ENVIRONMENTS FOR LARGE-SCALE OPTIMIZATION

We have developed an environment for solving large-scale optimization problems in which the user is required to provide code only for the ev aluation of a partially separable function. The development of this en vironment is important because it eliminates the need to provide the g radient and sparsity pattern; in all other approaches the user is requ ired to provide the gradient. In the current environment, the ADIFOR ( Automatic Differentiation in Fortran) tool is used to compute the grad ient of the function; a limited-memory variable-metric method is used to solve the optimization problem.