Environments for Large-Scale Optimization

Ali Bouaricha and Jorge J. Moré

We are developing an environment (ELSO) for large scale optimization problems that only requires the user to provide code to evaluate a partially separable function. This novel approach eliminates the need to provide the gradient and sparsity pattern; in all other approaches the user is required to provide the gradient and (for a Newton method) the sparsity pattern. This will provide a unique capability that is not available elsewhere.

The research on environment for large scale optimization is a joint project of Argonne National Laboratory and the NSF Center for Research on Parallel Computation

The schematic below shows that in a typical situation the user provides the function, ADIFOR computes the gradient and sparsity pattern, and our trust region Newton method solves the problem.

Recent Papers and Technical Reports

• A. Bouaricha and Jorge J. Moré, Impact of partial separability on large-scale optimization, Preprint MCS-P487-0195, November 1995. Postscript Version.

• Christian H. Bischof, Ali Bouaricha, Peyvand Khademi, and Jorge J. Moré, Computing gradients in large-scale optimization using automatic differentiation, Preprint MCS-P488-0195, January 1995. Postscript Version.

• Brett M. Averick, and Jorge J. Moré, Evaluation of large-scale optimization problems on vector and parallel architectures. SIAM Journal on Optimization, 4 (1994), 708-721.

• Brett M. Averick, Christian H. Bischof, Alan Carle, Jorge J. Moré, and Andreas Griewank, Computing large sparse Jacobian matrices using automatic differentiation, SIAM Journal on Scientific and Statistical Computing, 15 (1994), 285-294.