Fast direct solver for integral equations on massively parallel architectures

The Arlene code developed at the CEA (French Atomic Energy Agency) solves 3D electromagnetic problems in the frequency domain using a classical RWG Galerkin formulation of surface integral equations. Complex structures made up of metallic and dielectric materials with all configurations of junctions and discrete symmetries can be handled. This massively parallel code efficiently exploits the thousands of cores of the TERA100 cluster of CEA and also runs on a hybrid partition with GPU accelerators. The direct solve techniques compared to the iterative ones have always been preferred due to a good accuracy requirement, a poor matrix conditioning and high efficiency for multiple incidence problems. However direct methods become prohibitive for electrically large objects.