Non-subdifferentiability optimality and mean value theorems via new relative subdifferentials

TL;DR

This paper introduces new relative subdifferentials for non-subdifferentiable functions, establishing optimality conditions and mean value theorems that improve upon existing results in optimization theory.

Contribution

It develops novel relative subdifferentials and sum rules, providing sharper optimality conditions and mean value theorems for non-subdifferentiable optimization problems.

Findings

New relative subdifferentials for lower semicontinuous functions

Fuzzy sum rule and sum rule for relative subdifferentials

Sharper optimality conditions and mean value theorems

Abstract

Motivated by the optimality principles for non-subdifferentiable optimization problems, we introduce new relative subdifferentials and examine some properties for relatively lower semicontinuous functions including $\epsilon$-regular subdifferential and limiting subdifferential relative to a set. The fuzzy sum rule for the relative $\epsilon$-regular subdifferentials and the sum rule for the relative limiting subdifferentials are established. We utilize these relative subdifferentials to establish optimality conditions for non-subdifferentiable optimization problems under mild constraint qualifications. Examples are given to demonstrate that the optimality conditions obtained work better and sharper than some existing results. We also provide different versions of mean value theorems via the relative subdifferentials and employ them to characterize the equivalences between the convexity relative to a set and the monotonicity of the relative subdifferentials of a non-subdifferentiable function.