CRPC-TR97768-S						October 1997

Title: Automatic Differentiation and Navier-Stokes Computations

Author: Paul Hovland, Bijan Mohammadi, and Christian Bischof

Submitted September 1998; Also available as Argonne Techreport 
ANL/MCS-P687-0997

Abstract:
We describe the use of automatic differentiation (AD) in, and its 
application to, a compressible Navier-Stokes model. Within the solver, AD 
is used to accelerate convergence by more than an order of magnitude. 
Outside the solver, AD is used to compute the derivatives needed for 
optimization. We emphasize the potential for performance gains if the 
programmer does not treat AD as a black box, but instead utilizes 
high-level knowledge about the nature of the application.

------------------------------------------------------------------------------
Paul Hovland		Christian Bischof
hovland@mcs.anl.gov	bischof@mcs.anl.gov
Mathematics and Computer Science Division
Argonne National Laboratory

Bijan Mohammadi
bijan.mohammadi@inria.fr
Mathematics Department
University of Montpellier II and INRIA