Applications of the ADIFOR 3.0 Automatic Adjoint Generation Tool at the NASA Langley Research Center - Pending NASA Approval
by
Lawrence L. Green
MDOB / NASA Langley Research Center, Hampton, Virginia, 23681, USA
Coauthors: Mike Park (JIAFS / George Washington University), Dave Raney (DCB, NASA LaRC)

Accurate and efficient derivative calculations are very important in trade studies, sensitivity analyses, and gradient-based single and multidisciplinary optimizations. Over the past few years, staff at the NASA Langley Research Center (NASA LaRC) have been using the automatic differentiation (AD) tool known as ADIFOR (Automatic Differentiation in FORTRAN) and its subsequent revisions (ADIFOR 2.0 and ADIFOR 3.0) to generate sensitivity analyses for applications in a variety of disciplines. ADIFOR transforms an existing FORTRAN computer code into new code which performs the original functional analysis and user-specified sensitivity derivative analyses by systematic implementation of the chain rule of differentiation, in both forward (direct) and reverse (adjoint) modes of differentiation. The proposed talk will present a brief summary of ADIFOR applications at NASA LaRC, with emphasis on the recent ADIFOR 3.0 use for (1) aerodynamic shape optimization by means of nonlinear computational fluid dynamics codes and (2) control placement effectiveness. Sample adjoint results are shown in the extended abstract given below.

http://homepages.feis.herts.ac.uk/~comqun/AD2000/Ext_Abstracts/lgreen.ps

Date received: December 29, 1999