On the dual coordinate ascent approach for nonlinear networks

Dual coordinate ascent algorithms are popular methods for solving convex separable cost network flow problems due to their inherent simplicity and implementation performance. The principal disadvantage of the coordinate ascent approach is the inability to determine a good feasible solution without solving to optimality. This paper proposes a finitely convergent method that can be used to truncate the coordinate ascent algorithm with an •-optimal solution. Computational experience indicates that good feasible solutions can be found quickly with this approach.