Delayed Propagation of Derivatives in a Two-dimensional Aircraft Design Optimization Problem.




In computational fluid dynamics (CFD), the use of high-performance computers is often indispensable, simply to manage the high request for computation time and storage. We consider a particular two-dimensional aircraft design problem and report on the use of automatic differentiation (AD) to obtain accurate derivatives rather than approximations based on numerical differentiation. More precisely, we apply the AD system ADIFOR to the large-scale CFD solver TFS and quantify the computational savings of a technique starting the derivative computations in a delayed manner over a black box AD approach.