T. Peterka, J. Kwan, A. Pope, H. Finkel, K. Heitmann, S. Habib, "Meshing the Universe: Identifying Voids in Cosmological Simulations Through in Situ Parallel Voronoi Tessellation," Preprint ANL/MCS-P2087-0512, May 2012. [pdf]
Mesh tessellations are effective constructs for the visualization and analysis of point data, because they transform sparse discrete samples into dense and continuous functions. We present a prototype method for computing a Voronoi tessellation in parallel from large particle datasets; the same method, in principle, is applicable to the Delaunay. Computing large tessellations is computationally intensive and must be constructed in parallel on a distributed-memory supercomputer in order to satisfy time and memory constraints. We perform the mesh computation and analysis in situ with the simulation in order to minimize storage pressure and generate early results, specifically identifying voids in cosmological data. We demonstrate performance and scalability in a single time step, and we also compute time-varying tessellations to better understand the temporal dynamics of voids.