquake hazard maps and the design of safer, more economical earthquake-resistant buildings and infrastructure. In view of its broad applicability, the unprecedented simulation power afforded by this research could trigger numerous breakthroughs in the commercial and defense sectors. Four graduate research assistants will receive cross-disciplinary training, and undergraduate students will participate through the National Center for Supercomputing Application?s SPIN (Students Pushing INnovation) program.This collaborative project among scientists at the University of Illinois and the University of Tennessee will develop parallel software for efficient earthquake simulations on exascale supercomputers. State-of-the-art systems now resolve seismic response at frequencies up to 1 Hz, but design engineers require resolution up to 10 Hz. Synchronization barriers and load balancing across subdomains may limit performance on exascale systems. This project will replace the standard bulk synchronous parallel model and Domain Decomposition Method (DDM) with scalable, barrier-free asynchronous solvers and space-time adaptive meshing without DDM to meet the seismic resolution requirements on exascale platforms. It will develop Shear Transition Zone models for fault-gouge physics and use pseudo-time methods to extend asynchronous hyperbolic solvers to address elliptic (and eventually parabolic) systems. These extensions will enable the first regional, full-cycle seismic simulations, covering fast earthquake events and much slower crustal motion between earthquakes, as well as the use of asynchronous exascale solvers in most PDE-based scientific and engineering applications. The asynchronous solution technology will support more reliable earthquake hazard maps and the design of safer, more economical earthquake-resistant buildings and infrastructure. In view of its broad applicability, the unprecedented simulation power afforded by this research could trigger numerous breakthroughs in the commercial and defense sectors. Four graduate research assistants will receive cross-disciplinary training, and undergraduate students will participate through the National Center for Supercomputing Application?s SPIN (Students Pushing INnovation) program