- This event has passed.
Parallel Octree Algorithms for Adaptive Meshing, Search, and Visualization - Carsten Burstedde
February 6 @ 10:30 am - 11:30 am MST
Professor for Scientific Computing
Universität Bonn, Germany
Thursday, February 6th, at 10:30 AM in CCP Room 259
Abstract: The distributed, leaf-only encoding and storage of octrees has been established as a fast and scalable approach to adaptive mesh refinement, coarsening, and partitioning on parallel computing platforms. While the fundamentals of the above algorithms are explained relatively quickly, further functionality necessary for PDE solves or spatial searches are technically more involved. We can classify such algorithms as those based on creating a so-called ghost layer of remote-process neighbor leaves and those that operate purely on top-down tree traversals. We will discuss a few example algorithms of both classes and illustrate their purpose for encoding finite element discretizations, searching for geometric objects in a mesh, and parallel visualization.
Biography: Carsten’s research interests revolve around the construction of scalable parallel algorithms for adaptive mesh refinement, partitioning, computational geometry, and visualization. The basic technique relies on space-filling curves and leaf-only octree encoding, extended to a forest of octrees, which allows for a mesh representation that is both flexible and economic. Carsten enjoys collaborating on interdisciplinary science and simulation challenges.