Conjugate derivative of a multivalued mapping and the differentiability of the maximum under connected constraints (Q1074504): Difference between revisions

The problem of the directional differentiability of the maximum under connected constraints, i.e., of a function of the form \(\phi (x)=\max_{y\in \alpha (x)}f(x,y)\), where \(\alpha\) is a multivalued mapping, arises in the investigation of various extremal problems. It was discussed by many authors under various assumptions on function f and mapping \(\alpha\). As a rule, the answer to the indicated problem is given in terms of some or other sets of admissible or tangent directions, not always simple to determine. In this work the differentiability of \(\phi\) is studied under the assumption that the mapping \(\alpha\) is convex and admits a certain special approximation by its conjugate derivative; the latter is essentially the operator which puts into correspondence each vector with the directional derivative of the support function of the mapping \(\alpha\), evaluated on this vector. In this paper we show that the class of maps which have a conjugate derivative is sufficiently rich, and rules for computing this derivative are indicated. The approach proposed here permits to reduce the calculation of the derivative of the maximum function under connected constraints to the case where the constraints are not connected. This approach does not require the determination of sets of admissible and tangent directions and is particularly convenient in the case where the mapping is given by its support function. It permits to investigate classes of convex and concave functions f that were not considered before.