Recursive Piecewise-Linear Approximation Methods for Nonlinear Networks

The use of recursive piecewise-linear approximations in network optimization problems with convex separable objectives allows the utilization of fast solution methods for the corresponding linear network subproblems. Piecewise-linear approximations also enjoy many advantages over other types of approximations, including the ability to utilize simultaneously information from infeasible as well as feasible points (so that results of Lagrangian relaxations may be directly employed in the approximation), guaranteed decrease in the objective without the need for any line search, and easily computed and tight bounds on the optimal value. The speed of this approach will be exemplified by the presentation of computational results for large-scale nonlinear networks arising from econometric and water supply system applications.