Sensitivity computations by algorithmic differentiation of a high-order cfd code based on spectral differences
Résumé
We compute flow sensitivities by differentiating a high-order computational fluid dynamics code. Our fully discrete approach relies on automatic differentiation (AD) of the original source code. We obtain two transformed codes by using the AD tool Tapenade (INRIA), one for each differentiation mode: tangent and adjoint. Both differentiated codes are tested against each other by computing sensitivities in an unsteady test case. The results from both codes agree to within machine accuracy, and compare well with those approximated by finite differences. We compare execution times and discuss the encountered technical difficulties due to 1) the code parallelism and 2) the memory overhead caused by unsteady problems.
Origine : Fichiers éditeurs autorisés sur une archive ouverte
Loading...