New Directions in Modeling Tools for Optimization
20 June 2001
The designs of current modeling languages for large-scale optimization have their roots in linear and local nonlinear programming. Objectives and constraints are described by numerical expressions in the decision variables. Expressions are converted to a linear part represented by a sparse matrix of coefficients plus (if necessary) a nonlinear part represented by an operation list or graph. The data structure for the nonlinear part is designed to facilitate function evaluations at points generated by "solver" software that implements optimization methods.