Over the last decade the OSIRIS Consortium has developed a state-of-the-art, fully explicit, multi-dimensional, fully parallelized, fully relativistic, PIC code called OSIRIS [hemker:00,fonseca:02] and a set of sophisticated diagnostic and visualization packages [martins:05,fonseca:08]. The charge deposit is done locally using a charge conserving scheme, particles are updated using the Boris push, and parallelization is done using domain decomposition (this can be done in 1D, 2D, and 3D) with MPI. Dynamic load balancing taking into account both the field solve and particle push has been implemented. Higher order particle shapes as well as transmitting boundary conditions have additionally been implemented. It is used to study plasma based acceleration, fast ignition, and relativistic shocks.
OSIRIS also contains physics packages beyond the traditional PIC method. These include tunnel and impact ionization as well as a binary collision operator [nanbu:98, peano:09]. Very recently, we incorporated a framework proposed by Cohen et al. [Cohen:09] for coupling weakly collisional and collisional regions of plasmas. In this technique particles are used throughout the entire region including a relativistically correct collision operator; however in very dense regions where the plasma is collisionally dominated the electric field is solved for using an Ohm’s law. The magnetic field is updated using Faraday’s law. This allows one to take much larger timesteps that the inverse plasma frequency in the high density region. This technique combined with OSIRIS’s demonstrated parallel scalability and the use of splines will permit the watershed fast ignition simulation to be described later.
Importantly, OSIRIS also performs well on a single processor while maintaining high parallel efficiency, i.e., >80% for strong scaling (fixed problem size) and >95% for weak scaling (fixed size per processor) on greater than 300,000 processors of the Argonne JuGene Intrepid Computer. OSIRIS has also demonstrated very good scaling on a variety of other platforms including Franklin at NERSC, Atlas at LLNL, and DAWSON at UCLA. (see below). It also has been modified so that it can very effectively on Vector units. The key calculations are done in single precision without loss of accuracy. This is done by references the particle positions with respect to the cell corner. This capability will be tuned on Blue Waters.