SI2-SSE: Collaborative Research: Extending the Practicality and Scalability of LibMesh-Based Unstructured, Adaptive Finite Element Computations

以科学和工程模拟为重点的网络基础设施的开发和部署一直是并将继续是美国科学和工程进步的关键，对于大型超级计算环境中使用的软件尤其如此。因此，美国要继续在科学计算领域保持领导地位和进步，软件基础设施的发展至关重要，以实现现代计算和软件工程策略，模拟复杂的科学和工程系统。曾经这样的软件是libMesh有限元库。libMesh被美国和世界各地的数百个研究小组使用。libMesh可以利用大规模超级计算基础设施来模拟科学和工程系统。这项工作将更新libMesh软件库，以使用最先进的算法，从而在世界上最大的超级计算机上实现强大的模拟，并进一步提高可以使用libMesh成功建模的系统的复杂性。此外，该库将得到加强，以支持用户应用程序利用现代计算机架构，包括新兴的众核架构。这将使libMesh作为科学和工程模拟的基础工具和计算算法的教学工具得以继续使用。libMesh有限元库是开源工具的一个突出例子，支持自适应网格细化、卓越求解器包的接口以及复杂有限元模型的大型并行超级计算机解决方案。libMesh支持数百名用户和许多应用程序求解各种学科的偏微分方程，包括固体力学，流体力学，磁流体力学，高超音速，核工程，燃烧和声学，仅举几例。经过十多年的成功合作开源开发，该库有望保持其作为一个突出的开源有限元软件包的地位。要做到这一点，libMesh必须支持新兴的许多核心架构，利用最先进的可扩展算法，并与工程分析中使用的复杂网格底层几何图形接口。这项工作通过扩展和增强libMesh有限元库直接解决了这些问题。这些扩展将通过与世界级求解器库的接口无缝地提供现代解算算法，使用开放的软件库促进与底层几何表示的交互，并通过尖端的软件工程原理和设计有效地利用现代计算硬件。同时，开发的接口将允许建模内核开发的灵活性，并保持libMesh社区和科学社区的低门槛。这些崇高的目标将通过设计隐藏底层算法复杂性的可用接口和对现代计算架构的广泛测试来实现，以确保为libMesh社区提供性能和可扩展性。

The development and deployment of cyberinfrastructure focused on scientific and engineering simulation has been, and continues to be, essential to the progress of science and engineering in the U.S. This is particularly true for software used in large scale supercomputing environments. Thus, for the U.S. to continue leadership and advancement in scientific computing, it is crucial that software infrastructure advance to enable modern computational and software engineering strategies for simulating complex scientific and engineering systems. Once such piece of software is the libMesh finite element library. libMesh is used by hundreds of research groups in the U.S. and around the world. Critically, libMesh can utilize large scale supercomputing infrastructure for simulating scientific and engineering systems. This work will update the libMesh software library to use state-of-the-art algorithms that will enable robust simulations on the largest supercomputers in the world and further advance the complexity of systems that can be successfully modeled using libMesh. Furthermore, the library will be enhanced to support user applications to leverage modern computer architectures, including emerging many-core architectures. This will enable the continued use of libMesh as both a fundamental tool of scientific and engineering simulation and as an educational tool for computational algorithms.The libMesh finite element library is a prominent example of an open-source tool supporting adaptive mesh refinement, interfaces to preeminent solver packages, and solutions on large parallel supercomputers of complex finite element models. libMesh supports hundreds of users and many applications in solving partial differential equations across a variety of disciplines including solid mechanics, fluids mechanics, magnetohydrodynamics, hypersonics, nuclear engineering, combustion, and acoustics, to name a few examples. Following over a decade of successful collaborative open-source development, the library is poised to maintain its place as a prominent open-source finite element package. To do so, libMesh must be made to support emerging many core architectures, leverage the most advanced scalable algorithms, and interface with geometry underlying the complex meshes used in engineering analysis. The work addresses these issues directly by extending and enhancing the libMesh finite element library. The extensions will seamlessly make available modern solution algorithms through interfaces to world class solver libraries, facilitate the interaction with underlying geometric representations using openly available software libraries, and efficiently utilize modern computing hardware through cutting-edge software engineering principles and designs. Simultaneously, the developed interfaces will allow for flexibility of development of modeling kernels and maintain the low the barrier of entry that libMesh has always had for both the libMesh community as well as the scientific community in general. Such lofty goals will be attained by designing usable interfaces that hide the complexity of the underlying algorithms and extensive testing on modern computing architectures to ensure performance and scalability is delivered to the libMesh community.