Extreme scale supercomputing will soon offer a million times the computing power of a desktop – and as drastic upgrade as that from a slide rule to a desktop computer in the 1990s. I believe this will revolutionize how aerospace engineers work. It will enable them to rapidly and confidently tune and optimize their designs. But this revolution can only happen through innovating our computational algorithms.

In Computational Fluid Dynamics, high-fidelity simulations such as Detached and Large Eddy Simulation can often reliably predict the performance of aerospace vehicles and engines. But with today’s algorithms, these simulations take days if not weeks. With today’s optimization algorithms, it may take months if not years for us to reach a good design. Can we shorten each high fidelity CFD simulation to minutes, by innovating how we solve PDEs, better utilizing the skyrocketing concurrency in supercomputers? Better, can we shorten an entire high fidelity optimization to minutes by innovating how we do optimization, again utilizing more concurrency than we currently can? Even better, can we shorten a high fidelity optimization with hundreds of design parameters to minutes, by computing high fidelity design gradients, even when the simulations are turbulent and chaotic, and gradients in the traditional sense would diverge?

I believe that the answers are yes, yes and yes. In this talk, I will show you why I believe so, and discuss how we can advance aerospace design into the age of extreme-scale supercomputing.