A Fenchel-Lagrange duality approach for a bilevel programming problem with extremal-value function (Q635806)

The bilevel problem corresponds to a two-player game in which a leader announces first a strategy \(x\) to minimize his objective function \(F\), and where a follower reacts optimally by selecting a strategy \(y(x)\) to minimize his objective function \(f\). Furthermore, the reaction of the follower to the strategies of the leader is supposed to have no impact on the constraints of the leader. Adopting the Fenchel-Lagrange duality introduced by Bot and Wanka for ordinary convex programming problems, the authors give the so-called Fenchel-Lagrange dual of the bilevel problem. First, they show that a strong duality holds between the two problems. Then, they provide necessary and sufficient optimality conditions for the bilevel problem and its dual. These optimality conditions are given in terms of conjugate functions and subdifferentials, respectively. Finally, the authors show that the resolution of the dual problem is equivalent to the resolution of a one-level convex minimization problem whose objective function is expressed in terms of the conjugates of the functions involved. This approach may give a possibility to solve indirectly the bilevel problem via a one-level convex minimization problem.