Duality for convex infinite optimization on linear spaces (Q2080839)

Let \(X\) be a real linear space, \(T\) an infinite index set, and \(f,f_{t}:X\rightarrow \overline{\mathbb{R}}\) be proper convex functions. The problem \((P)\) under consideration is \(\inf \{f(x):f_{t}(x)\leq 0, \,t\in T\}\), and its conic-Lagrangian dual problem \((D)\) is \(\sup_{\lambda \in \mathbb{R}_{+}^{(T)}}\inf_{x\in M}\{f(x)+\sum_{t\in T}\lambda _{t}f_{t}(x)\}\); here \(\mathbb{R}_{+}^{(T)}\) denotes the set of functions \(\lambda =(\lambda _{t})_{t\in T}:T\rightarrow \mathbb{R}_{+}\) whith finite support \(\operatorname{supp}\lambda :=\{t\in T:\lambda _{t}>0\}\), and \(M:=\bigcap_{t\in T}\operatorname{dom}f_{t}\). Let \(v:\mathbb{R}^{T}\rightarrow \overline{\mathbb{R}}\) be the ordinary value function associated with \((P)\), defined by \(v(y):=\inf \{f(x):f_{t}(x)\leq y_{t},\,t\in T\}\), and \(\overline{v}\) be the lower semicontinuous hull of \(v\) with respect to the product topology on \(\mathbb{R }^{T}\). The main result states that, if \(\overline{v}(0)<+\infty\) or \(\sup (D)>-\infty\), then the limiting formula \(\sup (D)=\sup_{\epsilon >0,\,H\in \mathcal{F(T)}}\inf \{f(x):x\in M, \,f_{t}(x)\leq \varepsilon, \,t\in H\}\) holds, with \(\mathcal{F(T)}\) denoting the set of non-empty finite subsets of \(T\). This formula corrects an earlier limiting formula due to \textit{D. F. Karney} [Math. Program. 27, 75-82 (1983; Zbl 0527.49027)], which the authors show to fail, by means of an example, even in linear semi-infinite programming. The authors also obtain a limiting formula for the ordinary Lagrangian dual corresponding to the problem obtained from \((P)\) by replacing its constraints with the equivalent single constraint \(h(x)\leq 0\), with \(h:=\sup_{t\in T}f_{t}\). Combining this limiting formula with a classical duality theorem for ordinary convex problems, they prove that the equality \(v(0)=\lim_{\varepsilon \downarrow 0}\inf \{f(x):f_{t}(x)\leq \varepsilon,\, t\in T\}\) holds provided that the strong Slater condition (that is, the existence of \(a\in \operatorname{dom}f\) such that \(\sup_{t\in T}f_{t}(a)<0\)) is satisfied.