Integrating Automatic Differentiation with Object-oriented Toolkits for High-performance Scientific Computing
by
Paul Hovland
Argonne National Laboratory
Coauthors: Jason Abate (University of Texas), Lisa Grignon (University of North Carolina), Lois McInnes (Argonne National Laboratory), Boyana Norris (Argonne National Laboratory)

Many advanced computational science applications make use of object-oriented numerical toolkits designed for high-performance scientific computing. Integrating automatic differentiation with such toolkits is a natural combination, because AD provides the derivatives required for such tasks as optimization or the solution of nonlinear systems of equations, while the well-defined APIs of the toolkit make the AD process more completely automatic. We illustrate this synergy using two numerical toolkits, the Portable, Extensible Toolkit for Scientific Computing (PETSc) and the Toolkit for Advanced Optimization. For the applications considered, the analytic derivatives provided by AD provide greater robustness and faster convergence (in terms of iterations) than finite difference approximations.

Extended abstract for AD and Toolkits (postscript)

Date received: February 11, 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 # cads-75.