Empowering Computational Science and Engineering via Automatic Differentiation

Purdue University, West Lafayette IN

Investigators

Abstract

The following is a revised summary of the specific tasks that will be addressed through this effort: Develop automatic sparsity detection techniques for Jacobians and Hessians. Implement the techniques in the Automatic Differentiation tool ADOL-C. Further develop the coloring software package COLPACK and enhance it with new bicoloring algorithms. Apply Automatic Differentiation to a class of problems in chromatographic separation techniques in chemical engineering. Develop interfaces between the software packages COLPACK and ADOL-C, to enable its wider applicability to optimization codes. Develop efficient algorithms for the directed acyclic graph reversal problem in adjoint computation. Contribute to education and training in combinatorial scientific computing and Automatic Differentiation by developing a graduate level course, organizing mini courses and tutorials in conferences, and by writing expository articles and book chapters. The PIs will make every attempt to meet the scope and level of effort of the revised project.