Performing parameter scans in parallel

On some clusters, a single compute node may contain several GPUs. In this case, it can sometimes be useful to run a separate PIC simulation on each GPU, as part of a parameter scan (where e.g. each separate PIC simulation is run with a different value of the laser intensity).

However, launching these separate simulations with separate Python scripts may not be efficient, since it is possible that all simulations will be run on the same GPU, while the other GPUs will be left idle.

Therefore, FBPIC provides the possibility to launch these different simulations as a single command with mpirun, whereby each MPI rank will perform a separate PIC simulation, and FBPIC will make sure that each simulation runs on a separate GPU. This feature is activated by setting use_all_mpi_ranks=False in the Simulation object, in order to instruct FBPIC to use only one MPI rank per simulation instead of all MPI ranks for a single simulation.

Here is an example on how to structure the input script and specify the parameter to be varied between the separate simulations: parametric_script.py