Automatic differentiation for optimum design, applied to sonic boom reduction

We propose a methodology for adjoint-based optimum design that combines hand-coding with Automatic Differentiation in the reverse mode, therefore managing to keep the memory cost acceptable. This methodology also involves improvements to the reverse mode differentiation, based on dataflow information on the original program. We validate the method on a real-size 3D optimum design problem: reducing the sonic boom under a supersonic aircraft.