Imbedded Dual-Number Automatic Differentiation for Computational Fluid Dynamics Sensitivity Analysis

Abstract

Dual number automatic differentiation was applied to two computational fluid dynamics codes, one written specifically for this purpose and one “legacy” fortran code. Results for the simple case of a fully developed laminar flow in a channel validated the approach in computing derivatives with respect to both a fluid property and a geometric dimension. DNAD was also implemented into the JET fortran program which is available with a popular turbulence modeling textbook. Mean centerline velocity derivatives for a self-similar round jet with respect to all applicable turbulence model closure coefficients for k-ω and k-ε models were obtained.