POSTER: Experiences with Automatic Differentiation and the Livermore Solver for Ordinary Differential Equations (LSODE)
by
Eileen P. Kennelly
ENSCO, Inc. 445 Pineda Court, Melbourne, FL USA.

Generating the solutions of a large system of non-linear ordinary differential equations (ODEs) requires computationally intensive numerical algorithms. Frequently ODE solvers are found within large models which in themselves require many input parameters. In order to characterize the relationship between a model's input parameters and output quantities, a Sensitivity Analysis may be performed on the entire model. This analysis requires the calculation of many intermediate derivatives. These derivatives can be calculated using the technique of Automatic Differentiation (AD). In this study, the ADIFOR tool (Automatic Differentiation in FORTRAN), is used in conjunction with the Livermore Solver for Ordinary Differential Equations (LSODE) to perform a Sensitivity Analysis on an Atmospheric Chemistry model. This presentation will show that AD techniques should not be applied blindly to ODE solvers as this may lead to surprising and erroneous results.

Date received: March 13, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # caev-03.