Dice Question Streamline Icon: https://streamlinehq.com

Strictness of inclusion between convex and general polynomial bilevel value-function classes

Determine whether, for each mode in {optimistic, pessimistic} and each box type in {bounded, unbounded}, the inclusion C^{mode}_{boxtype} ⊆ P^{mode}_{boxtype} is strict or not, where P^{mode}_{boxtype} denotes the class of value functions realizable by polynomial bilevel problems with the specified mode and box constraints and no convexity requirement on the lower-level objective, and C^{mode}_{boxtype} denotes the corresponding class when the lower-level objective is convex in the lower-level variable.

Information Square Streamline Icon: https://streamlinehq.com

Background

The paper defines P{mode}_{boxtype} as the set of value functions arising from polynomial bilevel problems with specified optimistic or pessimistic mode and a lower-level feasible set that is a box (bounded or unbounded), without imposing convexity on the lower-level objective. The class C{mode}_{boxtype} is defined similarly but requires the lower-level objective to be convex in the lower-level variable.

By construction, imposing convexity on the lower-level objective yields C{mode}_{boxtype} ⊆ P{mode}_{boxtype}. The authors show large expressivity for C{mode}_{boxtype} (e.g., inclusion of piecewise polynomial functions) but explicitly note that they do not know whether the inclusion is strict.

References

From \Cref{def:class-general-value-functions} and \Cref{def:class-lower-level-convex}, we clearly have that: \begin{equation*} \cC{mode}_{boxtype} \subseteq \cP{mode}_{boxtype}, \end{equation*} for $mode \in {optim,pessim}, boxtype \in {unbound,bounded}$. However, we do not know if this inclusion is strict.

Geometric and computational hardness of bilevel programming (2407.12372 - Bolte et al., 17 Jul 2024) in Section 2.3 (Polynomial bilevel problems with convex lower-level)