This will be the third in a series of workshops devoted to Automatic Differentiation. The first meeting took place in Breckenridge in 1991, and the second meeting in Santa Fe in 1996.
The previous conferences have created a momentum. They motivated further research and development. They gathered researchers from different fields, from mathematics to scientific programming and to software engineering. They built a feeling of community between these researchers. We need a new gathering to continue this impulse, renew the motivation, and present the recent ideas and results to scientists and practitioners of scientific computing.
The domains covered by these conferences on Automatic Differentiation (AD) reflect these different horizons. They reflect the mathematical aspects, such as the differentiation of iterative processes, and the analysis of non-smooth code. They cover the scientific programming aspects, such as the use of adjoints in optimization and the propagation of rounding errors. They also cover implementation problems, such as the complexity of the different differentiation modes, memory management problems, and static compile-time analysis to improve differentiated programs.
Since the last meeting in 1996, research has progressed considerably. AD is being applied to larger and larger programs, for example for optimum shape design. In many such applications, modeling was restricted to simulations at various parameters settings. With the help of AD techniques, this trial and error approach can be replaced by a more efficient optimization with respect to modes, design, and control parameters. AD tools have also grown in variety, power and complexity, so that presentations and general discussions will be highly profitable.
The objective of this third conference is to continue the work started in the previous meetings, and present the audience with recent developments. A special emphasis will be put on the relationships and synergies between AD tools and other software tools such as compilers and parallelizers. Also, the maturation of AD tools motivates a special session on the use of AD in an industrial context. Topics of this conference include, but are not limited to: