Subdifferential test for optimality (Q2249811): Difference between revisions

Let \(X\) be a real Banach space with topological dual \(X^*\) and let \(\partial f: X\rightrightarrows X^*\) be a general subdifferential mapping for a lower semicontinuous functions \(f\) with the following properties: 1. \(\partial f\) coincides with the usual convex subdifferential whenever \(f\) is convex, 2. \(\partial(f- x^*)(x)=\partial f(x)- x^*\) for every \(x\in X\) and \(x^*\in X^*\), 3. if \(\varphi\) is convex and \(f+\varphi\) admits a local minimum at \(\overline x\), then \(0\in\partial f(\overline x)+\partial\varphi(\overline x)\). The authors provide a first-order necessary and sufficient condition for optimality of lower semicontinuous functions in terms of this subdifferential. The proof is based on the so-called \(\varepsilon\)-enlargement of \(\partial f\) given by \[ \widetilde\partial_\varepsilon f(\overline x):= \{x^*_\varepsilon\in\partial f(x_\varepsilon):\| x_\varepsilon-\overline x\|\leq\varepsilon, |f(x_\varepsilon)- f(\overline x)|\leq\varepsilon, \langle x^*_\varepsilon, x_\varepsilon-\overline x\rangle\leq\varepsilon\} \] and its relation \(f'(\overline x;d)\leq \text{inf}_{\varepsilon>0}\sup\langle\widetilde\partial_\varepsilon f(\overline x), d\rangle\) to the lower Dini directional derivative \(f'(\overline x;\cdot)\) of \(f\). As a consequence of this result it can be pointed out that the subdifferential mapping \(\partial f\) is in any case monotone absorbing, and hence maximal monotone if the function \(f\) is convex.