SC²S Colloquium - April 24, 2017
|Date:||April 24, 2017|
|Time:||2:30 pm, s.t.|
Severin Reiz: Black Box Hierarchical Approximations for SPD Matrices
N-body methods are vastly used in many disciplines: Simulations of gravity and coulomb-potentials, waves and scattering or fluids and transport; in data analysis the same problem occurs in machine learning, geostatistics and image analysis. A common approach is a far-field approximation by Barnes-Hut and the Fast Multipole Method. In most codes hierarchical domain decomposition is based on geometric distances in the original space. In this project we derived a geometry-oblivious partitioning scheme for SPD matrices from various problem domains. This leads to hierarchical low rank plus sparse matrices - an algebraic generalization from previously mentioned N-body methods. A parallel treecode is developed for a black box fast approximate matrix multiplication.
Jeeta Ann Chacko: Shallow Water Equation Simulation with Sparse Grid Combination Technique
Scientific applications are affected by the curse of dimensionality. Sparse grids can be used to reduce grid points drastically but they are difficult to parallelize. The sparse grid combination technique can be used to combine lower resolution anisotropic Cartesian grids to generate a sparse grid instead of directly using sparse grids. The shallow water equation is a hyperbolic PDE that is used to describe the changes in height and horizontal velocities of a fluid. The shallow water equation simulation is implemented using the sparse grid combination technique as a parallelization scheme in this project. The solution is compared with an existing implementation of shallow water equation simulation to evaluate the accuracy.