The problem is identical to Rabault et al but the optimization step uses GP. Solver instructure is taken from paper's code base. Search over the control space is done in parallel using only MPI as threading does not play well with FEniCS.

• FEniCS stack
• deap
• gmsh

• Do a dry-run of genetic_control.py in serial first to get everything compiled (there are some JIT issues which go away this way)

• penalize lift
• take into cost