On the relations between different duals assigned to composed optimization problems (Q2466796): Difference between revisions

In this paper the authors construct some dual problems to an optimization problem, where both objective and constraint functions are composed functions: \[ \inf_{x\in {\mathcal A}}f(F(x)), \qquad k:{\mathbb R}^n\to \overline{\mathbb R},\tag{P} \] where \[ {\mathcal A}=\{x\in X:\; g(G(x))\leq_{{\mathbb R}^k_+} 0 \} \] and \(X\) is a nonempty subset of \({\mathbb R}^n.\) In order to do this they apply the Fenchel-Rockafellar duality concept based on conjugacy and perturbation. In particular, they embed the original optimization problem \[ \inf_{x\in {\mathbb R}^n} k(x) \] in a family of perturbed problems \[ \inf_{x\in{\mathbb R}^n}\Phi(x,p),\qquad \Phi:{\mathbb R}^n\times {\mathbb R}^s\to \overline{\mathbb R}, \] such that \(\Phi(x,0)=k(x),\) for every \(x\in {\mathbb R}^n.\) The problem \[ \sup_{p^*\in {{\mathbb R}^s}}\{-\Phi^*(0,p^*)\}, \] where \(\Phi^*(x^*,p^*)\) is the conjugate function of the perturbation \(\Phi,\) defines a dual problem of (P). By investigating special perturbation functions, in Section 3 they study the Lagrange dual problem, the Fenchel dual problem, and the Fenchel-Lagrange dual problem, looking for the relations between the optimal objective values of the three dual problems, and they show that the optimal objective values of these dual problems are less than or equal to the optimal objective value of the primal problem. In Section 4, they investigate the relations between the optimal objective values of the three dual problems; moreover, they prove a strong duality result. In the last section they consider some special cases of the original problem, and show that the duality concepts introduced in the paper generalize some results obtained in the past.