A Reduced Gradient Algorithm for Nonlinear Network Problems

Reduced gradient algorithms are particularly efficient for problems with linear constraints. Here we consider the specializaton of one such method to problems with a nonlinear objective and linear network constraints. The focus is on software design and computational testing. We propose extensions of data structures used in linear network codes These extensions deal with the possibility of varying several nonbasic variables simultaneously Computational results are presented for problems having up to 666 nodes. Results show that our procedure requires far fewer iterations than a networkoriented speciallzatmn of the convex simplex algorithm Which code requires less computation time is problem dependent.