Challenges and Opportunities in Using Automatic Differentiation with Object-Oriented Toolkits for Scientific Computing
The increased use of object-oriented toolkits in large-scale scientific simulation presents new opportunities and challenges for the use of automatic (or algorithmic) differentiation (AD) techniques, especially in the context of optimization. Because object-oriented toolkits use well-defined interfaces and data structures, there is potential for simplifying the AD process. Furthermore, derivative computation can be improved by exploiting high-level information about numerical and computational abstractions. However, challenges to the successful use of AD with these toolkits also exist. Among the greatest challenges is balancing the desire to limit the scope of the AD process with the desire to minimize the work required of a user. They discuss their experiences in integrating AD with the PETSc, PVODE, and TAO toolkits and the plans for future research and development in this area.
- Research Organization:
- Lawrence Livermore National Laboratory (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- US Department of Energy (US)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 15005668
- Report Number(s):
- UCRL-JC-143410; TRN: US200324%%8
- Resource Relation:
- Journal Volume: 30; Conference: 1st Sandia Workshop on Large-Scale PDE-Constrained Optimization, Santa Fe, NM (US), 04/04/2001--04/06/2001; Other Information: PBD: 17 Apr 2001
- Country of Publication:
- United States
- Language:
- English
Similar Records
Developing a derivative-enhanced object-oriented toolkit for scientific computations.
On combining computational differentiation and toolkits for parallel scientific computing.