next up previous
Next: About this document ...

Rolf Roth
Discrete adjoint techniques for flow optimization

Technische Universitaet Darmstadt
Nonlinear Optimization
Dolivostrasse 15
64293 Darmstadt
rroth@mathematik.tu-darmstadt.de
Stefan Ulbrich

We describe a systematic way to generate adjoint code by applying an efficient sparsity exploiting forward mode of Automatic Differentiation to the original code. The result is a linear system for the adjoint that can be solved by taking advantage of the original code. The structure of a parallel solver can be used for the AD, which gives rise to a parallel distribution of the adjoint linear system. Reusage of a multigrid structure is also possible.

This procedure has been applied to the finite volume, parallel, block-structured, multigrid flow solver FASTEST, which can also run Large Eddy Simulations and is written in Fortran. Numerical results of engineering applications will be presented.





root 2010-03-02