CRPC-TR98738						February 1998

Title: Construction and Evaluation of an Incremental Iterative Version of
a Parallel Multigrid CFD Code via Automatic Differentiation for Shape
Optimization

Authors: Alan Carle and Mike Fagan

Submitted February 1998

Abstract:
Automatic differentiation (AD) is a technique for augmenting computer
codes to compute derivatives of a subset of their outputs with respect to
a subset of their inputs. AD has been shown to provide accurate, but
inefficient, sensitivity-enhanced CFD codes for use in aerodynamic shape
optimization. To address the inefficiency problem, several special purpose
techniques have been suggested. One such technique is the incremental
iterative (II) method. This report describes and evaluates the use of the
II method in AD-augmented code for OVERFLOW, a parallel, multigrid CFD
code.

------------------------------------------------------------------------------
Louis Thomas		Sean Suchter			Adam Rifkin
lthomas@cs.caltech.edu	triangle@cs.caltech.edu		adam@cs.caltech.edu
Department of Computer Science
California Institute of Technology