Exact characterization for subdifferentials of a special optimal value function (Q1749781)

Let \(X\) be a real Banach space, \(U\subset X\) be a bounded closed convex set and \(S\subset X\) be a closed set. Based on the Minkowski-function \[ \rho_U(u):= \text{inf}\{t>0\mid u\in tU\}, \] the authors regard the special optimal value function \(\Gamma_{S|U}: X\to\mathbb{R}\) according to \[ \Gamma_{S|U}(x):= \text{inf}\{\rho_U(s-x)\mid s\in S\}= \text{inf}\{t>0\mid S\cap(x+ tU)\neq\varnothing\}. \] Clearly, if \(U\) is the closed unit ball, then \(\Gamma_{S|U|(x)}(x)\) reduces to the distance function \(d_S(x)\). If \(U=\{0\}\), then \(\Gamma_{S|U}(x)\) reduces to the indicator function \(I_S(x)\). If \(S\) is convex, the function \(\Gamma_{X|U}\) is convex. However, if \(S\) is not convex, \(\Gamma_{S|U}\) is not necessary convex. Formulas for the representation of the Fréchet and the proximal subdifferentials of \(\Gamma_{S|U}\) at points \(x\in S\) in terms of the Fréchet and the proximal normal cones of \(S\) at \(x\) and a sublevel set of the support function of \(U\) are known. In the present paper, without additional assumptions, similar representations are provided for the case \(x\not\in S\). The subdifferentials are represented as the intersection of the associated normal cone of an enlarged set of \(S\) and a level set of the support function of \(U\). The assertions improve known results.