Directional subdifferential of the value function

TL;DR

This paper develops upper estimates for the directional subdifferentials of the value function in parametric optimization, extending sensitivity analysis and generalizing Danskin's theorem to directional perturbations.

Contribution

It introduces new upper bounds for directional subdifferentials and characterizes directional Lipschitzness, broadening the understanding of value function sensitivity under directional perturbations.

Findings

Derived upper estimates for directional limiting and singular subdifferentials.

Provided a characterization of directional Lipschitzness via subdifferentials.

Extended existing sensitivity results, including Danskin's theorem, to directional cases.

Abstract

The directional subdifferential of the value function gives an estimate on how much the optimal value changes under a perturbation in a certain direction. In this paper we derive upper estimates for the directional limiting and singular subdifferential of the value function for a very general parametric optimization problem. We obtain a characterization for the directional Lipschitzness of a locally lower semicontinuous function in terms of the directional subdifferentials. Based on this characterization and the derived upper estimate for the directional singular subdifferential, we are able to obtain a sufficient condition for the directional Lipschitzness of the value function. Finally, we specify these results for various cases when all functions involved are smooth, when the perturbation is additive, when the constraint is independent of the parameter, or when the constraints are equalities and inequalities. Our results extend the corresponding results on the sensitivity of the value function to allow directional perturbations. Even in the case of full perturbations, our results recover or even extend some existing results, including the Danskin's theorem.