Abstract

A parallel framework for performing aerodynamic design optimizations on unstructured meshes is described. The approach utilizes a discrete adjoint formulation which has previously been implemented in a sequential environment and is based on the three-dimensional Reynolds-averaged Navier-Stokes equations coupled with a one-equation turbulence model. Here, only the inviscid terms are treated in order to develop a basic foundation for a multiprocessor design methodology. A parallel version of the adjoint solver is developed using a library of MPI-based linear and nonlinear solvers known as PETSc, while a shared-memory approach is taken for the mesh movement and gradient evaluation codes. Parallel efficiencies are demonstrated and the linearization of the residual is shown to remain valid. 2.