XPS:FULL:DSD: A novel framework for developing highly scalable and energy efficient guaranteed quality mesh generation for 3D and 4D finite element analysis

Old Dominion University Research Foundation, Norfolk VA

Investigators

Abstract

Computational science is one of the "three pillars," complementing traditional theoretical and physical experimental studies in science and engineering. Parallel finite element mesh generation is a critical building block for this pillar and is becoming even more relevant for a growing number of engineering and life science applications. This project will set up a trajectory to deliver the first exascale-era unstructured finite element (FE) guaranteed quality Delaunay mesh generation. Why is FE mesh generation an important computational science tool? Many partial differential equations (PDEs) that are used to model complex multi-scale phenomena such as blood flow in the human body can only be solved by numerical approximation techniques. These techniques require the approximation of the domain by tessellating it into simpler geometric shapes such as triangles and tetrahedra in two and three dimensions, respectively. This project's focus is in 3D and 4D tessellation methods for life science applications such as blood flow simulations for Cerebro-Vascular Disease (CVD, or stroke), one of the leading natural causes of death in the US. In order to deliver exascale-era mesh generation, this project is set to achieve billion-way concurrency using substantially less electric power than today's state-of-the-art methods. This goal will be achieved by focusing on the following three objectives: (1) Integration of multiple parallel Delaunay mesh generation methods into a telescopic framework. (2) Development of application-specific models that describe the inherent concurrency and data access patterns of this framework. (3) Development of domain-specific energy-efficient and component-level (core and memory) power scaling for massively parallel mesh generation methods. This project will have broader impact in many other life science applications such as those related to the President's BRAIN Initiative. For example, the mesh generation techniques from this project can be customized to perform (by 2020) computer simulations to understand the circuitry of the human brain - an important milestone to help understand diseases like Parkinson's and Alzheimer's, which are expected to increase with the aging of the U.S. population. Finally, this project will contribute via the PI's MERIT outreach program for STEM education in K-12 and college students. The PI's goal is to "mentor, excite, and retain" students and help them to identify STEM areas of study that will transform them into responsible college graduates. The PI's MERIT Freshman Seminars (as opposed to traditional Freshman Seminars) re-connect students in the context of Research Experiences for Undergraduates (REU) activities (based on students interests) with highly visible national priorities such as the President's BRAIN Initiative. The MERIT program covers a variety of topics to reach as many students as possible with diverse interests and background. The goal is to prepare computational scientists to be entrepreneurs capable of understanding the ethical, economic, and research challenges in health care we face today.