 |
||
|
||
Home > TERATEC FORUM > Workshop 6
Algorithms and parallel software
HPDDM, a high-performance unified framework for domain decomposition methods
Frédéric NATAF, CNRS
HPDDM implements one- and two-level Restricted Additive Schwarz methods, the Finite Element Tearing and Interconnecting (FETI) method, and the Balancing Domain Decomposition (BDD) method and Optimized Schwarz methods. These methods can be enhanced with deflation vectors computed automatically by the framework using Generalized Eigenvalue problems on the Overlap (GenEO) or local Dirichlet-to-Neumann operators. This code has been proven to be efficient for solving various elliptic problems such as scalar diffusion equations (2 billions unknowns for a 3D heterogeneous problem in 180s using 16 000 threads), the systems of linear elasticity (300 millions unknowns for a 3D heterogeneous problem in 36s using 8 000 cores) or of Stokes, but also frequency domain problems like Helmholtz or Maxwell equations.
HPDDM (https://github.com/hpddm/) is a header-only library written in C++11 with MPI and OpenMP for parallelism. While its interface relies on plain old data objects, it requires a modern C++ compiler.
 |
Frédéric Nataf is senior scientist at CNRS in France in laboratory J.L.Lions. His field of expertise is in high performance scientific computing (domain decomposition methods/approximate factorizations), absorbing/PML boundary conditions and inverse problems. He coauthored nearly one hundred papers on these subjects. He developed the theory of optimized Schwarz methods and very recently the GENEO coarse space. This last method enables the solving of very large highly heterogeneous problems on large scale computers. |
