SC²S Colloquium - July 9, 2013

From Sccswiki
Jump to navigation Jump to search
Date: July 9, 2013
Room: 02.07.023
Time: 2:30 pm, s.t.

Volker Jacht: Implementation and Benchmarking of Sparse-Matrix Solvers for Two-Phase Flow Applications

The Gesellschaft für Anlagen- und Reaktorsicherheit (GRS) mbh performs safety analysis for nuclear power plants. For the simulation of two-phase flow phenomena in the cooling system of a nuclear power plant, the program ATHLET is being developed. The cooling system is modeled as a pipe network and discretized with the finite volume method resulting in sparse linear systems of equations of dimension 1.000-100.000. Within the scope of this thesis, the open-source tool PETSc, which provides many state-of-the-art solution methods, was integrated into ATHLET. All new solvers were benchmarked, revealing that some of them deliver a great speedup for ATHLET’s performance.

Thomas Hörmann: Parallel Algorithms for Sparse Grids in X10

The efficient integration of high dimensional functions is still an unsolved problem. For standard approaches we have an exponential dependency of the number of grid points. This is called the curse of dimensions. By using sparse grids we can overcome this to some extent. In this thesis we discuss how to integrate those functions by hierarchizing sparse grids. In order to solve them we show selected parallel algorithms for sparse grids. They are realized in X10, a programming language with implicit features concerning concurrency and distribution.